diff options
| author | Laurențiu Nicola <lnicola@dend.ro> | 2024-05-19 11:20:26 +0300 |
|---|---|---|
| committer | Laurențiu Nicola <lnicola@dend.ro> | 2024-05-19 11:20:26 +0300 |
| commit | 6ead205843eabf8121107b2f95c55be857f195e6 (patch) | |
| tree | ec4e9aabbf7b678dd04192d806e1ee7cc8987263 /compiler/rustc_parse/src/parser | |
| parent | 2018426f569829bf5bb8866e16957eaef508f770 (diff) | |
| parent | 6579ed89f0fcc26da71afdd11d30d63f6f812a0a (diff) | |
| download | rust-6ead205843eabf8121107b2f95c55be857f195e6.tar.gz rust-6ead205843eabf8121107b2f95c55be857f195e6.zip | |
Merge from rust-lang/rust
Diffstat (limited to 'compiler/rustc_parse/src/parser')
| -rw-r--r-- | compiler/rustc_parse/src/parser/attr.rs | 445 | ||||
| -rw-r--r-- | compiler/rustc_parse/src/parser/attr_wrapper.rs | 465 | ||||
| -rw-r--r-- | compiler/rustc_parse/src/parser/diagnostics.rs | 3056 | ||||
| -rw-r--r-- | compiler/rustc_parse/src/parser/expr.rs | 4047 | ||||
| -rw-r--r-- | compiler/rustc_parse/src/parser/generics.rs | 518 | ||||
| -rw-r--r-- | compiler/rustc_parse/src/parser/item.rs | 2881 | ||||
| -rw-r--r-- | compiler/rustc_parse/src/parser/mod.rs | 1643 | ||||
| -rw-r--r-- | compiler/rustc_parse/src/parser/mut_visit/tests.rs | 71 | ||||
| -rw-r--r-- | compiler/rustc_parse/src/parser/nonterminal.rs | 214 | ||||
| -rw-r--r-- | compiler/rustc_parse/src/parser/pat.rs | 1426 | ||||
| -rw-r--r-- | compiler/rustc_parse/src/parser/path.rs | 967 | ||||
| -rw-r--r-- | compiler/rustc_parse/src/parser/stmt.rs | 853 | ||||
| -rw-r--r-- | compiler/rustc_parse/src/parser/tests.rs | 1422 | ||||
| -rw-r--r-- | compiler/rustc_parse/src/parser/tokenstream/tests.rs | 108 | ||||
| -rw-r--r-- | compiler/rustc_parse/src/parser/ty.rs | 1214 |
15 files changed, 19330 insertions, 0 deletions
diff --git a/compiler/rustc_parse/src/parser/attr.rs b/compiler/rustc_parse/src/parser/attr.rs new file mode 100644 index 00000000000..a57eb70c705 --- /dev/null +++ b/compiler/rustc_parse/src/parser/attr.rs @@ -0,0 +1,445 @@ +use crate::errors; +use crate::fluent_generated as fluent; +use crate::maybe_whole; + +use super::{AttrWrapper, Capturing, FnParseMode, ForceCollect, Parser, PathStyle}; +use rustc_ast as ast; +use rustc_ast::attr; +use rustc_ast::token::{self, Delimiter}; +use rustc_errors::{codes::*, Diag, PResult}; +use rustc_span::{sym, BytePos, Span}; +use thin_vec::ThinVec; +use tracing::debug; + +// Public for rustfmt usage +#[derive(Debug)] +pub enum InnerAttrPolicy { + Permitted, + Forbidden(Option<InnerAttrForbiddenReason>), +} + +#[derive(Clone, Copy, Debug)] +pub enum InnerAttrForbiddenReason { + InCodeBlock, + AfterOuterDocComment { prev_doc_comment_span: Span }, + AfterOuterAttribute { prev_outer_attr_sp: Span }, +} + +enum OuterAttributeType { + DocComment, + DocBlockComment, + Attribute, +} + +impl<'a> Parser<'a> { + /// Parses attributes that appear before an item. + pub(super) fn parse_outer_attributes(&mut self) -> PResult<'a, AttrWrapper> { + let mut outer_attrs = ast::AttrVec::new(); + let mut just_parsed_doc_comment = false; + let start_pos = self.num_bump_calls; + loop { + let attr = if self.check(&token::Pound) { + let prev_outer_attr_sp = outer_attrs.last().map(|attr| attr.span); + + let inner_error_reason = if just_parsed_doc_comment { + Some(InnerAttrForbiddenReason::AfterOuterDocComment { + prev_doc_comment_span: prev_outer_attr_sp.unwrap(), + }) + } else { + prev_outer_attr_sp.map(|prev_outer_attr_sp| { + InnerAttrForbiddenReason::AfterOuterAttribute { prev_outer_attr_sp } + }) + }; + let inner_parse_policy = InnerAttrPolicy::Forbidden(inner_error_reason); + just_parsed_doc_comment = false; + Some(self.parse_attribute(inner_parse_policy)?) + } else if let token::DocComment(comment_kind, attr_style, data) = self.token.kind { + if attr_style != ast::AttrStyle::Outer { + let span = self.token.span; + let mut err = self + .dcx() + .struct_span_err(span, fluent::parse_inner_doc_comment_not_permitted); + err.code(E0753); + if let Some(replacement_span) = self.annotate_following_item_if_applicable( + &mut err, + span, + match comment_kind { + token::CommentKind::Line => OuterAttributeType::DocComment, + token::CommentKind::Block => OuterAttributeType::DocBlockComment, + }, + ) { + err.note(fluent::parse_note); + err.span_suggestion_verbose( + replacement_span, + fluent::parse_suggestion, + "", + rustc_errors::Applicability::MachineApplicable, + ); + } + err.emit(); + } + self.bump(); + just_parsed_doc_comment = true; + // Always make an outer attribute - this allows us to recover from a misplaced + // inner attribute. + Some(attr::mk_doc_comment( + &self.psess.attr_id_generator, + comment_kind, + ast::AttrStyle::Outer, + data, + self.prev_token.span, + )) + } else { + None + }; + + if let Some(attr) = attr { + if attr.style == ast::AttrStyle::Outer { + outer_attrs.push(attr); + } + } else { + break; + } + } + Ok(AttrWrapper::new(outer_attrs, start_pos)) + } + + /// Matches `attribute = # ! [ meta_item ]`. + /// `inner_parse_policy` prescribes how to handle inner attributes. + // Public for rustfmt usage. + pub fn parse_attribute( + &mut self, + inner_parse_policy: InnerAttrPolicy, + ) -> PResult<'a, ast::Attribute> { + debug!( + "parse_attribute: inner_parse_policy={:?} self.token={:?}", + inner_parse_policy, self.token + ); + let lo = self.token.span; + // Attributes can't have attributes of their own [Editor's note: not with that attitude] + self.collect_tokens_no_attrs(|this| { + assert!(this.eat(&token::Pound), "parse_attribute called in non-attribute position"); + + let style = + if this.eat(&token::Not) { ast::AttrStyle::Inner } else { ast::AttrStyle::Outer }; + + this.expect(&token::OpenDelim(Delimiter::Bracket))?; + let item = this.parse_attr_item(false)?; + this.expect(&token::CloseDelim(Delimiter::Bracket))?; + let attr_sp = lo.to(this.prev_token.span); + + // Emit error if inner attribute is encountered and forbidden. + if style == ast::AttrStyle::Inner { + this.error_on_forbidden_inner_attr(attr_sp, inner_parse_policy); + } + + Ok(attr::mk_attr_from_item(&self.psess.attr_id_generator, item, None, style, attr_sp)) + }) + } + + fn annotate_following_item_if_applicable( + &self, + err: &mut Diag<'_>, + span: Span, + attr_type: OuterAttributeType, + ) -> Option<Span> { + let mut snapshot = self.create_snapshot_for_diagnostic(); + let lo = span.lo() + + BytePos(match attr_type { + OuterAttributeType::Attribute => 1, + _ => 2, + }); + let hi = lo + BytePos(1); + let replacement_span = span.with_lo(lo).with_hi(hi); + if let OuterAttributeType::DocBlockComment | OuterAttributeType::DocComment = attr_type { + snapshot.bump(); + } + loop { + // skip any other attributes, we want the item + if snapshot.token.kind == token::Pound { + if let Err(err) = snapshot.parse_attribute(InnerAttrPolicy::Permitted) { + err.cancel(); + return Some(replacement_span); + } + } else { + break; + } + } + match snapshot.parse_item_common( + AttrWrapper::empty(), + true, + false, + FnParseMode { req_name: |_| true, req_body: true }, + ForceCollect::No, + ) { + Ok(Some(item)) => { + // FIXME(#100717) + err.arg("item", item.kind.descr()); + err.span_label(item.span, fluent::parse_label_does_not_annotate_this); + err.span_suggestion_verbose( + replacement_span, + fluent::parse_sugg_change_inner_to_outer, + match attr_type { + OuterAttributeType::Attribute => "", + OuterAttributeType::DocBlockComment => "*", + OuterAttributeType::DocComment => "/", + }, + rustc_errors::Applicability::MachineApplicable, + ); + return None; + } + Err(item_err) => { + item_err.cancel(); + } + Ok(None) => {} + } + Some(replacement_span) + } + + pub(super) fn error_on_forbidden_inner_attr(&self, attr_sp: Span, policy: InnerAttrPolicy) { + if let InnerAttrPolicy::Forbidden(reason) = policy { + let mut diag = match reason.as_ref().copied() { + Some(InnerAttrForbiddenReason::AfterOuterDocComment { prev_doc_comment_span }) => { + self.dcx() + .struct_span_err( + attr_sp, + fluent::parse_inner_attr_not_permitted_after_outer_doc_comment, + ) + .with_span_label(attr_sp, fluent::parse_label_attr) + .with_span_label( + prev_doc_comment_span, + fluent::parse_label_prev_doc_comment, + ) + } + Some(InnerAttrForbiddenReason::AfterOuterAttribute { prev_outer_attr_sp }) => self + .dcx() + .struct_span_err( + attr_sp, + fluent::parse_inner_attr_not_permitted_after_outer_attr, + ) + .with_span_label(attr_sp, fluent::parse_label_attr) + .with_span_label(prev_outer_attr_sp, fluent::parse_label_prev_attr), + Some(InnerAttrForbiddenReason::InCodeBlock) | None => { + self.dcx().struct_span_err(attr_sp, fluent::parse_inner_attr_not_permitted) + } + }; + + diag.note(fluent::parse_inner_attr_explanation); + if self + .annotate_following_item_if_applicable( + &mut diag, + attr_sp, + OuterAttributeType::Attribute, + ) + .is_some() + { + diag.note(fluent::parse_outer_attr_explanation); + }; + diag.emit(); + } + } + + /// Parses an inner part of an attribute (the path and following tokens). + /// The tokens must be either a delimited token stream, or empty token stream, + /// or the "legacy" key-value form. + /// PATH `(` TOKEN_STREAM `)` + /// PATH `[` TOKEN_STREAM `]` + /// PATH `{` TOKEN_STREAM `}` + /// PATH + /// PATH `=` UNSUFFIXED_LIT + /// The delimiters or `=` are still put into the resulting token stream. + pub fn parse_attr_item(&mut self, capture_tokens: bool) -> PResult<'a, ast::AttrItem> { + maybe_whole!(self, NtMeta, |attr| attr.into_inner()); + + let do_parse = |this: &mut Self| { + let path = this.parse_path(PathStyle::Mod)?; + let args = this.parse_attr_args()?; + Ok(ast::AttrItem { path, args, tokens: None }) + }; + // Attr items don't have attributes + if capture_tokens { self.collect_tokens_no_attrs(do_parse) } else { do_parse(self) } + } + + /// Parses attributes that appear after the opening of an item. These should + /// be preceded by an exclamation mark, but we accept and warn about one + /// terminated by a semicolon. + /// + /// Matches `inner_attrs*`. + pub(crate) fn parse_inner_attributes(&mut self) -> PResult<'a, ast::AttrVec> { + let mut attrs = ast::AttrVec::new(); + loop { + let start_pos: u32 = self.num_bump_calls.try_into().unwrap(); + // Only try to parse if it is an inner attribute (has `!`). + let attr = if self.check(&token::Pound) && self.look_ahead(1, |t| t == &token::Not) { + Some(self.parse_attribute(InnerAttrPolicy::Permitted)?) + } else if let token::DocComment(comment_kind, attr_style, data) = self.token.kind { + if attr_style == ast::AttrStyle::Inner { + self.bump(); + Some(attr::mk_doc_comment( + &self.psess.attr_id_generator, + comment_kind, + attr_style, + data, + self.prev_token.span, + )) + } else { + None + } + } else { + None + }; + if let Some(attr) = attr { + let end_pos: u32 = self.num_bump_calls.try_into().unwrap(); + // If we are currently capturing tokens, mark the location of this inner attribute. + // If capturing ends up creating a `LazyAttrTokenStream`, we will include + // this replace range with it, removing the inner attribute from the final + // `AttrTokenStream`. Inner attributes are stored in the parsed AST note. + // During macro expansion, they are selectively inserted back into the + // token stream (the first inner attribute is removed each time we invoke the + // corresponding macro). + let range = start_pos..end_pos; + if let Capturing::Yes = self.capture_state.capturing { + self.capture_state.inner_attr_ranges.insert(attr.id, (range, vec![])); + } + attrs.push(attr); + } else { + break; + } + } + Ok(attrs) + } + + // Note: must be unsuffixed. + pub(crate) fn parse_unsuffixed_meta_item_lit(&mut self) -> PResult<'a, ast::MetaItemLit> { + let lit = self.parse_meta_item_lit()?; + debug!("checking if {:?} is unsuffixed", lit); + + if !lit.kind.is_unsuffixed() { + self.dcx().emit_err(errors::SuffixedLiteralInAttribute { span: lit.span }); + } + + Ok(lit) + } + + /// Parses `cfg_attr(pred, attr_item_list)` where `attr_item_list` is comma-delimited. + pub fn parse_cfg_attr(&mut self) -> PResult<'a, (ast::MetaItem, Vec<(ast::AttrItem, Span)>)> { + let cfg_predicate = self.parse_meta_item()?; + self.expect(&token::Comma)?; + + // Presumably, the majority of the time there will only be one attr. + let mut expanded_attrs = Vec::with_capacity(1); + while self.token.kind != token::Eof { + let lo = self.token.span; + let item = self.parse_attr_item(true)?; + expanded_attrs.push((item, lo.to(self.prev_token.span))); + if !self.eat(&token::Comma) { + break; + } + } + + Ok((cfg_predicate, expanded_attrs)) + } + + /// Matches `COMMASEP(meta_item_inner)`. + pub(crate) fn parse_meta_seq_top(&mut self) -> PResult<'a, ThinVec<ast::NestedMetaItem>> { + // Presumably, the majority of the time there will only be one attr. + let mut nmis = ThinVec::with_capacity(1); + while self.token.kind != token::Eof { + nmis.push(self.parse_meta_item_inner()?); + if !self.eat(&token::Comma) { + break; + } + } + Ok(nmis) + } + + /// Parse a meta item per RFC 1559. + /// + /// ```ebnf + /// MetaItem = SimplePath ( '=' UNSUFFIXED_LIT | '(' MetaSeq? ')' )? ; + /// MetaSeq = MetaItemInner (',' MetaItemInner)* ','? ; + /// ``` + pub fn parse_meta_item(&mut self) -> PResult<'a, ast::MetaItem> { + // We can't use `maybe_whole` here because it would bump in the `None` + // case, which we don't want. + if let token::Interpolated(nt) = &self.token.kind + && let token::NtMeta(attr_item) = &**nt + { + match attr_item.meta(attr_item.path.span) { + Some(meta) => { + self.bump(); + return Ok(meta); + } + None => self.unexpected()?, + } + } + + let lo = self.token.span; + let path = self.parse_path(PathStyle::Mod)?; + let kind = self.parse_meta_item_kind()?; + let span = lo.to(self.prev_token.span); + Ok(ast::MetaItem { path, kind, span }) + } + + pub(crate) fn parse_meta_item_kind(&mut self) -> PResult<'a, ast::MetaItemKind> { + Ok(if self.eat(&token::Eq) { + ast::MetaItemKind::NameValue(self.parse_unsuffixed_meta_item_lit()?) + } else if self.check(&token::OpenDelim(Delimiter::Parenthesis)) { + let (list, _) = self.parse_paren_comma_seq(|p| p.parse_meta_item_inner())?; + ast::MetaItemKind::List(list) + } else { + ast::MetaItemKind::Word + }) + } + + /// Parse an inner meta item per RFC 1559. + /// + /// ```ebnf + /// MetaItemInner = UNSUFFIXED_LIT | MetaItem ; + /// ``` + fn parse_meta_item_inner(&mut self) -> PResult<'a, ast::NestedMetaItem> { + match self.parse_unsuffixed_meta_item_lit() { + Ok(lit) => return Ok(ast::NestedMetaItem::Lit(lit)), + Err(err) => err.cancel(), // we provide a better error below + } + + match self.parse_meta_item() { + Ok(mi) => return Ok(ast::NestedMetaItem::MetaItem(mi)), + Err(err) => err.cancel(), // we provide a better error below + } + + let mut err = errors::InvalidMetaItem { + span: self.token.span, + token: self.token.clone(), + quote_ident_sugg: None, + }; + + // Suggest quoting idents, e.g. in `#[cfg(key = value)]`. We don't use `Token::ident` and + // don't `uninterpolate` the token to avoid suggesting anything butchered or questionable + // when macro metavariables are involved. + if self.prev_token == token::Eq + && let token::Ident(..) = self.token.kind + { + let before = self.token.span.shrink_to_lo(); + while let token::Ident(..) = self.token.kind { + self.bump(); + } + err.quote_ident_sugg = Some(errors::InvalidMetaItemQuoteIdentSugg { + before, + after: self.prev_token.span.shrink_to_hi(), + }); + } + + Err(self.dcx().create_err(err)) + } +} + +/// The attributes are complete if all attributes are either a doc comment or a builtin attribute other than `cfg_attr` +pub fn is_complete(attrs: &[ast::Attribute]) -> bool { + attrs.iter().all(|attr| { + attr.is_doc_comment() + || attr.ident().is_some_and(|ident| { + ident.name != sym::cfg_attr && rustc_feature::is_builtin_attr_name(ident.name) + }) + }) +} diff --git a/compiler/rustc_parse/src/parser/attr_wrapper.rs b/compiler/rustc_parse/src/parser/attr_wrapper.rs new file mode 100644 index 00000000000..62c8f9f5dac --- /dev/null +++ b/compiler/rustc_parse/src/parser/attr_wrapper.rs @@ -0,0 +1,465 @@ +use super::{Capturing, FlatToken, ForceCollect, Parser, ReplaceRange, TokenCursor, TrailingToken}; +use rustc_ast::token::{self, Delimiter, Token, TokenKind}; +use rustc_ast::tokenstream::{AttrTokenStream, AttrTokenTree, AttributesData, DelimSpacing}; +use rustc_ast::tokenstream::{DelimSpan, LazyAttrTokenStream, Spacing, ToAttrTokenStream}; +use rustc_ast::{self as ast}; +use rustc_ast::{AttrVec, Attribute, HasAttrs, HasTokens}; +use rustc_errors::PResult; +use rustc_session::parse::ParseSess; +use rustc_span::{sym, Span, DUMMY_SP}; + +use std::ops::Range; + +/// A wrapper type to ensure that the parser handles outer attributes correctly. +/// When we parse outer attributes, we need to ensure that we capture tokens +/// for the attribute target. This allows us to perform cfg-expansion on +/// a token stream before we invoke a derive proc-macro. +/// +/// This wrapper prevents direct access to the underlying `ast::AttrVec>`. +/// Parsing code can only get access to the underlying attributes +/// by passing an `AttrWrapper` to `collect_tokens_trailing_tokens`. +/// This makes it difficult to accidentally construct an AST node +/// (which stores an `ast::AttrVec`) without first collecting tokens. +/// +/// This struct has its own module, to ensure that the parser code +/// cannot directly access the `attrs` field +#[derive(Debug, Clone)] +pub struct AttrWrapper { + attrs: AttrVec, + // The start of the outer attributes in the token cursor. + // This allows us to create a `ReplaceRange` for the entire attribute + // target, including outer attributes. + start_pos: usize, +} + +impl AttrWrapper { + pub(super) fn new(attrs: AttrVec, start_pos: usize) -> AttrWrapper { + AttrWrapper { attrs, start_pos } + } + pub fn empty() -> AttrWrapper { + AttrWrapper { attrs: AttrVec::new(), start_pos: usize::MAX } + } + + pub(crate) fn take_for_recovery(self, psess: &ParseSess) -> AttrVec { + psess.dcx.span_delayed_bug( + self.attrs.get(0).map(|attr| attr.span).unwrap_or(DUMMY_SP), + "AttrVec is taken for recovery but no error is produced", + ); + + self.attrs + } + + /// Prepend `self.attrs` to `attrs`. + // FIXME: require passing an NT to prevent misuse of this method + pub(crate) fn prepend_to_nt_inner(self, attrs: &mut AttrVec) { + let mut self_attrs = self.attrs; + std::mem::swap(attrs, &mut self_attrs); + attrs.extend(self_attrs); + } + + pub fn is_empty(&self) -> bool { + self.attrs.is_empty() + } + + pub fn is_complete(&self) -> bool { + crate::parser::attr::is_complete(&self.attrs) + } +} + +/// Returns `true` if `attrs` contains a `cfg` or `cfg_attr` attribute +fn has_cfg_or_cfg_attr(attrs: &[Attribute]) -> bool { + // NOTE: Builtin attributes like `cfg` and `cfg_attr` cannot be renamed via imports. + // Therefore, the absence of a literal `cfg` or `cfg_attr` guarantees that + // we don't need to do any eager expansion. + attrs.iter().any(|attr| { + attr.ident().is_some_and(|ident| ident.name == sym::cfg || ident.name == sym::cfg_attr) + }) +} + +// Produces a `TokenStream` on-demand. Using `cursor_snapshot` +// and `num_calls`, we can reconstruct the `TokenStream` seen +// by the callback. This allows us to avoid producing a `TokenStream` +// if it is never needed - for example, a captured `macro_rules!` +// argument that is never passed to a proc macro. +// In practice token stream creation happens rarely compared to +// calls to `collect_tokens` (see some statistics in #78736), +// so we are doing as little up-front work as possible. +// +// This also makes `Parser` very cheap to clone, since +// there is no intermediate collection buffer to clone. +#[derive(Clone)] +struct LazyAttrTokenStreamImpl { + start_token: (Token, Spacing), + cursor_snapshot: TokenCursor, + num_calls: usize, + break_last_token: bool, + replace_ranges: Box<[ReplaceRange]>, +} + +impl ToAttrTokenStream for LazyAttrTokenStreamImpl { + fn to_attr_token_stream(&self) -> AttrTokenStream { + // The token produced by the final call to `{,inlined_}next` was not + // actually consumed by the callback. The combination of chaining the + // initial token and using `take` produces the desired result - we + // produce an empty `TokenStream` if no calls were made, and omit the + // final token otherwise. + let mut cursor_snapshot = self.cursor_snapshot.clone(); + let tokens = + std::iter::once((FlatToken::Token(self.start_token.0.clone()), self.start_token.1)) + .chain(std::iter::repeat_with(|| { + let token = cursor_snapshot.next(); + (FlatToken::Token(token.0), token.1) + })) + .take(self.num_calls); + + if !self.replace_ranges.is_empty() { + let mut tokens: Vec<_> = tokens.collect(); + let mut replace_ranges = self.replace_ranges.to_vec(); + replace_ranges.sort_by_key(|(range, _)| range.start); + + #[cfg(debug_assertions)] + { + for [(range, tokens), (next_range, next_tokens)] in replace_ranges.array_windows() { + assert!( + range.end <= next_range.start || range.end >= next_range.end, + "Replace ranges should either be disjoint or nested: ({:?}, {:?}) ({:?}, {:?})", + range, + tokens, + next_range, + next_tokens, + ); + } + } + + // Process the replace ranges, starting from the highest start + // position and working our way back. If have tokens like: + // + // `#[cfg(FALSE)] struct Foo { #[cfg(FALSE)] field: bool }` + // + // Then we will generate replace ranges for both + // the `#[cfg(FALSE)] field: bool` and the entire + // `#[cfg(FALSE)] struct Foo { #[cfg(FALSE)] field: bool }` + // + // By starting processing from the replace range with the greatest + // start position, we ensure that any replace range which encloses + // another replace range will capture the *replaced* tokens for the inner + // range, not the original tokens. + for (range, new_tokens) in replace_ranges.into_iter().rev() { + assert!(!range.is_empty(), "Cannot replace an empty range: {range:?}"); + // Replace ranges are only allowed to decrease the number of tokens. + assert!( + range.len() >= new_tokens.len(), + "Range {range:?} has greater len than {new_tokens:?}" + ); + + // Replace any removed tokens with `FlatToken::Empty`. + // This keeps the total length of `tokens` constant throughout the + // replacement process, allowing us to use all of the `ReplaceRanges` entries + // without adjusting indices. + let filler = std::iter::repeat((FlatToken::Empty, Spacing::Alone)) + .take(range.len() - new_tokens.len()); + + tokens.splice( + (range.start as usize)..(range.end as usize), + new_tokens.into_iter().chain(filler), + ); + } + make_token_stream(tokens.into_iter(), self.break_last_token) + } else { + make_token_stream(tokens, self.break_last_token) + } + } +} + +impl<'a> Parser<'a> { + /// Records all tokens consumed by the provided callback, + /// including the current token. These tokens are collected + /// into a `LazyAttrTokenStream`, and returned along with the result + /// of the callback. + /// + /// Note: If your callback consumes an opening delimiter + /// (including the case where you call `collect_tokens` + /// when the current token is an opening delimiter), + /// you must also consume the corresponding closing delimiter. + /// + /// That is, you can consume + /// `something ([{ }])` or `([{}])`, but not `([{}]` + /// + /// This restriction shouldn't be an issue in practice, + /// since this function is used to record the tokens for + /// a parsed AST item, which always has matching delimiters. + pub fn collect_tokens_trailing_token<R: HasAttrs + HasTokens>( + &mut self, + attrs: AttrWrapper, + force_collect: ForceCollect, + f: impl FnOnce(&mut Self, ast::AttrVec) -> PResult<'a, (R, TrailingToken)>, + ) -> PResult<'a, R> { + // We only bail out when nothing could possibly observe the collected tokens: + // 1. We cannot be force collecting tokens (since force-collecting requires tokens + // by definition + if matches!(force_collect, ForceCollect::No) + // None of our outer attributes can require tokens (e.g. a proc-macro) + && attrs.is_complete() + // If our target supports custom inner attributes, then we cannot bail + // out early, since we may need to capture tokens for a custom inner attribute + // invocation. + && !R::SUPPORTS_CUSTOM_INNER_ATTRS + // Never bail out early in `capture_cfg` mode, since there might be `#[cfg]` + // or `#[cfg_attr]` attributes. + && !self.capture_cfg + { + return Ok(f(self, attrs.attrs)?.0); + } + + let start_token = (self.token.clone(), self.token_spacing); + let cursor_snapshot = self.token_cursor.clone(); + let start_pos = self.num_bump_calls; + + let has_outer_attrs = !attrs.attrs.is_empty(); + let prev_capturing = std::mem::replace(&mut self.capture_state.capturing, Capturing::Yes); + let replace_ranges_start = self.capture_state.replace_ranges.len(); + + let ret = f(self, attrs.attrs); + + self.capture_state.capturing = prev_capturing; + + let (mut ret, trailing) = ret?; + + // When we're not in `capture-cfg` mode, then bail out early if: + // 1. Our target doesn't support tokens at all (e.g we're parsing an `NtIdent`) + // so there's nothing for us to do. + // 2. Our target already has tokens set (e.g. we've parsed something + // like `#[my_attr] $item`. The actual parsing code takes care of prepending + // any attributes to the nonterminal, so we don't need to modify the + // already captured tokens. + // Note that this check is independent of `force_collect`- if we already + // have tokens, or can't even store them, then there's never a need to + // force collection of new tokens. + if !self.capture_cfg && matches!(ret.tokens_mut(), None | Some(Some(_))) { + return Ok(ret); + } + + // This is very similar to the bail out check at the start of this function. + // Now that we've parsed an AST node, we have more information available. + if matches!(force_collect, ForceCollect::No) + // We now have inner attributes available, so this check is more precise + // than `attrs.is_complete()` at the start of the function. + // As a result, we don't need to check `R::SUPPORTS_CUSTOM_INNER_ATTRS` + && crate::parser::attr::is_complete(ret.attrs()) + // Subtle: We call `has_cfg_or_cfg_attr` with the attrs from `ret`. + // This ensures that we consider inner attributes (e.g. `#![cfg]`), + // which require us to have tokens available + // We also call `has_cfg_or_cfg_attr` at the beginning of this function, + // but we only bail out if there's no possibility of inner attributes + // (!R::SUPPORTS_CUSTOM_INNER_ATTRS) + // We only capture about `#[cfg]` or `#[cfg_attr]` in `capture_cfg` + // mode - during normal parsing, we don't need any special capturing + // for those attributes, since they're builtin. + && !(self.capture_cfg && has_cfg_or_cfg_attr(ret.attrs())) + { + return Ok(ret); + } + + let mut inner_attr_replace_ranges = Vec::new(); + // Take the captured ranges for any inner attributes that we parsed. + for inner_attr in ret.attrs().iter().filter(|a| a.style == ast::AttrStyle::Inner) { + if let Some(attr_range) = self.capture_state.inner_attr_ranges.remove(&inner_attr.id) { + inner_attr_replace_ranges.push(attr_range); + } else { + self.dcx().span_delayed_bug(inner_attr.span, "Missing token range for attribute"); + } + } + + let replace_ranges_end = self.capture_state.replace_ranges.len(); + + let mut end_pos = self.num_bump_calls; + + let mut captured_trailing = false; + + // Capture a trailing token if requested by the callback 'f' + match trailing { + TrailingToken::None => {} + TrailingToken::Gt => { + assert_eq!(self.token.kind, token::Gt); + } + TrailingToken::Semi => { + assert_eq!(self.token.kind, token::Semi); + end_pos += 1; + captured_trailing = true; + } + TrailingToken::MaybeComma => { + if self.token.kind == token::Comma { + end_pos += 1; + captured_trailing = true; + } + } + } + + // If we 'broke' the last token (e.g. breaking a '>>' token to two '>' tokens), + // then extend the range of captured tokens to include it, since the parser + // was not actually bumped past it. When the `LazyAttrTokenStream` gets converted + // into an `AttrTokenStream`, we will create the proper token. + if self.break_last_token { + assert!(!captured_trailing, "Cannot set break_last_token and have trailing token"); + end_pos += 1; + } + + let num_calls = end_pos - start_pos; + + // If we have no attributes, then we will never need to + // use any replace ranges. + let replace_ranges: Box<[ReplaceRange]> = if ret.attrs().is_empty() && !self.capture_cfg { + Box::new([]) + } else { + // Grab any replace ranges that occur *inside* the current AST node. + // We will perform the actual replacement when we convert the `LazyAttrTokenStream` + // to an `AttrTokenStream`. + let start_calls: u32 = start_pos.try_into().unwrap(); + self.capture_state.replace_ranges[replace_ranges_start..replace_ranges_end] + .iter() + .cloned() + .chain(inner_attr_replace_ranges.iter().cloned()) + .map(|(range, tokens)| { + ((range.start - start_calls)..(range.end - start_calls), tokens) + }) + .collect() + }; + + let tokens = LazyAttrTokenStream::new(LazyAttrTokenStreamImpl { + start_token, + num_calls, + cursor_snapshot, + break_last_token: self.break_last_token, + replace_ranges, + }); + + // If we support tokens at all + if let Some(target_tokens) = ret.tokens_mut() { + if target_tokens.is_none() { + // Store se our newly captured tokens into the AST node + *target_tokens = Some(tokens.clone()); + } + } + + let final_attrs = ret.attrs(); + + // If `capture_cfg` is set and we're inside a recursive call to + // `collect_tokens_trailing_token`, then we need to register a replace range + // if we have `#[cfg]` or `#[cfg_attr]`. This allows us to run eager cfg-expansion + // on the captured token stream. + if self.capture_cfg + && matches!(self.capture_state.capturing, Capturing::Yes) + && has_cfg_or_cfg_attr(final_attrs) + { + let attr_data = AttributesData { attrs: final_attrs.iter().cloned().collect(), tokens }; + + // Replace the entire AST node that we just parsed, including attributes, + // with a `FlatToken::AttrTarget`. If this AST node is inside an item + // that has `#[derive]`, then this will allow us to cfg-expand this + // AST node. + let start_pos = if has_outer_attrs { attrs.start_pos } else { start_pos }; + let new_tokens = vec![(FlatToken::AttrTarget(attr_data), Spacing::Alone)]; + + assert!(!self.break_last_token, "Should not have unglued last token with cfg attr"); + let range: Range<u32> = (start_pos.try_into().unwrap())..(end_pos.try_into().unwrap()); + self.capture_state.replace_ranges.push((range, new_tokens)); + self.capture_state.replace_ranges.extend(inner_attr_replace_ranges); + } + + // Only clear our `replace_ranges` when we're finished capturing entirely. + if matches!(self.capture_state.capturing, Capturing::No) { + self.capture_state.replace_ranges.clear(); + // We don't clear `inner_attr_ranges`, as doing so repeatedly + // had a measurable performance impact. Most inner attributes that + // we insert will get removed - when we drop the parser, we'll free + // up the memory used by any attributes that we didn't remove from the map. + } + Ok(ret) + } +} + +/// Converts a flattened iterator of tokens (including open and close delimiter tokens) +/// into a `TokenStream`, creating a `TokenTree::Delimited` for each matching pair +/// of open and close delims. +fn make_token_stream( + mut iter: impl Iterator<Item = (FlatToken, Spacing)>, + break_last_token: bool, +) -> AttrTokenStream { + #[derive(Debug)] + struct FrameData { + // This is `None` for the first frame, `Some` for all others. + open_delim_sp: Option<(Delimiter, Span, Spacing)>, + inner: Vec<AttrTokenTree>, + } + let mut stack = vec![FrameData { open_delim_sp: None, inner: vec![] }]; + let mut token_and_spacing = iter.next(); + while let Some((token, spacing)) = token_and_spacing { + match token { + FlatToken::Token(Token { kind: TokenKind::OpenDelim(delim), span }) => { + stack + .push(FrameData { open_delim_sp: Some((delim, span, spacing)), inner: vec![] }); + } + FlatToken::Token(Token { kind: TokenKind::CloseDelim(delim), span }) => { + let frame_data = stack + .pop() + .unwrap_or_else(|| panic!("Token stack was empty for token: {token:?}")); + + let (open_delim, open_sp, open_spacing) = frame_data.open_delim_sp.unwrap(); + assert_eq!( + open_delim, delim, + "Mismatched open/close delims: open={open_delim:?} close={span:?}" + ); + let dspan = DelimSpan::from_pair(open_sp, span); + let dspacing = DelimSpacing::new(open_spacing, spacing); + let stream = AttrTokenStream::new(frame_data.inner); + let delimited = AttrTokenTree::Delimited(dspan, dspacing, delim, stream); + stack + .last_mut() + .unwrap_or_else(|| panic!("Bottom token frame is missing for token: {token:?}")) + .inner + .push(delimited); + } + FlatToken::Token(token) => stack + .last_mut() + .expect("Bottom token frame is missing!") + .inner + .push(AttrTokenTree::Token(token, spacing)), + FlatToken::AttrTarget(data) => stack + .last_mut() + .expect("Bottom token frame is missing!") + .inner + .push(AttrTokenTree::Attributes(data)), + FlatToken::Empty => {} + } + token_and_spacing = iter.next(); + } + let mut final_buf = stack.pop().expect("Missing final buf!"); + if break_last_token { + let last_token = final_buf.inner.pop().unwrap(); + if let AttrTokenTree::Token(last_token, spacing) = last_token { + let unglued_first = last_token.kind.break_two_token_op().unwrap().0; + + // An 'unglued' token is always two ASCII characters + let mut first_span = last_token.span.shrink_to_lo(); + first_span = first_span.with_hi(first_span.lo() + rustc_span::BytePos(1)); + + final_buf + .inner + .push(AttrTokenTree::Token(Token::new(unglued_first, first_span), spacing)); + } else { + panic!("Unexpected last token {last_token:?}") + } + } + AttrTokenStream::new(final_buf.inner) +} + +// Some types are used a lot. Make sure they don't unintentionally get bigger. +#[cfg(target_pointer_width = "64")] +mod size_asserts { + use super::*; + use rustc_data_structures::static_assert_size; + // tidy-alphabetical-start + static_assert_size!(AttrWrapper, 16); + static_assert_size!(LazyAttrTokenStreamImpl, 104); + // tidy-alphabetical-end +} diff --git a/compiler/rustc_parse/src/parser/diagnostics.rs b/compiler/rustc_parse/src/parser/diagnostics.rs new file mode 100644 index 00000000000..ac12787f2ef --- /dev/null +++ b/compiler/rustc_parse/src/parser/diagnostics.rs @@ -0,0 +1,3056 @@ +use super::pat::Expected; +use super::{ + BlockMode, CommaRecoveryMode, Parser, PathStyle, Restrictions, SemiColonMode, SeqSep, + TokenExpectType, TokenType, +}; +use crate::errors::{ + AmbiguousPlus, AsyncMoveBlockIn2015, AttributeOnParamType, BadQPathStage2, BadTypePlus, + BadTypePlusSub, ColonAsSemi, ComparisonOperatorsCannotBeChained, + ComparisonOperatorsCannotBeChainedSugg, ConstGenericWithoutBraces, + ConstGenericWithoutBracesSugg, DocCommentDoesNotDocumentAnything, DocCommentOnParamType, + DoubleColonInBound, ExpectedIdentifier, ExpectedSemi, ExpectedSemiSugg, + GenericParamsWithoutAngleBrackets, GenericParamsWithoutAngleBracketsSugg, + HelpIdentifierStartsWithNumber, HelpUseLatestEdition, InInTypo, IncorrectAwait, + IncorrectSemicolon, IncorrectUseOfAwait, PatternMethodParamWithoutBody, QuestionMarkInType, + QuestionMarkInTypeSugg, SelfParamNotFirst, StructLiteralBodyWithoutPath, + StructLiteralBodyWithoutPathSugg, StructLiteralNeedingParens, StructLiteralNeedingParensSugg, + SuggAddMissingLetStmt, SuggEscapeIdentifier, SuggRemoveComma, TernaryOperator, + UnexpectedConstInGenericParam, UnexpectedConstParamDeclaration, + UnexpectedConstParamDeclarationSugg, UnmatchedAngleBrackets, UseEqInstead, WrapType, +}; +use crate::fluent_generated as fluent; +use crate::parser; +use crate::parser::attr::InnerAttrPolicy; +use ast::token::IdentIsRaw; +use rustc_ast as ast; +use rustc_ast::ptr::P; +use rustc_ast::token::{self, Delimiter, Lit, LitKind, Token, TokenKind}; +use rustc_ast::tokenstream::AttrTokenTree; +use rustc_ast::util::parser::AssocOp; +use rustc_ast::{ + AngleBracketedArg, AngleBracketedArgs, AnonConst, AttrVec, BinOpKind, BindingMode, Block, + BlockCheckMode, Expr, ExprKind, GenericArg, Generics, HasTokens, Item, ItemKind, Param, Pat, + PatKind, Path, PathSegment, QSelf, Recovered, Ty, TyKind, +}; +use rustc_ast_pretty::pprust; +use rustc_data_structures::fx::FxHashSet; +use rustc_errors::{ + pluralize, Applicability, Diag, DiagCtxt, ErrorGuaranteed, FatalError, PErr, PResult, + Subdiagnostic, +}; +use rustc_session::errors::ExprParenthesesNeeded; +use rustc_span::source_map::Spanned; +use rustc_span::symbol::{kw, sym, Ident}; +use rustc_span::{BytePos, Span, SpanSnippetError, Symbol, DUMMY_SP}; +use std::mem::take; +use std::ops::{Deref, DerefMut}; +use thin_vec::{thin_vec, ThinVec}; + +/// Creates a placeholder argument. +pub(super) fn dummy_arg(ident: Ident, guar: ErrorGuaranteed) -> Param { + let pat = P(Pat { + id: ast::DUMMY_NODE_ID, + kind: PatKind::Ident(BindingMode::NONE, ident, None), + span: ident.span, + tokens: None, + }); + let ty = Ty { kind: TyKind::Err(guar), span: ident.span, id: ast::DUMMY_NODE_ID, tokens: None }; + Param { + attrs: AttrVec::default(), + id: ast::DUMMY_NODE_ID, + pat, + span: ident.span, + ty: P(ty), + is_placeholder: false, + } +} + +pub(super) trait RecoverQPath: Sized + 'static { + const PATH_STYLE: PathStyle = PathStyle::Expr; + fn to_ty(&self) -> Option<P<Ty>>; + fn recovered(qself: Option<P<QSelf>>, path: ast::Path) -> Self; +} + +impl RecoverQPath for Ty { + const PATH_STYLE: PathStyle = PathStyle::Type; + fn to_ty(&self) -> Option<P<Ty>> { + Some(P(self.clone())) + } + fn recovered(qself: Option<P<QSelf>>, path: ast::Path) -> Self { + Self { + span: path.span, + kind: TyKind::Path(qself, path), + id: ast::DUMMY_NODE_ID, + tokens: None, + } + } +} + +impl RecoverQPath for Pat { + const PATH_STYLE: PathStyle = PathStyle::Pat; + fn to_ty(&self) -> Option<P<Ty>> { + self.to_ty() + } + fn recovered(qself: Option<P<QSelf>>, path: ast::Path) -> Self { + Self { + span: path.span, + kind: PatKind::Path(qself, path), + id: ast::DUMMY_NODE_ID, + tokens: None, + } + } +} + +impl RecoverQPath for Expr { + fn to_ty(&self) -> Option<P<Ty>> { + self.to_ty() + } + fn recovered(qself: Option<P<QSelf>>, path: ast::Path) -> Self { + Self { + span: path.span, + kind: ExprKind::Path(qself, path), + attrs: AttrVec::new(), + id: ast::DUMMY_NODE_ID, + tokens: None, + } + } +} + +/// Control whether the closing delimiter should be consumed when calling `Parser::consume_block`. +pub(crate) enum ConsumeClosingDelim { + Yes, + No, +} + +#[derive(Clone, Copy)] +pub enum AttemptLocalParseRecovery { + Yes, + No, +} + +impl AttemptLocalParseRecovery { + pub fn yes(&self) -> bool { + match self { + AttemptLocalParseRecovery::Yes => true, + AttemptLocalParseRecovery::No => false, + } + } + + pub fn no(&self) -> bool { + match self { + AttemptLocalParseRecovery::Yes => false, + AttemptLocalParseRecovery::No => true, + } + } +} + +/// Information for emitting suggestions and recovering from +/// C-style `i++`, `--i`, etc. +#[derive(Debug, Copy, Clone)] +struct IncDecRecovery { + /// Is this increment/decrement its own statement? + standalone: IsStandalone, + /// Is this an increment or decrement? + op: IncOrDec, + /// Is this pre- or postfix? + fixity: UnaryFixity, +} + +/// Is an increment or decrement expression its own statement? +#[derive(Debug, Copy, Clone)] +enum IsStandalone { + /// It's standalone, i.e., its own statement. + Standalone, + /// It's a subexpression, i.e., *not* standalone. + Subexpr, +} + +#[derive(Debug, Copy, Clone, PartialEq, Eq)] +enum IncOrDec { + Inc, + Dec, +} + +#[derive(Debug, Copy, Clone, PartialEq, Eq)] +enum UnaryFixity { + Pre, + Post, +} + +impl IncOrDec { + fn chr(&self) -> char { + match self { + Self::Inc => '+', + Self::Dec => '-', + } + } + + fn name(&self) -> &'static str { + match self { + Self::Inc => "increment", + Self::Dec => "decrement", + } + } +} + +impl std::fmt::Display for UnaryFixity { + fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result { + match self { + Self::Pre => write!(f, "prefix"), + Self::Post => write!(f, "postfix"), + } + } +} + +struct MultiSugg { + msg: String, + patches: Vec<(Span, String)>, + applicability: Applicability, +} + +impl MultiSugg { + fn emit(self, err: &mut Diag<'_>) { + err.multipart_suggestion(self.msg, self.patches, self.applicability); + } + + fn emit_verbose(self, err: &mut Diag<'_>) { + err.multipart_suggestion_verbose(self.msg, self.patches, self.applicability); + } +} + +/// SnapshotParser is used to create a snapshot of the parser +/// without causing duplicate errors being emitted when the `Parser` +/// is dropped. +pub struct SnapshotParser<'a> { + parser: Parser<'a>, +} + +impl<'a> Deref for SnapshotParser<'a> { + type Target = Parser<'a>; + + fn deref(&self) -> &Self::Target { + &self.parser + } +} + +impl<'a> DerefMut for SnapshotParser<'a> { + fn deref_mut(&mut self) -> &mut Self::Target { + &mut self.parser + } +} + +impl<'a> Parser<'a> { + pub fn dcx(&self) -> &'a DiagCtxt { + &self.psess.dcx + } + + /// Replace `self` with `snapshot.parser`. + pub(super) fn restore_snapshot(&mut self, snapshot: SnapshotParser<'a>) { + *self = snapshot.parser; + } + + /// Create a snapshot of the `Parser`. + pub fn create_snapshot_for_diagnostic(&self) -> SnapshotParser<'a> { + let snapshot = self.clone(); + SnapshotParser { parser: snapshot } + } + + pub(super) fn span_to_snippet(&self, span: Span) -> Result<String, SpanSnippetError> { + self.psess.source_map().span_to_snippet(span) + } + + /// Emits an error with suggestions if an identifier was expected but not found. + /// + /// Returns a possibly recovered identifier. + pub(super) fn expected_ident_found( + &mut self, + recover: bool, + ) -> PResult<'a, (Ident, IdentIsRaw)> { + if let TokenKind::DocComment(..) = self.prev_token.kind { + return Err(self.dcx().create_err(DocCommentDoesNotDocumentAnything { + span: self.prev_token.span, + missing_comma: None, + })); + } + + let valid_follow = &[ + TokenKind::Eq, + TokenKind::Colon, + TokenKind::Comma, + TokenKind::Semi, + TokenKind::PathSep, + TokenKind::OpenDelim(Delimiter::Brace), + TokenKind::OpenDelim(Delimiter::Parenthesis), + TokenKind::CloseDelim(Delimiter::Brace), + TokenKind::CloseDelim(Delimiter::Parenthesis), + ]; + + let mut recovered_ident = None; + // we take this here so that the correct original token is retained in + // the diagnostic, regardless of eager recovery. + let bad_token = self.token.clone(); + + // suggest prepending a keyword in identifier position with `r#` + let suggest_raw = if let Some((ident, IdentIsRaw::No)) = self.token.ident() + && ident.is_raw_guess() + && self.look_ahead(1, |t| valid_follow.contains(&t.kind)) + { + recovered_ident = Some((ident, IdentIsRaw::Yes)); + + // `Symbol::to_string()` is different from `Symbol::into_diag_arg()`, + // which uses `Symbol::to_ident_string()` and "helpfully" adds an implicit `r#` + let ident_name = ident.name.to_string(); + + Some(SuggEscapeIdentifier { span: ident.span.shrink_to_lo(), ident_name }) + } else { + None + }; + + let suggest_remove_comma = + if self.token == token::Comma && self.look_ahead(1, |t| t.is_ident()) { + if recover { + self.bump(); + recovered_ident = self.ident_or_err(false).ok(); + }; + + Some(SuggRemoveComma { span: bad_token.span }) + } else { + None + }; + + let help_cannot_start_number = self.is_lit_bad_ident().map(|(len, valid_portion)| { + let (invalid, valid) = self.token.span.split_at(len as u32); + + recovered_ident = Some((Ident::new(valid_portion, valid), IdentIsRaw::No)); + + HelpIdentifierStartsWithNumber { num_span: invalid } + }); + + let err = ExpectedIdentifier { + span: bad_token.span, + token: bad_token, + suggest_raw, + suggest_remove_comma, + help_cannot_start_number, + }; + let mut err = self.dcx().create_err(err); + + // if the token we have is a `<` + // it *might* be a misplaced generic + // FIXME: could we recover with this? + if self.token == token::Lt { + // all keywords that could have generic applied + let valid_prev_keywords = + [kw::Fn, kw::Type, kw::Struct, kw::Enum, kw::Union, kw::Trait]; + + // If we've expected an identifier, + // and the current token is a '<' + // if the previous token is a valid keyword + // that might use a generic, then suggest a correct + // generic placement (later on) + let maybe_keyword = self.prev_token.clone(); + if valid_prev_keywords.into_iter().any(|x| maybe_keyword.is_keyword(x)) { + // if we have a valid keyword, attempt to parse generics + // also obtain the keywords symbol + match self.parse_generics() { + Ok(generic) => { + if let TokenKind::Ident(symbol, _) = maybe_keyword.kind { + let ident_name = symbol; + // at this point, we've found something like + // `fn <T>id` + // and current token should be Ident with the item name (i.e. the function name) + // if there is a `<` after the fn name, then don't show a suggestion, show help + + if !self.look_ahead(1, |t| *t == token::Lt) + && let Ok(snippet) = + self.psess.source_map().span_to_snippet(generic.span) + { + err.multipart_suggestion_verbose( + format!("place the generic parameter name after the {ident_name} name"), + vec![ + (self.token.span.shrink_to_hi(), snippet), + (generic.span, String::new()) + ], + Applicability::MaybeIncorrect, + ); + } else { + err.help(format!( + "place the generic parameter name after the {ident_name} name" + )); + } + } + } + Err(err) => { + // if there's an error parsing the generics, + // then don't do a misplaced generics suggestion + // and emit the expected ident error instead; + err.cancel(); + } + } + } + } + + if let Some(recovered_ident) = recovered_ident + && recover + { + err.emit(); + Ok(recovered_ident) + } else { + Err(err) + } + } + + pub(super) fn expected_ident_found_err(&mut self) -> Diag<'a> { + self.expected_ident_found(false).unwrap_err() + } + + /// Checks if the current token is a integer or float literal and looks like + /// it could be a invalid identifier with digits at the start. + /// + /// Returns the number of characters (bytes) composing the invalid portion + /// of the identifier and the valid portion of the identifier. + pub(super) fn is_lit_bad_ident(&mut self) -> Option<(usize, Symbol)> { + // ensure that the integer literal is followed by a *invalid* + // suffix: this is how we know that it is a identifier with an + // invalid beginning. + if let token::Literal(Lit { + kind: token::LitKind::Integer | token::LitKind::Float, + symbol, + suffix: Some(suffix), // no suffix makes it a valid literal + }) = self.token.kind + && rustc_ast::MetaItemLit::from_token(&self.token).is_none() + { + Some((symbol.as_str().len(), suffix)) + } else { + None + } + } + + pub(super) fn expected_one_of_not_found( + &mut self, + edible: &[TokenKind], + inedible: &[TokenKind], + ) -> PResult<'a, Recovered> { + debug!("expected_one_of_not_found(edible: {:?}, inedible: {:?})", edible, inedible); + fn tokens_to_string(tokens: &[TokenType]) -> String { + let mut i = tokens.iter(); + // This might be a sign we need a connect method on `Iterator`. + let b = i.next().map_or_else(String::new, |t| t.to_string()); + i.enumerate().fold(b, |mut b, (i, a)| { + if tokens.len() > 2 && i == tokens.len() - 2 { + b.push_str(", or "); + } else if tokens.len() == 2 && i == tokens.len() - 2 { + b.push_str(" or "); + } else { + b.push_str(", "); + } + b.push_str(&a.to_string()); + b + }) + } + + self.expected_tokens.extend(edible.iter().chain(inedible).cloned().map(TokenType::Token)); + let mut expected = self + .expected_tokens + .iter() + .filter(|token| { + // Filter out suggestions that suggest the same token which was found and deemed incorrect. + fn is_ident_eq_keyword(found: &TokenKind, expected: &TokenType) -> bool { + if let TokenKind::Ident(current_sym, _) = found + && let TokenType::Keyword(suggested_sym) = expected + { + return current_sym == suggested_sym; + } + false + } + + if **token != parser::TokenType::Token(self.token.kind.clone()) { + let eq = is_ident_eq_keyword(&self.token.kind, &token); + // If the suggestion is a keyword and the found token is an ident, + // the content of which are equal to the suggestion's content, + // we can remove that suggestion (see the `return false` below). + + // If this isn't the case however, and the suggestion is a token the + // content of which is the same as the found token's, we remove it as well. + if !eq { + if let TokenType::Token(kind) = &token { + if kind == &self.token.kind { + return false; + } + } + return true; + } + } + false + }) + .cloned() + .collect::<Vec<_>>(); + expected.sort_by_cached_key(|x| x.to_string()); + expected.dedup(); + + let sm = self.psess.source_map(); + + // Special-case "expected `;`" errors. + if expected.contains(&TokenType::Token(token::Semi)) { + // If the user is trying to write a ternary expression, recover it and + // return an Err to prevent a cascade of irrelevant diagnostics. + if self.prev_token == token::Question + && let Err(e) = self.maybe_recover_from_ternary_operator() + { + return Err(e); + } + + if self.token.span == DUMMY_SP || self.prev_token.span == DUMMY_SP { + // Likely inside a macro, can't provide meaningful suggestions. + } else if !sm.is_multiline(self.prev_token.span.until(self.token.span)) { + // The current token is in the same line as the prior token, not recoverable. + } else if [token::Comma, token::Colon].contains(&self.token.kind) + && self.prev_token.kind == token::CloseDelim(Delimiter::Parenthesis) + { + // Likely typo: The current token is on a new line and is expected to be + // `.`, `;`, `?`, or an operator after a close delimiter token. + // + // let a = std::process::Command::new("echo") + // .arg("1") + // ,arg("2") + // ^ + // https://github.com/rust-lang/rust/issues/72253 + } else if self.look_ahead(1, |t| { + t == &token::CloseDelim(Delimiter::Brace) + || t.can_begin_expr() && t.kind != token::Colon + }) && [token::Comma, token::Colon].contains(&self.token.kind) + { + // Likely typo: `,` → `;` or `:` → `;`. This is triggered if the current token is + // either `,` or `:`, and the next token could either start a new statement or is a + // block close. For example: + // + // let x = 32: + // let y = 42; + let guar = self.dcx().emit_err(ExpectedSemi { + span: self.token.span, + token: self.token.clone(), + unexpected_token_label: None, + sugg: ExpectedSemiSugg::ChangeToSemi(self.token.span), + }); + self.bump(); + return Ok(Recovered::Yes(guar)); + } else if self.look_ahead(0, |t| { + t == &token::CloseDelim(Delimiter::Brace) + || ((t.can_begin_expr() || t.can_begin_item()) + && t != &token::Semi + && t != &token::Pound) + // Avoid triggering with too many trailing `#` in raw string. + || (sm.is_multiline( + self.prev_token.span.shrink_to_hi().until(self.token.span.shrink_to_lo()), + ) && t == &token::Pound) + }) && !expected.contains(&TokenType::Token(token::Comma)) + { + // Missing semicolon typo. This is triggered if the next token could either start a + // new statement or is a block close. For example: + // + // let x = 32 + // let y = 42; + let span = self.prev_token.span.shrink_to_hi(); + let guar = self.dcx().emit_err(ExpectedSemi { + span, + token: self.token.clone(), + unexpected_token_label: Some(self.token.span), + sugg: ExpectedSemiSugg::AddSemi(span), + }); + return Ok(Recovered::Yes(guar)); + } + } + + if self.token.kind == TokenKind::EqEq + && self.prev_token.is_ident() + && expected.iter().any(|tok| matches!(tok, TokenType::Token(TokenKind::Eq))) + { + // Likely typo: `=` → `==` in let expr or enum item + return Err(self.dcx().create_err(UseEqInstead { span: self.token.span })); + } + + if self.token.is_keyword(kw::Move) && self.prev_token.is_keyword(kw::Async) { + // The 2015 edition is in use because parsing of `async move` has failed. + let span = self.prev_token.span.to(self.token.span); + return Err(self.dcx().create_err(AsyncMoveBlockIn2015 { span })); + } + + let expect = tokens_to_string(&expected); + let actual = super::token_descr(&self.token); + let (msg_exp, (label_sp, label_exp)) = if expected.len() > 1 { + let fmt = format!("expected one of {expect}, found {actual}"); + let short_expect = if expected.len() > 6 { + format!("{} possible tokens", expected.len()) + } else { + expect + }; + (fmt, (self.prev_token.span.shrink_to_hi(), format!("expected one of {short_expect}"))) + } else if expected.is_empty() { + ( + format!("unexpected token: {actual}"), + (self.prev_token.span, "unexpected token after this".to_string()), + ) + } else { + ( + format!("expected {expect}, found {actual}"), + (self.prev_token.span.shrink_to_hi(), format!("expected {expect}")), + ) + }; + self.last_unexpected_token_span = Some(self.token.span); + // FIXME: translation requires list formatting (for `expect`) + let mut err = self.dcx().struct_span_err(self.token.span, msg_exp); + + // Look for usages of '=>' where '>=' was probably intended + if self.token == token::FatArrow + && expected + .iter() + .any(|tok| matches!(tok, TokenType::Operator | TokenType::Token(TokenKind::Le))) + && !expected.iter().any(|tok| { + matches!( + tok, + TokenType::Token(TokenKind::FatArrow) | TokenType::Token(TokenKind::Comma) + ) + }) + { + err.span_suggestion( + self.token.span, + "you might have meant to write a \"greater than or equal to\" comparison", + ">=", + Applicability::MaybeIncorrect, + ); + } + + if let TokenKind::Ident(symbol, _) = &self.prev_token.kind { + if ["def", "fun", "func", "function"].contains(&symbol.as_str()) { + err.span_suggestion_short( + self.prev_token.span, + format!("write `fn` instead of `{symbol}` to declare a function"), + "fn", + Applicability::MachineApplicable, + ); + } + } + + if let TokenKind::Ident(prev, _) = &self.prev_token.kind + && let TokenKind::Ident(cur, _) = &self.token.kind + { + let concat = Symbol::intern(&format!("{prev}{cur}")); + let ident = Ident::new(concat, DUMMY_SP); + if ident.is_used_keyword() || ident.is_reserved() || ident.is_raw_guess() { + let span = self.prev_token.span.to(self.token.span); + err.span_suggestion_verbose( + span, + format!("consider removing the space to spell keyword `{concat}`"), + concat, + Applicability::MachineApplicable, + ); + } + } + + // Try to detect an intended c-string literal while using a pre-2021 edition. The heuristic + // here is to identify a cooked, uninterpolated `c` id immediately followed by a string, or + // a cooked, uninterpolated `cr` id immediately followed by a string or a `#`, in an edition + // where c-string literals are not allowed. There is the very slight possibility of a false + // positive for a `cr#` that wasn't intended to start a c-string literal, but identifying + // that in the parser requires unbounded lookahead, so we only add a hint to the existing + // error rather than replacing it entirely. + if ((self.prev_token.kind == TokenKind::Ident(sym::c, IdentIsRaw::No) + && matches!(&self.token.kind, TokenKind::Literal(token::Lit { kind: token::Str, .. }))) + || (self.prev_token.kind == TokenKind::Ident(sym::cr, IdentIsRaw::No) + && matches!( + &self.token.kind, + TokenKind::Literal(token::Lit { kind: token::Str, .. }) | token::Pound + ))) + && self.prev_token.span.hi() == self.token.span.lo() + && !self.token.span.at_least_rust_2021() + { + err.note("you may be trying to write a c-string literal"); + err.note("c-string literals require Rust 2021 or later"); + err.subdiagnostic(self.dcx(), HelpUseLatestEdition::new()); + } + + // `pub` may be used for an item or `pub(crate)` + if self.prev_token.is_ident_named(sym::public) + && (self.token.can_begin_item() + || self.token.kind == TokenKind::OpenDelim(Delimiter::Parenthesis)) + { + err.span_suggestion_short( + self.prev_token.span, + "write `pub` instead of `public` to make the item public", + "pub", + Applicability::MachineApplicable, + ); + } + + if let token::DocComment(kind, style, _) = self.token.kind { + // We have something like `expr //!val` where the user likely meant `expr // !val` + let pos = self.token.span.lo() + BytePos(2); + let span = self.token.span.with_lo(pos).with_hi(pos); + err.span_suggestion_verbose( + span, + format!( + "add a space before {} to write a regular comment", + match (kind, style) { + (token::CommentKind::Line, ast::AttrStyle::Inner) => "`!`", + (token::CommentKind::Block, ast::AttrStyle::Inner) => "`!`", + (token::CommentKind::Line, ast::AttrStyle::Outer) => "the last `/`", + (token::CommentKind::Block, ast::AttrStyle::Outer) => "the last `*`", + }, + ), + " ".to_string(), + Applicability::MachineApplicable, + ); + } + + let sp = if self.token == token::Eof { + // This is EOF; don't want to point at the following char, but rather the last token. + self.prev_token.span + } else { + label_sp + }; + + if self.check_too_many_raw_str_terminators(&mut err) { + if expected.contains(&TokenType::Token(token::Semi)) && self.eat(&token::Semi) { + let guar = err.emit(); + return Ok(Recovered::Yes(guar)); + } else { + return Err(err); + } + } + + if self.prev_token.span == DUMMY_SP { + // Account for macro context where the previous span might not be + // available to avoid incorrect output (#54841). + err.span_label(self.token.span, label_exp); + } else if !sm.is_multiline(self.token.span.shrink_to_hi().until(sp.shrink_to_lo())) { + // When the spans are in the same line, it means that the only content between + // them is whitespace, point at the found token in that case: + // + // X | () => { syntax error }; + // | ^^^^^ expected one of 8 possible tokens here + // + // instead of having: + // + // X | () => { syntax error }; + // | -^^^^^ unexpected token + // | | + // | expected one of 8 possible tokens here + err.span_label(self.token.span, label_exp); + } else { + err.span_label(sp, label_exp); + err.span_label(self.token.span, "unexpected token"); + } + Err(err) + } + + /// The user has written `#[attr] expr` which is unsupported. (#106020) + pub(super) fn attr_on_non_tail_expr(&self, expr: &Expr) -> ErrorGuaranteed { + // Missing semicolon typo error. + let span = self.prev_token.span.shrink_to_hi(); + let mut err = self.dcx().create_err(ExpectedSemi { + span, + token: self.token.clone(), + unexpected_token_label: Some(self.token.span), + sugg: ExpectedSemiSugg::AddSemi(span), + }); + let attr_span = match &expr.attrs[..] { + [] => unreachable!(), + [only] => only.span, + [first, rest @ ..] => { + for attr in rest { + err.span_label(attr.span, ""); + } + first.span + } + }; + err.span_label( + attr_span, + format!( + "only `;` terminated statements or tail expressions are allowed after {}", + if expr.attrs.len() == 1 { "this attribute" } else { "these attributes" }, + ), + ); + if self.token == token::Pound + && self.look_ahead(1, |t| t.kind == token::OpenDelim(Delimiter::Bracket)) + { + // We have + // #[attr] + // expr + // #[not_attr] + // other_expr + err.span_label(span, "expected `;` here"); + err.multipart_suggestion( + "alternatively, consider surrounding the expression with a block", + vec![ + (expr.span.shrink_to_lo(), "{ ".to_string()), + (expr.span.shrink_to_hi(), " }".to_string()), + ], + Applicability::MachineApplicable, + ); + + // Special handling for `#[cfg(...)]` chains + let mut snapshot = self.create_snapshot_for_diagnostic(); + if let [attr] = &expr.attrs[..] + && let ast::AttrKind::Normal(attr_kind) = &attr.kind + && let [segment] = &attr_kind.item.path.segments[..] + && segment.ident.name == sym::cfg + && let Some(args_span) = attr_kind.item.args.span() + && let next_attr = match snapshot.parse_attribute(InnerAttrPolicy::Forbidden(None)) + { + Ok(next_attr) => next_attr, + Err(inner_err) => { + inner_err.cancel(); + return err.emit(); + } + } + && let ast::AttrKind::Normal(next_attr_kind) = next_attr.kind + && let Some(next_attr_args_span) = next_attr_kind.item.args.span() + && let [next_segment] = &next_attr_kind.item.path.segments[..] + && segment.ident.name == sym::cfg + { + let next_expr = match snapshot.parse_expr() { + Ok(next_expr) => next_expr, + Err(inner_err) => { + inner_err.cancel(); + return err.emit(); + } + }; + // We have for sure + // #[cfg(..)] + // expr + // #[cfg(..)] + // other_expr + // So we suggest using `if cfg!(..) { expr } else if cfg!(..) { other_expr }`. + let margin = self.psess.source_map().span_to_margin(next_expr.span).unwrap_or(0); + let sugg = vec![ + (attr.span.with_hi(segment.span().hi()), "if cfg!".to_string()), + (args_span.shrink_to_hi().with_hi(attr.span.hi()), " {".to_string()), + (expr.span.shrink_to_lo(), " ".to_string()), + ( + next_attr.span.with_hi(next_segment.span().hi()), + "} else if cfg!".to_string(), + ), + ( + next_attr_args_span.shrink_to_hi().with_hi(next_attr.span.hi()), + " {".to_string(), + ), + (next_expr.span.shrink_to_lo(), " ".to_string()), + (next_expr.span.shrink_to_hi(), format!("\n{}}}", " ".repeat(margin))), + ]; + err.multipart_suggestion( + "it seems like you are trying to provide different expressions depending on \ + `cfg`, consider using `if cfg!(..)`", + sugg, + Applicability::MachineApplicable, + ); + } + } + err.emit() + } + + fn check_too_many_raw_str_terminators(&mut self, err: &mut Diag<'_>) -> bool { + let sm = self.psess.source_map(); + match (&self.prev_token.kind, &self.token.kind) { + ( + TokenKind::Literal(Lit { + kind: LitKind::StrRaw(n_hashes) | LitKind::ByteStrRaw(n_hashes), + .. + }), + TokenKind::Pound, + ) if !sm.is_multiline( + self.prev_token.span.shrink_to_hi().until(self.token.span.shrink_to_lo()), + ) => + { + let n_hashes: u8 = *n_hashes; + err.primary_message("too many `#` when terminating raw string"); + let str_span = self.prev_token.span; + let mut span = self.token.span; + let mut count = 0; + while self.token.kind == TokenKind::Pound + && !sm.is_multiline(span.shrink_to_hi().until(self.token.span.shrink_to_lo())) + { + span = span.with_hi(self.token.span.hi()); + self.bump(); + count += 1; + } + err.span(span); + err.span_suggestion( + span, + format!("remove the extra `#`{}", pluralize!(count)), + "", + Applicability::MachineApplicable, + ); + err.span_label( + str_span, + format!("this raw string started with {n_hashes} `#`{}", pluralize!(n_hashes)), + ); + true + } + _ => false, + } + } + + pub fn maybe_suggest_struct_literal( + &mut self, + lo: Span, + s: BlockCheckMode, + maybe_struct_name: token::Token, + can_be_struct_literal: bool, + ) -> Option<PResult<'a, P<Block>>> { + if self.token.is_ident() && self.look_ahead(1, |t| t == &token::Colon) { + // We might be having a struct literal where people forgot to include the path: + // fn foo() -> Foo { + // field: value, + // } + debug!(?maybe_struct_name, ?self.token); + let mut snapshot = self.create_snapshot_for_diagnostic(); + let path = Path { + segments: ThinVec::new(), + span: self.prev_token.span.shrink_to_lo(), + tokens: None, + }; + let struct_expr = snapshot.parse_expr_struct(None, path, false); + let block_tail = self.parse_block_tail(lo, s, AttemptLocalParseRecovery::No); + return Some(match (struct_expr, block_tail) { + (Ok(expr), Err(err)) => { + // We have encountered the following: + // fn foo() -> Foo { + // field: value, + // } + // Suggest: + // fn foo() -> Foo { Path { + // field: value, + // } } + let guar = err.delay_as_bug(); + self.restore_snapshot(snapshot); + let mut tail = self.mk_block( + thin_vec![self.mk_stmt_err(expr.span, guar)], + s, + lo.to(self.prev_token.span), + ); + tail.could_be_bare_literal = true; + if maybe_struct_name.is_ident() && can_be_struct_literal { + // Account for `if Example { a: one(), }.is_pos() {}`. + // expand `before` so that we take care of module path such as: + // `foo::Bar { ... } ` + // we expect to suggest `(foo::Bar { ... })` instead of `foo::(Bar { ... })` + let sm = self.psess.source_map(); + let before = maybe_struct_name.span.shrink_to_lo(); + if let Ok(extend_before) = sm.span_extend_prev_while(before, |t| { + t.is_alphanumeric() || t == ':' || t == '_' + }) { + Err(self.dcx().create_err(StructLiteralNeedingParens { + span: maybe_struct_name.span.to(expr.span), + sugg: StructLiteralNeedingParensSugg { + before: extend_before.shrink_to_lo(), + after: expr.span.shrink_to_hi(), + }, + })) + } else { + return None; + } + } else { + self.dcx().emit_err(StructLiteralBodyWithoutPath { + span: expr.span, + sugg: StructLiteralBodyWithoutPathSugg { + before: expr.span.shrink_to_lo(), + after: expr.span.shrink_to_hi(), + }, + }); + Ok(tail) + } + } + (Err(err), Ok(tail)) => { + // We have a block tail that contains a somehow valid type ascription expr. + err.cancel(); + Ok(tail) + } + (Err(snapshot_err), Err(err)) => { + // We don't know what went wrong, emit the normal error. + snapshot_err.cancel(); + self.consume_block(Delimiter::Brace, ConsumeClosingDelim::Yes); + Err(err) + } + (Ok(_), Ok(mut tail)) => { + tail.could_be_bare_literal = true; + Ok(tail) + } + }); + } + None + } + + pub(super) fn recover_closure_body( + &mut self, + mut err: Diag<'a>, + before: token::Token, + prev: token::Token, + token: token::Token, + lo: Span, + decl_hi: Span, + ) -> PResult<'a, P<Expr>> { + err.span_label(lo.to(decl_hi), "while parsing the body of this closure"); + let guar = match before.kind { + token::OpenDelim(Delimiter::Brace) + if !matches!(token.kind, token::OpenDelim(Delimiter::Brace)) => + { + // `{ || () }` should have been `|| { () }` + err.multipart_suggestion( + "you might have meant to open the body of the closure, instead of enclosing \ + the closure in a block", + vec![ + (before.span, String::new()), + (prev.span.shrink_to_hi(), " {".to_string()), + ], + Applicability::MaybeIncorrect, + ); + let guar = err.emit(); + self.eat_to_tokens(&[&token::CloseDelim(Delimiter::Brace)]); + guar + } + token::OpenDelim(Delimiter::Parenthesis) + if !matches!(token.kind, token::OpenDelim(Delimiter::Brace)) => + { + // We are within a function call or tuple, we can emit the error + // and recover. + self.eat_to_tokens(&[&token::CloseDelim(Delimiter::Parenthesis), &token::Comma]); + + err.multipart_suggestion_verbose( + "you might have meant to open the body of the closure", + vec![ + (prev.span.shrink_to_hi(), " {".to_string()), + (self.token.span.shrink_to_lo(), "}".to_string()), + ], + Applicability::MaybeIncorrect, + ); + err.emit() + } + _ if !matches!(token.kind, token::OpenDelim(Delimiter::Brace)) => { + // We don't have a heuristic to correctly identify where the block + // should be closed. + err.multipart_suggestion_verbose( + "you might have meant to open the body of the closure", + vec![(prev.span.shrink_to_hi(), " {".to_string())], + Applicability::HasPlaceholders, + ); + return Err(err); + } + _ => return Err(err), + }; + Ok(self.mk_expr_err(lo.to(self.token.span), guar)) + } + + /// Eats and discards tokens until one of `kets` is encountered. Respects token trees, + /// passes through any errors encountered. Used for error recovery. + pub(super) fn eat_to_tokens(&mut self, kets: &[&TokenKind]) { + if let Err(err) = + self.parse_seq_to_before_tokens(kets, SeqSep::none(), TokenExpectType::Expect, |p| { + Ok(p.parse_token_tree()) + }) + { + err.cancel(); + } + } + + /// This function checks if there are trailing angle brackets and produces + /// a diagnostic to suggest removing them. + /// + /// ```ignore (diagnostic) + /// let _ = [1, 2, 3].into_iter().collect::<Vec<usize>>>>(); + /// ^^ help: remove extra angle brackets + /// ``` + /// + /// If `true` is returned, then trailing brackets were recovered, tokens were consumed + /// up until one of the tokens in 'end' was encountered, and an error was emitted. + pub(super) fn check_trailing_angle_brackets( + &mut self, + segment: &PathSegment, + end: &[&TokenKind], + ) -> Option<ErrorGuaranteed> { + if !self.may_recover() { + return None; + } + + // This function is intended to be invoked after parsing a path segment where there are two + // cases: + // + // 1. A specific token is expected after the path segment. + // eg. `x.foo(`, `x.foo::<u32>(` (parenthesis - method call), + // `Foo::`, or `Foo::<Bar>::` (mod sep - continued path). + // 2. No specific token is expected after the path segment. + // eg. `x.foo` (field access) + // + // This function is called after parsing `.foo` and before parsing the token `end` (if + // present). This includes any angle bracket arguments, such as `.foo::<u32>` or + // `Foo::<Bar>`. + + // We only care about trailing angle brackets if we previously parsed angle bracket + // arguments. This helps stop us incorrectly suggesting that extra angle brackets be + // removed in this case: + // + // `x.foo >> (3)` (where `x.foo` is a `u32` for example) + // + // This case is particularly tricky as we won't notice it just looking at the tokens - + // it will appear the same (in terms of upcoming tokens) as below (since the `::<u32>` will + // have already been parsed): + // + // `x.foo::<u32>>>(3)` + let parsed_angle_bracket_args = + segment.args.as_ref().is_some_and(|args| args.is_angle_bracketed()); + + debug!( + "check_trailing_angle_brackets: parsed_angle_bracket_args={:?}", + parsed_angle_bracket_args, + ); + if !parsed_angle_bracket_args { + return None; + } + + // Keep the span at the start so we can highlight the sequence of `>` characters to be + // removed. + let lo = self.token.span; + + // We need to look-ahead to see if we have `>` characters without moving the cursor forward + // (since we might have the field access case and the characters we're eating are + // actual operators and not trailing characters - ie `x.foo >> 3`). + let mut position = 0; + + // We can encounter `>` or `>>` tokens in any order, so we need to keep track of how + // many of each (so we can correctly pluralize our error messages) and continue to + // advance. + let mut number_of_shr = 0; + let mut number_of_gt = 0; + while self.look_ahead(position, |t| { + trace!("check_trailing_angle_brackets: t={:?}", t); + if *t == token::BinOp(token::BinOpToken::Shr) { + number_of_shr += 1; + true + } else if *t == token::Gt { + number_of_gt += 1; + true + } else { + false + } + }) { + position += 1; + } + + // If we didn't find any trailing `>` characters, then we have nothing to error about. + debug!( + "check_trailing_angle_brackets: number_of_gt={:?} number_of_shr={:?}", + number_of_gt, number_of_shr, + ); + if number_of_gt < 1 && number_of_shr < 1 { + return None; + } + + // Finally, double check that we have our end token as otherwise this is the + // second case. + if self.look_ahead(position, |t| { + trace!("check_trailing_angle_brackets: t={:?}", t); + end.contains(&&t.kind) + }) { + // Eat from where we started until the end token so that parsing can continue + // as if we didn't have those extra angle brackets. + self.eat_to_tokens(end); + let span = lo.until(self.token.span); + + let num_extra_brackets = number_of_gt + number_of_shr * 2; + return Some(self.dcx().emit_err(UnmatchedAngleBrackets { span, num_extra_brackets })); + } + None + } + + /// Check if a method call with an intended turbofish has been written without surrounding + /// angle brackets. + pub(super) fn check_turbofish_missing_angle_brackets(&mut self, segment: &mut PathSegment) { + if !self.may_recover() { + return; + } + + if token::PathSep == self.token.kind && segment.args.is_none() { + let snapshot = self.create_snapshot_for_diagnostic(); + self.bump(); + let lo = self.token.span; + match self.parse_angle_args(None) { + Ok(args) => { + let span = lo.to(self.prev_token.span); + // Detect trailing `>` like in `x.collect::Vec<_>>()`. + let mut trailing_span = self.prev_token.span.shrink_to_hi(); + while self.token.kind == token::BinOp(token::Shr) + || self.token.kind == token::Gt + { + trailing_span = trailing_span.to(self.token.span); + self.bump(); + } + if self.token.kind == token::OpenDelim(Delimiter::Parenthesis) { + // Recover from bad turbofish: `foo.collect::Vec<_>()`. + segment.args = Some(AngleBracketedArgs { args, span }.into()); + + self.dcx().emit_err(GenericParamsWithoutAngleBrackets { + span, + sugg: GenericParamsWithoutAngleBracketsSugg { + left: span.shrink_to_lo(), + right: trailing_span, + }, + }); + } else { + // This doesn't look like an invalid turbofish, can't recover parse state. + self.restore_snapshot(snapshot); + } + } + Err(err) => { + // We couldn't parse generic parameters, unlikely to be a turbofish. Rely on + // generic parse error instead. + err.cancel(); + self.restore_snapshot(snapshot); + } + } + } + } + + /// When writing a turbofish with multiple type parameters missing the leading `::`, we will + /// encounter a parse error when encountering the first `,`. + pub(super) fn check_mistyped_turbofish_with_multiple_type_params( + &mut self, + mut e: Diag<'a>, + expr: &mut P<Expr>, + ) -> PResult<'a, ErrorGuaranteed> { + if let ExprKind::Binary(binop, _, _) = &expr.kind + && let ast::BinOpKind::Lt = binop.node + && self.eat(&token::Comma) + { + let x = self.parse_seq_to_before_end( + &token::Gt, + SeqSep::trailing_allowed(token::Comma), + |p| match p.parse_generic_arg(None)? { + Some(arg) => Ok(arg), + // If we didn't eat a generic arg, then we should error. + None => p.unexpected_any(), + }, + ); + match x { + Ok((_, _, Recovered::No)) => { + if self.eat(&token::Gt) { + // We made sense of it. Improve the error message. + e.span_suggestion_verbose( + binop.span.shrink_to_lo(), + fluent::parse_sugg_turbofish_syntax, + "::", + Applicability::MaybeIncorrect, + ); + match self.parse_expr() { + Ok(_) => { + // The subsequent expression is valid. Mark + // `expr` as erroneous and emit `e` now, but + // return `Ok` so parsing can continue. + let guar = e.emit(); + *expr = self.mk_expr_err(expr.span.to(self.prev_token.span), guar); + return Ok(guar); + } + Err(err) => { + err.cancel(); + } + } + } + } + Ok((_, _, Recovered::Yes(_))) => {} + Err(err) => { + err.cancel(); + } + } + } + Err(e) + } + + /// Suggest add the missing `let` before the identifier in stmt + /// `a: Ty = 1` -> `let a: Ty = 1` + pub(super) fn suggest_add_missing_let_for_stmt(&mut self, err: &mut Diag<'a>) { + if self.token == token::Colon { + let prev_span = self.prev_token.span.shrink_to_lo(); + let snapshot = self.create_snapshot_for_diagnostic(); + self.bump(); + match self.parse_ty() { + Ok(_) => { + if self.token == token::Eq { + let sugg = SuggAddMissingLetStmt { span: prev_span }; + sugg.add_to_diag(err); + } + } + Err(e) => { + e.cancel(); + } + } + self.restore_snapshot(snapshot); + } + } + + /// Check to see if a pair of chained operators looks like an attempt at chained comparison, + /// e.g. `1 < x <= 3`. If so, suggest either splitting the comparison into two, or + /// parenthesising the leftmost comparison. The return value indicates if recovery happened. + fn attempt_chained_comparison_suggestion( + &mut self, + err: &mut ComparisonOperatorsCannotBeChained, + inner_op: &Expr, + outer_op: &Spanned<AssocOp>, + ) -> bool { + if let ExprKind::Binary(op, l1, r1) = &inner_op.kind { + if let ExprKind::Field(_, ident) = l1.kind + && ident.as_str().parse::<i32>().is_err() + && !matches!(r1.kind, ExprKind::Lit(_)) + { + // The parser has encountered `foo.bar<baz`, the likelihood of the turbofish + // suggestion being the only one to apply is high. + return false; + } + return match (op.node, &outer_op.node) { + // `x == y == z` + (BinOpKind::Eq, AssocOp::Equal) | + // `x < y < z` and friends. + (BinOpKind::Lt, AssocOp::Less | AssocOp::LessEqual) | + (BinOpKind::Le, AssocOp::LessEqual | AssocOp::Less) | + // `x > y > z` and friends. + (BinOpKind::Gt, AssocOp::Greater | AssocOp::GreaterEqual) | + (BinOpKind::Ge, AssocOp::GreaterEqual | AssocOp::Greater) => { + let expr_to_str = |e: &Expr| { + self.span_to_snippet(e.span) + .unwrap_or_else(|_| pprust::expr_to_string(e)) + }; + err.chaining_sugg = Some(ComparisonOperatorsCannotBeChainedSugg::SplitComparison { + span: inner_op.span.shrink_to_hi(), + middle_term: expr_to_str(r1), + }); + false // Keep the current parse behavior, where the AST is `(x < y) < z`. + } + // `x == y < z` + (BinOpKind::Eq, AssocOp::Less | AssocOp::LessEqual | AssocOp::Greater | AssocOp::GreaterEqual) => { + // Consume `z`/outer-op-rhs. + let snapshot = self.create_snapshot_for_diagnostic(); + match self.parse_expr() { + Ok(r2) => { + // We are sure that outer-op-rhs could be consumed, the suggestion is + // likely correct. + err.chaining_sugg = Some(ComparisonOperatorsCannotBeChainedSugg::Parenthesize { + left: r1.span.shrink_to_lo(), + right: r2.span.shrink_to_hi(), + }); + true + } + Err(expr_err) => { + expr_err.cancel(); + self.restore_snapshot(snapshot); + true + } + } + } + // `x > y == z` + (BinOpKind::Lt | BinOpKind::Le | BinOpKind::Gt | BinOpKind::Ge, AssocOp::Equal) => { + let snapshot = self.create_snapshot_for_diagnostic(); + // At this point it is always valid to enclose the lhs in parentheses, no + // further checks are necessary. + match self.parse_expr() { + Ok(_) => { + err.chaining_sugg = Some(ComparisonOperatorsCannotBeChainedSugg::Parenthesize { + left: l1.span.shrink_to_lo(), + right: r1.span.shrink_to_hi(), + }); + true + } + Err(expr_err) => { + expr_err.cancel(); + self.restore_snapshot(snapshot); + false + } + } + } + _ => false + }; + } + false + } + + /// Produces an error if comparison operators are chained (RFC #558). + /// We only need to check the LHS, not the RHS, because all comparison ops have same + /// precedence (see `fn precedence`) and are left-associative (see `fn fixity`). + /// + /// This can also be hit if someone incorrectly writes `foo<bar>()` when they should have used + /// the turbofish (`foo::<bar>()`) syntax. We attempt some heuristic recovery if that is the + /// case. + /// + /// Keep in mind that given that `outer_op.is_comparison()` holds and comparison ops are left + /// associative we can infer that we have: + /// + /// ```text + /// outer_op + /// / \ + /// inner_op r2 + /// / \ + /// l1 r1 + /// ``` + pub(super) fn check_no_chained_comparison( + &mut self, + inner_op: &Expr, + outer_op: &Spanned<AssocOp>, + ) -> PResult<'a, Option<P<Expr>>> { + debug_assert!( + outer_op.node.is_comparison(), + "check_no_chained_comparison: {:?} is not comparison", + outer_op.node, + ); + + let mk_err_expr = + |this: &Self, span, guar| Ok(Some(this.mk_expr(span, ExprKind::Err(guar)))); + + match &inner_op.kind { + ExprKind::Binary(op, l1, r1) if op.node.is_comparison() => { + let mut err = ComparisonOperatorsCannotBeChained { + span: vec![op.span, self.prev_token.span], + suggest_turbofish: None, + help_turbofish: None, + chaining_sugg: None, + }; + + // Include `<` to provide this recommendation even in a case like + // `Foo<Bar<Baz<Qux, ()>>>` + if op.node == BinOpKind::Lt && outer_op.node == AssocOp::Less + || outer_op.node == AssocOp::Greater + { + if outer_op.node == AssocOp::Less { + let snapshot = self.create_snapshot_for_diagnostic(); + self.bump(); + // So far we have parsed `foo<bar<`, consume the rest of the type args. + let modifiers = + [(token::Lt, 1), (token::Gt, -1), (token::BinOp(token::Shr), -2)]; + self.consume_tts(1, &modifiers); + + if !&[token::OpenDelim(Delimiter::Parenthesis), token::PathSep] + .contains(&self.token.kind) + { + // We don't have `foo< bar >(` or `foo< bar >::`, so we rewind the + // parser and bail out. + self.restore_snapshot(snapshot); + } + } + return if token::PathSep == self.token.kind { + // We have some certainty that this was a bad turbofish at this point. + // `foo< bar >::` + if let ExprKind::Binary(o, ..) = inner_op.kind + && o.node == BinOpKind::Lt + { + err.suggest_turbofish = Some(op.span.shrink_to_lo()); + } else { + err.help_turbofish = Some(()); + } + + let snapshot = self.create_snapshot_for_diagnostic(); + self.bump(); // `::` + + // Consume the rest of the likely `foo<bar>::new()` or return at `foo<bar>`. + match self.parse_expr() { + Ok(_) => { + // 99% certain that the suggestion is correct, continue parsing. + let guar = self.dcx().emit_err(err); + // FIXME: actually check that the two expressions in the binop are + // paths and resynthesize new fn call expression instead of using + // `ExprKind::Err` placeholder. + mk_err_expr(self, inner_op.span.to(self.prev_token.span), guar) + } + Err(expr_err) => { + expr_err.cancel(); + // Not entirely sure now, but we bubble the error up with the + // suggestion. + self.restore_snapshot(snapshot); + Err(self.dcx().create_err(err)) + } + } + } else if token::OpenDelim(Delimiter::Parenthesis) == self.token.kind { + // We have high certainty that this was a bad turbofish at this point. + // `foo< bar >(` + if let ExprKind::Binary(o, ..) = inner_op.kind + && o.node == BinOpKind::Lt + { + err.suggest_turbofish = Some(op.span.shrink_to_lo()); + } else { + err.help_turbofish = Some(()); + } + // Consume the fn call arguments. + match self.consume_fn_args() { + Err(()) => Err(self.dcx().create_err(err)), + Ok(()) => { + let guar = self.dcx().emit_err(err); + // FIXME: actually check that the two expressions in the binop are + // paths and resynthesize new fn call expression instead of using + // `ExprKind::Err` placeholder. + mk_err_expr(self, inner_op.span.to(self.prev_token.span), guar) + } + } + } else { + if !matches!(l1.kind, ExprKind::Lit(_)) + && !matches!(r1.kind, ExprKind::Lit(_)) + { + // All we know is that this is `foo < bar >` and *nothing* else. Try to + // be helpful, but don't attempt to recover. + err.help_turbofish = Some(()); + } + + // If it looks like a genuine attempt to chain operators (as opposed to a + // misformatted turbofish, for instance), suggest a correct form. + let recovered = self + .attempt_chained_comparison_suggestion(&mut err, inner_op, outer_op); + if recovered { + let guar = self.dcx().emit_err(err); + mk_err_expr(self, inner_op.span.to(self.prev_token.span), guar) + } else { + // These cases cause too many knock-down errors, bail out (#61329). + Err(self.dcx().create_err(err)) + } + }; + } + let recovered = + self.attempt_chained_comparison_suggestion(&mut err, inner_op, outer_op); + let guar = self.dcx().emit_err(err); + if recovered { + return mk_err_expr(self, inner_op.span.to(self.prev_token.span), guar); + } + } + _ => {} + } + Ok(None) + } + + fn consume_fn_args(&mut self) -> Result<(), ()> { + let snapshot = self.create_snapshot_for_diagnostic(); + self.bump(); // `(` + + // Consume the fn call arguments. + let modifiers = [ + (token::OpenDelim(Delimiter::Parenthesis), 1), + (token::CloseDelim(Delimiter::Parenthesis), -1), + ]; + self.consume_tts(1, &modifiers); + + if self.token.kind == token::Eof { + // Not entirely sure that what we consumed were fn arguments, rollback. + self.restore_snapshot(snapshot); + Err(()) + } else { + // 99% certain that the suggestion is correct, continue parsing. + Ok(()) + } + } + + pub(super) fn maybe_report_ambiguous_plus(&mut self, impl_dyn_multi: bool, ty: &Ty) { + if impl_dyn_multi { + self.dcx().emit_err(AmbiguousPlus { sum_ty: pprust::ty_to_string(ty), span: ty.span }); + } + } + + /// Swift lets users write `Ty?` to mean `Option<Ty>`. Parse the construct and recover from it. + pub(super) fn maybe_recover_from_question_mark(&mut self, ty: P<Ty>) -> P<Ty> { + if self.token == token::Question { + self.bump(); + let guar = self.dcx().emit_err(QuestionMarkInType { + span: self.prev_token.span, + sugg: QuestionMarkInTypeSugg { + left: ty.span.shrink_to_lo(), + right: self.prev_token.span, + }, + }); + self.mk_ty(ty.span.to(self.prev_token.span), TyKind::Err(guar)) + } else { + ty + } + } + + /// Rust has no ternary operator (`cond ? then : else`). Parse it and try + /// to recover from it if `then` and `else` are valid expressions. Returns + /// an err if this appears to be a ternary expression. + pub(super) fn maybe_recover_from_ternary_operator(&mut self) -> PResult<'a, ()> { + if self.prev_token != token::Question { + return PResult::Ok(()); + } + + let lo = self.prev_token.span.lo(); + let snapshot = self.create_snapshot_for_diagnostic(); + + if match self.parse_expr() { + Ok(_) => true, + Err(err) => { + err.cancel(); + // The colon can sometimes be mistaken for type + // ascription. Catch when this happens and continue. + self.token == token::Colon + } + } { + if self.eat_noexpect(&token::Colon) { + match self.parse_expr() { + Ok(_) => { + return Err(self + .dcx() + .create_err(TernaryOperator { span: self.token.span.with_lo(lo) })); + } + Err(err) => { + err.cancel(); + } + }; + } + } + self.restore_snapshot(snapshot); + Ok(()) + } + + pub(super) fn maybe_recover_from_bad_type_plus(&mut self, ty: &Ty) -> PResult<'a, ()> { + // Do not add `+` to expected tokens. + if !self.token.is_like_plus() { + return Ok(()); + } + + self.bump(); // `+` + let bounds = self.parse_generic_bounds()?; + let sum_span = ty.span.to(self.prev_token.span); + + let sub = match &ty.kind { + TyKind::Ref(lifetime, mut_ty) => { + let sum_with_parens = pprust::to_string(|s| { + s.s.word("&"); + s.print_opt_lifetime(lifetime); + s.print_mutability(mut_ty.mutbl, false); + s.popen(); + s.print_type(&mut_ty.ty); + if !bounds.is_empty() { + s.word(" + "); + s.print_type_bounds(&bounds); + } + s.pclose() + }); + + BadTypePlusSub::AddParen { sum_with_parens, span: sum_span } + } + TyKind::Ptr(..) | TyKind::BareFn(..) => BadTypePlusSub::ForgotParen { span: sum_span }, + _ => BadTypePlusSub::ExpectPath { span: sum_span }, + }; + + self.dcx().emit_err(BadTypePlus { ty: pprust::ty_to_string(ty), span: sum_span, sub }); + + Ok(()) + } + + pub(super) fn recover_from_prefix_increment( + &mut self, + operand_expr: P<Expr>, + op_span: Span, + start_stmt: bool, + ) -> PResult<'a, P<Expr>> { + let standalone = if start_stmt { IsStandalone::Standalone } else { IsStandalone::Subexpr }; + let kind = IncDecRecovery { standalone, op: IncOrDec::Inc, fixity: UnaryFixity::Pre }; + self.recover_from_inc_dec(operand_expr, kind, op_span) + } + + pub(super) fn recover_from_postfix_increment( + &mut self, + operand_expr: P<Expr>, + op_span: Span, + start_stmt: bool, + ) -> PResult<'a, P<Expr>> { + let kind = IncDecRecovery { + standalone: if start_stmt { IsStandalone::Standalone } else { IsStandalone::Subexpr }, + op: IncOrDec::Inc, + fixity: UnaryFixity::Post, + }; + self.recover_from_inc_dec(operand_expr, kind, op_span) + } + + pub(super) fn recover_from_postfix_decrement( + &mut self, + operand_expr: P<Expr>, + op_span: Span, + start_stmt: bool, + ) -> PResult<'a, P<Expr>> { + let kind = IncDecRecovery { + standalone: if start_stmt { IsStandalone::Standalone } else { IsStandalone::Subexpr }, + op: IncOrDec::Dec, + fixity: UnaryFixity::Post, + }; + self.recover_from_inc_dec(operand_expr, kind, op_span) + } + + fn recover_from_inc_dec( + &mut self, + base: P<Expr>, + kind: IncDecRecovery, + op_span: Span, + ) -> PResult<'a, P<Expr>> { + let mut err = self.dcx().struct_span_err( + op_span, + format!("Rust has no {} {} operator", kind.fixity, kind.op.name()), + ); + err.span_label(op_span, format!("not a valid {} operator", kind.fixity)); + + let help_base_case = |mut err: Diag<'_, _>, base| { + err.help(format!("use `{}= 1` instead", kind.op.chr())); + err.emit(); + Ok(base) + }; + + // (pre, post) + let spans = match kind.fixity { + UnaryFixity::Pre => (op_span, base.span.shrink_to_hi()), + UnaryFixity::Post => (base.span.shrink_to_lo(), op_span), + }; + + match kind.standalone { + IsStandalone::Standalone => { + self.inc_dec_standalone_suggest(kind, spans).emit_verbose(&mut err) + } + IsStandalone::Subexpr => { + let Ok(base_src) = self.span_to_snippet(base.span) else { + return help_base_case(err, base); + }; + match kind.fixity { + UnaryFixity::Pre => { + self.prefix_inc_dec_suggest(base_src, kind, spans).emit(&mut err) + } + UnaryFixity::Post => { + // won't suggest since we can not handle the precedences + // for example: `a + b++` has been parsed (a + b)++ and we can not suggest here + if !matches!(base.kind, ExprKind::Binary(_, _, _)) { + self.postfix_inc_dec_suggest(base_src, kind, spans).emit(&mut err) + } + } + } + } + } + Err(err) + } + + fn prefix_inc_dec_suggest( + &mut self, + base_src: String, + kind: IncDecRecovery, + (pre_span, post_span): (Span, Span), + ) -> MultiSugg { + MultiSugg { + msg: format!("use `{}= 1` instead", kind.op.chr()), + patches: vec![ + (pre_span, "{ ".to_string()), + (post_span, format!(" {}= 1; {} }}", kind.op.chr(), base_src)), + ], + applicability: Applicability::MachineApplicable, + } + } + + fn postfix_inc_dec_suggest( + &mut self, + base_src: String, + kind: IncDecRecovery, + (pre_span, post_span): (Span, Span), + ) -> MultiSugg { + let tmp_var = if base_src.trim() == "tmp" { "tmp_" } else { "tmp" }; + MultiSugg { + msg: format!("use `{}= 1` instead", kind.op.chr()), + patches: vec![ + (pre_span, format!("{{ let {tmp_var} = ")), + (post_span, format!("; {} {}= 1; {} }}", base_src, kind.op.chr(), tmp_var)), + ], + applicability: Applicability::HasPlaceholders, + } + } + + fn inc_dec_standalone_suggest( + &mut self, + kind: IncDecRecovery, + (pre_span, post_span): (Span, Span), + ) -> MultiSugg { + let mut patches = Vec::new(); + + if !pre_span.is_empty() { + patches.push((pre_span, String::new())); + } + + patches.push((post_span, format!(" {}= 1", kind.op.chr()))); + MultiSugg { + msg: format!("use `{}= 1` instead", kind.op.chr()), + patches, + applicability: Applicability::MachineApplicable, + } + } + + /// Tries to recover from associated item paths like `[T]::AssocItem` / `(T, U)::AssocItem`. + /// Attempts to convert the base expression/pattern/type into a type, parses the `::AssocItem` + /// tail, and combines them into a `<Ty>::AssocItem` expression/pattern/type. + pub(super) fn maybe_recover_from_bad_qpath<T: RecoverQPath>( + &mut self, + base: P<T>, + ) -> PResult<'a, P<T>> { + if !self.may_recover() { + return Ok(base); + } + + // Do not add `::` to expected tokens. + if self.token == token::PathSep { + if let Some(ty) = base.to_ty() { + return self.maybe_recover_from_bad_qpath_stage_2(ty.span, ty); + } + } + Ok(base) + } + + /// Given an already parsed `Ty`, parses the `::AssocItem` tail and + /// combines them into a `<Ty>::AssocItem` expression/pattern/type. + pub(super) fn maybe_recover_from_bad_qpath_stage_2<T: RecoverQPath>( + &mut self, + ty_span: Span, + ty: P<Ty>, + ) -> PResult<'a, P<T>> { + self.expect(&token::PathSep)?; + + let mut path = ast::Path { segments: ThinVec::new(), span: DUMMY_SP, tokens: None }; + self.parse_path_segments(&mut path.segments, T::PATH_STYLE, None)?; + path.span = ty_span.to(self.prev_token.span); + + self.dcx().emit_err(BadQPathStage2 { + span: ty_span, + wrap: WrapType { lo: ty_span.shrink_to_lo(), hi: ty_span.shrink_to_hi() }, + }); + + let path_span = ty_span.shrink_to_hi(); // Use an empty path since `position == 0`. + Ok(P(T::recovered(Some(P(QSelf { ty, path_span, position: 0 })), path))) + } + + /// This function gets called in places where a semicolon is NOT expected and if there's a + /// semicolon it emits the appropriate error and returns true. + pub fn maybe_consume_incorrect_semicolon(&mut self, previous_item: Option<&Item>) -> bool { + if self.token.kind != TokenKind::Semi { + return false; + } + + // Check previous item to add it to the diagnostic, for example to say + // `enum declarations are not followed by a semicolon` + let err = match previous_item { + Some(previous_item) => { + let name = match previous_item.kind { + // Say "braced struct" because tuple-structs and + // braceless-empty-struct declarations do take a semicolon. + ItemKind::Struct(..) => "braced struct", + _ => previous_item.kind.descr(), + }; + IncorrectSemicolon { span: self.token.span, name, show_help: true } + } + None => IncorrectSemicolon { span: self.token.span, name: "", show_help: false }, + }; + self.dcx().emit_err(err); + + self.bump(); + true + } + + /// Creates a `Diag` for an unexpected token `t` and tries to recover if it is a + /// closing delimiter. + pub(super) fn unexpected_try_recover(&mut self, t: &TokenKind) -> PResult<'a, Recovered> { + let token_str = pprust::token_kind_to_string(t); + let this_token_str = super::token_descr(&self.token); + let (prev_sp, sp) = match (&self.token.kind, self.subparser_name) { + // Point at the end of the macro call when reaching end of macro arguments. + (token::Eof, Some(_)) => { + let sp = self.prev_token.span.shrink_to_hi(); + (sp, sp) + } + // We don't want to point at the following span after DUMMY_SP. + // This happens when the parser finds an empty TokenStream. + _ if self.prev_token.span == DUMMY_SP => (self.token.span, self.token.span), + // EOF, don't want to point at the following char, but rather the last token. + (token::Eof, None) => (self.prev_token.span, self.token.span), + _ => (self.prev_token.span.shrink_to_hi(), self.token.span), + }; + let msg = format!( + "expected `{}`, found {}", + token_str, + match (&self.token.kind, self.subparser_name) { + (token::Eof, Some(origin)) => format!("end of {origin}"), + _ => this_token_str, + }, + ); + let mut err = self.dcx().struct_span_err(sp, msg); + let label_exp = format!("expected `{token_str}`"); + let sm = self.psess.source_map(); + if !sm.is_multiline(prev_sp.until(sp)) { + // When the spans are in the same line, it means that the only content + // between them is whitespace, point only at the found token. + err.span_label(sp, label_exp); + } else { + err.span_label(prev_sp, label_exp); + err.span_label(sp, "unexpected token"); + } + Err(err) + } + + pub(super) fn expect_semi(&mut self) -> PResult<'a, ()> { + if self.eat(&token::Semi) || self.recover_colon_as_semi() { + return Ok(()); + } + self.expect(&token::Semi).map(drop) // Error unconditionally + } + + pub(super) fn recover_colon_as_semi(&mut self) -> bool { + let line_idx = |span: Span| { + self.psess + .source_map() + .span_to_lines(span) + .ok() + .and_then(|lines| Some(lines.lines.get(0)?.line_index)) + }; + + if self.may_recover() + && self.token == token::Colon + && self.look_ahead(1, |next| line_idx(self.token.span) < line_idx(next.span)) + { + self.dcx().emit_err(ColonAsSemi { + span: self.token.span, + type_ascription: self.psess.unstable_features.is_nightly_build().then_some(()), + }); + self.bump(); + return true; + } + + false + } + + /// Consumes alternative await syntaxes like `await!(<expr>)`, `await <expr>`, + /// `await? <expr>`, `await(<expr>)`, and `await { <expr> }`. + pub(super) fn recover_incorrect_await_syntax( + &mut self, + lo: Span, + await_sp: Span, + ) -> PResult<'a, P<Expr>> { + let (hi, expr, is_question) = if self.token == token::Not { + // Handle `await!(<expr>)`. + self.recover_await_macro()? + } else { + self.recover_await_prefix(await_sp)? + }; + let (sp, guar) = self.error_on_incorrect_await(lo, hi, &expr, is_question); + let expr = self.mk_expr_err(lo.to(sp), guar); + self.maybe_recover_from_bad_qpath(expr) + } + + fn recover_await_macro(&mut self) -> PResult<'a, (Span, P<Expr>, bool)> { + self.expect(&token::Not)?; + self.expect(&token::OpenDelim(Delimiter::Parenthesis))?; + let expr = self.parse_expr()?; + self.expect(&token::CloseDelim(Delimiter::Parenthesis))?; + Ok((self.prev_token.span, expr, false)) + } + + fn recover_await_prefix(&mut self, await_sp: Span) -> PResult<'a, (Span, P<Expr>, bool)> { + let is_question = self.eat(&token::Question); // Handle `await? <expr>`. + let expr = if self.token == token::OpenDelim(Delimiter::Brace) { + // Handle `await { <expr> }`. + // This needs to be handled separately from the next arm to avoid + // interpreting `await { <expr> }?` as `<expr>?.await`. + self.parse_expr_block(None, self.token.span, BlockCheckMode::Default) + } else { + self.parse_expr() + } + .map_err(|mut err| { + err.span_label(await_sp, "while parsing this incorrect await expression"); + err + })?; + Ok((expr.span, expr, is_question)) + } + + fn error_on_incorrect_await( + &self, + lo: Span, + hi: Span, + expr: &Expr, + is_question: bool, + ) -> (Span, ErrorGuaranteed) { + let span = lo.to(hi); + let applicability = match expr.kind { + ExprKind::Try(_) => Applicability::MaybeIncorrect, // `await <expr>?` + _ => Applicability::MachineApplicable, + }; + + let guar = self.dcx().emit_err(IncorrectAwait { + span, + sugg_span: (span, applicability), + expr: self.span_to_snippet(expr.span).unwrap_or_else(|_| pprust::expr_to_string(expr)), + question_mark: if is_question { "?" } else { "" }, + }); + + (span, guar) + } + + /// If encountering `future.await()`, consumes and emits an error. + pub(super) fn recover_from_await_method_call(&mut self) { + if self.token == token::OpenDelim(Delimiter::Parenthesis) + && self.look_ahead(1, |t| t == &token::CloseDelim(Delimiter::Parenthesis)) + { + // future.await() + let lo = self.token.span; + self.bump(); // ( + let span = lo.to(self.token.span); + self.bump(); // ) + + self.dcx().emit_err(IncorrectUseOfAwait { span }); + } + } + + pub(super) fn try_macro_suggestion(&mut self) -> PResult<'a, P<Expr>> { + let is_try = self.token.is_keyword(kw::Try); + let is_questionmark = self.look_ahead(1, |t| t == &token::Not); //check for ! + let is_open = self.look_ahead(2, |t| t == &token::OpenDelim(Delimiter::Parenthesis)); //check for ( + + if is_try && is_questionmark && is_open { + let lo = self.token.span; + self.bump(); //remove try + self.bump(); //remove ! + let try_span = lo.to(self.token.span); //we take the try!( span + self.bump(); //remove ( + let is_empty = self.token == token::CloseDelim(Delimiter::Parenthesis); //check if the block is empty + self.consume_block(Delimiter::Parenthesis, ConsumeClosingDelim::No); //eat the block + let hi = self.token.span; + self.bump(); //remove ) + let mut err = self.dcx().struct_span_err(lo.to(hi), "use of deprecated `try` macro"); + err.note("in the 2018 edition `try` is a reserved keyword, and the `try!()` macro is deprecated"); + let prefix = if is_empty { "" } else { "alternatively, " }; + if !is_empty { + err.multipart_suggestion( + "you can use the `?` operator instead", + vec![(try_span, "".to_owned()), (hi, "?".to_owned())], + Applicability::MachineApplicable, + ); + } + err.span_suggestion(lo.shrink_to_lo(), format!("{prefix}you can still access the deprecated `try!()` macro using the \"raw identifier\" syntax"), "r#", Applicability::MachineApplicable); + let guar = err.emit(); + Ok(self.mk_expr_err(lo.to(hi), guar)) + } else { + Err(self.expected_expression_found()) // The user isn't trying to invoke the try! macro + } + } + + /// When trying to close a generics list and encountering code like + /// ```text + /// impl<S: Into<std::borrow::Cow<'static, str>> From<S> for Canonical {} + /// // ^ missing > here + /// ``` + /// we provide a structured suggestion on the error from `expect_gt`. + pub(super) fn expect_gt_or_maybe_suggest_closing_generics( + &mut self, + params: &[ast::GenericParam], + ) -> PResult<'a, ()> { + let Err(mut err) = self.expect_gt() else { + return Ok(()); + }; + // Attempt to find places where a missing `>` might belong. + if let [.., ast::GenericParam { bounds, .. }] = params + && let Some(poly) = bounds + .iter() + .filter_map(|bound| match bound { + ast::GenericBound::Trait(poly, _) => Some(poly), + _ => None, + }) + .last() + { + err.span_suggestion_verbose( + poly.span.shrink_to_hi(), + "you might have meant to end the type parameters here", + ">", + Applicability::MaybeIncorrect, + ); + } + Err(err) + } + + pub(super) fn recover_seq_parse_error( + &mut self, + delim: Delimiter, + lo: Span, + err: PErr<'a>, + ) -> P<Expr> { + let guar = err.emit(); + // Recover from parse error, callers expect the closing delim to be consumed. + self.consume_block(delim, ConsumeClosingDelim::Yes); + self.mk_expr(lo.to(self.prev_token.span), ExprKind::Err(guar)) + } + + /// Eats tokens until we can be relatively sure we reached the end of the + /// statement. This is something of a best-effort heuristic. + /// + /// We terminate when we find an unmatched `}` (without consuming it). + pub(super) fn recover_stmt(&mut self) { + self.recover_stmt_(SemiColonMode::Ignore, BlockMode::Ignore) + } + + /// If `break_on_semi` is `Break`, then we will stop consuming tokens after + /// finding (and consuming) a `;` outside of `{}` or `[]` (note that this is + /// approximate -- it can mean we break too early due to macros, but that + /// should only lead to sub-optimal recovery, not inaccurate parsing). + /// + /// If `break_on_block` is `Break`, then we will stop consuming tokens + /// after finding (and consuming) a brace-delimited block. + pub(super) fn recover_stmt_( + &mut self, + break_on_semi: SemiColonMode, + break_on_block: BlockMode, + ) { + let mut brace_depth = 0; + let mut bracket_depth = 0; + let mut in_block = false; + debug!("recover_stmt_ enter loop (semi={:?}, block={:?})", break_on_semi, break_on_block); + loop { + debug!("recover_stmt_ loop {:?}", self.token); + match self.token.kind { + token::OpenDelim(Delimiter::Brace) => { + brace_depth += 1; + self.bump(); + if break_on_block == BlockMode::Break && brace_depth == 1 && bracket_depth == 0 + { + in_block = true; + } + } + token::OpenDelim(Delimiter::Bracket) => { + bracket_depth += 1; + self.bump(); + } + token::CloseDelim(Delimiter::Brace) => { + if brace_depth == 0 { + debug!("recover_stmt_ return - close delim {:?}", self.token); + break; + } + brace_depth -= 1; + self.bump(); + if in_block && bracket_depth == 0 && brace_depth == 0 { + debug!("recover_stmt_ return - block end {:?}", self.token); + break; + } + } + token::CloseDelim(Delimiter::Bracket) => { + bracket_depth -= 1; + if bracket_depth < 0 { + bracket_depth = 0; + } + self.bump(); + } + token::Eof => { + debug!("recover_stmt_ return - Eof"); + break; + } + token::Semi => { + self.bump(); + if break_on_semi == SemiColonMode::Break + && brace_depth == 0 + && bracket_depth == 0 + { + debug!("recover_stmt_ return - Semi"); + break; + } + } + token::Comma + if break_on_semi == SemiColonMode::Comma + && brace_depth == 0 + && bracket_depth == 0 => + { + break; + } + _ => self.bump(), + } + } + } + + pub(super) fn check_for_for_in_in_typo(&mut self, in_span: Span) { + if self.eat_keyword(kw::In) { + // a common typo: `for _ in in bar {}` + self.dcx().emit_err(InInTypo { + span: self.prev_token.span, + sugg_span: in_span.until(self.prev_token.span), + }); + } + } + + pub(super) fn eat_incorrect_doc_comment_for_param_type(&mut self) { + if let token::DocComment(..) = self.token.kind { + self.dcx().emit_err(DocCommentOnParamType { span: self.token.span }); + self.bump(); + } else if self.token == token::Pound + && self.look_ahead(1, |t| *t == token::OpenDelim(Delimiter::Bracket)) + { + let lo = self.token.span; + // Skip every token until next possible arg. + while self.token != token::CloseDelim(Delimiter::Bracket) { + self.bump(); + } + let sp = lo.to(self.token.span); + self.bump(); + self.dcx().emit_err(AttributeOnParamType { span: sp }); + } + } + + pub(super) fn parameter_without_type( + &mut self, + err: &mut Diag<'_>, + pat: P<ast::Pat>, + require_name: bool, + first_param: bool, + ) -> Option<Ident> { + // If we find a pattern followed by an identifier, it could be an (incorrect) + // C-style parameter declaration. + if self.check_ident() + && self.look_ahead(1, |t| { + *t == token::Comma || *t == token::CloseDelim(Delimiter::Parenthesis) + }) + { + // `fn foo(String s) {}` + let ident = self.parse_ident().unwrap(); + let span = pat.span.with_hi(ident.span.hi()); + + err.span_suggestion( + span, + "declare the type after the parameter binding", + "<identifier>: <type>", + Applicability::HasPlaceholders, + ); + return Some(ident); + } else if require_name + && (self.token == token::Comma + || self.token == token::Lt + || self.token == token::CloseDelim(Delimiter::Parenthesis)) + { + let rfc_note = "anonymous parameters are removed in the 2018 edition (see RFC 1685)"; + + let (ident, self_sugg, param_sugg, type_sugg, self_span, param_span, type_span) = + match pat.kind { + PatKind::Ident(_, ident, _) => ( + ident, + "self: ", + ": TypeName".to_string(), + "_: ", + pat.span.shrink_to_lo(), + pat.span.shrink_to_hi(), + pat.span.shrink_to_lo(), + ), + // Also catches `fn foo(&a)`. + PatKind::Ref(ref inner_pat, mutab) + if matches!(inner_pat.clone().into_inner().kind, PatKind::Ident(..)) => + { + match inner_pat.clone().into_inner().kind { + PatKind::Ident(_, ident, _) => { + let mutab = mutab.prefix_str(); + ( + ident, + "self: ", + format!("{ident}: &{mutab}TypeName"), + "_: ", + pat.span.shrink_to_lo(), + pat.span, + pat.span.shrink_to_lo(), + ) + } + _ => unreachable!(), + } + } + _ => { + // Otherwise, try to get a type and emit a suggestion. + if let Some(ty) = pat.to_ty() { + err.span_suggestion_verbose( + pat.span, + "explicitly ignore the parameter name", + format!("_: {}", pprust::ty_to_string(&ty)), + Applicability::MachineApplicable, + ); + err.note(rfc_note); + } + + return None; + } + }; + + // `fn foo(a, b) {}`, `fn foo(a<x>, b<y>) {}` or `fn foo(usize, usize) {}` + if first_param { + err.span_suggestion( + self_span, + "if this is a `self` type, give it a parameter name", + self_sugg, + Applicability::MaybeIncorrect, + ); + } + // Avoid suggesting that `fn foo(HashMap<u32>)` is fixed with a change to + // `fn foo(HashMap: TypeName<u32>)`. + if self.token != token::Lt { + err.span_suggestion( + param_span, + "if this is a parameter name, give it a type", + param_sugg, + Applicability::HasPlaceholders, + ); + } + err.span_suggestion( + type_span, + "if this is a type, explicitly ignore the parameter name", + type_sugg, + Applicability::MachineApplicable, + ); + err.note(rfc_note); + + // Don't attempt to recover by using the `X` in `X<Y>` as the parameter name. + return if self.token == token::Lt { None } else { Some(ident) }; + } + None + } + + pub(super) fn recover_arg_parse(&mut self) -> PResult<'a, (P<ast::Pat>, P<ast::Ty>)> { + let pat = self.parse_pat_no_top_alt(Some(Expected::ArgumentName), None)?; + self.expect(&token::Colon)?; + let ty = self.parse_ty()?; + + self.dcx().emit_err(PatternMethodParamWithoutBody { span: pat.span }); + + // Pretend the pattern is `_`, to avoid duplicate errors from AST validation. + let pat = + P(Pat { kind: PatKind::Wild, span: pat.span, id: ast::DUMMY_NODE_ID, tokens: None }); + Ok((pat, ty)) + } + + pub(super) fn recover_bad_self_param(&mut self, mut param: Param) -> PResult<'a, Param> { + let span = param.pat.span; + let guar = self.dcx().emit_err(SelfParamNotFirst { span }); + param.ty.kind = TyKind::Err(guar); + Ok(param) + } + + pub(super) fn consume_block(&mut self, delim: Delimiter, consume_close: ConsumeClosingDelim) { + let mut brace_depth = 0; + loop { + if self.eat(&token::OpenDelim(delim)) { + brace_depth += 1; + } else if self.check(&token::CloseDelim(delim)) { + if brace_depth == 0 { + if let ConsumeClosingDelim::Yes = consume_close { + // Some of the callers of this method expect to be able to parse the + // closing delimiter themselves, so we leave it alone. Otherwise we advance + // the parser. + self.bump(); + } + return; + } else { + self.bump(); + brace_depth -= 1; + continue; + } + } else if self.token == token::Eof { + return; + } else { + self.bump(); + } + } + } + + pub(super) fn expected_expression_found(&self) -> Diag<'a> { + let (span, msg) = match (&self.token.kind, self.subparser_name) { + (&token::Eof, Some(origin)) => { + let sp = self.prev_token.span.shrink_to_hi(); + (sp, format!("expected expression, found end of {origin}")) + } + _ => ( + self.token.span, + format!("expected expression, found {}", super::token_descr(&self.token)), + ), + }; + let mut err = self.dcx().struct_span_err(span, msg); + let sp = self.psess.source_map().start_point(self.token.span); + if let Some(sp) = self.psess.ambiguous_block_expr_parse.borrow().get(&sp) { + err.subdiagnostic(self.dcx(), ExprParenthesesNeeded::surrounding(*sp)); + } + err.span_label(span, "expected expression"); + + // Walk the chain of macro expansions for the current token to point at how the original + // code was interpreted. This helps the user realize when a macro argument of one type is + // later reinterpreted as a different type, like `$x:expr` being reinterpreted as `$x:pat` + // in a subsequent macro invocation (#71039). + let mut tok = self.token.clone(); + let mut labels = vec![]; + while let TokenKind::Interpolated(nt) = &tok.kind { + let tokens = nt.tokens(); + labels.push(nt.clone()); + if let Some(tokens) = tokens + && let tokens = tokens.to_attr_token_stream() + && let tokens = tokens.0.deref() + && let [AttrTokenTree::Token(token, _)] = &tokens[..] + { + tok = token.clone(); + } else { + break; + } + } + let mut iter = labels.into_iter().peekable(); + let mut show_link = false; + while let Some(nt) = iter.next() { + let descr = nt.descr(); + if let Some(next) = iter.peek() { + let next_descr = next.descr(); + if next_descr != descr { + err.span_label(next.use_span(), format!("this is expected to be {next_descr}")); + err.span_label( + nt.use_span(), + format!( + "this is interpreted as {}, but it is expected to be {}", + next_descr, descr, + ), + ); + show_link = true; + } + } + } + if show_link { + err.note( + "when forwarding a matched fragment to another macro-by-example, matchers in the \ + second macro will see an opaque AST of the fragment type, not the underlying \ + tokens", + ); + } + err + } + + fn consume_tts( + &mut self, + mut acc: i64, // `i64` because malformed code can have more closing delims than opening. + // Not using `FxHashMap` due to `token::TokenKind: !Eq + !Hash`. + modifier: &[(token::TokenKind, i64)], + ) { + while acc > 0 { + if let Some((_, val)) = modifier.iter().find(|(t, _)| *t == self.token.kind) { + acc += *val; + } + if self.token.kind == token::Eof { + break; + } + self.bump(); + } + } + + /// Replace duplicated recovered parameters with `_` pattern to avoid unnecessary errors. + /// + /// This is necessary because at this point we don't know whether we parsed a function with + /// anonymous parameters or a function with names but no types. In order to minimize + /// unnecessary errors, we assume the parameters are in the shape of `fn foo(a, b, c)` where + /// the parameters are *names* (so we don't emit errors about not being able to find `b` in + /// the local scope), but if we find the same name multiple times, like in `fn foo(i8, i8)`, + /// we deduplicate them to not complain about duplicated parameter names. + pub(super) fn deduplicate_recovered_params_names(&self, fn_inputs: &mut ThinVec<Param>) { + let mut seen_inputs = FxHashSet::default(); + for input in fn_inputs.iter_mut() { + let opt_ident = if let (PatKind::Ident(_, ident, _), TyKind::Err(_)) = + (&input.pat.kind, &input.ty.kind) + { + Some(*ident) + } else { + None + }; + if let Some(ident) = opt_ident { + if seen_inputs.contains(&ident) { + input.pat.kind = PatKind::Wild; + } + seen_inputs.insert(ident); + } + } + } + + /// Handle encountering a symbol in a generic argument list that is not a `,` or `>`. In this + /// case, we emit an error and try to suggest enclosing a const argument in braces if it looks + /// like the user has forgotten them. + pub fn handle_ambiguous_unbraced_const_arg( + &mut self, + args: &mut ThinVec<AngleBracketedArg>, + ) -> PResult<'a, bool> { + // If we haven't encountered a closing `>`, then the argument is malformed. + // It's likely that the user has written a const expression without enclosing it + // in braces, so we try to recover here. + let arg = args.pop().unwrap(); + // FIXME: for some reason using `unexpected` or `expected_one_of_not_found` has + // adverse side-effects to subsequent errors and seems to advance the parser. + // We are causing this error here exclusively in case that a `const` expression + // could be recovered from the current parser state, even if followed by more + // arguments after a comma. + let mut err = self.dcx().struct_span_err( + self.token.span, + format!("expected one of `,` or `>`, found {}", super::token_descr(&self.token)), + ); + err.span_label(self.token.span, "expected one of `,` or `>`"); + match self.recover_const_arg(arg.span(), err) { + Ok(arg) => { + args.push(AngleBracketedArg::Arg(arg)); + if self.eat(&token::Comma) { + return Ok(true); // Continue + } + } + Err(err) => { + args.push(arg); + // We will emit a more generic error later. + err.delay_as_bug(); + } + } + return Ok(false); // Don't continue. + } + + /// Attempt to parse a generic const argument that has not been enclosed in braces. + /// There are a limited number of expressions that are permitted without being encoded + /// in braces: + /// - Literals. + /// - Single-segment paths (i.e. standalone generic const parameters). + /// All other expressions that can be parsed will emit an error suggesting the expression be + /// wrapped in braces. + pub fn handle_unambiguous_unbraced_const_arg(&mut self) -> PResult<'a, P<Expr>> { + let start = self.token.span; + let expr = self.parse_expr_res(Restrictions::CONST_EXPR, None).map_err(|mut err| { + err.span_label( + start.shrink_to_lo(), + "while parsing a const generic argument starting here", + ); + err + })?; + if !self.expr_is_valid_const_arg(&expr) { + self.dcx().emit_err(ConstGenericWithoutBraces { + span: expr.span, + sugg: ConstGenericWithoutBracesSugg { + left: expr.span.shrink_to_lo(), + right: expr.span.shrink_to_hi(), + }, + }); + } + Ok(expr) + } + + fn recover_const_param_decl(&mut self, ty_generics: Option<&Generics>) -> Option<GenericArg> { + let snapshot = self.create_snapshot_for_diagnostic(); + let param = match self.parse_const_param(AttrVec::new()) { + Ok(param) => param, + Err(err) => { + err.cancel(); + self.restore_snapshot(snapshot); + return None; + } + }; + + let ident = param.ident.to_string(); + let sugg = match (ty_generics, self.psess.source_map().span_to_snippet(param.span())) { + (Some(Generics { params, span: impl_generics, .. }), Ok(snippet)) => { + Some(match ¶ms[..] { + [] => UnexpectedConstParamDeclarationSugg::AddParam { + impl_generics: *impl_generics, + incorrect_decl: param.span(), + snippet, + ident, + }, + [.., generic] => UnexpectedConstParamDeclarationSugg::AppendParam { + impl_generics_end: generic.span().shrink_to_hi(), + incorrect_decl: param.span(), + snippet, + ident, + }, + }) + } + _ => None, + }; + let guar = + self.dcx().emit_err(UnexpectedConstParamDeclaration { span: param.span(), sugg }); + + let value = self.mk_expr_err(param.span(), guar); + Some(GenericArg::Const(AnonConst { id: ast::DUMMY_NODE_ID, value })) + } + + pub fn recover_const_param_declaration( + &mut self, + ty_generics: Option<&Generics>, + ) -> PResult<'a, Option<GenericArg>> { + // We have to check for a few different cases. + if let Some(arg) = self.recover_const_param_decl(ty_generics) { + return Ok(Some(arg)); + } + + // We haven't consumed `const` yet. + let start = self.token.span; + self.bump(); // `const` + + // Detect and recover from the old, pre-RFC2000 syntax for const generics. + let mut err = UnexpectedConstInGenericParam { span: start, to_remove: None }; + if self.check_const_arg() { + err.to_remove = Some(start.until(self.token.span)); + self.dcx().emit_err(err); + Ok(Some(GenericArg::Const(self.parse_const_arg()?))) + } else { + let after_kw_const = self.token.span; + self.recover_const_arg(after_kw_const, self.dcx().create_err(err)).map(Some) + } + } + + /// Try to recover from possible generic const argument without `{` and `}`. + /// + /// When encountering code like `foo::< bar + 3 >` or `foo::< bar - baz >` we suggest + /// `foo::<{ bar + 3 }>` and `foo::<{ bar - baz }>`, respectively. We only provide a suggestion + /// if we think that the resulting expression would be well formed. + pub fn recover_const_arg(&mut self, start: Span, mut err: Diag<'a>) -> PResult<'a, GenericArg> { + let is_op_or_dot = AssocOp::from_token(&self.token) + .and_then(|op| { + if let AssocOp::Greater + | AssocOp::Less + | AssocOp::ShiftRight + | AssocOp::GreaterEqual + // Don't recover from `foo::<bar = baz>`, because this could be an attempt to + // assign a value to a defaulted generic parameter. + | AssocOp::Assign + | AssocOp::AssignOp(_) = op + { + None + } else { + Some(op) + } + }) + .is_some() + || self.token.kind == TokenKind::Dot; + // This will be true when a trait object type `Foo +` or a path which was a `const fn` with + // type params has been parsed. + let was_op = + matches!(self.prev_token.kind, token::BinOp(token::Plus | token::Shr) | token::Gt); + if !is_op_or_dot && !was_op { + // We perform these checks and early return to avoid taking a snapshot unnecessarily. + return Err(err); + } + let snapshot = self.create_snapshot_for_diagnostic(); + if is_op_or_dot { + self.bump(); + } + match self.parse_expr_res(Restrictions::CONST_EXPR, None) { + Ok(expr) => { + // Find a mistake like `MyTrait<Assoc == S::Assoc>`. + if token::EqEq == snapshot.token.kind { + err.span_suggestion( + snapshot.token.span, + "if you meant to use an associated type binding, replace `==` with `=`", + "=", + Applicability::MaybeIncorrect, + ); + let guar = err.emit(); + let value = self.mk_expr_err(start.to(expr.span), guar); + return Ok(GenericArg::Const(AnonConst { id: ast::DUMMY_NODE_ID, value })); + } else if token::Colon == snapshot.token.kind + && expr.span.lo() == snapshot.token.span.hi() + && matches!(expr.kind, ExprKind::Path(..)) + { + // Find a mistake like "foo::var:A". + err.span_suggestion( + snapshot.token.span, + "write a path separator here", + "::", + Applicability::MaybeIncorrect, + ); + let guar = err.emit(); + return Ok(GenericArg::Type( + self.mk_ty(start.to(expr.span), TyKind::Err(guar)), + )); + } else if token::Comma == self.token.kind || self.token.kind.should_end_const_arg() + { + // Avoid the following output by checking that we consumed a full const arg: + // help: expressions must be enclosed in braces to be used as const generic + // arguments + // | + // LL | let sr: Vec<{ (u32, _, _) = vec![] }; + // | ^ ^ + return Ok(self.dummy_const_arg_needs_braces(err, start.to(expr.span))); + } + } + Err(err) => { + err.cancel(); + } + } + self.restore_snapshot(snapshot); + Err(err) + } + + /// Try to recover from an unbraced const argument whose first token [could begin a type][ty]. + /// + /// [ty]: token::Token::can_begin_type + pub(crate) fn recover_unbraced_const_arg_that_can_begin_ty( + &mut self, + mut snapshot: SnapshotParser<'a>, + ) -> Option<P<ast::Expr>> { + match snapshot.parse_expr_res(Restrictions::CONST_EXPR, None) { + // Since we don't know the exact reason why we failed to parse the type or the + // expression, employ a simple heuristic to weed out some pathological cases. + Ok(expr) if let token::Comma | token::Gt = snapshot.token.kind => { + self.restore_snapshot(snapshot); + Some(expr) + } + Ok(_) => None, + Err(err) => { + err.cancel(); + None + } + } + } + + /// Creates a dummy const argument, and reports that the expression must be enclosed in braces + pub fn dummy_const_arg_needs_braces(&self, mut err: Diag<'a>, span: Span) -> GenericArg { + err.multipart_suggestion( + "expressions must be enclosed in braces to be used as const generic \ + arguments", + vec![(span.shrink_to_lo(), "{ ".to_string()), (span.shrink_to_hi(), " }".to_string())], + Applicability::MaybeIncorrect, + ); + let guar = err.emit(); + let value = self.mk_expr_err(span, guar); + GenericArg::Const(AnonConst { id: ast::DUMMY_NODE_ID, value }) + } + + /// Some special error handling for the "top-level" patterns in a match arm, + /// `for` loop, `let`, &c. (in contrast to subpatterns within such). + pub(crate) fn maybe_recover_colon_colon_in_pat_typo( + &mut self, + mut first_pat: P<Pat>, + expected: Option<Expected>, + ) -> P<Pat> { + if token::Colon != self.token.kind { + return first_pat; + } + if !matches!(first_pat.kind, PatKind::Ident(_, _, None) | PatKind::Path(..)) + || !self.look_ahead(1, |token| token.is_ident() && !token.is_reserved_ident()) + { + let mut snapshot_type = self.create_snapshot_for_diagnostic(); + snapshot_type.bump(); // `:` + match snapshot_type.parse_ty() { + Err(inner_err) => { + inner_err.cancel(); + } + Ok(ty) => { + let Err(mut err) = self.expected_one_of_not_found(&[], &[]) else { + return first_pat; + }; + err.span_label(ty.span, "specifying the type of a pattern isn't supported"); + self.restore_snapshot(snapshot_type); + let span = first_pat.span.to(ty.span); + first_pat = self.mk_pat(span, PatKind::Wild); + err.emit(); + } + } + return first_pat; + } + // The pattern looks like it might be a path with a `::` -> `:` typo: + // `match foo { bar:baz => {} }` + let colon_span = self.token.span; + // We only emit "unexpected `:`" error here if we can successfully parse the + // whole pattern correctly in that case. + let mut snapshot_pat = self.create_snapshot_for_diagnostic(); + let mut snapshot_type = self.create_snapshot_for_diagnostic(); + + // Create error for "unexpected `:`". + match self.expected_one_of_not_found(&[], &[]) { + Err(mut err) => { + // Skip the `:`. + snapshot_pat.bump(); + snapshot_type.bump(); + match snapshot_pat.parse_pat_no_top_alt(expected, None) { + Err(inner_err) => { + inner_err.cancel(); + } + Ok(mut pat) => { + // We've parsed the rest of the pattern. + let new_span = first_pat.span.to(pat.span); + let mut show_sugg = false; + // Try to construct a recovered pattern. + match &mut pat.kind { + PatKind::Struct(qself @ None, path, ..) + | PatKind::TupleStruct(qself @ None, path, _) + | PatKind::Path(qself @ None, path) => match &first_pat.kind { + PatKind::Ident(_, ident, _) => { + path.segments.insert(0, PathSegment::from_ident(*ident)); + path.span = new_span; + show_sugg = true; + first_pat = pat; + } + PatKind::Path(old_qself, old_path) => { + path.segments = old_path + .segments + .iter() + .cloned() + .chain(take(&mut path.segments)) + .collect(); + path.span = new_span; + *qself = old_qself.clone(); + first_pat = pat; + show_sugg = true; + } + _ => {} + }, + PatKind::Ident(BindingMode::NONE, ident, None) => { + match &first_pat.kind { + PatKind::Ident(_, old_ident, _) => { + let path = PatKind::Path( + None, + Path { + span: new_span, + segments: thin_vec![ + PathSegment::from_ident(*old_ident), + PathSegment::from_ident(*ident), + ], + tokens: None, + }, + ); + first_pat = self.mk_pat(new_span, path); + show_sugg = true; + } + PatKind::Path(old_qself, old_path) => { + let mut segments = old_path.segments.clone(); + segments.push(PathSegment::from_ident(*ident)); + let path = PatKind::Path( + old_qself.clone(), + Path { span: new_span, segments, tokens: None }, + ); + first_pat = self.mk_pat(new_span, path); + show_sugg = true; + } + _ => {} + } + } + _ => {} + } + if show_sugg { + err.span_suggestion_verbose( + colon_span.until(self.look_ahead(1, |t| t.span)), + "maybe write a path separator here", + "::", + Applicability::MaybeIncorrect, + ); + } else { + first_pat = self.mk_pat(new_span, PatKind::Wild); + } + self.restore_snapshot(snapshot_pat); + } + } + match snapshot_type.parse_ty() { + Err(inner_err) => { + inner_err.cancel(); + } + Ok(ty) => { + err.span_label(ty.span, "specifying the type of a pattern isn't supported"); + self.restore_snapshot(snapshot_type); + let new_span = first_pat.span.to(ty.span); + first_pat = self.mk_pat(new_span, PatKind::Wild); + } + } + err.emit(); + } + _ => { + // Carry on as if we had not done anything. This should be unreachable. + } + }; + first_pat + } + + pub(crate) fn maybe_recover_unexpected_block_label(&mut self) -> bool { + // Check for `'a : {` + if !(self.check_lifetime() + && self.look_ahead(1, |tok| tok.kind == token::Colon) + && self.look_ahead(2, |tok| tok.kind == token::OpenDelim(Delimiter::Brace))) + { + return false; + } + let label = self.eat_label().expect("just checked if a label exists"); + self.bump(); // eat `:` + let span = label.ident.span.to(self.prev_token.span); + self.dcx() + .struct_span_err(span, "block label not supported here") + .with_span_label(span, "not supported here") + .with_tool_only_span_suggestion( + label.ident.span.until(self.token.span), + "remove this block label", + "", + Applicability::MachineApplicable, + ) + .emit(); + true + } + + /// Some special error handling for the "top-level" patterns in a match arm, + /// `for` loop, `let`, &c. (in contrast to subpatterns within such). + pub(crate) fn maybe_recover_unexpected_comma( + &mut self, + lo: Span, + rt: CommaRecoveryMode, + ) -> PResult<'a, ()> { + if self.token != token::Comma { + return Ok(()); + } + + // An unexpected comma after a top-level pattern is a clue that the + // user (perhaps more accustomed to some other language) forgot the + // parentheses in what should have been a tuple pattern; return a + // suggestion-enhanced error here rather than choking on the comma later. + let comma_span = self.token.span; + self.bump(); + if let Err(err) = self.skip_pat_list() { + // We didn't expect this to work anyway; we just wanted to advance to the + // end of the comma-sequence so we know the span to suggest parenthesizing. + err.cancel(); + } + let seq_span = lo.to(self.prev_token.span); + let mut err = self.dcx().struct_span_err(comma_span, "unexpected `,` in pattern"); + if let Ok(seq_snippet) = self.span_to_snippet(seq_span) { + err.multipart_suggestion( + format!( + "try adding parentheses to match on a tuple{}", + if let CommaRecoveryMode::LikelyTuple = rt { "" } else { "..." }, + ), + vec![ + (seq_span.shrink_to_lo(), "(".to_string()), + (seq_span.shrink_to_hi(), ")".to_string()), + ], + Applicability::MachineApplicable, + ); + if let CommaRecoveryMode::EitherTupleOrPipe = rt { + err.span_suggestion( + seq_span, + "...or a vertical bar to match on multiple alternatives", + seq_snippet.replace(',', " |"), + Applicability::MachineApplicable, + ); + } + } + Err(err) + } + + pub(crate) fn maybe_recover_bounds_doubled_colon(&mut self, ty: &Ty) -> PResult<'a, ()> { + let TyKind::Path(qself, path) = &ty.kind else { return Ok(()) }; + let qself_position = qself.as_ref().map(|qself| qself.position); + for (i, segments) in path.segments.windows(2).enumerate() { + if qself_position.is_some_and(|pos| i < pos) { + continue; + } + if let [a, b] = segments { + let (a_span, b_span) = (a.span(), b.span()); + let between_span = a_span.shrink_to_hi().to(b_span.shrink_to_lo()); + if self.span_to_snippet(between_span).as_deref() == Ok(":: ") { + return Err(self.dcx().create_err(DoubleColonInBound { + span: path.span.shrink_to_hi(), + between: between_span, + })); + } + } + } + Ok(()) + } + + /// Check for exclusive ranges written as `..<` + pub(crate) fn maybe_err_dotdotlt_syntax(&self, maybe_lt: Token, mut err: PErr<'a>) -> PErr<'a> { + if maybe_lt == token::Lt + && (self.expected_tokens.contains(&TokenType::Token(token::Gt)) + || matches!(self.token.kind, token::Literal(..))) + { + err.span_suggestion( + maybe_lt.span, + "remove the `<` to write an exclusive range", + "", + Applicability::MachineApplicable, + ); + } + err + } + + /// This checks if this is a conflict marker, depending of the parameter passed. + /// + /// * `>>>>>` + /// * `=====` + /// * `<<<<<` + /// + pub fn is_vcs_conflict_marker( + &mut self, + long_kind: &TokenKind, + short_kind: &TokenKind, + ) -> bool { + (0..3).all(|i| self.look_ahead(i, |tok| tok == long_kind)) + && self.look_ahead(3, |tok| tok == short_kind) + } + + fn conflict_marker(&mut self, long_kind: &TokenKind, short_kind: &TokenKind) -> Option<Span> { + if self.is_vcs_conflict_marker(long_kind, short_kind) { + let lo = self.token.span; + for _ in 0..4 { + self.bump(); + } + return Some(lo.to(self.prev_token.span)); + } + None + } + + pub fn recover_vcs_conflict_marker(&mut self) { + if let Err(err) = self.err_vcs_conflict_marker() { + err.emit(); + FatalError.raise(); + } + } + + pub fn err_vcs_conflict_marker(&mut self) -> PResult<'a, ()> { + let Some(start) = self.conflict_marker(&TokenKind::BinOp(token::Shl), &TokenKind::Lt) + else { + return Ok(()); + }; + let mut spans = Vec::with_capacity(3); + spans.push(start); + let mut middlediff3 = None; + let mut middle = None; + let mut end = None; + loop { + if self.token.kind == TokenKind::Eof { + break; + } + if let Some(span) = self.conflict_marker(&TokenKind::OrOr, &TokenKind::BinOp(token::Or)) + { + middlediff3 = Some(span); + } + if let Some(span) = self.conflict_marker(&TokenKind::EqEq, &TokenKind::Eq) { + middle = Some(span); + } + if let Some(span) = self.conflict_marker(&TokenKind::BinOp(token::Shr), &TokenKind::Gt) + { + spans.push(span); + end = Some(span); + break; + } + self.bump(); + } + let mut err = self.dcx().struct_span_err(spans, "encountered diff marker"); + err.span_label(start, "after this is the code before the merge"); + if let Some(middle) = middlediff3 { + err.span_label(middle, ""); + } + if let Some(middle) = middle { + err.span_label(middle, ""); + } + if let Some(end) = end { + err.span_label(end, "above this are the incoming code changes"); + } + err.help( + "if you're having merge conflicts after pulling new code, the top section is the code \ + you already had and the bottom section is the remote code", + ); + err.help( + "if you're in the middle of a rebase, the top section is the code being rebased onto \ + and the bottom section is the code coming from the current commit being rebased", + ); + err.note( + "for an explanation on these markers from the `git` documentation, visit \ + <https://git-scm.com/book/en/v2/Git-Tools-Advanced-Merging#_checking_out_conflicts>", + ); + Err(err) + } + + /// Parse and throw away a parenthesized comma separated + /// sequence of patterns until `)` is reached. + fn skip_pat_list(&mut self) -> PResult<'a, ()> { + while !self.check(&token::CloseDelim(Delimiter::Parenthesis)) { + self.parse_pat_no_top_alt(None, None)?; + if !self.eat(&token::Comma) { + return Ok(()); + } + } + Ok(()) + } +} diff --git a/compiler/rustc_parse/src/parser/expr.rs b/compiler/rustc_parse/src/parser/expr.rs new file mode 100644 index 00000000000..d2d21624150 --- /dev/null +++ b/compiler/rustc_parse/src/parser/expr.rs @@ -0,0 +1,4047 @@ +// ignore-tidy-filelength +use super::diagnostics::SnapshotParser; +use super::pat::{CommaRecoveryMode, Expected, RecoverColon, RecoverComma}; +use super::ty::{AllowPlus, RecoverQPath, RecoverReturnSign}; +use super::{ + AttrWrapper, BlockMode, ClosureSpans, ForceCollect, Parser, PathStyle, Restrictions, + SemiColonMode, SeqSep, TokenExpectType, TokenType, Trailing, TrailingToken, +}; + +use crate::errors; +use crate::maybe_recover_from_interpolated_ty_qpath; +use ast::mut_visit::{noop_visit_expr, MutVisitor}; +use ast::token::IdentIsRaw; +use ast::{CoroutineKind, ForLoopKind, GenBlockKind, MatchKind, Pat, Path, PathSegment, Recovered}; +use core::mem; +use core::ops::ControlFlow; +use rustc_ast::ptr::P; +use rustc_ast::token::{self, Delimiter, Token, TokenKind}; +use rustc_ast::util::case::Case; +use rustc_ast::util::classify; +use rustc_ast::util::parser::{prec_let_scrutinee_needs_par, AssocOp, Fixity}; +use rustc_ast::visit::{walk_expr, Visitor}; +use rustc_ast::{self as ast, AttrStyle, AttrVec, CaptureBy, ExprField, UnOp, DUMMY_NODE_ID}; +use rustc_ast::{AnonConst, BinOp, BinOpKind, FnDecl, FnRetTy, MacCall, Param, Ty, TyKind}; +use rustc_ast::{Arm, BlockCheckMode, Expr, ExprKind, Label, Movability, RangeLimits}; +use rustc_ast::{ClosureBinder, MetaItemLit, StmtKind}; +use rustc_ast_pretty::pprust; +use rustc_data_structures::stack::ensure_sufficient_stack; +use rustc_errors::{Applicability, Diag, PResult, StashKey, Subdiagnostic}; +use rustc_lexer::unescape::unescape_char; +use rustc_macros::Subdiagnostic; +use rustc_session::errors::{report_lit_error, ExprParenthesesNeeded}; +use rustc_session::lint::builtin::BREAK_WITH_LABEL_AND_LOOP; +use rustc_session::lint::BuiltinLintDiag; +use rustc_span::source_map::{self, Spanned}; +use rustc_span::symbol::{kw, sym, Ident, Symbol}; +use rustc_span::{BytePos, ErrorGuaranteed, Pos, Span}; +use thin_vec::{thin_vec, ThinVec}; + +/// Possibly accepts an `token::Interpolated` expression (a pre-parsed expression +/// dropped into the token stream, which happens while parsing the result of +/// macro expansion). Placement of these is not as complex as I feared it would +/// be. The important thing is to make sure that lookahead doesn't balk at +/// `token::Interpolated` tokens. +macro_rules! maybe_whole_expr { + ($p:expr) => { + if let token::Interpolated(nt) = &$p.token.kind { + match &**nt { + token::NtExpr(e) | token::NtLiteral(e) => { + let e = e.clone(); + $p.bump(); + return Ok(e); + } + token::NtPath(path) => { + let path = (**path).clone(); + $p.bump(); + return Ok($p.mk_expr($p.prev_token.span, ExprKind::Path(None, path))); + } + token::NtBlock(block) => { + let block = block.clone(); + $p.bump(); + return Ok($p.mk_expr($p.prev_token.span, ExprKind::Block(block, None))); + } + _ => {} + }; + } + }; +} + +#[derive(Debug)] +pub(super) enum LhsExpr { + NotYetParsed, + AttributesParsed(AttrWrapper), + AlreadyParsed { expr: P<Expr>, starts_statement: bool }, +} + +impl From<Option<AttrWrapper>> for LhsExpr { + /// Converts `Some(attrs)` into `LhsExpr::AttributesParsed(attrs)` + /// and `None` into `LhsExpr::NotYetParsed`. + /// + /// This conversion does not allocate. + fn from(o: Option<AttrWrapper>) -> Self { + if let Some(attrs) = o { LhsExpr::AttributesParsed(attrs) } else { LhsExpr::NotYetParsed } + } +} + +impl From<P<Expr>> for LhsExpr { + /// Converts the `expr: P<Expr>` into `LhsExpr::AlreadyParsed { expr, starts_statement: false }`. + /// + /// This conversion does not allocate. + fn from(expr: P<Expr>) -> Self { + LhsExpr::AlreadyParsed { expr, starts_statement: false } + } +} + +#[derive(Debug)] +enum DestructuredFloat { + /// 1e2 + Single(Symbol, Span), + /// 1. + TrailingDot(Symbol, Span, Span), + /// 1.2 | 1.2e3 + MiddleDot(Symbol, Span, Span, Symbol, Span), + /// Invalid + Error, +} + +impl<'a> Parser<'a> { + /// Parses an expression. + #[inline] + pub fn parse_expr(&mut self) -> PResult<'a, P<Expr>> { + self.current_closure.take(); + + self.parse_expr_res(Restrictions::empty(), None) + } + + /// Parses an expression, forcing tokens to be collected + pub fn parse_expr_force_collect(&mut self) -> PResult<'a, P<Expr>> { + self.collect_tokens_no_attrs(|this| this.parse_expr()) + } + + pub fn parse_expr_anon_const(&mut self) -> PResult<'a, AnonConst> { + self.parse_expr().map(|value| AnonConst { id: DUMMY_NODE_ID, value }) + } + + fn parse_expr_catch_underscore(&mut self, restrictions: Restrictions) -> PResult<'a, P<Expr>> { + match self.parse_expr_res(restrictions, None) { + Ok(expr) => Ok(expr), + Err(err) => match self.token.ident() { + Some((Ident { name: kw::Underscore, .. }, IdentIsRaw::No)) + if self.may_recover() && self.look_ahead(1, |t| t == &token::Comma) => + { + // Special-case handling of `foo(_, _, _)` + let guar = err.emit(); + self.bump(); + Ok(self.mk_expr(self.prev_token.span, ExprKind::Err(guar))) + } + _ => Err(err), + }, + } + } + + /// Parses a sequence of expressions delimited by parentheses. + fn parse_expr_paren_seq(&mut self) -> PResult<'a, ThinVec<P<Expr>>> { + self.parse_paren_comma_seq(|p| p.parse_expr_catch_underscore(Restrictions::empty())) + .map(|(r, _)| r) + } + + /// Parses an expression, subject to the given restrictions. + #[inline] + pub(super) fn parse_expr_res( + &mut self, + r: Restrictions, + already_parsed_attrs: Option<AttrWrapper>, + ) -> PResult<'a, P<Expr>> { + self.with_res(r, |this| this.parse_expr_assoc(already_parsed_attrs)) + } + + /// Parses an associative expression. + /// + /// This parses an expression accounting for associativity and precedence of the operators in + /// the expression. + #[inline] + fn parse_expr_assoc( + &mut self, + already_parsed_attrs: Option<AttrWrapper>, + ) -> PResult<'a, P<Expr>> { + self.parse_expr_assoc_with(0, already_parsed_attrs.into()) + } + + /// Parses an associative expression with operators of at least `min_prec` precedence. + pub(super) fn parse_expr_assoc_with( + &mut self, + min_prec: usize, + lhs: LhsExpr, + ) -> PResult<'a, P<Expr>> { + let mut starts_stmt = false; + let mut lhs = if let LhsExpr::AlreadyParsed { expr, starts_statement } = lhs { + starts_stmt = starts_statement; + expr + } else { + let attrs = match lhs { + LhsExpr::AttributesParsed(attrs) => Some(attrs), + _ => None, + }; + if self.token.is_range_separator() { + return self.parse_expr_prefix_range(attrs); + } else { + self.parse_expr_prefix(attrs)? + } + }; + + if !self.should_continue_as_assoc_expr(&lhs) { + return Ok(lhs); + } + + self.expected_tokens.push(TokenType::Operator); + while let Some(op) = self.check_assoc_op() { + let lhs_span = self.interpolated_or_expr_span(&lhs); + let cur_op_span = self.token.span; + let restrictions = if op.node.is_assign_like() { + self.restrictions & Restrictions::NO_STRUCT_LITERAL + } else { + self.restrictions + }; + let prec = op.node.precedence(); + if prec < min_prec { + break; + } + // Check for deprecated `...` syntax + if self.token == token::DotDotDot && op.node == AssocOp::DotDotEq { + self.err_dotdotdot_syntax(self.token.span); + } + + if self.token == token::LArrow { + self.err_larrow_operator(self.token.span); + } + + self.bump(); + if op.node.is_comparison() { + if let Some(expr) = self.check_no_chained_comparison(&lhs, &op)? { + return Ok(expr); + } + } + + // Look for JS' `===` and `!==` and recover + if (op.node == AssocOp::Equal || op.node == AssocOp::NotEqual) + && self.token.kind == token::Eq + && self.prev_token.span.hi() == self.token.span.lo() + { + let sp = op.span.to(self.token.span); + let sugg = match op.node { + AssocOp::Equal => "==", + AssocOp::NotEqual => "!=", + _ => unreachable!(), + } + .into(); + let invalid = format!("{sugg}="); + self.dcx().emit_err(errors::InvalidComparisonOperator { + span: sp, + invalid: invalid.clone(), + sub: errors::InvalidComparisonOperatorSub::Correctable { + span: sp, + invalid, + correct: sugg, + }, + }); + self.bump(); + } + + // Look for PHP's `<>` and recover + if op.node == AssocOp::Less + && self.token.kind == token::Gt + && self.prev_token.span.hi() == self.token.span.lo() + { + let sp = op.span.to(self.token.span); + self.dcx().emit_err(errors::InvalidComparisonOperator { + span: sp, + invalid: "<>".into(), + sub: errors::InvalidComparisonOperatorSub::Correctable { + span: sp, + invalid: "<>".into(), + correct: "!=".into(), + }, + }); + self.bump(); + } + + // Look for C++'s `<=>` and recover + if op.node == AssocOp::LessEqual + && self.token.kind == token::Gt + && self.prev_token.span.hi() == self.token.span.lo() + { + let sp = op.span.to(self.token.span); + self.dcx().emit_err(errors::InvalidComparisonOperator { + span: sp, + invalid: "<=>".into(), + sub: errors::InvalidComparisonOperatorSub::Spaceship(sp), + }); + self.bump(); + } + + if self.prev_token == token::BinOp(token::Plus) + && self.token == token::BinOp(token::Plus) + && self.prev_token.span.between(self.token.span).is_empty() + { + let op_span = self.prev_token.span.to(self.token.span); + // Eat the second `+` + self.bump(); + lhs = self.recover_from_postfix_increment(lhs, op_span, starts_stmt)?; + continue; + } + + if self.prev_token == token::BinOp(token::Minus) + && self.token == token::BinOp(token::Minus) + && self.prev_token.span.between(self.token.span).is_empty() + && !self.look_ahead(1, |tok| tok.can_begin_expr()) + { + let op_span = self.prev_token.span.to(self.token.span); + // Eat the second `-` + self.bump(); + lhs = self.recover_from_postfix_decrement(lhs, op_span, starts_stmt)?; + continue; + } + + let op = op.node; + // Special cases: + if op == AssocOp::As { + lhs = self.parse_assoc_op_cast(lhs, lhs_span, ExprKind::Cast)?; + continue; + } else if op == AssocOp::DotDot || op == AssocOp::DotDotEq { + // If we didn't have to handle `x..`/`x..=`, it would be pretty easy to + // generalise it to the Fixity::None code. + lhs = self.parse_expr_range(prec, lhs, op, cur_op_span)?; + break; + } + + let fixity = op.fixity(); + let prec_adjustment = match fixity { + Fixity::Right => 0, + Fixity::Left => 1, + // We currently have no non-associative operators that are not handled above by + // the special cases. The code is here only for future convenience. + Fixity::None => 1, + }; + let rhs = self.with_res(restrictions - Restrictions::STMT_EXPR, |this| { + this.parse_expr_assoc_with(prec + prec_adjustment, LhsExpr::NotYetParsed) + })?; + + self.error_ambiguous_outer_attrs(&lhs, lhs_span, rhs.span); + let span = lhs_span.to(rhs.span); + + lhs = match op { + AssocOp::Add + | AssocOp::Subtract + | AssocOp::Multiply + | AssocOp::Divide + | AssocOp::Modulus + | AssocOp::LAnd + | AssocOp::LOr + | AssocOp::BitXor + | AssocOp::BitAnd + | AssocOp::BitOr + | AssocOp::ShiftLeft + | AssocOp::ShiftRight + | AssocOp::Equal + | AssocOp::Less + | AssocOp::LessEqual + | AssocOp::NotEqual + | AssocOp::Greater + | AssocOp::GreaterEqual => { + let ast_op = op.to_ast_binop().unwrap(); + let binary = self.mk_binary(source_map::respan(cur_op_span, ast_op), lhs, rhs); + self.mk_expr(span, binary) + } + AssocOp::Assign => self.mk_expr(span, ExprKind::Assign(lhs, rhs, cur_op_span)), + AssocOp::AssignOp(k) => { + let aop = match k { + token::Plus => BinOpKind::Add, + token::Minus => BinOpKind::Sub, + token::Star => BinOpKind::Mul, + token::Slash => BinOpKind::Div, + token::Percent => BinOpKind::Rem, + token::Caret => BinOpKind::BitXor, + token::And => BinOpKind::BitAnd, + token::Or => BinOpKind::BitOr, + token::Shl => BinOpKind::Shl, + token::Shr => BinOpKind::Shr, + }; + let aopexpr = self.mk_assign_op(source_map::respan(cur_op_span, aop), lhs, rhs); + self.mk_expr(span, aopexpr) + } + AssocOp::As | AssocOp::DotDot | AssocOp::DotDotEq => { + self.dcx().span_bug(span, "AssocOp should have been handled by special case") + } + }; + + if let Fixity::None = fixity { + break; + } + } + + Ok(lhs) + } + + fn should_continue_as_assoc_expr(&mut self, lhs: &Expr) -> bool { + match (self.expr_is_complete(lhs), AssocOp::from_token(&self.token)) { + // Semi-statement forms are odd: + // See https://github.com/rust-lang/rust/issues/29071 + (true, None) => false, + (false, _) => true, // Continue parsing the expression. + // An exhaustive check is done in the following block, but these are checked first + // because they *are* ambiguous but also reasonable looking incorrect syntax, so we + // want to keep their span info to improve diagnostics in these cases in a later stage. + (true, Some(AssocOp::Multiply)) | // `{ 42 } *foo = bar;` or `{ 42 } * 3` + (true, Some(AssocOp::Subtract)) | // `{ 42 } -5` + (true, Some(AssocOp::Add)) | // `{ 42 } + 42` (unary plus) + (true, Some(AssocOp::LAnd)) | // `{ 42 } &&x` (#61475) or `{ 42 } && if x { 1 } else { 0 }` + (true, Some(AssocOp::LOr)) | // `{ 42 } || 42` ("logical or" or closure) + (true, Some(AssocOp::BitOr)) // `{ 42 } | 42` or `{ 42 } |x| 42` + => { + // These cases are ambiguous and can't be identified in the parser alone. + // + // Bitwise AND is left out because guessing intent is hard. We can make + // suggestions based on the assumption that double-refs are rarely intentional, + // and closures are distinct enough that they don't get mixed up with their + // return value. + let sp = self.psess.source_map().start_point(self.token.span); + self.psess.ambiguous_block_expr_parse.borrow_mut().insert(sp, lhs.span); + false + } + (true, Some(op)) if !op.can_continue_expr_unambiguously() => false, + (true, Some(_)) => { + self.error_found_expr_would_be_stmt(lhs); + true + } + } + } + + /// We've found an expression that would be parsed as a statement, + /// but the next token implies this should be parsed as an expression. + /// For example: `if let Some(x) = x { x } else { 0 } / 2`. + fn error_found_expr_would_be_stmt(&self, lhs: &Expr) { + self.dcx().emit_err(errors::FoundExprWouldBeStmt { + span: self.token.span, + token: self.token.clone(), + suggestion: ExprParenthesesNeeded::surrounding(lhs.span), + }); + } + + fn error_ambiguous_outer_attrs(&self, lhs: &P<Expr>, lhs_span: Span, rhs_span: Span) { + if let Some(attr) = lhs.attrs.iter().find(|a| a.style == AttrStyle::Outer) { + self.dcx().emit_err(errors::AmbiguousOuterAttributes { + span: attr.span.to(rhs_span), + sugg: errors::WrapInParentheses::Expression { + left: attr.span.shrink_to_lo(), + right: lhs_span.shrink_to_hi(), + }, + }); + } + } + + /// Possibly translate the current token to an associative operator. + /// The method does not advance the current token. + /// + /// Also performs recovery for `and` / `or` which are mistaken for `&&` and `||` respectively. + pub fn check_assoc_op(&self) -> Option<Spanned<AssocOp>> { + let (op, span) = match (AssocOp::from_token(&self.token), self.token.ident()) { + // When parsing const expressions, stop parsing when encountering `>`. + ( + Some( + AssocOp::ShiftRight + | AssocOp::Greater + | AssocOp::GreaterEqual + | AssocOp::AssignOp(token::BinOpToken::Shr), + ), + _, + ) if self.restrictions.contains(Restrictions::CONST_EXPR) => { + return None; + } + // When recovering patterns as expressions, stop parsing when encountering an assignment `=`, an alternative `|`, or a range `..`. + ( + Some( + AssocOp::Assign + | AssocOp::AssignOp(_) + | AssocOp::BitOr + | AssocOp::DotDot + | AssocOp::DotDotEq, + ), + _, + ) if self.restrictions.contains(Restrictions::IS_PAT) => { + return None; + } + (Some(op), _) => (op, self.token.span), + (None, Some((Ident { name: sym::and, span }, IdentIsRaw::No))) + if self.may_recover() => + { + self.dcx().emit_err(errors::InvalidLogicalOperator { + span: self.token.span, + incorrect: "and".into(), + sub: errors::InvalidLogicalOperatorSub::Conjunction(self.token.span), + }); + (AssocOp::LAnd, span) + } + (None, Some((Ident { name: sym::or, span }, IdentIsRaw::No))) if self.may_recover() => { + self.dcx().emit_err(errors::InvalidLogicalOperator { + span: self.token.span, + incorrect: "or".into(), + sub: errors::InvalidLogicalOperatorSub::Disjunction(self.token.span), + }); + (AssocOp::LOr, span) + } + _ => return None, + }; + Some(source_map::respan(span, op)) + } + + /// Checks if this expression is a successfully parsed statement. + fn expr_is_complete(&self, e: &Expr) -> bool { + self.restrictions.contains(Restrictions::STMT_EXPR) && classify::expr_is_complete(e) + } + + /// Parses `x..y`, `x..=y`, and `x..`/`x..=`. + /// The other two variants are handled in `parse_prefix_range_expr` below. + fn parse_expr_range( + &mut self, + prec: usize, + lhs: P<Expr>, + op: AssocOp, + cur_op_span: Span, + ) -> PResult<'a, P<Expr>> { + let rhs = if self.is_at_start_of_range_notation_rhs() { + let maybe_lt = self.token.clone(); + Some( + self.parse_expr_assoc_with(prec + 1, LhsExpr::NotYetParsed) + .map_err(|err| self.maybe_err_dotdotlt_syntax(maybe_lt, err))?, + ) + } else { + None + }; + let rhs_span = rhs.as_ref().map_or(cur_op_span, |x| x.span); + self.error_ambiguous_outer_attrs(&lhs, lhs.span, rhs_span); + let span = lhs.span.to(rhs_span); + let limits = + if op == AssocOp::DotDot { RangeLimits::HalfOpen } else { RangeLimits::Closed }; + let range = self.mk_range(Some(lhs), rhs, limits); + Ok(self.mk_expr(span, range)) + } + + fn is_at_start_of_range_notation_rhs(&self) -> bool { + if self.token.can_begin_expr() { + // Parse `for i in 1.. { }` as infinite loop, not as `for i in (1..{})`. + if self.token == token::OpenDelim(Delimiter::Brace) { + return !self.restrictions.contains(Restrictions::NO_STRUCT_LITERAL); + } + true + } else { + false + } + } + + /// Parses prefix-forms of range notation: `..expr`, `..`, `..=expr`. + fn parse_expr_prefix_range(&mut self, attrs: Option<AttrWrapper>) -> PResult<'a, P<Expr>> { + // Check for deprecated `...` syntax. + if self.token == token::DotDotDot { + self.err_dotdotdot_syntax(self.token.span); + } + + debug_assert!( + self.token.is_range_separator(), + "parse_prefix_range_expr: token {:?} is not DotDot/DotDotEq", + self.token + ); + + let limits = match self.token.kind { + token::DotDot => RangeLimits::HalfOpen, + _ => RangeLimits::Closed, + }; + let op = AssocOp::from_token(&self.token); + // FIXME: `parse_prefix_range_expr` is called when the current + // token is `DotDot`, `DotDotDot`, or `DotDotEq`. If we haven't already + // parsed attributes, then trying to parse them here will always fail. + // We should figure out how we want attributes on range expressions to work. + let attrs = self.parse_or_use_outer_attributes(attrs)?; + self.collect_tokens_for_expr(attrs, |this, attrs| { + let lo = this.token.span; + let maybe_lt = this.look_ahead(1, |t| t.clone()); + this.bump(); + let (span, opt_end) = if this.is_at_start_of_range_notation_rhs() { + // RHS must be parsed with more associativity than the dots. + this.parse_expr_assoc_with(op.unwrap().precedence() + 1, LhsExpr::NotYetParsed) + .map(|x| (lo.to(x.span), Some(x))) + .map_err(|err| this.maybe_err_dotdotlt_syntax(maybe_lt, err))? + } else { + (lo, None) + }; + let range = this.mk_range(None, opt_end, limits); + Ok(this.mk_expr_with_attrs(span, range, attrs)) + }) + } + + /// Parses a prefix-unary-operator expr. + fn parse_expr_prefix(&mut self, attrs: Option<AttrWrapper>) -> PResult<'a, P<Expr>> { + let attrs = self.parse_or_use_outer_attributes(attrs)?; + let lo = self.token.span; + + macro_rules! make_it { + ($this:ident, $attrs:expr, |this, _| $body:expr) => { + $this.collect_tokens_for_expr($attrs, |$this, attrs| { + let (hi, ex) = $body?; + Ok($this.mk_expr_with_attrs(lo.to(hi), ex, attrs)) + }) + }; + } + + let this = self; + + // Note: when adding new unary operators, don't forget to adjust TokenKind::can_begin_expr() + match this.token.uninterpolate().kind { + // `!expr` + token::Not => make_it!(this, attrs, |this, _| this.parse_expr_unary(lo, UnOp::Not)), + // `~expr` + token::Tilde => make_it!(this, attrs, |this, _| this.recover_tilde_expr(lo)), + // `-expr` + token::BinOp(token::Minus) => { + make_it!(this, attrs, |this, _| this.parse_expr_unary(lo, UnOp::Neg)) + } + // `*expr` + token::BinOp(token::Star) => { + make_it!(this, attrs, |this, _| this.parse_expr_unary(lo, UnOp::Deref)) + } + // `&expr` and `&&expr` + token::BinOp(token::And) | token::AndAnd => { + make_it!(this, attrs, |this, _| this.parse_expr_borrow(lo)) + } + // `+lit` + token::BinOp(token::Plus) if this.look_ahead(1, |tok| tok.is_numeric_lit()) => { + let mut err = errors::LeadingPlusNotSupported { + span: lo, + remove_plus: None, + add_parentheses: None, + }; + + // a block on the LHS might have been intended to be an expression instead + if let Some(sp) = this.psess.ambiguous_block_expr_parse.borrow().get(&lo) { + err.add_parentheses = Some(ExprParenthesesNeeded::surrounding(*sp)); + } else { + err.remove_plus = Some(lo); + } + this.dcx().emit_err(err); + + this.bump(); + this.parse_expr_prefix(None) + } + // Recover from `++x`: + token::BinOp(token::Plus) + if this.look_ahead(1, |t| *t == token::BinOp(token::Plus)) => + { + let starts_stmt = this.prev_token == token::Semi + || this.prev_token == token::CloseDelim(Delimiter::Brace); + let pre_span = this.token.span.to(this.look_ahead(1, |t| t.span)); + // Eat both `+`s. + this.bump(); + this.bump(); + + let operand_expr = this.parse_expr_dot_or_call(Default::default())?; + this.recover_from_prefix_increment(operand_expr, pre_span, starts_stmt) + } + token::Ident(..) if this.token.is_keyword(kw::Box) => { + make_it!(this, attrs, |this, _| this.parse_expr_box(lo)) + } + token::Ident(..) if this.may_recover() && this.is_mistaken_not_ident_negation() => { + make_it!(this, attrs, |this, _| this.recover_not_expr(lo)) + } + _ => return this.parse_expr_dot_or_call(Some(attrs)), + } + } + + fn parse_expr_prefix_common(&mut self, lo: Span) -> PResult<'a, (Span, P<Expr>)> { + self.bump(); + let expr = self.parse_expr_prefix(None)?; + let span = self.interpolated_or_expr_span(&expr); + Ok((lo.to(span), expr)) + } + + fn parse_expr_unary(&mut self, lo: Span, op: UnOp) -> PResult<'a, (Span, ExprKind)> { + let (span, expr) = self.parse_expr_prefix_common(lo)?; + Ok((span, self.mk_unary(op, expr))) + } + + /// Recover on `~expr` in favor of `!expr`. + fn recover_tilde_expr(&mut self, lo: Span) -> PResult<'a, (Span, ExprKind)> { + self.dcx().emit_err(errors::TildeAsUnaryOperator(lo)); + + self.parse_expr_unary(lo, UnOp::Not) + } + + /// Parse `box expr` - this syntax has been removed, but we still parse this + /// for now to provide a more useful error + fn parse_expr_box(&mut self, box_kw: Span) -> PResult<'a, (Span, ExprKind)> { + let (span, _) = self.parse_expr_prefix_common(box_kw)?; + let inner_span = span.with_lo(box_kw.hi()); + let code = self.psess.source_map().span_to_snippet(inner_span).unwrap(); + let guar = self.dcx().emit_err(errors::BoxSyntaxRemoved { span: span, code: code.trim() }); + Ok((span, ExprKind::Err(guar))) + } + + fn is_mistaken_not_ident_negation(&self) -> bool { + let token_cannot_continue_expr = |t: &Token| match t.uninterpolate().kind { + // These tokens can start an expression after `!`, but + // can't continue an expression after an ident + token::Ident(name, is_raw) => token::ident_can_begin_expr(name, t.span, is_raw), + token::Literal(..) | token::Pound => true, + _ => t.is_whole_expr(), + }; + self.token.is_ident_named(sym::not) && self.look_ahead(1, token_cannot_continue_expr) + } + + /// Recover on `not expr` in favor of `!expr`. + fn recover_not_expr(&mut self, lo: Span) -> PResult<'a, (Span, ExprKind)> { + let negated_token = self.look_ahead(1, |t| t.clone()); + + let sub_diag = if negated_token.is_numeric_lit() { + errors::NotAsNegationOperatorSub::SuggestNotBitwise + } else if negated_token.is_bool_lit() { + errors::NotAsNegationOperatorSub::SuggestNotLogical + } else { + errors::NotAsNegationOperatorSub::SuggestNotDefault + }; + + self.dcx().emit_err(errors::NotAsNegationOperator { + negated: negated_token.span, + negated_desc: super::token_descr(&negated_token), + // Span the `not` plus trailing whitespace to avoid + // trailing whitespace after the `!` in our suggestion + sub: sub_diag( + self.psess.source_map().span_until_non_whitespace(lo.to(negated_token.span)), + ), + }); + + self.parse_expr_unary(lo, UnOp::Not) + } + + /// Returns the span of expr if it was not interpolated, or the span of the interpolated token. + fn interpolated_or_expr_span(&self, expr: &Expr) -> Span { + match self.prev_token.kind { + TokenKind::NtIdent(..) | TokenKind::NtLifetime(..) | TokenKind::Interpolated(..) => { + self.prev_token.span + } + _ => expr.span, + } + } + + fn parse_assoc_op_cast( + &mut self, + lhs: P<Expr>, + lhs_span: Span, + expr_kind: fn(P<Expr>, P<Ty>) -> ExprKind, + ) -> PResult<'a, P<Expr>> { + let mk_expr = |this: &mut Self, lhs: P<Expr>, rhs: P<Ty>| { + this.error_ambiguous_outer_attrs(&lhs, lhs_span, rhs.span); + this.mk_expr(lhs_span.to(rhs.span), expr_kind(lhs, rhs)) + }; + + // Save the state of the parser before parsing type normally, in case there is a + // LessThan comparison after this cast. + let parser_snapshot_before_type = self.clone(); + let cast_expr = match self.parse_as_cast_ty() { + Ok(rhs) => mk_expr(self, lhs, rhs), + Err(type_err) => { + if !self.may_recover() { + return Err(type_err); + } + + // Rewind to before attempting to parse the type with generics, to recover + // from situations like `x as usize < y` in which we first tried to parse + // `usize < y` as a type with generic arguments. + let parser_snapshot_after_type = mem::replace(self, parser_snapshot_before_type); + + // Check for typo of `'a: loop { break 'a }` with a missing `'`. + match (&lhs.kind, &self.token.kind) { + ( + // `foo: ` + ExprKind::Path(None, ast::Path { segments, .. }), + token::Ident(kw::For | kw::Loop | kw::While, IdentIsRaw::No), + ) if segments.len() == 1 => { + let snapshot = self.create_snapshot_for_diagnostic(); + let label = Label { + ident: Ident::from_str_and_span( + &format!("'{}", segments[0].ident), + segments[0].ident.span, + ), + }; + match self.parse_expr_labeled(label, false) { + Ok(expr) => { + type_err.cancel(); + self.dcx().emit_err(errors::MalformedLoopLabel { + span: label.ident.span, + correct_label: label.ident, + }); + return Ok(expr); + } + Err(err) => { + err.cancel(); + self.restore_snapshot(snapshot); + } + } + } + _ => {} + } + + match self.parse_path(PathStyle::Expr) { + Ok(path) => { + let span_after_type = parser_snapshot_after_type.token.span; + let expr = mk_expr( + self, + lhs, + self.mk_ty(path.span, TyKind::Path(None, path.clone())), + ); + + let args_span = self.look_ahead(1, |t| t.span).to(span_after_type); + let suggestion = errors::ComparisonOrShiftInterpretedAsGenericSugg { + left: expr.span.shrink_to_lo(), + right: expr.span.shrink_to_hi(), + }; + + match self.token.kind { + token::Lt => { + self.dcx().emit_err(errors::ComparisonInterpretedAsGeneric { + comparison: self.token.span, + r#type: path, + args: args_span, + suggestion, + }) + } + token::BinOp(token::Shl) => { + self.dcx().emit_err(errors::ShiftInterpretedAsGeneric { + shift: self.token.span, + r#type: path, + args: args_span, + suggestion, + }) + } + _ => { + // We can end up here even without `<` being the next token, for + // example because `parse_ty_no_plus` returns `Err` on keywords, + // but `parse_path` returns `Ok` on them due to error recovery. + // Return original error and parser state. + *self = parser_snapshot_after_type; + return Err(type_err); + } + }; + + // Successfully parsed the type path leaving a `<` yet to parse. + type_err.cancel(); + + // Keep `x as usize` as an expression in AST and continue parsing. + expr + } + Err(path_err) => { + // Couldn't parse as a path, return original error and parser state. + path_err.cancel(); + *self = parser_snapshot_after_type; + return Err(type_err); + } + } + } + }; + + self.parse_and_disallow_postfix_after_cast(cast_expr) + } + + /// Parses a postfix operators such as `.`, `?`, or index (`[]`) after a cast, + /// then emits an error and returns the newly parsed tree. + /// The resulting parse tree for `&x as T[0]` has a precedence of `((&x) as T)[0]`. + fn parse_and_disallow_postfix_after_cast( + &mut self, + cast_expr: P<Expr>, + ) -> PResult<'a, P<Expr>> { + if let ExprKind::Type(_, _) = cast_expr.kind { + panic!("ExprKind::Type must not be parsed"); + } + + let span = cast_expr.span; + + let with_postfix = self.parse_expr_dot_or_call_with_(cast_expr, span)?; + + // Check if an illegal postfix operator has been added after the cast. + // If the resulting expression is not a cast, it is an illegal postfix operator. + if !matches!(with_postfix.kind, ExprKind::Cast(_, _)) { + let msg = format!( + "cast cannot be followed by {}", + match with_postfix.kind { + ExprKind::Index(..) => "indexing", + ExprKind::Try(_) => "`?`", + ExprKind::Field(_, _) => "a field access", + ExprKind::MethodCall(_) => "a method call", + ExprKind::Call(_, _) => "a function call", + ExprKind::Await(_, _) => "`.await`", + ExprKind::Match(_, _, MatchKind::Postfix) => "a postfix match", + ExprKind::Err(_) => return Ok(with_postfix), + _ => unreachable!("parse_dot_or_call_expr_with_ shouldn't produce this"), + } + ); + let mut err = self.dcx().struct_span_err(span, msg); + + let suggest_parens = |err: &mut Diag<'_>| { + let suggestions = vec![ + (span.shrink_to_lo(), "(".to_string()), + (span.shrink_to_hi(), ")".to_string()), + ]; + err.multipart_suggestion( + "try surrounding the expression in parentheses", + suggestions, + Applicability::MachineApplicable, + ); + }; + + suggest_parens(&mut err); + + err.emit(); + }; + Ok(with_postfix) + } + + /// Parse `& mut? <expr>` or `& raw [ const | mut ] <expr>`. + fn parse_expr_borrow(&mut self, lo: Span) -> PResult<'a, (Span, ExprKind)> { + self.expect_and()?; + let has_lifetime = self.token.is_lifetime() && self.look_ahead(1, |t| t != &token::Colon); + let lifetime = has_lifetime.then(|| self.expect_lifetime()); // For recovery, see below. + let (borrow_kind, mutbl) = self.parse_borrow_modifiers(lo); + let expr = if self.token.is_range_separator() { + self.parse_expr_prefix_range(None) + } else { + self.parse_expr_prefix(None) + }?; + let hi = self.interpolated_or_expr_span(&expr); + let span = lo.to(hi); + if let Some(lt) = lifetime { + self.error_remove_borrow_lifetime(span, lt.ident.span); + } + Ok((span, ExprKind::AddrOf(borrow_kind, mutbl, expr))) + } + + fn error_remove_borrow_lifetime(&self, span: Span, lt_span: Span) { + self.dcx().emit_err(errors::LifetimeInBorrowExpression { span, lifetime_span: lt_span }); + } + + /// Parse `mut?` or `raw [ const | mut ]`. + fn parse_borrow_modifiers(&mut self, lo: Span) -> (ast::BorrowKind, ast::Mutability) { + if self.check_keyword(kw::Raw) && self.look_ahead(1, Token::is_mutability) { + // `raw [ const | mut ]`. + let found_raw = self.eat_keyword(kw::Raw); + assert!(found_raw); + let mutability = self.parse_const_or_mut().unwrap(); + self.psess.gated_spans.gate(sym::raw_ref_op, lo.to(self.prev_token.span)); + (ast::BorrowKind::Raw, mutability) + } else { + // `mut?` + (ast::BorrowKind::Ref, self.parse_mutability()) + } + } + + /// Parses `a.b` or `a(13)` or `a[4]` or just `a`. + fn parse_expr_dot_or_call(&mut self, attrs: Option<AttrWrapper>) -> PResult<'a, P<Expr>> { + let attrs = self.parse_or_use_outer_attributes(attrs)?; + self.collect_tokens_for_expr(attrs, |this, attrs| { + let base = this.parse_expr_bottom()?; + let span = this.interpolated_or_expr_span(&base); + this.parse_expr_dot_or_call_with(base, span, attrs) + }) + } + + pub(super) fn parse_expr_dot_or_call_with( + &mut self, + e0: P<Expr>, + lo: Span, + mut attrs: ast::AttrVec, + ) -> PResult<'a, P<Expr>> { + // Stitch the list of outer attributes onto the return value. + // A little bit ugly, but the best way given the current code + // structure + let res = ensure_sufficient_stack( + // this expr demonstrates the recursion it guards against + || self.parse_expr_dot_or_call_with_(e0, lo), + ); + if attrs.is_empty() { + res + } else { + res.map(|expr| { + expr.map(|mut expr| { + attrs.extend(expr.attrs); + expr.attrs = attrs; + expr + }) + }) + } + } + + fn parse_expr_dot_or_call_with_(&mut self, mut e: P<Expr>, lo: Span) -> PResult<'a, P<Expr>> { + loop { + let has_question = + if self.prev_token.kind == TokenKind::Ident(kw::Return, IdentIsRaw::No) { + // we are using noexpect here because we don't expect a `?` directly after a `return` + // which could be suggested otherwise + self.eat_noexpect(&token::Question) + } else { + self.eat(&token::Question) + }; + if has_question { + // `expr?` + e = self.mk_expr(lo.to(self.prev_token.span), ExprKind::Try(e)); + continue; + } + let has_dot = if self.prev_token.kind == TokenKind::Ident(kw::Return, IdentIsRaw::No) { + // we are using noexpect here because we don't expect a `.` directly after a `return` + // which could be suggested otherwise + self.eat_noexpect(&token::Dot) + } else if self.token.kind == TokenKind::RArrow && self.may_recover() { + // Recovery for `expr->suffix`. + self.bump(); + let span = self.prev_token.span; + self.dcx().emit_err(errors::ExprRArrowCall { span }); + true + } else { + self.eat(&token::Dot) + }; + if has_dot { + // expr.f + e = self.parse_dot_suffix_expr(lo, e)?; + continue; + } + if self.expr_is_complete(&e) { + return Ok(e); + } + e = match self.token.kind { + token::OpenDelim(Delimiter::Parenthesis) => self.parse_expr_fn_call(lo, e), + token::OpenDelim(Delimiter::Bracket) => self.parse_expr_index(lo, e)?, + _ => return Ok(e), + } + } + } + + pub fn parse_dot_suffix_expr(&mut self, lo: Span, base: P<Expr>) -> PResult<'a, P<Expr>> { + // At this point we've consumed something like `expr.` and `self.token` holds the token + // after the dot. + match self.token.uninterpolate().kind { + token::Ident(..) => self.parse_dot_suffix(base, lo), + token::Literal(token::Lit { kind: token::Integer, symbol, suffix }) => { + let ident_span = self.token.span; + self.bump(); + Ok(self.mk_expr_tuple_field_access(lo, ident_span, base, symbol, suffix)) + } + token::Literal(token::Lit { kind: token::Float, symbol, suffix }) => { + Ok(match self.break_up_float(symbol, self.token.span) { + // 1e2 + DestructuredFloat::Single(sym, _sp) => { + // `foo.1e2`: a single complete dot access, fully consumed. We end up with + // the `1e2` token in `self.prev_token` and the following token in + // `self.token`. + let ident_span = self.token.span; + self.bump(); + self.mk_expr_tuple_field_access(lo, ident_span, base, sym, suffix) + } + // 1. + DestructuredFloat::TrailingDot(sym, ident_span, dot_span) => { + // `foo.1.`: a single complete dot access and the start of another. + // We end up with the `sym` (`1`) token in `self.prev_token` and a dot in + // `self.token`. + assert!(suffix.is_none()); + self.token = Token::new(token::Ident(sym, IdentIsRaw::No), ident_span); + self.bump_with((Token::new(token::Dot, dot_span), self.token_spacing)); + self.mk_expr_tuple_field_access(lo, ident_span, base, sym, None) + } + // 1.2 | 1.2e3 + DestructuredFloat::MiddleDot( + sym1, + ident1_span, + _dot_span, + sym2, + ident2_span, + ) => { + // `foo.1.2` (or `foo.1.2e3`): two complete dot accesses. We end up with + // the `sym2` (`2` or `2e3`) token in `self.prev_token` and the following + // token in `self.token`. + let next_token2 = + Token::new(token::Ident(sym2, IdentIsRaw::No), ident2_span); + self.bump_with((next_token2, self.token_spacing)); + self.bump(); + let base1 = + self.mk_expr_tuple_field_access(lo, ident1_span, base, sym1, None); + self.mk_expr_tuple_field_access(lo, ident2_span, base1, sym2, suffix) + } + DestructuredFloat::Error => base, + }) + } + _ => { + self.error_unexpected_after_dot(); + Ok(base) + } + } + } + + fn error_unexpected_after_dot(&self) { + let actual = pprust::token_to_string(&self.token); + let span = self.token.span; + let sm = self.psess.source_map(); + let (span, actual) = match (&self.token.kind, self.subparser_name) { + (token::Eof, Some(_)) if let Ok(actual) = sm.span_to_snippet(sm.next_point(span)) => { + (span.shrink_to_hi(), actual.into()) + } + _ => (span, actual), + }; + self.dcx().emit_err(errors::UnexpectedTokenAfterDot { span, actual }); + } + + // We need an identifier or integer, but the next token is a float. + // Break the float into components to extract the identifier or integer. + // FIXME: With current `TokenCursor` it's hard to break tokens into more than 2 + // parts unless those parts are processed immediately. `TokenCursor` should either + // support pushing "future tokens" (would be also helpful to `break_and_eat`), or + // we should break everything including floats into more basic proc-macro style + // tokens in the lexer (probably preferable). + // See also `TokenKind::break_two_token_op` which does similar splitting of `>>` into `>`. + fn break_up_float(&self, float: Symbol, span: Span) -> DestructuredFloat { + #[derive(Debug)] + enum FloatComponent { + IdentLike(String), + Punct(char), + } + use FloatComponent::*; + + let float_str = float.as_str(); + let mut components = Vec::new(); + let mut ident_like = String::new(); + for c in float_str.chars() { + if c == '_' || c.is_ascii_alphanumeric() { + ident_like.push(c); + } else if matches!(c, '.' | '+' | '-') { + if !ident_like.is_empty() { + components.push(IdentLike(mem::take(&mut ident_like))); + } + components.push(Punct(c)); + } else { + panic!("unexpected character in a float token: {c:?}") + } + } + if !ident_like.is_empty() { + components.push(IdentLike(ident_like)); + } + + // With proc macros the span can refer to anything, the source may be too short, + // or too long, or non-ASCII. It only makes sense to break our span into components + // if its underlying text is identical to our float literal. + let can_take_span_apart = + || self.span_to_snippet(span).as_deref() == Ok(float_str).as_deref(); + + match &*components { + // 1e2 + [IdentLike(i)] => { + DestructuredFloat::Single(Symbol::intern(i), span) + } + // 1. + [IdentLike(i), Punct('.')] => { + let (ident_span, dot_span) = if can_take_span_apart() { + let (span, ident_len) = (span.data(), BytePos::from_usize(i.len())); + let ident_span = span.with_hi(span.lo + ident_len); + let dot_span = span.with_lo(span.lo + ident_len); + (ident_span, dot_span) + } else { + (span, span) + }; + let symbol = Symbol::intern(i); + DestructuredFloat::TrailingDot(symbol, ident_span, dot_span) + } + // 1.2 | 1.2e3 + [IdentLike(i1), Punct('.'), IdentLike(i2)] => { + let (ident1_span, dot_span, ident2_span) = if can_take_span_apart() { + let (span, ident1_len) = (span.data(), BytePos::from_usize(i1.len())); + let ident1_span = span.with_hi(span.lo + ident1_len); + let dot_span = span + .with_lo(span.lo + ident1_len) + .with_hi(span.lo + ident1_len + BytePos(1)); + let ident2_span = self.token.span.with_lo(span.lo + ident1_len + BytePos(1)); + (ident1_span, dot_span, ident2_span) + } else { + (span, span, span) + }; + let symbol1 = Symbol::intern(i1); + let symbol2 = Symbol::intern(i2); + DestructuredFloat::MiddleDot(symbol1, ident1_span, dot_span, symbol2, ident2_span) + } + // 1e+ | 1e- (recovered) + [IdentLike(_), Punct('+' | '-')] | + // 1e+2 | 1e-2 + [IdentLike(_), Punct('+' | '-'), IdentLike(_)] | + // 1.2e+ | 1.2e- + [IdentLike(_), Punct('.'), IdentLike(_), Punct('+' | '-')] | + // 1.2e+3 | 1.2e-3 + [IdentLike(_), Punct('.'), IdentLike(_), Punct('+' | '-'), IdentLike(_)] => { + // See the FIXME about `TokenCursor` above. + self.error_unexpected_after_dot(); + DestructuredFloat::Error + } + _ => panic!("unexpected components in a float token: {components:?}"), + } + } + + /// Parse the field access used in offset_of, matched by `$(e:expr)+`. + /// Currently returns a list of idents. However, it should be possible in + /// future to also do array indices, which might be arbitrary expressions. + fn parse_floating_field_access(&mut self) -> PResult<'a, P<[Ident]>> { + let mut fields = Vec::new(); + let mut trailing_dot = None; + + loop { + // This is expected to use a metavariable $(args:expr)+, but the builtin syntax + // could be called directly. Calling `parse_expr` allows this function to only + // consider `Expr`s. + let expr = self.parse_expr()?; + let mut current = &expr; + let start_idx = fields.len(); + loop { + match current.kind { + ExprKind::Field(ref left, right) => { + // Field access is read right-to-left. + fields.insert(start_idx, right); + trailing_dot = None; + current = left; + } + // Parse this both to give helpful error messages and to + // verify it can be done with this parser setup. + ExprKind::Index(ref left, ref _right, span) => { + self.dcx().emit_err(errors::ArrayIndexInOffsetOf(span)); + current = left; + } + ExprKind::Lit(token::Lit { + kind: token::Float | token::Integer, + symbol, + suffix, + }) => { + if let Some(suffix) = suffix { + self.expect_no_tuple_index_suffix(current.span, suffix); + } + match self.break_up_float(symbol, current.span) { + // 1e2 + DestructuredFloat::Single(sym, sp) => { + trailing_dot = None; + fields.insert(start_idx, Ident::new(sym, sp)); + } + // 1. + DestructuredFloat::TrailingDot(sym, sym_span, dot_span) => { + assert!(suffix.is_none()); + trailing_dot = Some(dot_span); + fields.insert(start_idx, Ident::new(sym, sym_span)); + } + // 1.2 | 1.2e3 + DestructuredFloat::MiddleDot( + symbol1, + span1, + _dot_span, + symbol2, + span2, + ) => { + trailing_dot = None; + fields.insert(start_idx, Ident::new(symbol2, span2)); + fields.insert(start_idx, Ident::new(symbol1, span1)); + } + DestructuredFloat::Error => { + trailing_dot = None; + fields.insert(start_idx, Ident::new(symbol, self.prev_token.span)); + } + } + break; + } + ExprKind::Path(None, Path { ref segments, .. }) => { + match &segments[..] { + [PathSegment { ident, args: None, .. }] => { + trailing_dot = None; + fields.insert(start_idx, *ident) + } + _ => { + self.dcx().emit_err(errors::InvalidOffsetOf(current.span)); + break; + } + } + break; + } + _ => { + self.dcx().emit_err(errors::InvalidOffsetOf(current.span)); + break; + } + } + } + + if matches!(self.token.kind, token::CloseDelim(..) | token::Comma) { + break; + } else if trailing_dot.is_none() { + // This loop should only repeat if there is a trailing dot. + self.dcx().emit_err(errors::InvalidOffsetOf(self.token.span)); + break; + } + } + if let Some(dot) = trailing_dot { + self.dcx().emit_err(errors::InvalidOffsetOf(dot)); + } + Ok(fields.into_iter().collect()) + } + + fn mk_expr_tuple_field_access( + &mut self, + lo: Span, + ident_span: Span, + base: P<Expr>, + field: Symbol, + suffix: Option<Symbol>, + ) -> P<Expr> { + if let Some(suffix) = suffix { + self.expect_no_tuple_index_suffix(ident_span, suffix); + } + self.mk_expr(lo.to(ident_span), ExprKind::Field(base, Ident::new(field, ident_span))) + } + + /// Parse a function call expression, `expr(...)`. + fn parse_expr_fn_call(&mut self, lo: Span, fun: P<Expr>) -> P<Expr> { + let snapshot = if self.token.kind == token::OpenDelim(Delimiter::Parenthesis) { + Some((self.create_snapshot_for_diagnostic(), fun.kind.clone())) + } else { + None + }; + let open_paren = self.token.span; + + let seq = self + .parse_expr_paren_seq() + .map(|args| self.mk_expr(lo.to(self.prev_token.span), self.mk_call(fun, args))); + match self.maybe_recover_struct_lit_bad_delims(lo, open_paren, seq, snapshot) { + Ok(expr) => expr, + Err(err) => self.recover_seq_parse_error(Delimiter::Parenthesis, lo, err), + } + } + + /// If we encounter a parser state that looks like the user has written a `struct` literal with + /// parentheses instead of braces, recover the parser state and provide suggestions. + #[instrument(skip(self, seq, snapshot), level = "trace")] + fn maybe_recover_struct_lit_bad_delims( + &mut self, + lo: Span, + open_paren: Span, + seq: PResult<'a, P<Expr>>, + snapshot: Option<(SnapshotParser<'a>, ExprKind)>, + ) -> PResult<'a, P<Expr>> { + match (self.may_recover(), seq, snapshot) { + (true, Err(err), Some((mut snapshot, ExprKind::Path(None, path)))) => { + snapshot.bump(); // `(` + match snapshot.parse_struct_fields(path.clone(), false, Delimiter::Parenthesis) { + Ok((fields, ..)) + if snapshot.eat(&token::CloseDelim(Delimiter::Parenthesis)) => + { + // We are certain we have `Enum::Foo(a: 3, b: 4)`, suggest + // `Enum::Foo { a: 3, b: 4 }` or `Enum::Foo(3, 4)`. + self.restore_snapshot(snapshot); + let close_paren = self.prev_token.span; + let span = lo.to(close_paren); + // filter shorthand fields + let fields: Vec<_> = + fields.into_iter().filter(|field| !field.is_shorthand).collect(); + + let guar = if !fields.is_empty() && + // `token.kind` should not be compared here. + // This is because the `snapshot.token.kind` is treated as the same as + // that of the open delim in `TokenTreesReader::parse_token_tree`, even + // if they are different. + self.span_to_snippet(close_paren).is_ok_and(|snippet| snippet == ")") + { + err.cancel(); + self.dcx() + .create_err(errors::ParenthesesWithStructFields { + span, + r#type: path, + braces_for_struct: errors::BracesForStructLiteral { + first: open_paren, + second: close_paren, + }, + no_fields_for_fn: errors::NoFieldsForFnCall { + fields: fields + .into_iter() + .map(|field| field.span.until(field.expr.span)) + .collect(), + }, + }) + .emit() + } else { + err.emit() + }; + Ok(self.mk_expr_err(span, guar)) + } + Ok(_) => Err(err), + Err(err2) => { + err2.cancel(); + Err(err) + } + } + } + (_, seq, _) => seq, + } + } + + /// Parse an indexing expression `expr[...]`. + fn parse_expr_index(&mut self, lo: Span, base: P<Expr>) -> PResult<'a, P<Expr>> { + let prev_span = self.prev_token.span; + let open_delim_span = self.token.span; + self.bump(); // `[` + let index = self.parse_expr()?; + self.suggest_missing_semicolon_before_array(prev_span, open_delim_span)?; + self.expect(&token::CloseDelim(Delimiter::Bracket))?; + Ok(self.mk_expr( + lo.to(self.prev_token.span), + self.mk_index(base, index, open_delim_span.to(self.prev_token.span)), + )) + } + + /// Assuming we have just parsed `.`, continue parsing into an expression. + fn parse_dot_suffix(&mut self, self_arg: P<Expr>, lo: Span) -> PResult<'a, P<Expr>> { + if self.token.uninterpolated_span().at_least_rust_2018() && self.eat_keyword(kw::Await) { + return Ok(self.mk_await_expr(self_arg, lo)); + } + + // Post-fix match + if self.eat_keyword(kw::Match) { + let match_span = self.prev_token.span; + self.psess.gated_spans.gate(sym::postfix_match, match_span); + return self.parse_match_block(lo, match_span, self_arg, MatchKind::Postfix); + } + + let fn_span_lo = self.token.span; + let mut seg = self.parse_path_segment(PathStyle::Expr, None)?; + self.check_trailing_angle_brackets(&seg, &[&token::OpenDelim(Delimiter::Parenthesis)]); + self.check_turbofish_missing_angle_brackets(&mut seg); + + if self.check(&token::OpenDelim(Delimiter::Parenthesis)) { + // Method call `expr.f()` + let args = self.parse_expr_paren_seq()?; + let fn_span = fn_span_lo.to(self.prev_token.span); + let span = lo.to(self.prev_token.span); + Ok(self.mk_expr( + span, + ExprKind::MethodCall(Box::new(ast::MethodCall { + seg, + receiver: self_arg, + args, + span: fn_span, + })), + )) + } else { + // Field access `expr.f` + if let Some(args) = seg.args { + self.dcx().emit_err(errors::FieldExpressionWithGeneric(args.span())); + } + + let span = lo.to(self.prev_token.span); + Ok(self.mk_expr(span, ExprKind::Field(self_arg, seg.ident))) + } + } + + /// At the bottom (top?) of the precedence hierarchy, + /// Parses things like parenthesized exprs, macros, `return`, etc. + /// + /// N.B., this does not parse outer attributes, and is private because it only works + /// correctly if called from `parse_dot_or_call_expr()`. + fn parse_expr_bottom(&mut self) -> PResult<'a, P<Expr>> { + maybe_recover_from_interpolated_ty_qpath!(self, true); + maybe_whole_expr!(self); + + // Outer attributes are already parsed and will be + // added to the return value after the fact. + + let restrictions = self.restrictions; + self.with_res(restrictions - Restrictions::ALLOW_LET, |this| { + // Note: when adding new syntax here, don't forget to adjust `TokenKind::can_begin_expr()`. + let lo = this.token.span; + if let token::Literal(_) = this.token.kind { + // This match arm is a special-case of the `_` match arm below and + // could be removed without changing functionality, but it's faster + // to have it here, especially for programs with large constants. + this.parse_expr_lit() + } else if this.check(&token::OpenDelim(Delimiter::Parenthesis)) { + this.parse_expr_tuple_parens(restrictions) + } else if this.check(&token::OpenDelim(Delimiter::Brace)) { + this.parse_expr_block(None, lo, BlockCheckMode::Default) + } else if this.check(&token::BinOp(token::Or)) || this.check(&token::OrOr) { + this.parse_expr_closure().map_err(|mut err| { + // If the input is something like `if a { 1 } else { 2 } | if a { 3 } else { 4 }` + // then suggest parens around the lhs. + if let Some(sp) = this.psess.ambiguous_block_expr_parse.borrow().get(&lo) { + err.subdiagnostic(this.dcx(), ExprParenthesesNeeded::surrounding(*sp)); + } + err + }) + } else if this.check(&token::OpenDelim(Delimiter::Bracket)) { + this.parse_expr_array_or_repeat(Delimiter::Bracket) + } else if this.is_builtin() { + this.parse_expr_builtin() + } else if this.check_path() { + this.parse_expr_path_start() + } else if this.check_keyword(kw::Move) + || this.check_keyword(kw::Static) + || this.check_const_closure() + { + this.parse_expr_closure() + } else if this.eat_keyword(kw::If) { + this.parse_expr_if() + } else if this.check_keyword(kw::For) { + if this.choose_generics_over_qpath(1) { + this.parse_expr_closure() + } else { + assert!(this.eat_keyword(kw::For)); + this.parse_expr_for(None, this.prev_token.span) + } + } else if this.eat_keyword(kw::While) { + this.parse_expr_while(None, this.prev_token.span) + } else if let Some(label) = this.eat_label() { + this.parse_expr_labeled(label, true) + } else if this.eat_keyword(kw::Loop) { + let sp = this.prev_token.span; + this.parse_expr_loop(None, this.prev_token.span).map_err(|mut err| { + err.span_label(sp, "while parsing this `loop` expression"); + err + }) + } else if this.eat_keyword(kw::Match) { + let match_sp = this.prev_token.span; + this.parse_expr_match().map_err(|mut err| { + err.span_label(match_sp, "while parsing this `match` expression"); + err + }) + } else if this.eat_keyword(kw::Unsafe) { + let sp = this.prev_token.span; + this.parse_expr_block(None, lo, BlockCheckMode::Unsafe(ast::UserProvided)).map_err( + |mut err| { + err.span_label(sp, "while parsing this `unsafe` expression"); + err + }, + ) + } else if this.check_inline_const(0) { + this.parse_const_block(lo.to(this.token.span), false) + } else if this.may_recover() && this.is_do_catch_block() { + this.recover_do_catch() + } else if this.is_try_block() { + this.expect_keyword(kw::Try)?; + this.parse_try_block(lo) + } else if this.eat_keyword(kw::Return) { + this.parse_expr_return() + } else if this.eat_keyword(kw::Continue) { + this.parse_expr_continue(lo) + } else if this.eat_keyword(kw::Break) { + this.parse_expr_break() + } else if this.eat_keyword(kw::Yield) { + this.parse_expr_yield() + } else if this.is_do_yeet() { + this.parse_expr_yeet() + } else if this.eat_keyword(kw::Become) { + this.parse_expr_become() + } else if this.check_keyword(kw::Let) { + this.parse_expr_let(restrictions) + } else if this.eat_keyword(kw::Underscore) { + Ok(this.mk_expr(this.prev_token.span, ExprKind::Underscore)) + } else if this.token.uninterpolated_span().at_least_rust_2018() { + // `Span::at_least_rust_2018()` is somewhat expensive; don't get it repeatedly. + if this.token.uninterpolated_span().at_least_rust_2024() + // check for `gen {}` and `gen move {}` + // or `async gen {}` and `async gen move {}` + && (this.is_gen_block(kw::Gen, 0) + || (this.check_keyword(kw::Async) && this.is_gen_block(kw::Gen, 1))) + { + // FIXME: (async) gen closures aren't yet parsed. + this.parse_gen_block() + } else if this.check_keyword(kw::Async) { + // FIXME(gen_blocks): Parse `gen async` and suggest swap + if this.is_gen_block(kw::Async, 0) { + // Check for `async {` and `async move {`, + this.parse_gen_block() + } else { + this.parse_expr_closure() + } + } else if this.eat_keyword_noexpect(kw::Await) { + this.recover_incorrect_await_syntax(lo, this.prev_token.span) + } else { + this.parse_expr_lit() + } + } else { + this.parse_expr_lit() + } + }) + } + + fn parse_expr_lit(&mut self) -> PResult<'a, P<Expr>> { + let lo = self.token.span; + match self.parse_opt_token_lit() { + Some((token_lit, _)) => { + let expr = self.mk_expr(lo.to(self.prev_token.span), ExprKind::Lit(token_lit)); + self.maybe_recover_from_bad_qpath(expr) + } + None => self.try_macro_suggestion(), + } + } + + fn parse_expr_tuple_parens(&mut self, restrictions: Restrictions) -> PResult<'a, P<Expr>> { + let lo = self.token.span; + self.expect(&token::OpenDelim(Delimiter::Parenthesis))?; + let (es, trailing_comma) = match self.parse_seq_to_end( + &token::CloseDelim(Delimiter::Parenthesis), + SeqSep::trailing_allowed(token::Comma), + |p| p.parse_expr_catch_underscore(restrictions.intersection(Restrictions::ALLOW_LET)), + ) { + Ok(x) => x, + Err(err) => { + return Ok(self.recover_seq_parse_error(Delimiter::Parenthesis, lo, err)); + } + }; + let kind = if es.len() == 1 && matches!(trailing_comma, Trailing::No) { + // `(e)` is parenthesized `e`. + ExprKind::Paren(es.into_iter().next().unwrap()) + } else { + // `(e,)` is a tuple with only one field, `e`. + ExprKind::Tup(es) + }; + let expr = self.mk_expr(lo.to(self.prev_token.span), kind); + self.maybe_recover_from_bad_qpath(expr) + } + + fn parse_expr_array_or_repeat(&mut self, close_delim: Delimiter) -> PResult<'a, P<Expr>> { + let lo = self.token.span; + self.bump(); // `[` or other open delim + + let close = &token::CloseDelim(close_delim); + let kind = if self.eat(close) { + // Empty vector + ExprKind::Array(ThinVec::new()) + } else { + // Non-empty vector + let first_expr = self.parse_expr()?; + if self.eat(&token::Semi) { + // Repeating array syntax: `[ 0; 512 ]` + let count = self.parse_expr_anon_const()?; + self.expect(close)?; + ExprKind::Repeat(first_expr, count) + } else if self.eat(&token::Comma) { + // Vector with two or more elements. + let sep = SeqSep::trailing_allowed(token::Comma); + let (mut exprs, _) = self.parse_seq_to_end(close, sep, |p| p.parse_expr())?; + exprs.insert(0, first_expr); + ExprKind::Array(exprs) + } else { + // Vector with one element + self.expect(close)?; + ExprKind::Array(thin_vec![first_expr]) + } + }; + let expr = self.mk_expr(lo.to(self.prev_token.span), kind); + self.maybe_recover_from_bad_qpath(expr) + } + + fn parse_expr_path_start(&mut self) -> PResult<'a, P<Expr>> { + let maybe_eq_tok = self.prev_token.clone(); + let (qself, path) = if self.eat_lt() { + let lt_span = self.prev_token.span; + let (qself, path) = self.parse_qpath(PathStyle::Expr).map_err(|mut err| { + // Suggests using '<=' if there is an error parsing qpath when the previous token + // is an '=' token. Only emits suggestion if the '<' token and '=' token are + // directly adjacent (i.e. '=<') + if maybe_eq_tok.kind == TokenKind::Eq && maybe_eq_tok.span.hi() == lt_span.lo() { + let eq_lt = maybe_eq_tok.span.to(lt_span); + err.span_suggestion(eq_lt, "did you mean", "<=", Applicability::Unspecified); + } + err + })?; + (Some(qself), path) + } else { + (None, self.parse_path(PathStyle::Expr)?) + }; + + // `!`, as an operator, is prefix, so we know this isn't that. + let (span, kind) = if self.eat(&token::Not) { + // MACRO INVOCATION expression + if qself.is_some() { + self.dcx().emit_err(errors::MacroInvocationWithQualifiedPath(path.span)); + } + let lo = path.span; + let mac = P(MacCall { path, args: self.parse_delim_args()? }); + (lo.to(self.prev_token.span), ExprKind::MacCall(mac)) + } else if self.check(&token::OpenDelim(Delimiter::Brace)) + && let Some(expr) = self.maybe_parse_struct_expr(&qself, &path) + { + if qself.is_some() { + self.psess.gated_spans.gate(sym::more_qualified_paths, path.span); + } + return expr; + } else { + (path.span, ExprKind::Path(qself, path)) + }; + + let expr = self.mk_expr(span, kind); + self.maybe_recover_from_bad_qpath(expr) + } + + /// Parse `'label: $expr`. The label is already parsed. + pub(super) fn parse_expr_labeled( + &mut self, + label_: Label, + mut consume_colon: bool, + ) -> PResult<'a, P<Expr>> { + let lo = label_.ident.span; + let label = Some(label_); + let ate_colon = self.eat(&token::Colon); + let expr = if self.eat_keyword(kw::While) { + self.parse_expr_while(label, lo) + } else if self.eat_keyword(kw::For) { + self.parse_expr_for(label, lo) + } else if self.eat_keyword(kw::Loop) { + self.parse_expr_loop(label, lo) + } else if self.check_noexpect(&token::OpenDelim(Delimiter::Brace)) + || self.token.is_whole_block() + { + self.parse_expr_block(label, lo, BlockCheckMode::Default) + } else if !ate_colon + && self.may_recover() + && (matches!(self.token.kind, token::CloseDelim(_) | token::Comma) + || self.token.is_punct()) + && could_be_unclosed_char_literal(label_.ident) + { + let (lit, _) = + self.recover_unclosed_char(label_.ident, Parser::mk_token_lit_char, |self_| { + self_.dcx().create_err(errors::UnexpectedTokenAfterLabel { + span: self_.token.span, + remove_label: None, + enclose_in_block: None, + }) + }); + consume_colon = false; + Ok(self.mk_expr(lo, ExprKind::Lit(lit))) + } else if !ate_colon + && (self.check_noexpect(&TokenKind::Comma) || self.check_noexpect(&TokenKind::Gt)) + { + // We're probably inside of a `Path<'a>` that needs a turbofish + let guar = self.dcx().emit_err(errors::UnexpectedTokenAfterLabel { + span: self.token.span, + remove_label: None, + enclose_in_block: None, + }); + consume_colon = false; + Ok(self.mk_expr_err(lo, guar)) + } else { + let mut err = errors::UnexpectedTokenAfterLabel { + span: self.token.span, + remove_label: None, + enclose_in_block: None, + }; + + // Continue as an expression in an effort to recover on `'label: non_block_expr`. + let expr = self.parse_expr().map(|expr| { + let span = expr.span; + + let found_labeled_breaks = { + struct FindLabeledBreaksVisitor; + + impl<'ast> Visitor<'ast> for FindLabeledBreaksVisitor { + type Result = ControlFlow<()>; + fn visit_expr(&mut self, ex: &'ast Expr) -> ControlFlow<()> { + if let ExprKind::Break(Some(_label), _) = ex.kind { + ControlFlow::Break(()) + } else { + walk_expr(self, ex) + } + } + } + + FindLabeledBreaksVisitor.visit_expr(&expr).is_break() + }; + + // Suggestion involves adding a labeled block. + // + // If there are no breaks that may use this label, suggest removing the label and + // recover to the unmodified expression. + if !found_labeled_breaks { + err.remove_label = Some(lo.until(span)); + + return expr; + } + + err.enclose_in_block = Some(errors::UnexpectedTokenAfterLabelSugg { + left: span.shrink_to_lo(), + right: span.shrink_to_hi(), + }); + + // Replace `'label: non_block_expr` with `'label: {non_block_expr}` in order to suppress future errors about `break 'label`. + let stmt = self.mk_stmt(span, StmtKind::Expr(expr)); + let blk = self.mk_block(thin_vec![stmt], BlockCheckMode::Default, span); + self.mk_expr(span, ExprKind::Block(blk, label)) + }); + + self.dcx().emit_err(err); + expr + }?; + + if !ate_colon && consume_colon { + self.dcx().emit_err(errors::RequireColonAfterLabeledExpression { + span: expr.span, + label: lo, + label_end: lo.shrink_to_hi(), + }); + } + + Ok(expr) + } + + /// Emit an error when a char is parsed as a lifetime or label because of a missing quote. + pub(super) fn recover_unclosed_char<L>( + &self, + ident: Ident, + mk_lit_char: impl FnOnce(Symbol, Span) -> L, + err: impl FnOnce(&Self) -> Diag<'a>, + ) -> L { + assert!(could_be_unclosed_char_literal(ident)); + self.dcx() + .try_steal_modify_and_emit_err(ident.span, StashKey::LifetimeIsChar, |err| { + err.span_suggestion_verbose( + ident.span.shrink_to_hi(), + "add `'` to close the char literal", + "'", + Applicability::MaybeIncorrect, + ); + }) + .unwrap_or_else(|| { + err(self) + .with_span_suggestion_verbose( + ident.span.shrink_to_hi(), + "add `'` to close the char literal", + "'", + Applicability::MaybeIncorrect, + ) + .emit() + }); + let name = ident.without_first_quote().name; + mk_lit_char(name, ident.span) + } + + /// Recover on the syntax `do catch { ... }` suggesting `try { ... }` instead. + fn recover_do_catch(&mut self) -> PResult<'a, P<Expr>> { + let lo = self.token.span; + + self.bump(); // `do` + self.bump(); // `catch` + + let span = lo.to(self.prev_token.span); + self.dcx().emit_err(errors::DoCatchSyntaxRemoved { span }); + + self.parse_try_block(lo) + } + + /// Parse an expression if the token can begin one. + fn parse_expr_opt(&mut self) -> PResult<'a, Option<P<Expr>>> { + Ok(if self.token.can_begin_expr() { Some(self.parse_expr()?) } else { None }) + } + + /// Parse `"return" expr?`. + fn parse_expr_return(&mut self) -> PResult<'a, P<Expr>> { + let lo = self.prev_token.span; + let kind = ExprKind::Ret(self.parse_expr_opt()?); + let expr = self.mk_expr(lo.to(self.prev_token.span), kind); + self.maybe_recover_from_bad_qpath(expr) + } + + /// Parse `"do" "yeet" expr?`. + fn parse_expr_yeet(&mut self) -> PResult<'a, P<Expr>> { + let lo = self.token.span; + + self.bump(); // `do` + self.bump(); // `yeet` + + let kind = ExprKind::Yeet(self.parse_expr_opt()?); + + let span = lo.to(self.prev_token.span); + self.psess.gated_spans.gate(sym::yeet_expr, span); + let expr = self.mk_expr(span, kind); + self.maybe_recover_from_bad_qpath(expr) + } + + /// Parse `"become" expr`, with `"become"` token already eaten. + fn parse_expr_become(&mut self) -> PResult<'a, P<Expr>> { + let lo = self.prev_token.span; + let kind = ExprKind::Become(self.parse_expr()?); + let span = lo.to(self.prev_token.span); + self.psess.gated_spans.gate(sym::explicit_tail_calls, span); + let expr = self.mk_expr(span, kind); + self.maybe_recover_from_bad_qpath(expr) + } + + /// Parse `"break" (('label (:? expr)?) | expr?)` with `"break"` token already eaten. + /// If the label is followed immediately by a `:` token, the label and `:` are + /// parsed as part of the expression (i.e. a labeled loop). The language team has + /// decided in #87026 to require parentheses as a visual aid to avoid confusion if + /// the break expression of an unlabeled break is a labeled loop (as in + /// `break 'lbl: loop {}`); a labeled break with an unlabeled loop as its value + /// expression only gets a warning for compatibility reasons; and a labeled break + /// with a labeled loop does not even get a warning because there is no ambiguity. + fn parse_expr_break(&mut self) -> PResult<'a, P<Expr>> { + let lo = self.prev_token.span; + let mut label = self.eat_label(); + let kind = if self.token == token::Colon + && let Some(label) = label.take() + { + // The value expression can be a labeled loop, see issue #86948, e.g.: + // `loop { break 'label: loop { break 'label 42; }; }` + let lexpr = self.parse_expr_labeled(label, true)?; + self.dcx().emit_err(errors::LabeledLoopInBreak { + span: lexpr.span, + sub: errors::WrapInParentheses::Expression { + left: lexpr.span.shrink_to_lo(), + right: lexpr.span.shrink_to_hi(), + }, + }); + Some(lexpr) + } else if self.token != token::OpenDelim(Delimiter::Brace) + || !self.restrictions.contains(Restrictions::NO_STRUCT_LITERAL) + { + let mut expr = self.parse_expr_opt()?; + if let Some(expr) = &mut expr { + if label.is_some() + && matches!( + expr.kind, + ExprKind::While(_, _, None) + | ExprKind::ForLoop { label: None, .. } + | ExprKind::Loop(_, None, _) + | ExprKind::Block(_, None) + ) + { + self.psess.buffer_lint_with_diagnostic( + BREAK_WITH_LABEL_AND_LOOP, + lo.to(expr.span), + ast::CRATE_NODE_ID, + "this labeled break expression is easy to confuse with an unlabeled break with a labeled value expression", + BuiltinLintDiag::BreakWithLabelAndLoop(expr.span), + ); + } + + // Recover `break label aaaaa` + if self.may_recover() + && let ExprKind::Path(None, p) = &expr.kind + && let [segment] = &*p.segments + && let &ast::PathSegment { ident, args: None, .. } = segment + && let Some(next) = self.parse_expr_opt()? + { + label = Some(self.recover_ident_into_label(ident)); + *expr = next; + } + } + + expr + } else { + None + }; + let expr = self.mk_expr(lo.to(self.prev_token.span), ExprKind::Break(label, kind)); + self.maybe_recover_from_bad_qpath(expr) + } + + /// Parse `"continue" label?`. + fn parse_expr_continue(&mut self, lo: Span) -> PResult<'a, P<Expr>> { + let mut label = self.eat_label(); + + // Recover `continue label` -> `continue 'label` + if self.may_recover() + && label.is_none() + && let Some((ident, _)) = self.token.ident() + { + self.bump(); + label = Some(self.recover_ident_into_label(ident)); + } + + let kind = ExprKind::Continue(label); + Ok(self.mk_expr(lo.to(self.prev_token.span), kind)) + } + + /// Parse `"yield" expr?`. + fn parse_expr_yield(&mut self) -> PResult<'a, P<Expr>> { + let lo = self.prev_token.span; + let kind = ExprKind::Yield(self.parse_expr_opt()?); + let span = lo.to(self.prev_token.span); + self.psess.gated_spans.gate(sym::yield_expr, span); + let expr = self.mk_expr(span, kind); + self.maybe_recover_from_bad_qpath(expr) + } + + /// Parse `builtin # ident(args,*)`. + fn parse_expr_builtin(&mut self) -> PResult<'a, P<Expr>> { + self.parse_builtin(|this, lo, ident| { + Ok(match ident.name { + sym::offset_of => Some(this.parse_expr_offset_of(lo)?), + sym::type_ascribe => Some(this.parse_expr_type_ascribe(lo)?), + _ => None, + }) + }) + } + + pub(crate) fn parse_builtin<T>( + &mut self, + parse: impl FnOnce(&mut Parser<'a>, Span, Ident) -> PResult<'a, Option<T>>, + ) -> PResult<'a, T> { + let lo = self.token.span; + + self.bump(); // `builtin` + self.bump(); // `#` + + let Some((ident, IdentIsRaw::No)) = self.token.ident() else { + let err = self.dcx().create_err(errors::ExpectedBuiltinIdent { span: self.token.span }); + return Err(err); + }; + self.psess.gated_spans.gate(sym::builtin_syntax, ident.span); + self.bump(); + + self.expect(&TokenKind::OpenDelim(Delimiter::Parenthesis))?; + let ret = if let Some(res) = parse(self, lo, ident)? { + Ok(res) + } else { + let err = self.dcx().create_err(errors::UnknownBuiltinConstruct { + span: lo.to(ident.span), + name: ident.name, + }); + return Err(err); + }; + self.expect(&TokenKind::CloseDelim(Delimiter::Parenthesis))?; + + ret + } + + /// Built-in macro for `offset_of!` expressions. + pub(crate) fn parse_expr_offset_of(&mut self, lo: Span) -> PResult<'a, P<Expr>> { + let container = self.parse_ty()?; + self.expect(&TokenKind::Comma)?; + + let fields = self.parse_floating_field_access()?; + let trailing_comma = self.eat_noexpect(&TokenKind::Comma); + + if let Err(mut e) = + self.expect_one_of(&[], &[TokenKind::CloseDelim(Delimiter::Parenthesis)]) + { + if trailing_comma { + e.note("unexpected third argument to offset_of"); + } else { + e.note("offset_of expects dot-separated field and variant names"); + } + e.emit(); + } + + // Eat tokens until the macro call ends. + if self.may_recover() { + while !matches!(self.token.kind, token::CloseDelim(..) | token::Eof) { + self.bump(); + } + } + + let span = lo.to(self.token.span); + Ok(self.mk_expr(span, ExprKind::OffsetOf(container, fields))) + } + + /// Built-in macro for type ascription expressions. + pub(crate) fn parse_expr_type_ascribe(&mut self, lo: Span) -> PResult<'a, P<Expr>> { + let expr = self.parse_expr()?; + self.expect(&token::Comma)?; + let ty = self.parse_ty()?; + let span = lo.to(self.token.span); + Ok(self.mk_expr(span, ExprKind::Type(expr, ty))) + } + + /// Returns a string literal if the next token is a string literal. + /// In case of error returns `Some(lit)` if the next token is a literal with a wrong kind, + /// and returns `None` if the next token is not literal at all. + pub fn parse_str_lit(&mut self) -> Result<ast::StrLit, Option<MetaItemLit>> { + match self.parse_opt_meta_item_lit() { + Some(lit) => match lit.kind { + ast::LitKind::Str(symbol_unescaped, style) => Ok(ast::StrLit { + style, + symbol: lit.symbol, + suffix: lit.suffix, + span: lit.span, + symbol_unescaped, + }), + _ => Err(Some(lit)), + }, + None => Err(None), + } + } + + pub(crate) fn mk_token_lit_char(name: Symbol, span: Span) -> (token::Lit, Span) { + (token::Lit { symbol: name, suffix: None, kind: token::Char }, span) + } + + fn mk_meta_item_lit_char(name: Symbol, span: Span) -> MetaItemLit { + ast::MetaItemLit { + symbol: name, + suffix: None, + kind: ast::LitKind::Char(name.as_str().chars().next().unwrap_or('_')), + span, + } + } + + fn handle_missing_lit<L>( + &mut self, + mk_lit_char: impl FnOnce(Symbol, Span) -> L, + ) -> PResult<'a, L> { + let token = self.token.clone(); + let err = |self_: &Self| { + let msg = format!("unexpected token: {}", super::token_descr(&token)); + self_.dcx().struct_span_err(token.span, msg) + }; + // On an error path, eagerly consider a lifetime to be an unclosed character lit, if that + // makes sense. + if let Some(ident) = self.token.lifetime() + && could_be_unclosed_char_literal(ident) + { + let lt = self.expect_lifetime(); + Ok(self.recover_unclosed_char(lt.ident, mk_lit_char, err)) + } else { + Err(err(self)) + } + } + + pub(super) fn parse_token_lit(&mut self) -> PResult<'a, (token::Lit, Span)> { + self.parse_opt_token_lit() + .ok_or(()) + .or_else(|()| self.handle_missing_lit(Parser::mk_token_lit_char)) + } + + pub(super) fn parse_meta_item_lit(&mut self) -> PResult<'a, MetaItemLit> { + self.parse_opt_meta_item_lit() + .ok_or(()) + .or_else(|()| self.handle_missing_lit(Parser::mk_meta_item_lit_char)) + } + + fn recover_after_dot(&mut self) -> Option<Token> { + let mut recovered = None; + if self.token == token::Dot { + // Attempt to recover `.4` as `0.4`. We don't currently have any syntax where + // dot would follow an optional literal, so we do this unconditionally. + recovered = self.look_ahead(1, |next_token| { + if let token::Literal(token::Lit { kind: token::Integer, symbol, suffix }) = + next_token.kind + { + // If this integer looks like a float, then recover as such. + // + // We will never encounter the exponent part of a floating + // point literal here, since there's no use of the exponent + // syntax that also constitutes a valid integer, so we need + // not check for that. + if suffix.map_or(true, |s| s == sym::f32 || s == sym::f64) + && symbol.as_str().chars().all(|c| c.is_numeric() || c == '_') + && self.token.span.hi() == next_token.span.lo() + { + let s = String::from("0.") + symbol.as_str(); + let kind = TokenKind::lit(token::Float, Symbol::intern(&s), suffix); + return Some(Token::new(kind, self.token.span.to(next_token.span))); + } + } + None + }); + if let Some(token) = &recovered { + self.bump(); + self.dcx().emit_err(errors::FloatLiteralRequiresIntegerPart { + span: token.span, + correct: pprust::token_to_string(token).into_owned(), + }); + } + } + + recovered + } + + /// Matches `lit = true | false | token_lit`. + /// Returns `None` if the next token is not a literal. + pub(super) fn parse_opt_token_lit(&mut self) -> Option<(token::Lit, Span)> { + let recovered = self.recover_after_dot(); + let token = recovered.as_ref().unwrap_or(&self.token); + let span = token.span; + + token::Lit::from_token(token).map(|token_lit| { + self.bump(); + (token_lit, span) + }) + } + + /// Matches `lit = true | false | token_lit`. + /// Returns `None` if the next token is not a literal. + pub(super) fn parse_opt_meta_item_lit(&mut self) -> Option<MetaItemLit> { + let recovered = self.recover_after_dot(); + let token = recovered.as_ref().unwrap_or(&self.token); + match token::Lit::from_token(token) { + Some(lit) => { + match MetaItemLit::from_token_lit(lit, token.span) { + Ok(lit) => { + self.bump(); + Some(lit) + } + Err(err) => { + let span = token.uninterpolated_span(); + self.bump(); + let guar = report_lit_error(self.psess, err, lit, span); + // Pack possible quotes and prefixes from the original literal into + // the error literal's symbol so they can be pretty-printed faithfully. + let suffixless_lit = token::Lit::new(lit.kind, lit.symbol, None); + let symbol = Symbol::intern(&suffixless_lit.to_string()); + let lit = token::Lit::new(token::Err(guar), symbol, lit.suffix); + Some( + MetaItemLit::from_token_lit(lit, span) + .unwrap_or_else(|_| unreachable!()), + ) + } + } + } + None => None, + } + } + + pub(super) fn expect_no_tuple_index_suffix(&self, span: Span, suffix: Symbol) { + if [sym::i32, sym::u32, sym::isize, sym::usize].contains(&suffix) { + // #59553: warn instead of reject out of hand to allow the fix to percolate + // through the ecosystem when people fix their macros + self.dcx().emit_warn(errors::InvalidLiteralSuffixOnTupleIndex { + span, + suffix, + exception: Some(()), + }); + } else { + self.dcx().emit_err(errors::InvalidLiteralSuffixOnTupleIndex { + span, + suffix, + exception: None, + }); + } + } + + /// Matches `'-' lit | lit` (cf. `ast_validation::AstValidator::check_expr_within_pat`). + /// Keep this in sync with `Token::can_begin_literal_maybe_minus`. + pub fn parse_literal_maybe_minus(&mut self) -> PResult<'a, P<Expr>> { + maybe_whole_expr!(self); + + let lo = self.token.span; + let minus_present = self.eat(&token::BinOp(token::Minus)); + let (token_lit, span) = self.parse_token_lit()?; + let expr = self.mk_expr(span, ExprKind::Lit(token_lit)); + + if minus_present { + Ok(self.mk_expr(lo.to(self.prev_token.span), self.mk_unary(UnOp::Neg, expr))) + } else { + Ok(expr) + } + } + + fn is_array_like_block(&mut self) -> bool { + self.look_ahead(1, |t| matches!(t.kind, TokenKind::Ident(..) | TokenKind::Literal(_))) + && self.look_ahead(2, |t| t == &token::Comma) + && self.look_ahead(3, |t| t.can_begin_expr()) + } + + /// Emits a suggestion if it looks like the user meant an array but + /// accidentally used braces, causing the code to be interpreted as a block + /// expression. + fn maybe_suggest_brackets_instead_of_braces(&mut self, lo: Span) -> Option<P<Expr>> { + let mut snapshot = self.create_snapshot_for_diagnostic(); + match snapshot.parse_expr_array_or_repeat(Delimiter::Brace) { + Ok(arr) => { + let guar = self.dcx().emit_err(errors::ArrayBracketsInsteadOfSpaces { + span: arr.span, + sub: errors::ArrayBracketsInsteadOfSpacesSugg { + left: lo, + right: snapshot.prev_token.span, + }, + }); + + self.restore_snapshot(snapshot); + Some(self.mk_expr_err(arr.span, guar)) + } + Err(e) => { + e.cancel(); + None + } + } + } + + fn suggest_missing_semicolon_before_array( + &self, + prev_span: Span, + open_delim_span: Span, + ) -> PResult<'a, ()> { + if !self.may_recover() { + return Ok(()); + } + + if self.token.kind == token::Comma { + if !self.psess.source_map().is_multiline(prev_span.until(self.token.span)) { + return Ok(()); + } + let mut snapshot = self.create_snapshot_for_diagnostic(); + snapshot.bump(); + match snapshot.parse_seq_to_before_end( + &token::CloseDelim(Delimiter::Bracket), + SeqSep::trailing_allowed(token::Comma), + |p| p.parse_expr(), + ) { + Ok(_) + // When the close delim is `)`, `token.kind` is expected to be `token::CloseDelim(Delimiter::Parenthesis)`, + // but the actual `token.kind` is `token::CloseDelim(Delimiter::Bracket)`. + // This is because the `token.kind` of the close delim is treated as the same as + // that of the open delim in `TokenTreesReader::parse_token_tree`, even if the delimiters of them are different. + // Therefore, `token.kind` should not be compared here. + if snapshot + .span_to_snippet(snapshot.token.span) + .is_ok_and(|snippet| snippet == "]") => + { + return Err(self.dcx().create_err(errors::MissingSemicolonBeforeArray { + open_delim: open_delim_span, + semicolon: prev_span.shrink_to_hi(), + })); + } + Ok(_) => (), + Err(err) => err.cancel(), + } + } + Ok(()) + } + + /// Parses a block or unsafe block. + pub(super) fn parse_expr_block( + &mut self, + opt_label: Option<Label>, + lo: Span, + blk_mode: BlockCheckMode, + ) -> PResult<'a, P<Expr>> { + if self.may_recover() && self.is_array_like_block() { + if let Some(arr) = self.maybe_suggest_brackets_instead_of_braces(lo) { + return Ok(arr); + } + } + + if self.token.is_whole_block() { + self.dcx().emit_err(errors::InvalidBlockMacroSegment { + span: self.token.span, + context: lo.to(self.token.span), + wrap: errors::WrapInExplicitBlock { + lo: self.token.span.shrink_to_lo(), + hi: self.token.span.shrink_to_hi(), + }, + }); + } + + let (attrs, blk) = self.parse_block_common(lo, blk_mode, true)?; + Ok(self.mk_expr_with_attrs(blk.span, ExprKind::Block(blk, opt_label), attrs)) + } + + /// Parse a block which takes no attributes and has no label + fn parse_simple_block(&mut self) -> PResult<'a, P<Expr>> { + let blk = self.parse_block()?; + Ok(self.mk_expr(blk.span, ExprKind::Block(blk, None))) + } + + /// Parses a closure expression (e.g., `move |args| expr`). + fn parse_expr_closure(&mut self) -> PResult<'a, P<Expr>> { + let lo = self.token.span; + + let before = self.prev_token.clone(); + let binder = if self.check_keyword(kw::For) { + let lo = self.token.span; + let lifetime_defs = self.parse_late_bound_lifetime_defs()?; + let span = lo.to(self.prev_token.span); + + self.psess.gated_spans.gate(sym::closure_lifetime_binder, span); + + ClosureBinder::For { span, generic_params: lifetime_defs } + } else { + ClosureBinder::NotPresent + }; + + let constness = self.parse_closure_constness(); + + let movability = + if self.eat_keyword(kw::Static) { Movability::Static } else { Movability::Movable }; + + let coroutine_kind = if self.token.uninterpolated_span().at_least_rust_2018() { + self.parse_coroutine_kind(Case::Sensitive) + } else { + None + }; + + let capture_clause = self.parse_capture_clause()?; + let (fn_decl, fn_arg_span) = self.parse_fn_block_decl()?; + let decl_hi = self.prev_token.span; + let mut body = match fn_decl.output { + FnRetTy::Default(_) => { + let restrictions = + self.restrictions - Restrictions::STMT_EXPR - Restrictions::ALLOW_LET; + let prev = self.prev_token.clone(); + let token = self.token.clone(); + match self.parse_expr_res(restrictions, None) { + Ok(expr) => expr, + Err(err) => self.recover_closure_body(err, before, prev, token, lo, decl_hi)?, + } + } + _ => { + // If an explicit return type is given, require a block to appear (RFC 968). + let body_lo = self.token.span; + self.parse_expr_block(None, body_lo, BlockCheckMode::Default)? + } + }; + + match coroutine_kind { + Some(CoroutineKind::Async { span, .. }) => { + // Feature-gate `async ||` closures. + self.psess.gated_spans.gate(sym::async_closure, span); + } + Some(CoroutineKind::Gen { span, .. }) | Some(CoroutineKind::AsyncGen { span, .. }) => { + // Feature-gate `gen ||` and `async gen ||` closures. + // FIXME(gen_blocks): This perhaps should be a different gate. + self.psess.gated_spans.gate(sym::gen_blocks, span); + } + None => {} + } + + if self.token.kind == TokenKind::Semi + && matches!(self.token_cursor.stack.last(), Some((.., Delimiter::Parenthesis))) + && self.may_recover() + { + // It is likely that the closure body is a block but where the + // braces have been removed. We will recover and eat the next + // statements later in the parsing process. + body = self.mk_expr_err( + body.span, + self.dcx().span_delayed_bug(body.span, "recovered a closure body as a block"), + ); + } + + let body_span = body.span; + + let closure = self.mk_expr( + lo.to(body.span), + ExprKind::Closure(Box::new(ast::Closure { + binder, + capture_clause, + constness, + coroutine_kind, + movability, + fn_decl, + body, + fn_decl_span: lo.to(decl_hi), + fn_arg_span, + })), + ); + + // Disable recovery for closure body + let spans = + ClosureSpans { whole_closure: closure.span, closing_pipe: decl_hi, body: body_span }; + self.current_closure = Some(spans); + + Ok(closure) + } + + /// Parses an optional `move` prefix to a closure-like construct. + fn parse_capture_clause(&mut self) -> PResult<'a, CaptureBy> { + if self.eat_keyword(kw::Move) { + let move_kw_span = self.prev_token.span; + // Check for `move async` and recover + if self.check_keyword(kw::Async) { + let move_async_span = self.token.span.with_lo(self.prev_token.span.data().lo); + Err(self + .dcx() + .create_err(errors::AsyncMoveOrderIncorrect { span: move_async_span })) + } else { + Ok(CaptureBy::Value { move_kw: move_kw_span }) + } + } else { + Ok(CaptureBy::Ref) + } + } + + /// Parses the `|arg, arg|` header of a closure. + fn parse_fn_block_decl(&mut self) -> PResult<'a, (P<FnDecl>, Span)> { + let arg_start = self.token.span.lo(); + + let inputs = if self.eat(&token::OrOr) { + ThinVec::new() + } else { + self.expect(&token::BinOp(token::Or))?; + let args = self + .parse_seq_to_before_tokens( + &[&token::BinOp(token::Or), &token::OrOr], + SeqSep::trailing_allowed(token::Comma), + TokenExpectType::NoExpect, + |p| p.parse_fn_block_param(), + )? + .0; + self.expect_or()?; + args + }; + let arg_span = self.prev_token.span.with_lo(arg_start); + let output = + self.parse_ret_ty(AllowPlus::Yes, RecoverQPath::Yes, RecoverReturnSign::Yes)?; + + Ok((P(FnDecl { inputs, output }), arg_span)) + } + + /// Parses a parameter in a closure header (e.g., `|arg, arg|`). + fn parse_fn_block_param(&mut self) -> PResult<'a, Param> { + let lo = self.token.span; + let attrs = self.parse_outer_attributes()?; + self.collect_tokens_trailing_token(attrs, ForceCollect::No, |this, attrs| { + let pat = this.parse_pat_no_top_alt(Some(Expected::ParameterName), None)?; + let ty = if this.eat(&token::Colon) { + this.parse_ty()? + } else { + this.mk_ty(pat.span, TyKind::Infer) + }; + + Ok(( + Param { + attrs, + ty, + pat, + span: lo.to(this.prev_token.span), + id: DUMMY_NODE_ID, + is_placeholder: false, + }, + TrailingToken::MaybeComma, + )) + }) + } + + /// Parses an `if` expression (`if` token already eaten). + fn parse_expr_if(&mut self) -> PResult<'a, P<Expr>> { + let lo = self.prev_token.span; + let cond = self.parse_expr_cond()?; + self.parse_if_after_cond(lo, cond) + } + + fn parse_if_after_cond(&mut self, lo: Span, mut cond: P<Expr>) -> PResult<'a, P<Expr>> { + let cond_span = cond.span; + // Tries to interpret `cond` as either a missing expression if it's a block, + // or as an unfinished expression if it's a binop and the RHS is a block. + // We could probably add more recoveries here too... + let mut recover_block_from_condition = |this: &mut Self| { + let block = match &mut cond.kind { + ExprKind::Binary(Spanned { span: binop_span, .. }, _, right) + if let ExprKind::Block(_, None) = right.kind => + { + let guar = this.dcx().emit_err(errors::IfExpressionMissingThenBlock { + if_span: lo, + missing_then_block_sub: + errors::IfExpressionMissingThenBlockSub::UnfinishedCondition( + cond_span.shrink_to_lo().to(*binop_span), + ), + let_else_sub: None, + }); + std::mem::replace(right, this.mk_expr_err(binop_span.shrink_to_hi(), guar)) + } + ExprKind::Block(_, None) => { + let guar = this.dcx().emit_err(errors::IfExpressionMissingCondition { + if_span: lo.with_neighbor(cond.span).shrink_to_hi(), + block_span: self.psess.source_map().start_point(cond_span), + }); + std::mem::replace(&mut cond, this.mk_expr_err(cond_span.shrink_to_hi(), guar)) + } + _ => { + return None; + } + }; + if let ExprKind::Block(block, _) = &block.kind { + Some(block.clone()) + } else { + unreachable!() + } + }; + // Parse then block + let thn = if self.token.is_keyword(kw::Else) { + if let Some(block) = recover_block_from_condition(self) { + block + } else { + let let_else_sub = matches!(cond.kind, ExprKind::Let(..)) + .then(|| errors::IfExpressionLetSomeSub { if_span: lo.until(cond_span) }); + + let guar = self.dcx().emit_err(errors::IfExpressionMissingThenBlock { + if_span: lo, + missing_then_block_sub: errors::IfExpressionMissingThenBlockSub::AddThenBlock( + cond_span.shrink_to_hi(), + ), + let_else_sub, + }); + self.mk_block_err(cond_span.shrink_to_hi(), guar) + } + } else { + let attrs = self.parse_outer_attributes()?; // For recovery. + let maybe_fatarrow = self.token.clone(); + let block = if self.check(&token::OpenDelim(Delimiter::Brace)) { + self.parse_block()? + } else { + if let Some(block) = recover_block_from_condition(self) { + block + } else { + self.error_on_extra_if(&cond)?; + // Parse block, which will always fail, but we can add a nice note to the error + self.parse_block().map_err(|mut err| { + if self.prev_token == token::Semi + && self.token == token::AndAnd + && let maybe_let = self.look_ahead(1, |t| t.clone()) + && maybe_let.is_keyword(kw::Let) + { + err.span_suggestion( + self.prev_token.span, + "consider removing this semicolon to parse the `let` as part of the same chain", + "", + Applicability::MachineApplicable, + ).span_note( + self.token.span.to(maybe_let.span), + "you likely meant to continue parsing the let-chain starting here", + ); + } else { + // Look for usages of '=>' where '>=' might be intended + if maybe_fatarrow.kind == token::FatArrow { + err.span_suggestion( + maybe_fatarrow.span, + "you might have meant to write a \"greater than or equal to\" comparison", + ">=", + Applicability::MaybeIncorrect, + ); + } + err.span_note( + cond_span, + "the `if` expression is missing a block after this condition", + ); + } + err + })? + } + }; + self.error_on_if_block_attrs(lo, false, block.span, attrs); + block + }; + let els = if self.eat_keyword(kw::Else) { Some(self.parse_expr_else()?) } else { None }; + Ok(self.mk_expr(lo.to(self.prev_token.span), ExprKind::If(cond, thn, els))) + } + + /// Parses the condition of a `if` or `while` expression. + fn parse_expr_cond(&mut self) -> PResult<'a, P<Expr>> { + let mut cond = + self.parse_expr_res(Restrictions::NO_STRUCT_LITERAL | Restrictions::ALLOW_LET, None)?; + + CondChecker::new(self).visit_expr(&mut cond); + + if let ExprKind::Let(_, _, _, Recovered::No) = cond.kind { + // Remove the last feature gating of a `let` expression since it's stable. + self.psess.gated_spans.ungate_last(sym::let_chains, cond.span); + } + + Ok(cond) + } + + /// Parses a `let $pat = $expr` pseudo-expression. + fn parse_expr_let(&mut self, restrictions: Restrictions) -> PResult<'a, P<Expr>> { + let recovered = if !restrictions.contains(Restrictions::ALLOW_LET) { + let err = errors::ExpectedExpressionFoundLet { + span: self.token.span, + reason: ForbiddenLetReason::OtherForbidden, + missing_let: None, + comparison: None, + }; + if self.prev_token.kind == token::BinOp(token::Or) { + // This was part of a closure, the that part of the parser recover. + return Err(self.dcx().create_err(err)); + } else { + Recovered::Yes(self.dcx().emit_err(err)) + } + } else { + Recovered::No + }; + self.bump(); // Eat `let` token + let lo = self.prev_token.span; + let pat = self.parse_pat_allow_top_alt( + None, + RecoverComma::Yes, + RecoverColon::Yes, + CommaRecoveryMode::LikelyTuple, + )?; + if self.token == token::EqEq { + self.dcx().emit_err(errors::ExpectedEqForLetExpr { + span: self.token.span, + sugg_span: self.token.span, + }); + self.bump(); + } else { + self.expect(&token::Eq)?; + } + let expr = self.parse_expr_assoc_with(1 + prec_let_scrutinee_needs_par(), None.into())?; + let span = lo.to(expr.span); + Ok(self.mk_expr(span, ExprKind::Let(pat, expr, span, recovered))) + } + + /// Parses an `else { ... }` expression (`else` token already eaten). + fn parse_expr_else(&mut self) -> PResult<'a, P<Expr>> { + let else_span = self.prev_token.span; // `else` + let attrs = self.parse_outer_attributes()?; // For recovery. + let expr = if self.eat_keyword(kw::If) { + ensure_sufficient_stack(|| self.parse_expr_if())? + } else if self.check(&TokenKind::OpenDelim(Delimiter::Brace)) { + self.parse_simple_block()? + } else { + let snapshot = self.create_snapshot_for_diagnostic(); + let first_tok = super::token_descr(&self.token); + let first_tok_span = self.token.span; + match self.parse_expr() { + Ok(cond) + // Try to guess the difference between a "condition-like" vs + // "statement-like" expression. + // + // We are seeing the following code, in which $cond is neither + // ExprKind::Block nor ExprKind::If (the 2 cases wherein this + // would be valid syntax). + // + // if ... { + // } else $cond + // + // If $cond is "condition-like" such as ExprKind::Binary, we + // want to suggest inserting `if`. + // + // if ... { + // } else if a == b { + // ^^ + // } + // + // If $cond is "statement-like" such as ExprKind::While then we + // want to suggest wrapping in braces. + // + // if ... { + // } else { + // ^ + // while true {} + // } + // ^ + if self.check(&TokenKind::OpenDelim(Delimiter::Brace)) + && classify::expr_requires_semi_to_be_stmt(&cond) => + { + self.dcx().emit_err(errors::ExpectedElseBlock { + first_tok_span, + first_tok, + else_span, + condition_start: cond.span.shrink_to_lo(), + }); + self.parse_if_after_cond(cond.span.shrink_to_lo(), cond)? + } + Err(e) => { + e.cancel(); + self.restore_snapshot(snapshot); + self.parse_simple_block()? + }, + Ok(_) => { + self.restore_snapshot(snapshot); + self.parse_simple_block()? + }, + } + }; + self.error_on_if_block_attrs(else_span, true, expr.span, attrs); + Ok(expr) + } + + fn error_on_if_block_attrs( + &self, + ctx_span: Span, + is_ctx_else: bool, + branch_span: Span, + attrs: AttrWrapper, + ) { + if !attrs.is_empty() + && let [x0 @ xn] | [x0, .., xn] = &*attrs.take_for_recovery(self.psess) + { + let attributes = x0.span.to(xn.span); + let last = xn.span; + let ctx = if is_ctx_else { "else" } else { "if" }; + self.dcx().emit_err(errors::OuterAttributeNotAllowedOnIfElse { + last, + branch_span, + ctx_span, + ctx: ctx.to_string(), + attributes, + }); + } + } + + fn error_on_extra_if(&mut self, cond: &P<Expr>) -> PResult<'a, ()> { + if let ExprKind::Binary(Spanned { span: binop_span, node: binop }, _, right) = &cond.kind + && let BinOpKind::And = binop + && let ExprKind::If(cond, ..) = &right.kind + { + Err(self.dcx().create_err(errors::UnexpectedIfWithIf( + binop_span.shrink_to_hi().to(cond.span.shrink_to_lo()), + ))) + } else { + Ok(()) + } + } + + fn parse_for_head(&mut self) -> PResult<'a, (P<Pat>, P<Expr>)> { + let begin_paren = if self.token.kind == token::OpenDelim(Delimiter::Parenthesis) { + // Record whether we are about to parse `for (`. + // This is used below for recovery in case of `for ( $stuff ) $block` + // in which case we will suggest `for $stuff $block`. + let start_span = self.token.span; + let left = self.prev_token.span.between(self.look_ahead(1, |t| t.span)); + Some((start_span, left)) + } else { + None + }; + // Try to parse the pattern `for ($PAT) in $EXPR`. + let pat = match ( + self.parse_pat_allow_top_alt( + None, + RecoverComma::Yes, + RecoverColon::Yes, + CommaRecoveryMode::LikelyTuple, + ), + begin_paren, + ) { + (Ok(pat), _) => pat, // Happy path. + (Err(err), Some((start_span, left))) if self.eat_keyword(kw::In) => { + // We know for sure we have seen `for ($SOMETHING in`. In the happy path this would + // happen right before the return of this method. + let expr = match self.parse_expr_res(Restrictions::NO_STRUCT_LITERAL, None) { + Ok(expr) => expr, + Err(expr_err) => { + // We don't know what followed the `in`, so cancel and bubble up the + // original error. + expr_err.cancel(); + return Err(err); + } + }; + return if self.token.kind == token::CloseDelim(Delimiter::Parenthesis) { + // We know for sure we have seen `for ($SOMETHING in $EXPR)`, so we recover the + // parser state and emit a targeted suggestion. + let span = vec![start_span, self.token.span]; + let right = self.prev_token.span.between(self.look_ahead(1, |t| t.span)); + self.bump(); // ) + err.cancel(); + self.dcx().emit_err(errors::ParenthesesInForHead { + span, + // With e.g. `for (x) in y)` this would replace `(x) in y)` + // with `x) in y)` which is syntactically invalid. + // However, this is prevented before we get here. + sugg: errors::ParenthesesInForHeadSugg { left, right }, + }); + Ok((self.mk_pat(start_span.to(right), ast::PatKind::Wild), expr)) + } else { + Err(err) // Some other error, bubble up. + }; + } + (Err(err), _) => return Err(err), // Some other error, bubble up. + }; + if !self.eat_keyword(kw::In) { + self.error_missing_in_for_loop(); + } + self.check_for_for_in_in_typo(self.prev_token.span); + let expr = self.parse_expr_res(Restrictions::NO_STRUCT_LITERAL, None)?; + Ok((pat, expr)) + } + + /// Parses `for await? <src_pat> in <src_expr> <src_loop_block>` (`for` token already eaten). + fn parse_expr_for(&mut self, opt_label: Option<Label>, lo: Span) -> PResult<'a, P<Expr>> { + let is_await = + self.token.uninterpolated_span().at_least_rust_2018() && self.eat_keyword(kw::Await); + + if is_await { + self.psess.gated_spans.gate(sym::async_for_loop, self.prev_token.span); + } + + let kind = if is_await { ForLoopKind::ForAwait } else { ForLoopKind::For }; + + let (pat, expr) = self.parse_for_head()?; + // Recover from missing expression in `for` loop + if matches!(expr.kind, ExprKind::Block(..)) + && !matches!(self.token.kind, token::OpenDelim(Delimiter::Brace)) + && self.may_recover() + { + let guar = self + .dcx() + .emit_err(errors::MissingExpressionInForLoop { span: expr.span.shrink_to_lo() }); + let err_expr = self.mk_expr(expr.span, ExprKind::Err(guar)); + let block = self.mk_block(thin_vec![], BlockCheckMode::Default, self.prev_token.span); + return Ok(self.mk_expr( + lo.to(self.prev_token.span), + ExprKind::ForLoop { pat, iter: err_expr, body: block, label: opt_label, kind }, + )); + } + + let (attrs, loop_block) = self.parse_inner_attrs_and_block()?; + + let kind = ExprKind::ForLoop { pat, iter: expr, body: loop_block, label: opt_label, kind }; + + self.recover_loop_else("for", lo)?; + + Ok(self.mk_expr_with_attrs(lo.to(self.prev_token.span), kind, attrs)) + } + + /// Recovers from an `else` clause after a loop (`for...else`, `while...else`) + fn recover_loop_else(&mut self, loop_kind: &'static str, loop_kw: Span) -> PResult<'a, ()> { + if self.token.is_keyword(kw::Else) && self.may_recover() { + let else_span = self.token.span; + self.bump(); + let else_clause = self.parse_expr_else()?; + self.dcx().emit_err(errors::LoopElseNotSupported { + span: else_span.to(else_clause.span), + loop_kind, + loop_kw, + }); + } + Ok(()) + } + + fn error_missing_in_for_loop(&mut self) { + let (span, sub): (_, fn(_) -> _) = if self.token.is_ident_named(sym::of) { + // Possibly using JS syntax (#75311). + let span = self.token.span; + self.bump(); + (span, errors::MissingInInForLoopSub::InNotOf) + } else { + (self.prev_token.span.between(self.token.span), errors::MissingInInForLoopSub::AddIn) + }; + + self.dcx().emit_err(errors::MissingInInForLoop { span, sub: sub(span) }); + } + + /// Parses a `while` or `while let` expression (`while` token already eaten). + fn parse_expr_while(&mut self, opt_label: Option<Label>, lo: Span) -> PResult<'a, P<Expr>> { + let cond = self.parse_expr_cond().map_err(|mut err| { + err.span_label(lo, "while parsing the condition of this `while` expression"); + err + })?; + let (attrs, body) = self.parse_inner_attrs_and_block().map_err(|mut err| { + err.span_label(lo, "while parsing the body of this `while` expression"); + err.span_label(cond.span, "this `while` condition successfully parsed"); + err + })?; + + self.recover_loop_else("while", lo)?; + + Ok(self.mk_expr_with_attrs( + lo.to(self.prev_token.span), + ExprKind::While(cond, body, opt_label), + attrs, + )) + } + + /// Parses `loop { ... }` (`loop` token already eaten). + fn parse_expr_loop(&mut self, opt_label: Option<Label>, lo: Span) -> PResult<'a, P<Expr>> { + let loop_span = self.prev_token.span; + let (attrs, body) = self.parse_inner_attrs_and_block()?; + self.recover_loop_else("loop", lo)?; + Ok(self.mk_expr_with_attrs( + lo.to(self.prev_token.span), + ExprKind::Loop(body, opt_label, loop_span), + attrs, + )) + } + + pub(crate) fn eat_label(&mut self) -> Option<Label> { + self.token.lifetime().map(|ident| { + self.bump(); + Label { ident } + }) + } + + /// Parses a `match ... { ... }` expression (`match` token already eaten). + fn parse_expr_match(&mut self) -> PResult<'a, P<Expr>> { + let match_span = self.prev_token.span; + let scrutinee = self.parse_expr_res(Restrictions::NO_STRUCT_LITERAL, None)?; + + self.parse_match_block(match_span, match_span, scrutinee, MatchKind::Prefix) + } + + /// Parses the block of a `match expr { ... }` or a `expr.match { ... }` + /// expression. This is after the match token and scrutinee are eaten + fn parse_match_block( + &mut self, + lo: Span, + match_span: Span, + scrutinee: P<Expr>, + match_kind: MatchKind, + ) -> PResult<'a, P<Expr>> { + if let Err(mut e) = self.expect(&token::OpenDelim(Delimiter::Brace)) { + if self.token == token::Semi { + e.span_suggestion_short( + match_span, + "try removing this `match`", + "", + Applicability::MaybeIncorrect, // speculative + ); + } + if self.maybe_recover_unexpected_block_label() { + e.cancel(); + self.bump(); + } else { + return Err(e); + } + } + let attrs = self.parse_inner_attributes()?; + + let mut arms = ThinVec::new(); + while self.token != token::CloseDelim(Delimiter::Brace) { + match self.parse_arm() { + Ok(arm) => arms.push(arm), + Err(e) => { + // Recover by skipping to the end of the block. + let guar = e.emit(); + self.recover_stmt(); + let span = lo.to(self.token.span); + if self.token == token::CloseDelim(Delimiter::Brace) { + self.bump(); + } + // Always push at least one arm to make the match non-empty + arms.push(Arm { + attrs: Default::default(), + pat: self.mk_pat(span, ast::PatKind::Err(guar)), + guard: None, + body: Some(self.mk_expr_err(span, guar)), + span, + id: DUMMY_NODE_ID, + is_placeholder: false, + }); + return Ok(self.mk_expr_with_attrs( + span, + ExprKind::Match(scrutinee, arms, match_kind), + attrs, + )); + } + } + } + let hi = self.token.span; + self.bump(); + Ok(self.mk_expr_with_attrs(lo.to(hi), ExprKind::Match(scrutinee, arms, match_kind), attrs)) + } + + /// Attempt to recover from match arm body with statements and no surrounding braces. + fn parse_arm_body_missing_braces( + &mut self, + first_expr: &P<Expr>, + arrow_span: Span, + ) -> Option<(Span, ErrorGuaranteed)> { + if self.token.kind != token::Semi { + return None; + } + let start_snapshot = self.create_snapshot_for_diagnostic(); + let semi_sp = self.token.span; + self.bump(); // `;` + let mut stmts = + vec![self.mk_stmt(first_expr.span, ast::StmtKind::Expr(first_expr.clone()))]; + let err = |this: &Parser<'_>, stmts: Vec<ast::Stmt>| { + let span = stmts[0].span.to(stmts[stmts.len() - 1].span); + + let guar = this.dcx().emit_err(errors::MatchArmBodyWithoutBraces { + statements: span, + arrow: arrow_span, + num_statements: stmts.len(), + sub: if stmts.len() > 1 { + errors::MatchArmBodyWithoutBracesSugg::AddBraces { + left: span.shrink_to_lo(), + right: span.shrink_to_hi(), + } + } else { + errors::MatchArmBodyWithoutBracesSugg::UseComma { semicolon: semi_sp } + }, + }); + (span, guar) + }; + // We might have either a `,` -> `;` typo, or a block without braces. We need + // a more subtle parsing strategy. + loop { + if self.token.kind == token::CloseDelim(Delimiter::Brace) { + // We have reached the closing brace of the `match` expression. + return Some(err(self, stmts)); + } + if self.token.kind == token::Comma { + self.restore_snapshot(start_snapshot); + return None; + } + let pre_pat_snapshot = self.create_snapshot_for_diagnostic(); + match self.parse_pat_no_top_alt(None, None) { + Ok(_pat) => { + if self.token.kind == token::FatArrow { + // Reached arm end. + self.restore_snapshot(pre_pat_snapshot); + return Some(err(self, stmts)); + } + } + Err(err) => { + err.cancel(); + } + } + + self.restore_snapshot(pre_pat_snapshot); + match self.parse_stmt_without_recovery(true, ForceCollect::No) { + // Consume statements for as long as possible. + Ok(Some(stmt)) => { + stmts.push(stmt); + } + Ok(None) => { + self.restore_snapshot(start_snapshot); + break; + } + // We couldn't parse either yet another statement missing it's + // enclosing block nor the next arm's pattern or closing brace. + Err(stmt_err) => { + stmt_err.cancel(); + self.restore_snapshot(start_snapshot); + break; + } + } + } + None + } + + pub(super) fn parse_arm(&mut self) -> PResult<'a, Arm> { + let attrs = self.parse_outer_attributes()?; + self.collect_tokens_trailing_token(attrs, ForceCollect::No, |this, attrs| { + let lo = this.token.span; + let (pat, guard) = this.parse_match_arm_pat_and_guard()?; + + let span_before_body = this.prev_token.span; + let arm_body; + let is_fat_arrow = this.check(&token::FatArrow); + let is_almost_fat_arrow = TokenKind::FatArrow + .similar_tokens() + .is_some_and(|similar_tokens| similar_tokens.contains(&this.token.kind)); + + // this avoids the compiler saying that a `,` or `}` was expected even though + // the pattern isn't a never pattern (and thus an arm body is required) + let armless = (!is_fat_arrow && !is_almost_fat_arrow && pat.could_be_never_pattern()) + || matches!(this.token.kind, token::Comma | token::CloseDelim(Delimiter::Brace)); + + let mut result = if armless { + // A pattern without a body, allowed for never patterns. + arm_body = None; + this.expect_one_of(&[token::Comma], &[token::CloseDelim(Delimiter::Brace)]).map( + |x| { + // Don't gate twice + if !pat.contains_never_pattern() { + this.psess.gated_spans.gate(sym::never_patterns, pat.span); + } + x + }, + ) + } else { + if let Err(mut err) = this.expect(&token::FatArrow) { + // We might have a `=>` -> `=` or `->` typo (issue #89396). + if is_almost_fat_arrow { + err.span_suggestion( + this.token.span, + "use a fat arrow to start a match arm", + "=>", + Applicability::MachineApplicable, + ); + if matches!( + (&this.prev_token.kind, &this.token.kind), + (token::DotDotEq, token::Gt) + ) { + // `error_inclusive_range_match_arrow` handles cases like `0..=> {}`, + // so we suppress the error here + err.delay_as_bug(); + } else { + err.emit(); + } + this.bump(); + } else { + return Err(err); + } + } + let arrow_span = this.prev_token.span; + let arm_start_span = this.token.span; + + let expr = + this.parse_expr_res(Restrictions::STMT_EXPR, None).map_err(|mut err| { + err.span_label(arrow_span, "while parsing the `match` arm starting here"); + err + })?; + + let require_comma = !classify::expr_is_complete(&expr) + && this.token != token::CloseDelim(Delimiter::Brace); + + if !require_comma { + arm_body = Some(expr); + this.eat(&token::Comma); + Ok(Recovered::No) + } else if let Some((span, guar)) = + this.parse_arm_body_missing_braces(&expr, arrow_span) + { + let body = this.mk_expr_err(span, guar); + arm_body = Some(body); + Ok(Recovered::Yes(guar)) + } else { + let expr_span = expr.span; + arm_body = Some(expr); + this.expect_one_of(&[token::Comma], &[token::CloseDelim(Delimiter::Brace)]) + .map_err(|mut err| { + if this.token == token::FatArrow { + let sm = this.psess.source_map(); + if let Ok(expr_lines) = sm.span_to_lines(expr_span) + && let Ok(arm_start_lines) = sm.span_to_lines(arm_start_span) + && arm_start_lines.lines[0].end_col + == expr_lines.lines[0].end_col + && expr_lines.lines.len() == 2 + { + // We check whether there's any trailing code in the parse span, + // if there isn't, we very likely have the following: + // + // X | &Y => "y" + // | -- - missing comma + // | | + // | arrow_span + // X | &X => "x" + // | - ^^ self.token.span + // | | + // | parsed until here as `"y" & X` + err.span_suggestion_short( + arm_start_span.shrink_to_hi(), + "missing a comma here to end this `match` arm", + ",", + Applicability::MachineApplicable, + ); + } + } else { + err.span_label( + arrow_span, + "while parsing the `match` arm starting here", + ); + } + err + }) + } + }; + + let hi_span = arm_body.as_ref().map_or(span_before_body, |body| body.span); + let arm_span = lo.to(hi_span); + + // We want to recover: + // X | Some(_) => foo() + // | - missing comma + // X | None => "x" + // | ^^^^ self.token.span + // as well as: + // X | Some(!) + // | - missing comma + // X | None => "x" + // | ^^^^ self.token.span + // But we musn't recover + // X | pat[0] => {} + // | ^ self.token.span + let recover_missing_comma = arm_body.is_some() || pat.could_be_never_pattern(); + if recover_missing_comma { + result = result.or_else(|err| { + // FIXME(compiler-errors): We could also recover `; PAT =>` here + + // Try to parse a following `PAT =>`, if successful + // then we should recover. + let mut snapshot = this.create_snapshot_for_diagnostic(); + let pattern_follows = snapshot + .parse_pat_allow_top_alt( + None, + RecoverComma::Yes, + RecoverColon::Yes, + CommaRecoveryMode::EitherTupleOrPipe, + ) + .map_err(|err| err.cancel()) + .is_ok(); + if pattern_follows && snapshot.check(&TokenKind::FatArrow) { + err.cancel(); + let guar = this.dcx().emit_err(errors::MissingCommaAfterMatchArm { + span: arm_span.shrink_to_hi(), + }); + return Ok(Recovered::Yes(guar)); + } + Err(err) + }); + } + result?; + + Ok(( + ast::Arm { + attrs, + pat, + guard, + body: arm_body, + span: arm_span, + id: DUMMY_NODE_ID, + is_placeholder: false, + }, + TrailingToken::None, + )) + }) + } + + fn parse_match_arm_guard(&mut self) -> PResult<'a, Option<P<Expr>>> { + // Used to check the `let_chains` and `if_let_guard` features mostly by scanning + // `&&` tokens. + fn check_let_expr(expr: &Expr) -> (bool, bool) { + match &expr.kind { + ExprKind::Binary(BinOp { node: BinOpKind::And, .. }, lhs, rhs) => { + let lhs_rslt = check_let_expr(lhs); + let rhs_rslt = check_let_expr(rhs); + (lhs_rslt.0 || rhs_rslt.0, false) + } + ExprKind::Let(..) => (true, true), + _ => (false, true), + } + } + if !self.eat_keyword(kw::If) { + // No match arm guard present. + return Ok(None); + } + + let if_span = self.prev_token.span; + let mut cond = self.parse_match_guard_condition()?; + + CondChecker::new(self).visit_expr(&mut cond); + + let (has_let_expr, does_not_have_bin_op) = check_let_expr(&cond); + if has_let_expr { + if does_not_have_bin_op { + // Remove the last feature gating of a `let` expression since it's stable. + self.psess.gated_spans.ungate_last(sym::let_chains, cond.span); + } + let span = if_span.to(cond.span); + self.psess.gated_spans.gate(sym::if_let_guard, span); + } + Ok(Some(cond)) + } + + fn parse_match_arm_pat_and_guard(&mut self) -> PResult<'a, (P<Pat>, Option<P<Expr>>)> { + if self.token.kind == token::OpenDelim(Delimiter::Parenthesis) { + // Detect and recover from `($pat if $cond) => $arm`. + let left = self.token.span; + match self.parse_pat_allow_top_alt( + None, + RecoverComma::Yes, + RecoverColon::Yes, + CommaRecoveryMode::EitherTupleOrPipe, + ) { + Ok(pat) => Ok((pat, self.parse_match_arm_guard()?)), + Err(err) + if let prev_sp = self.prev_token.span + && let true = self.eat_keyword(kw::If) => + { + // We know for certain we've found `($pat if` so far. + let mut cond = match self.parse_match_guard_condition() { + Ok(cond) => cond, + Err(cond_err) => { + cond_err.cancel(); + return Err(err); + } + }; + err.cancel(); + CondChecker::new(self).visit_expr(&mut cond); + self.eat_to_tokens(&[&token::CloseDelim(Delimiter::Parenthesis)]); + self.expect(&token::CloseDelim(Delimiter::Parenthesis))?; + let right = self.prev_token.span; + self.dcx().emit_err(errors::ParenthesesInMatchPat { + span: vec![left, right], + sugg: errors::ParenthesesInMatchPatSugg { left, right }, + }); + Ok((self.mk_pat(left.to(prev_sp), ast::PatKind::Wild), Some(cond))) + } + Err(err) => Err(err), + } + } else { + // Regular parser flow: + let pat = self.parse_pat_allow_top_alt( + None, + RecoverComma::Yes, + RecoverColon::Yes, + CommaRecoveryMode::EitherTupleOrPipe, + )?; + Ok((pat, self.parse_match_arm_guard()?)) + } + } + + fn parse_match_guard_condition(&mut self) -> PResult<'a, P<Expr>> { + self.parse_expr_res(Restrictions::ALLOW_LET | Restrictions::IN_IF_GUARD, None).map_err( + |mut err| { + if self.prev_token == token::OpenDelim(Delimiter::Brace) { + let sugg_sp = self.prev_token.span.shrink_to_lo(); + // Consume everything within the braces, let's avoid further parse + // errors. + self.recover_stmt_(SemiColonMode::Ignore, BlockMode::Ignore); + let msg = "you might have meant to start a match arm after the match guard"; + if self.eat(&token::CloseDelim(Delimiter::Brace)) { + let applicability = if self.token.kind != token::FatArrow { + // We have high confidence that we indeed didn't have a struct + // literal in the match guard, but rather we had some operation + // that ended in a path, immediately followed by a block that was + // meant to be the match arm. + Applicability::MachineApplicable + } else { + Applicability::MaybeIncorrect + }; + err.span_suggestion_verbose(sugg_sp, msg, "=> ", applicability); + } + } + err + }, + ) + } + + pub(crate) fn is_builtin(&self) -> bool { + self.token.is_keyword(kw::Builtin) && self.look_ahead(1, |t| *t == token::Pound) + } + + /// Parses a `try {...}` expression (`try` token already eaten). + fn parse_try_block(&mut self, span_lo: Span) -> PResult<'a, P<Expr>> { + let (attrs, body) = self.parse_inner_attrs_and_block()?; + if self.eat_keyword(kw::Catch) { + Err(self.dcx().create_err(errors::CatchAfterTry { span: self.prev_token.span })) + } else { + let span = span_lo.to(body.span); + self.psess.gated_spans.gate(sym::try_blocks, span); + Ok(self.mk_expr_with_attrs(span, ExprKind::TryBlock(body), attrs)) + } + } + + fn is_do_catch_block(&self) -> bool { + self.token.is_keyword(kw::Do) + && self.is_keyword_ahead(1, &[kw::Catch]) + && self + .look_ahead(2, |t| *t == token::OpenDelim(Delimiter::Brace) || t.is_whole_block()) + && !self.restrictions.contains(Restrictions::NO_STRUCT_LITERAL) + } + + fn is_do_yeet(&self) -> bool { + self.token.is_keyword(kw::Do) && self.is_keyword_ahead(1, &[kw::Yeet]) + } + + fn is_try_block(&self) -> bool { + self.token.is_keyword(kw::Try) + && self + .look_ahead(1, |t| *t == token::OpenDelim(Delimiter::Brace) || t.is_whole_block()) + && self.token.uninterpolated_span().at_least_rust_2018() + } + + /// Parses an `async move? {...}` or `gen move? {...}` expression. + fn parse_gen_block(&mut self) -> PResult<'a, P<Expr>> { + let lo = self.token.span; + let kind = if self.eat_keyword(kw::Async) { + if self.eat_keyword(kw::Gen) { GenBlockKind::AsyncGen } else { GenBlockKind::Async } + } else { + assert!(self.eat_keyword(kw::Gen)); + GenBlockKind::Gen + }; + match kind { + GenBlockKind::Async => { + // `async` blocks are stable + } + GenBlockKind::Gen | GenBlockKind::AsyncGen => { + self.psess.gated_spans.gate(sym::gen_blocks, lo.to(self.prev_token.span)); + } + } + let capture_clause = self.parse_capture_clause()?; + let (attrs, body) = self.parse_inner_attrs_and_block()?; + let kind = ExprKind::Gen(capture_clause, body, kind); + Ok(self.mk_expr_with_attrs(lo.to(self.prev_token.span), kind, attrs)) + } + + fn is_gen_block(&self, kw: Symbol, lookahead: usize) -> bool { + self.is_keyword_ahead(lookahead, &[kw]) + && (( + // `async move {` + self.is_keyword_ahead(lookahead + 1, &[kw::Move]) + && self.look_ahead(lookahead + 2, |t| { + *t == token::OpenDelim(Delimiter::Brace) || t.is_whole_block() + }) + ) || ( + // `async {` + self.look_ahead(lookahead + 1, |t| { + *t == token::OpenDelim(Delimiter::Brace) || t.is_whole_block() + }) + )) + } + + pub(super) fn is_async_gen_block(&self) -> bool { + self.token.is_keyword(kw::Async) && self.is_gen_block(kw::Gen, 1) + } + + fn is_certainly_not_a_block(&self) -> bool { + self.look_ahead(1, |t| t.is_ident()) + && ( + // `{ ident, ` cannot start a block. + self.look_ahead(2, |t| t == &token::Comma) + || self.look_ahead(2, |t| t == &token::Colon) + && ( + // `{ ident: token, ` cannot start a block. + self.look_ahead(4, |t| t == &token::Comma) + // `{ ident: ` cannot start a block unless it's a type ascription + // `ident: Type`. + || self.look_ahead(3, |t| !t.can_begin_type()) + ) + ) + } + + fn maybe_parse_struct_expr( + &mut self, + qself: &Option<P<ast::QSelf>>, + path: &ast::Path, + ) -> Option<PResult<'a, P<Expr>>> { + let struct_allowed = !self.restrictions.contains(Restrictions::NO_STRUCT_LITERAL); + if struct_allowed || self.is_certainly_not_a_block() { + if let Err(err) = self.expect(&token::OpenDelim(Delimiter::Brace)) { + return Some(Err(err)); + } + let expr = self.parse_expr_struct(qself.clone(), path.clone(), true); + if let (Ok(expr), false) = (&expr, struct_allowed) { + // This is a struct literal, but we don't can't accept them here. + self.dcx().emit_err(errors::StructLiteralNotAllowedHere { + span: expr.span, + sub: errors::StructLiteralNotAllowedHereSugg { + left: path.span.shrink_to_lo(), + right: expr.span.shrink_to_hi(), + }, + }); + } + return Some(expr); + } + None + } + + pub(super) fn parse_struct_fields( + &mut self, + pth: ast::Path, + recover: bool, + close_delim: Delimiter, + ) -> PResult< + 'a, + ( + ThinVec<ExprField>, + ast::StructRest, + Option<ErrorGuaranteed>, /* async blocks are forbidden in Rust 2015 */ + ), + > { + let mut fields = ThinVec::new(); + let mut base = ast::StructRest::None; + let mut recovered_async = None; + let in_if_guard = self.restrictions.contains(Restrictions::IN_IF_GUARD); + + let async_block_err = |e: &mut Diag<'_>, span: Span| { + errors::AsyncBlockIn2015 { span }.add_to_diag(e); + errors::HelpUseLatestEdition::new().add_to_diag(e); + }; + + while self.token != token::CloseDelim(close_delim) { + if self.eat(&token::DotDot) || self.recover_struct_field_dots(close_delim) { + let exp_span = self.prev_token.span; + // We permit `.. }` on the left-hand side of a destructuring assignment. + if self.check(&token::CloseDelim(close_delim)) { + base = ast::StructRest::Rest(self.prev_token.span.shrink_to_hi()); + break; + } + match self.parse_expr() { + Ok(e) => base = ast::StructRest::Base(e), + Err(e) if recover => { + e.emit(); + self.recover_stmt(); + } + Err(e) => return Err(e), + } + self.recover_struct_comma_after_dotdot(exp_span); + break; + } + + // Peek the field's ident before parsing its expr in order to emit better diagnostics. + let peek = self + .token + .ident() + .filter(|(ident, is_raw)| { + (!ident.is_reserved() || matches!(is_raw, IdentIsRaw::Yes)) + && self.look_ahead(1, |tok| *tok == token::Colon) + }) + .map(|(ident, _)| ident); + + // We still want a field even if its expr didn't parse. + let field_ident = |this: &Self, guar: ErrorGuaranteed| { + peek.map(|ident| { + let span = ident.span; + ExprField { + ident, + span, + expr: this.mk_expr_err(span, guar), + is_shorthand: false, + attrs: AttrVec::new(), + id: DUMMY_NODE_ID, + is_placeholder: false, + } + }) + }; + + let parsed_field = match self.parse_expr_field() { + Ok(f) => Ok(f), + Err(mut e) => { + if pth == kw::Async { + async_block_err(&mut e, pth.span); + } else { + e.span_label(pth.span, "while parsing this struct"); + } + + if let Some((ident, _)) = self.token.ident() + && !self.token.is_reserved_ident() + && self.look_ahead(1, |t| { + AssocOp::from_token(t).is_some() + || matches!(t.kind, token::OpenDelim(_)) + || t.kind == token::Dot + }) + { + // Looks like they tried to write a shorthand, complex expression. + e.span_suggestion_verbose( + self.token.span.shrink_to_lo(), + "try naming a field", + &format!("{ident}: ",), + Applicability::MaybeIncorrect, + ); + } + if in_if_guard && close_delim == Delimiter::Brace { + return Err(e); + } + + if !recover { + return Err(e); + } + + let guar = e.emit(); + if pth == kw::Async { + recovered_async = Some(guar); + } + + // If the next token is a comma, then try to parse + // what comes next as additional fields, rather than + // bailing out until next `}`. + if self.token != token::Comma { + self.recover_stmt_(SemiColonMode::Comma, BlockMode::Ignore); + if self.token != token::Comma { + break; + } + } + + Err(guar) + } + }; + + let is_shorthand = parsed_field.as_ref().is_ok_and(|f| f.is_shorthand); + // A shorthand field can be turned into a full field with `:`. + // We should point this out. + self.check_or_expected(!is_shorthand, TokenType::Token(token::Colon)); + + match self.expect_one_of(&[token::Comma], &[token::CloseDelim(close_delim)]) { + Ok(_) => { + if let Ok(f) = parsed_field.or_else(|guar| field_ident(self, guar).ok_or(guar)) + { + // Only include the field if there's no parse error for the field name. + fields.push(f); + } + } + Err(mut e) => { + if pth == kw::Async { + async_block_err(&mut e, pth.span); + } else { + e.span_label(pth.span, "while parsing this struct"); + if peek.is_some() { + e.span_suggestion( + self.prev_token.span.shrink_to_hi(), + "try adding a comma", + ",", + Applicability::MachineApplicable, + ); + } + } + if !recover { + return Err(e); + } + let guar = e.emit(); + if pth == kw::Async { + recovered_async = Some(guar); + } else if let Some(f) = field_ident(self, guar) { + fields.push(f); + } + self.recover_stmt_(SemiColonMode::Comma, BlockMode::Ignore); + self.eat(&token::Comma); + } + } + } + Ok((fields, base, recovered_async)) + } + + /// Precondition: already parsed the '{'. + pub(super) fn parse_expr_struct( + &mut self, + qself: Option<P<ast::QSelf>>, + pth: ast::Path, + recover: bool, + ) -> PResult<'a, P<Expr>> { + let lo = pth.span; + let (fields, base, recovered_async) = + self.parse_struct_fields(pth.clone(), recover, Delimiter::Brace)?; + let span = lo.to(self.token.span); + self.expect(&token::CloseDelim(Delimiter::Brace))?; + let expr = if let Some(guar) = recovered_async { + ExprKind::Err(guar) + } else { + ExprKind::Struct(P(ast::StructExpr { qself, path: pth, fields, rest: base })) + }; + Ok(self.mk_expr(span, expr)) + } + + fn recover_struct_comma_after_dotdot(&mut self, span: Span) { + if self.token != token::Comma { + return; + } + self.dcx().emit_err(errors::CommaAfterBaseStruct { + span: span.to(self.prev_token.span), + comma: self.token.span, + }); + self.recover_stmt(); + } + + fn recover_struct_field_dots(&mut self, close_delim: Delimiter) -> bool { + if !self.look_ahead(1, |t| *t == token::CloseDelim(close_delim)) + && self.eat(&token::DotDotDot) + { + // recover from typo of `...`, suggest `..` + let span = self.prev_token.span; + self.dcx().emit_err(errors::MissingDotDot { token_span: span, sugg_span: span }); + return true; + } + false + } + + /// Converts an ident into 'label and emits an "expected a label, found an identifier" error. + fn recover_ident_into_label(&mut self, ident: Ident) -> Label { + // Convert `label` -> `'label`, + // so that nameres doesn't complain about non-existing label + let label = format!("'{}", ident.name); + let ident = Ident { name: Symbol::intern(&label), span: ident.span }; + + self.dcx().emit_err(errors::ExpectedLabelFoundIdent { + span: ident.span, + start: ident.span.shrink_to_lo(), + }); + + Label { ident } + } + + /// Parses `ident (COLON expr)?`. + fn parse_expr_field(&mut self) -> PResult<'a, ExprField> { + let attrs = self.parse_outer_attributes()?; + self.recover_vcs_conflict_marker(); + self.collect_tokens_trailing_token(attrs, ForceCollect::No, |this, attrs| { + let lo = this.token.span; + + // Check if a colon exists one ahead. This means we're parsing a fieldname. + let is_shorthand = !this.look_ahead(1, |t| t == &token::Colon || t == &token::Eq); + // Proactively check whether parsing the field will be incorrect. + let is_wrong = this.token.is_ident() + && !this.token.is_reserved_ident() + && !this.look_ahead(1, |t| { + t == &token::Colon + || t == &token::Eq + || t == &token::Comma + || t == &token::CloseDelim(Delimiter::Brace) + || t == &token::CloseDelim(Delimiter::Parenthesis) + }); + if is_wrong { + return Err(this.dcx().create_err(errors::ExpectedStructField { + span: this.look_ahead(1, |t| t.span), + ident_span: this.token.span, + token: this.look_ahead(1, |t| t.clone()), + })); + } + let (ident, expr) = if is_shorthand { + // Mimic `x: x` for the `x` field shorthand. + let ident = this.parse_ident_common(false)?; + let path = ast::Path::from_ident(ident); + (ident, this.mk_expr(ident.span, ExprKind::Path(None, path))) + } else { + let ident = this.parse_field_name()?; + this.error_on_eq_field_init(ident); + this.bump(); // `:` + (ident, this.parse_expr()?) + }; + + Ok(( + ast::ExprField { + ident, + span: lo.to(expr.span), + expr, + is_shorthand, + attrs, + id: DUMMY_NODE_ID, + is_placeholder: false, + }, + TrailingToken::MaybeComma, + )) + }) + } + + /// Check for `=`. This means the source incorrectly attempts to + /// initialize a field with an eq rather than a colon. + fn error_on_eq_field_init(&self, field_name: Ident) { + if self.token != token::Eq { + return; + } + + self.dcx().emit_err(errors::EqFieldInit { + span: self.token.span, + eq: field_name.span.shrink_to_hi().to(self.token.span), + }); + } + + fn err_dotdotdot_syntax(&self, span: Span) { + self.dcx().emit_err(errors::DotDotDot { span }); + } + + fn err_larrow_operator(&self, span: Span) { + self.dcx().emit_err(errors::LeftArrowOperator { span }); + } + + fn mk_assign_op(&self, binop: BinOp, lhs: P<Expr>, rhs: P<Expr>) -> ExprKind { + ExprKind::AssignOp(binop, lhs, rhs) + } + + fn mk_range( + &mut self, + start: Option<P<Expr>>, + end: Option<P<Expr>>, + limits: RangeLimits, + ) -> ExprKind { + if end.is_none() && limits == RangeLimits::Closed { + let guar = self.inclusive_range_with_incorrect_end(); + ExprKind::Err(guar) + } else { + ExprKind::Range(start, end, limits) + } + } + + fn mk_unary(&self, unop: UnOp, expr: P<Expr>) -> ExprKind { + ExprKind::Unary(unop, expr) + } + + fn mk_binary(&self, binop: BinOp, lhs: P<Expr>, rhs: P<Expr>) -> ExprKind { + ExprKind::Binary(binop, lhs, rhs) + } + + fn mk_index(&self, expr: P<Expr>, idx: P<Expr>, brackets_span: Span) -> ExprKind { + ExprKind::Index(expr, idx, brackets_span) + } + + fn mk_call(&self, f: P<Expr>, args: ThinVec<P<Expr>>) -> ExprKind { + ExprKind::Call(f, args) + } + + fn mk_await_expr(&mut self, self_arg: P<Expr>, lo: Span) -> P<Expr> { + let span = lo.to(self.prev_token.span); + let await_expr = self.mk_expr(span, ExprKind::Await(self_arg, self.prev_token.span)); + self.recover_from_await_method_call(); + await_expr + } + + pub(crate) fn mk_expr_with_attrs(&self, span: Span, kind: ExprKind, attrs: AttrVec) -> P<Expr> { + P(Expr { kind, span, attrs, id: DUMMY_NODE_ID, tokens: None }) + } + + pub(crate) fn mk_expr(&self, span: Span, kind: ExprKind) -> P<Expr> { + self.mk_expr_with_attrs(span, kind, AttrVec::new()) + } + + pub(super) fn mk_expr_err(&self, span: Span, guar: ErrorGuaranteed) -> P<Expr> { + self.mk_expr(span, ExprKind::Err(guar)) + } + + fn collect_tokens_for_expr( + &mut self, + attrs: AttrWrapper, + f: impl FnOnce(&mut Self, ast::AttrVec) -> PResult<'a, P<Expr>>, + ) -> PResult<'a, P<Expr>> { + self.collect_tokens_trailing_token(attrs, ForceCollect::No, |this, attrs| { + let res = f(this, attrs)?; + let trailing = if this.restrictions.contains(Restrictions::STMT_EXPR) + && this.token.kind == token::Semi + { + TrailingToken::Semi + } else if this.token.kind == token::Gt { + TrailingToken::Gt + } else { + // FIXME - pass this through from the place where we know + // we need a comma, rather than assuming that `#[attr] expr,` + // always captures a trailing comma + TrailingToken::MaybeComma + }; + Ok((res, trailing)) + }) + } +} + +/// Could this lifetime/label be an unclosed char literal? For example, `'a` +/// could be, but `'abc` could not. +pub(crate) fn could_be_unclosed_char_literal(ident: Ident) -> bool { + ident.name.as_str().starts_with('\'') + && unescape_char(ident.without_first_quote().name.as_str()).is_ok() +} + +/// Used to forbid `let` expressions in certain syntactic locations. +#[derive(Clone, Copy, Subdiagnostic)] +pub(crate) enum ForbiddenLetReason { + /// `let` is not valid and the source environment is not important + OtherForbidden, + /// A let chain with the `||` operator + #[note(parse_not_supported_or)] + NotSupportedOr(#[primary_span] Span), + /// A let chain with invalid parentheses + /// + /// For example, `let 1 = 1 && (expr && expr)` is allowed + /// but `(let 1 = 1 && (let 1 = 1 && (let 1 = 1))) && let a = 1` is not + #[note(parse_not_supported_parentheses)] + NotSupportedParentheses(#[primary_span] Span), +} + +/// Visitor to check for invalid/unstable use of `ExprKind::Let` that can't +/// easily be caught in parsing. For example: +/// +/// ```rust,ignore (example) +/// // Only know that the let isn't allowed once the `||` token is reached +/// if let Some(x) = y || true {} +/// // Only know that the let isn't allowed once the second `=` token is reached. +/// if let Some(x) = y && z = 1 {} +/// ``` +struct CondChecker<'a> { + parser: &'a Parser<'a>, + forbid_let_reason: Option<ForbiddenLetReason>, + missing_let: Option<errors::MaybeMissingLet>, + comparison: Option<errors::MaybeComparison>, +} + +impl<'a> CondChecker<'a> { + fn new(parser: &'a Parser<'a>) -> Self { + CondChecker { parser, forbid_let_reason: None, missing_let: None, comparison: None } + } +} + +impl MutVisitor for CondChecker<'_> { + fn visit_expr(&mut self, e: &mut P<Expr>) { + use ForbiddenLetReason::*; + + let span = e.span; + match e.kind { + ExprKind::Let(_, _, _, ref mut recovered @ Recovered::No) => { + if let Some(reason) = self.forbid_let_reason { + *recovered = Recovered::Yes(self.parser.dcx().emit_err( + errors::ExpectedExpressionFoundLet { + span, + reason, + missing_let: self.missing_let, + comparison: self.comparison, + }, + )); + } else { + self.parser.psess.gated_spans.gate(sym::let_chains, span); + } + } + ExprKind::Binary(Spanned { node: BinOpKind::And, .. }, _, _) => { + noop_visit_expr(e, self); + } + ExprKind::Binary(Spanned { node: BinOpKind::Or, span: or_span }, _, _) + if let None | Some(NotSupportedOr(_)) = self.forbid_let_reason => + { + let forbid_let_reason = self.forbid_let_reason; + self.forbid_let_reason = Some(NotSupportedOr(or_span)); + noop_visit_expr(e, self); + self.forbid_let_reason = forbid_let_reason; + } + ExprKind::Paren(ref inner) + if let None | Some(NotSupportedParentheses(_)) = self.forbid_let_reason => + { + let forbid_let_reason = self.forbid_let_reason; + self.forbid_let_reason = Some(NotSupportedParentheses(inner.span)); + noop_visit_expr(e, self); + self.forbid_let_reason = forbid_let_reason; + } + ExprKind::Assign(ref lhs, _, span) => { + let forbid_let_reason = self.forbid_let_reason; + self.forbid_let_reason = Some(OtherForbidden); + let missing_let = self.missing_let; + if let ExprKind::Binary(_, _, rhs) = &lhs.kind + && let ExprKind::Path(_, _) + | ExprKind::Struct(_) + | ExprKind::Call(_, _) + | ExprKind::Array(_) = rhs.kind + { + self.missing_let = + Some(errors::MaybeMissingLet { span: rhs.span.shrink_to_lo() }); + } + let comparison = self.comparison; + self.comparison = Some(errors::MaybeComparison { span: span.shrink_to_hi() }); + noop_visit_expr(e, self); + self.forbid_let_reason = forbid_let_reason; + self.missing_let = missing_let; + self.comparison = comparison; + } + ExprKind::Unary(_, _) + | ExprKind::Await(_, _) + | ExprKind::AssignOp(_, _, _) + | ExprKind::Range(_, _, _) + | ExprKind::Try(_) + | ExprKind::AddrOf(_, _, _) + | ExprKind::Binary(_, _, _) + | ExprKind::Field(_, _) + | ExprKind::Index(_, _, _) + | ExprKind::Call(_, _) + | ExprKind::MethodCall(_) + | ExprKind::Tup(_) + | ExprKind::Paren(_) => { + let forbid_let_reason = self.forbid_let_reason; + self.forbid_let_reason = Some(OtherForbidden); + noop_visit_expr(e, self); + self.forbid_let_reason = forbid_let_reason; + } + ExprKind::Cast(ref mut op, _) | ExprKind::Type(ref mut op, _) => { + let forbid_let_reason = self.forbid_let_reason; + self.forbid_let_reason = Some(OtherForbidden); + self.visit_expr(op); + self.forbid_let_reason = forbid_let_reason; + } + ExprKind::Let(_, _, _, Recovered::Yes(_)) + | ExprKind::Array(_) + | ExprKind::ConstBlock(_) + | ExprKind::Lit(_) + | ExprKind::If(_, _, _) + | ExprKind::While(_, _, _) + | ExprKind::ForLoop { .. } + | ExprKind::Loop(_, _, _) + | ExprKind::Match(_, _, _) + | ExprKind::Closure(_) + | ExprKind::Block(_, _) + | ExprKind::Gen(_, _, _) + | ExprKind::TryBlock(_) + | ExprKind::Underscore + | ExprKind::Path(_, _) + | ExprKind::Break(_, _) + | ExprKind::Continue(_) + | ExprKind::Ret(_) + | ExprKind::InlineAsm(_) + | ExprKind::OffsetOf(_, _) + | ExprKind::MacCall(_) + | ExprKind::Struct(_) + | ExprKind::Repeat(_, _) + | ExprKind::Yield(_) + | ExprKind::Yeet(_) + | ExprKind::Become(_) + | ExprKind::IncludedBytes(_) + | ExprKind::FormatArgs(_) + | ExprKind::Err(_) + | ExprKind::Dummy => { + // These would forbid any let expressions they contain already. + } + } + } +} diff --git a/compiler/rustc_parse/src/parser/generics.rs b/compiler/rustc_parse/src/parser/generics.rs new file mode 100644 index 00000000000..93a15c938ec --- /dev/null +++ b/compiler/rustc_parse/src/parser/generics.rs @@ -0,0 +1,518 @@ +use crate::errors::{ + self, MultipleWhereClauses, UnexpectedDefaultValueForLifetimeInGenericParameters, + UnexpectedSelfInGenericParameters, WhereClauseBeforeTupleStructBody, + WhereClauseBeforeTupleStructBodySugg, +}; + +use super::{ForceCollect, Parser, TrailingToken}; + +use ast::token::Delimiter; +use rustc_ast::token; +use rustc_ast::{ + self as ast, AttrVec, GenericBounds, GenericParam, GenericParamKind, TyKind, WhereClause, +}; +use rustc_errors::{Applicability, PResult}; +use rustc_span::symbol::{kw, Ident}; +use rustc_span::Span; +use thin_vec::ThinVec; + +enum PredicateOrStructBody { + Predicate(ast::WherePredicate), + StructBody(ThinVec<ast::FieldDef>), +} + +impl<'a> Parser<'a> { + /// Parses bounds of a lifetime parameter `BOUND + BOUND + BOUND`, possibly with trailing `+`. + /// + /// ```text + /// BOUND = LT_BOUND (e.g., `'a`) + /// ``` + fn parse_lt_param_bounds(&mut self) -> GenericBounds { + let mut lifetimes = Vec::new(); + while self.check_lifetime() { + lifetimes.push(ast::GenericBound::Outlives(self.expect_lifetime())); + + if !self.eat_plus() { + break; + } + } + lifetimes + } + + /// Matches `typaram = IDENT (`?` unbound)? optbounds ( EQ ty )?`. + fn parse_ty_param(&mut self, preceding_attrs: AttrVec) -> PResult<'a, GenericParam> { + let ident = self.parse_ident()?; + + // We might have a typo'd `Const` that was parsed as a type parameter. + if self.may_recover() + && ident.name.as_str().to_ascii_lowercase() == kw::Const.as_str() + && self.check_ident() + // `Const` followed by IDENT + { + return self.recover_const_param_with_mistyped_const(preceding_attrs, ident); + } + + // Parse optional colon and param bounds. + let mut colon_span = None; + let bounds = if self.eat(&token::Colon) { + colon_span = Some(self.prev_token.span); + // recover from `impl Trait` in type param bound + if self.token.is_keyword(kw::Impl) { + let impl_span = self.token.span; + let snapshot = self.create_snapshot_for_diagnostic(); + match self.parse_ty() { + Ok(p) => { + if let TyKind::ImplTrait(_, bounds, None) = &p.kind { + let span = impl_span.to(self.token.span.shrink_to_lo()); + let mut err = self.dcx().struct_span_err( + span, + "expected trait bound, found `impl Trait` type", + ); + err.span_label(span, "not a trait"); + if let [bound, ..] = &bounds[..] { + err.span_suggestion_verbose( + impl_span.until(bound.span()), + "use the trait bounds directly", + String::new(), + Applicability::MachineApplicable, + ); + } + return Err(err); + } + } + Err(err) => { + err.cancel(); + } + } + self.restore_snapshot(snapshot); + } + self.parse_generic_bounds()? + } else { + Vec::new() + }; + + let default = if self.eat(&token::Eq) { Some(self.parse_ty()?) } else { None }; + Ok(GenericParam { + ident, + id: ast::DUMMY_NODE_ID, + attrs: preceding_attrs, + bounds, + kind: GenericParamKind::Type { default }, + is_placeholder: false, + colon_span, + }) + } + + pub(crate) fn parse_const_param( + &mut self, + preceding_attrs: AttrVec, + ) -> PResult<'a, GenericParam> { + let const_span = self.token.span; + + self.expect_keyword(kw::Const)?; + let ident = self.parse_ident()?; + self.expect(&token::Colon)?; + let ty = self.parse_ty()?; + + // Parse optional const generics default value. + let default = if self.eat(&token::Eq) { Some(self.parse_const_arg()?) } else { None }; + + Ok(GenericParam { + ident, + id: ast::DUMMY_NODE_ID, + attrs: preceding_attrs, + bounds: Vec::new(), + kind: GenericParamKind::Const { ty, kw_span: const_span, default }, + is_placeholder: false, + colon_span: None, + }) + } + + pub(crate) fn recover_const_param_with_mistyped_const( + &mut self, + preceding_attrs: AttrVec, + mistyped_const_ident: Ident, + ) -> PResult<'a, GenericParam> { + let ident = self.parse_ident()?; + self.expect(&token::Colon)?; + let ty = self.parse_ty()?; + + // Parse optional const generics default value. + let default = if self.eat(&token::Eq) { Some(self.parse_const_arg()?) } else { None }; + + self.dcx() + .struct_span_err( + mistyped_const_ident.span, + format!("`const` keyword was mistyped as `{}`", mistyped_const_ident.as_str()), + ) + .with_span_suggestion_verbose( + mistyped_const_ident.span, + "use the `const` keyword", + kw::Const, + Applicability::MachineApplicable, + ) + .emit(); + + Ok(GenericParam { + ident, + id: ast::DUMMY_NODE_ID, + attrs: preceding_attrs, + bounds: Vec::new(), + kind: GenericParamKind::Const { ty, kw_span: mistyped_const_ident.span, default }, + is_placeholder: false, + colon_span: None, + }) + } + + /// Parses a (possibly empty) list of lifetime and type parameters, possibly including + /// a trailing comma and erroneous trailing attributes. + pub(super) fn parse_generic_params(&mut self) -> PResult<'a, ThinVec<ast::GenericParam>> { + let mut params = ThinVec::new(); + let mut done = false; + while !done { + let attrs = self.parse_outer_attributes()?; + let param = + self.collect_tokens_trailing_token(attrs, ForceCollect::No, |this, attrs| { + if this.eat_keyword_noexpect(kw::SelfUpper) { + // `Self` as a generic param is invalid. Here we emit the diagnostic and continue parsing + // as if `Self` never existed. + this.dcx().emit_err(UnexpectedSelfInGenericParameters { + span: this.prev_token.span, + }); + + this.eat(&token::Comma); + } + + let param = if this.check_lifetime() { + let lifetime = this.expect_lifetime(); + // Parse lifetime parameter. + let (colon_span, bounds) = if this.eat(&token::Colon) { + (Some(this.prev_token.span), this.parse_lt_param_bounds()) + } else { + (None, Vec::new()) + }; + + if this.check_noexpect(&token::Eq) + && this.look_ahead(1, |t| t.is_lifetime()) + { + let lo = this.token.span; + // Parse `= 'lifetime`. + this.bump(); // `=` + this.bump(); // `'lifetime` + let span = lo.to(this.prev_token.span); + this.dcx().emit_err( + UnexpectedDefaultValueForLifetimeInGenericParameters { span }, + ); + } + + Some(ast::GenericParam { + ident: lifetime.ident, + id: lifetime.id, + attrs, + bounds, + kind: ast::GenericParamKind::Lifetime, + is_placeholder: false, + colon_span, + }) + } else if this.check_keyword(kw::Const) { + // Parse const parameter. + Some(this.parse_const_param(attrs)?) + } else if this.check_ident() { + // Parse type parameter. + Some(this.parse_ty_param(attrs)?) + } else if this.token.can_begin_type() { + // Trying to write an associated type bound? (#26271) + let snapshot = this.create_snapshot_for_diagnostic(); + match this.parse_ty_where_predicate() { + Ok(where_predicate) => { + this.dcx().emit_err(errors::BadAssocTypeBounds { + span: where_predicate.span(), + }); + // FIXME - try to continue parsing other generics? + return Ok((None, TrailingToken::None)); + } + Err(err) => { + err.cancel(); + // FIXME - maybe we should overwrite 'self' outside of `collect_tokens`? + this.restore_snapshot(snapshot); + return Ok((None, TrailingToken::None)); + } + } + } else { + // Check for trailing attributes and stop parsing. + if !attrs.is_empty() { + if !params.is_empty() { + this.dcx() + .emit_err(errors::AttrAfterGeneric { span: attrs[0].span }); + } else { + this.dcx() + .emit_err(errors::AttrWithoutGenerics { span: attrs[0].span }); + } + } + return Ok((None, TrailingToken::None)); + }; + + if !this.eat(&token::Comma) { + done = true; + } + // We just ate the comma, so no need to use `TrailingToken` + Ok((param, TrailingToken::None)) + })?; + + if let Some(param) = param { + params.push(param); + } else { + break; + } + } + Ok(params) + } + + /// Parses a set of optional generic type parameter declarations. Where + /// clauses are not parsed here, and must be added later via + /// `parse_where_clause()`. + /// + /// matches generics = ( ) | ( < > ) | ( < typaramseq ( , )? > ) | ( < lifetimes ( , )? > ) + /// | ( < lifetimes , typaramseq ( , )? > ) + /// where typaramseq = ( typaram ) | ( typaram , typaramseq ) + pub(super) fn parse_generics(&mut self) -> PResult<'a, ast::Generics> { + let span_lo = self.token.span; + let (params, span) = if self.eat_lt() { + let params = self.parse_generic_params()?; + self.expect_gt_or_maybe_suggest_closing_generics(¶ms)?; + (params, span_lo.to(self.prev_token.span)) + } else { + (ThinVec::new(), self.prev_token.span.shrink_to_hi()) + }; + Ok(ast::Generics { + params, + where_clause: WhereClause { + has_where_token: false, + predicates: ThinVec::new(), + span: self.prev_token.span.shrink_to_hi(), + }, + span, + }) + } + + /// Parses an optional where-clause. + /// + /// ```ignore (only-for-syntax-highlight) + /// where T : Trait<U, V> + 'b, 'a : 'b + /// ``` + pub(super) fn parse_where_clause(&mut self) -> PResult<'a, WhereClause> { + self.parse_where_clause_common(None).map(|(clause, _)| clause) + } + + pub(super) fn parse_struct_where_clause( + &mut self, + struct_name: Ident, + body_insertion_point: Span, + ) -> PResult<'a, (WhereClause, Option<ThinVec<ast::FieldDef>>)> { + self.parse_where_clause_common(Some((struct_name, body_insertion_point))) + } + + fn parse_where_clause_common( + &mut self, + struct_: Option<(Ident, Span)>, + ) -> PResult<'a, (WhereClause, Option<ThinVec<ast::FieldDef>>)> { + let mut where_clause = WhereClause { + has_where_token: false, + predicates: ThinVec::new(), + span: self.prev_token.span.shrink_to_hi(), + }; + let mut tuple_struct_body = None; + + if !self.eat_keyword(kw::Where) { + return Ok((where_clause, None)); + } + where_clause.has_where_token = true; + let where_lo = self.prev_token.span; + + // We are considering adding generics to the `where` keyword as an alternative higher-rank + // parameter syntax (as in `where<'a>` or `where<T>`. To avoid that being a breaking + // change we parse those generics now, but report an error. + if self.choose_generics_over_qpath(0) { + let generics = self.parse_generics()?; + self.dcx().emit_err(errors::WhereOnGenerics { span: generics.span }); + } + + loop { + let where_sp = where_lo.to(self.prev_token.span); + let pred_lo = self.token.span; + if self.check_lifetime() && self.look_ahead(1, |t| !t.is_like_plus()) { + let lifetime = self.expect_lifetime(); + // Bounds starting with a colon are mandatory, but possibly empty. + self.expect(&token::Colon)?; + let bounds = self.parse_lt_param_bounds(); + where_clause.predicates.push(ast::WherePredicate::RegionPredicate( + ast::WhereRegionPredicate { + span: pred_lo.to(self.prev_token.span), + lifetime, + bounds, + }, + )); + } else if self.check_type() { + match self.parse_ty_where_predicate_or_recover_tuple_struct_body( + struct_, pred_lo, where_sp, + )? { + PredicateOrStructBody::Predicate(pred) => where_clause.predicates.push(pred), + PredicateOrStructBody::StructBody(body) => { + tuple_struct_body = Some(body); + break; + } + } + } else { + break; + } + + let prev_token = self.prev_token.span; + let ate_comma = self.eat(&token::Comma); + + if self.eat_keyword_noexpect(kw::Where) { + self.dcx().emit_err(MultipleWhereClauses { + span: self.token.span, + previous: pred_lo, + between: prev_token.shrink_to_hi().to(self.prev_token.span), + }); + } else if !ate_comma { + break; + } + } + + where_clause.span = where_lo.to(self.prev_token.span); + Ok((where_clause, tuple_struct_body)) + } + + fn parse_ty_where_predicate_or_recover_tuple_struct_body( + &mut self, + struct_: Option<(Ident, Span)>, + pred_lo: Span, + where_sp: Span, + ) -> PResult<'a, PredicateOrStructBody> { + let mut snapshot = None; + + if let Some(struct_) = struct_ + && self.may_recover() + && self.token.kind == token::OpenDelim(Delimiter::Parenthesis) + { + snapshot = Some((struct_, self.create_snapshot_for_diagnostic())); + }; + + match self.parse_ty_where_predicate() { + Ok(pred) => Ok(PredicateOrStructBody::Predicate(pred)), + Err(type_err) => { + let Some(((struct_name, body_insertion_point), mut snapshot)) = snapshot else { + return Err(type_err); + }; + + // Check if we might have encountered an out of place tuple struct body. + match snapshot.parse_tuple_struct_body() { + // Since we don't know the exact reason why we failed to parse the + // predicate (we might have stumbled upon something bogus like `(T): ?`), + // employ a simple heuristic to weed out some pathological cases: + // Look for a semicolon (strong indicator) or anything that might mark + // the end of the item (weak indicator) following the body. + Ok(body) + if matches!(snapshot.token.kind, token::Semi | token::Eof) + || snapshot.token.can_begin_item() => + { + type_err.cancel(); + + let body_sp = pred_lo.to(snapshot.prev_token.span); + let map = self.psess.source_map(); + + self.dcx().emit_err(WhereClauseBeforeTupleStructBody { + span: where_sp, + name: struct_name.span, + body: body_sp, + sugg: map.span_to_snippet(body_sp).ok().map(|body| { + WhereClauseBeforeTupleStructBodySugg { + left: body_insertion_point.shrink_to_hi(), + snippet: body, + right: map.end_point(where_sp).to(body_sp), + } + }), + }); + + self.restore_snapshot(snapshot); + Ok(PredicateOrStructBody::StructBody(body)) + } + Ok(_) => Err(type_err), + Err(body_err) => { + body_err.cancel(); + Err(type_err) + } + } + } + } + } + + fn parse_ty_where_predicate(&mut self) -> PResult<'a, ast::WherePredicate> { + let lo = self.token.span; + // Parse optional `for<'a, 'b>`. + // This `for` is parsed greedily and applies to the whole predicate, + // the bounded type can have its own `for` applying only to it. + // Examples: + // * `for<'a> Trait1<'a>: Trait2<'a /* ok */>` + // * `(for<'a> Trait1<'a>): Trait2<'a /* not ok */>` + // * `for<'a> for<'b> Trait1<'a, 'b>: Trait2<'a /* ok */, 'b /* not ok */>` + let lifetime_defs = self.parse_late_bound_lifetime_defs()?; + + // Parse type with mandatory colon and (possibly empty) bounds, + // or with mandatory equality sign and the second type. + let ty = self.parse_ty_for_where_clause()?; + if self.eat(&token::Colon) { + let bounds = self.parse_generic_bounds()?; + Ok(ast::WherePredicate::BoundPredicate(ast::WhereBoundPredicate { + span: lo.to(self.prev_token.span), + bound_generic_params: lifetime_defs, + bounded_ty: ty, + bounds, + })) + // FIXME: Decide what should be used here, `=` or `==`. + // FIXME: We are just dropping the binders in lifetime_defs on the floor here. + } else if self.eat(&token::Eq) || self.eat(&token::EqEq) { + let rhs_ty = self.parse_ty()?; + Ok(ast::WherePredicate::EqPredicate(ast::WhereEqPredicate { + span: lo.to(self.prev_token.span), + lhs_ty: ty, + rhs_ty, + })) + } else { + self.maybe_recover_bounds_doubled_colon(&ty)?; + self.unexpected_any() + } + } + + pub(super) fn choose_generics_over_qpath(&self, start: usize) -> bool { + // There's an ambiguity between generic parameters and qualified paths in impls. + // If we see `<` it may start both, so we have to inspect some following tokens. + // The following combinations can only start generics, + // but not qualified paths (with one exception): + // `<` `>` - empty generic parameters + // `<` `#` - generic parameters with attributes + // `<` (LIFETIME|IDENT) `>` - single generic parameter + // `<` (LIFETIME|IDENT) `,` - first generic parameter in a list + // `<` (LIFETIME|IDENT) `:` - generic parameter with bounds + // `<` (LIFETIME|IDENT) `=` - generic parameter with a default + // `<` const - generic const parameter + // `<` IDENT `?` - RECOVERY for `impl<T ?Bound` missing a `:`, meant to + // avoid the `T?` to `Option<T>` recovery for types. + // The only truly ambiguous case is + // `<` IDENT `>` `::` IDENT ... + // we disambiguate it in favor of generics (`impl<T> ::absolute::Path<T> { ... }`) + // because this is what almost always expected in practice, qualified paths in impls + // (`impl <Type>::AssocTy { ... }`) aren't even allowed by type checker at the moment. + self.look_ahead(start, |t| t == &token::Lt) + && (self.look_ahead(start + 1, |t| t == &token::Pound || t == &token::Gt) + || self.look_ahead(start + 1, |t| t.is_lifetime() || t.is_ident()) + && self.look_ahead(start + 2, |t| { + matches!(t.kind, token::Gt | token::Comma | token::Colon | token::Eq) + // Recovery-only branch -- this could be removed, + // since it only affects diagnostics currently. + || matches!(t.kind, token::Question) + }) + || self.is_keyword_ahead(start + 1, &[kw::Const])) + } +} diff --git a/compiler/rustc_parse/src/parser/item.rs b/compiler/rustc_parse/src/parser/item.rs new file mode 100644 index 00000000000..a46c104b6d9 --- /dev/null +++ b/compiler/rustc_parse/src/parser/item.rs @@ -0,0 +1,2881 @@ +use super::diagnostics::{dummy_arg, ConsumeClosingDelim}; +use super::ty::{AllowPlus, RecoverQPath, RecoverReturnSign}; +use super::{ + AttrWrapper, FollowedByType, ForceCollect, Parser, PathStyle, Trailing, TrailingToken, +}; +use crate::errors::{self, MacroExpandsToAdtField}; +use crate::fluent_generated as fluent; +use crate::maybe_whole; +use ast::token::IdentIsRaw; +use rustc_ast::ast::*; +use rustc_ast::ptr::P; +use rustc_ast::token::{self, Delimiter, TokenKind}; +use rustc_ast::tokenstream::{DelimSpan, TokenStream, TokenTree}; +use rustc_ast::util::case::Case; +use rustc_ast::{self as ast}; +use rustc_ast_pretty::pprust; +use rustc_errors::{codes::*, struct_span_code_err, Applicability, PResult, StashKey}; +use rustc_span::edit_distance::edit_distance; +use rustc_span::edition::Edition; +use rustc_span::source_map; +use rustc_span::symbol::{kw, sym, Ident, Symbol}; +use rustc_span::{Span, DUMMY_SP}; +use std::fmt::Write; +use std::mem; +use thin_vec::{thin_vec, ThinVec}; + +impl<'a> Parser<'a> { + /// Parses a source module as a crate. This is the main entry point for the parser. + pub fn parse_crate_mod(&mut self) -> PResult<'a, ast::Crate> { + let (attrs, items, spans) = self.parse_mod(&token::Eof)?; + Ok(ast::Crate { attrs, items, spans, id: DUMMY_NODE_ID, is_placeholder: false }) + } + + /// Parses a `mod <foo> { ... }` or `mod <foo>;` item. + fn parse_item_mod(&mut self, attrs: &mut AttrVec) -> PResult<'a, ItemInfo> { + let safety = self.parse_safety(Case::Sensitive); + self.expect_keyword(kw::Mod)?; + let id = self.parse_ident()?; + let mod_kind = if self.eat(&token::Semi) { + ModKind::Unloaded + } else { + self.expect(&token::OpenDelim(Delimiter::Brace))?; + let (inner_attrs, items, inner_span) = + self.parse_mod(&token::CloseDelim(Delimiter::Brace))?; + attrs.extend(inner_attrs); + ModKind::Loaded(items, Inline::Yes, inner_span) + }; + Ok((id, ItemKind::Mod(safety, mod_kind))) + } + + /// Parses the contents of a module (inner attributes followed by module items). + /// We exit once we hit `term` + pub fn parse_mod( + &mut self, + term: &TokenKind, + ) -> PResult<'a, (AttrVec, ThinVec<P<Item>>, ModSpans)> { + let lo = self.token.span; + let attrs = self.parse_inner_attributes()?; + + let post_attr_lo = self.token.span; + let mut items = ThinVec::new(); + while let Some(item) = self.parse_item(ForceCollect::No)? { + self.maybe_consume_incorrect_semicolon(Some(&item)); + items.push(item); + } + + if !self.eat(term) { + let token_str = super::token_descr(&self.token); + if !self.maybe_consume_incorrect_semicolon(items.last().map(|x| &**x)) { + let msg = format!("expected item, found {token_str}"); + let mut err = self.dcx().struct_span_err(self.token.span, msg); + let span = self.token.span; + if self.is_kw_followed_by_ident(kw::Let) { + err.span_label( + span, + "consider using `const` or `static` instead of `let` for global variables", + ); + } else { + err.span_label(span, "expected item") + .note("for a full list of items that can appear in modules, see <https://doc.rust-lang.org/reference/items.html>"); + }; + return Err(err); + } + } + + let inject_use_span = post_attr_lo.data().with_hi(post_attr_lo.lo()); + let mod_spans = ModSpans { inner_span: lo.to(self.prev_token.span), inject_use_span }; + Ok((attrs, items, mod_spans)) + } +} + +pub(super) type ItemInfo = (Ident, ItemKind); + +impl<'a> Parser<'a> { + pub fn parse_item(&mut self, force_collect: ForceCollect) -> PResult<'a, Option<P<Item>>> { + let fn_parse_mode = FnParseMode { req_name: |_| true, req_body: true }; + self.parse_item_(fn_parse_mode, force_collect).map(|i| i.map(P)) + } + + fn parse_item_( + &mut self, + fn_parse_mode: FnParseMode, + force_collect: ForceCollect, + ) -> PResult<'a, Option<Item>> { + self.recover_vcs_conflict_marker(); + let attrs = self.parse_outer_attributes()?; + self.recover_vcs_conflict_marker(); + self.parse_item_common(attrs, true, false, fn_parse_mode, force_collect) + } + + pub(super) fn parse_item_common( + &mut self, + attrs: AttrWrapper, + mac_allowed: bool, + attrs_allowed: bool, + fn_parse_mode: FnParseMode, + force_collect: ForceCollect, + ) -> PResult<'a, Option<Item>> { + maybe_whole!(self, NtItem, |item| { + attrs.prepend_to_nt_inner(&mut item.attrs); + Some(item.into_inner()) + }); + + let item = + self.collect_tokens_trailing_token(attrs, force_collect, |this: &mut Self, attrs| { + let item = + this.parse_item_common_(attrs, mac_allowed, attrs_allowed, fn_parse_mode); + Ok((item?, TrailingToken::None)) + })?; + + Ok(item) + } + + fn parse_item_common_( + &mut self, + mut attrs: AttrVec, + mac_allowed: bool, + attrs_allowed: bool, + fn_parse_mode: FnParseMode, + ) -> PResult<'a, Option<Item>> { + let lo = self.token.span; + let vis = self.parse_visibility(FollowedByType::No)?; + let mut def = self.parse_defaultness(); + let kind = self.parse_item_kind( + &mut attrs, + mac_allowed, + lo, + &vis, + &mut def, + fn_parse_mode, + Case::Sensitive, + )?; + if let Some((ident, kind)) = kind { + self.error_on_unconsumed_default(def, &kind); + let span = lo.to(self.prev_token.span); + let id = DUMMY_NODE_ID; + let item = Item { ident, attrs, id, kind, vis, span, tokens: None }; + return Ok(Some(item)); + } + + // At this point, we have failed to parse an item. + if !matches!(vis.kind, VisibilityKind::Inherited) { + self.dcx().emit_err(errors::VisibilityNotFollowedByItem { span: vis.span, vis }); + } + + if let Defaultness::Default(span) = def { + self.dcx().emit_err(errors::DefaultNotFollowedByItem { span }); + } + + if !attrs_allowed { + self.recover_attrs_no_item(&attrs)?; + } + Ok(None) + } + + /// Error in-case `default` was parsed in an in-appropriate context. + fn error_on_unconsumed_default(&self, def: Defaultness, kind: &ItemKind) { + if let Defaultness::Default(span) = def { + self.dcx().emit_err(errors::InappropriateDefault { + span, + article: kind.article(), + descr: kind.descr(), + }); + } + } + + /// Parses one of the items allowed by the flags. + fn parse_item_kind( + &mut self, + attrs: &mut AttrVec, + macros_allowed: bool, + lo: Span, + vis: &Visibility, + def: &mut Defaultness, + fn_parse_mode: FnParseMode, + case: Case, + ) -> PResult<'a, Option<ItemInfo>> { + let check_pub = def == &Defaultness::Final; + let mut def_ = || mem::replace(def, Defaultness::Final); + + let info = if self.eat_keyword_case(kw::Use, case) { + self.parse_use_item()? + } else if self.check_fn_front_matter(check_pub, case) { + // FUNCTION ITEM + let (ident, sig, generics, body) = + self.parse_fn(attrs, fn_parse_mode, lo, vis, case)?; + (ident, ItemKind::Fn(Box::new(Fn { defaultness: def_(), sig, generics, body }))) + } else if self.eat_keyword(kw::Extern) { + if self.eat_keyword(kw::Crate) { + // EXTERN CRATE + self.parse_item_extern_crate()? + } else { + // EXTERN BLOCK + self.parse_item_foreign_mod(attrs, Safety::Default)? + } + } else if self.is_unsafe_foreign_mod() { + // EXTERN BLOCK + let safety = self.parse_safety(Case::Sensitive); + self.expect_keyword(kw::Extern)?; + self.parse_item_foreign_mod(attrs, safety)? + } else if self.is_static_global() { + // STATIC ITEM + self.bump(); // `static` + let mutability = self.parse_mutability(); + let (ident, item) = self.parse_static_item(mutability)?; + (ident, ItemKind::Static(Box::new(item))) + } else if let Const::Yes(const_span) = self.parse_constness(Case::Sensitive) { + // CONST ITEM + if self.token.is_keyword(kw::Impl) { + // recover from `const impl`, suggest `impl const` + self.recover_const_impl(const_span, attrs, def_())? + } else { + self.recover_const_mut(const_span); + let (ident, generics, ty, expr) = self.parse_const_item()?; + ( + ident, + ItemKind::Const(Box::new(ConstItem { + defaultness: def_(), + generics, + ty, + expr, + })), + ) + } + } else if self.check_keyword(kw::Trait) || self.check_auto_or_unsafe_trait_item() { + // TRAIT ITEM + self.parse_item_trait(attrs, lo)? + } else if self.check_keyword(kw::Impl) + || self.check_keyword(kw::Unsafe) && self.is_keyword_ahead(1, &[kw::Impl]) + { + // IMPL ITEM + self.parse_item_impl(attrs, def_())? + } else if self.is_reuse_path_item() { + self.parse_item_delegation()? + } else if self.check_keyword(kw::Mod) + || self.check_keyword(kw::Unsafe) && self.is_keyword_ahead(1, &[kw::Mod]) + { + // MODULE ITEM + self.parse_item_mod(attrs)? + } else if self.eat_keyword(kw::Type) { + // TYPE ITEM + self.parse_type_alias(def_())? + } else if self.eat_keyword(kw::Enum) { + // ENUM ITEM + self.parse_item_enum()? + } else if self.eat_keyword(kw::Struct) { + // STRUCT ITEM + self.parse_item_struct()? + } else if self.is_kw_followed_by_ident(kw::Union) { + // UNION ITEM + self.bump(); // `union` + self.parse_item_union()? + } else if self.is_builtin() { + // BUILTIN# ITEM + return self.parse_item_builtin(); + } else if self.eat_keyword(kw::Macro) { + // MACROS 2.0 ITEM + self.parse_item_decl_macro(lo)? + } else if let IsMacroRulesItem::Yes { has_bang } = self.is_macro_rules_item() { + // MACRO_RULES ITEM + self.parse_item_macro_rules(vis, has_bang)? + } else if self.isnt_macro_invocation() + && (self.token.is_ident_named(sym::import) + || self.token.is_ident_named(sym::using) + || self.token.is_ident_named(sym::include) + || self.token.is_ident_named(sym::require)) + { + return self.recover_import_as_use(); + } else if self.isnt_macro_invocation() && vis.kind.is_pub() { + self.recover_missing_kw_before_item()?; + return Ok(None); + } else if self.isnt_macro_invocation() && case == Case::Sensitive { + _ = def_; + + // Recover wrong cased keywords + return self.parse_item_kind( + attrs, + macros_allowed, + lo, + vis, + def, + fn_parse_mode, + Case::Insensitive, + ); + } else if macros_allowed && self.check_path() { + // MACRO INVOCATION ITEM + (Ident::empty(), ItemKind::MacCall(P(self.parse_item_macro(vis)?))) + } else { + return Ok(None); + }; + Ok(Some(info)) + } + + fn recover_import_as_use(&mut self) -> PResult<'a, Option<ItemInfo>> { + let span = self.token.span; + let token_name = super::token_descr(&self.token); + let snapshot = self.create_snapshot_for_diagnostic(); + self.bump(); + match self.parse_use_item() { + Ok(u) => { + self.dcx().emit_err(errors::RecoverImportAsUse { span, token_name }); + Ok(Some(u)) + } + Err(e) => { + e.cancel(); + self.restore_snapshot(snapshot); + Ok(None) + } + } + } + + fn parse_use_item(&mut self) -> PResult<'a, ItemInfo> { + let tree = self.parse_use_tree()?; + if let Err(mut e) = self.expect_semi() { + match tree.kind { + UseTreeKind::Glob => { + e.note("the wildcard token must be last on the path"); + } + UseTreeKind::Nested { .. } => { + e.note("glob-like brace syntax must be last on the path"); + } + _ => (), + } + return Err(e); + } + Ok((Ident::empty(), ItemKind::Use(tree))) + } + + /// When parsing a statement, would the start of a path be an item? + pub(super) fn is_path_start_item(&mut self) -> bool { + self.is_kw_followed_by_ident(kw::Union) // no: `union::b`, yes: `union U { .. }` + || self.is_reuse_path_item() + || self.check_auto_or_unsafe_trait_item() // no: `auto::b`, yes: `auto trait X { .. }` + || self.is_async_fn() // no(2015): `async::b`, yes: `async fn` + || matches!(self.is_macro_rules_item(), IsMacroRulesItem::Yes{..}) // no: `macro_rules::b`, yes: `macro_rules! mac` + } + + fn is_reuse_path_item(&mut self) -> bool { + // no: `reuse ::path` for compatibility reasons with macro invocations + self.token.is_keyword(kw::Reuse) + && self.look_ahead(1, |t| t.is_path_start() && t.kind != token::PathSep) + } + + /// Are we sure this could not possibly be a macro invocation? + fn isnt_macro_invocation(&mut self) -> bool { + self.check_ident() && self.look_ahead(1, |t| *t != token::Not && *t != token::PathSep) + } + + /// Recover on encountering a struct or method definition where the user + /// forgot to add the `struct` or `fn` keyword after writing `pub`: `pub S {}`. + fn recover_missing_kw_before_item(&mut self) -> PResult<'a, ()> { + // Space between `pub` keyword and the identifier + // + // pub S {} + // ^^^ `sp` points here + let sp = self.prev_token.span.between(self.token.span); + let full_sp = self.prev_token.span.to(self.token.span); + let ident_sp = self.token.span; + + let ident = if self.look_ahead(1, |t| { + [ + token::Lt, + token::OpenDelim(Delimiter::Brace), + token::OpenDelim(Delimiter::Parenthesis), + ] + .contains(&t.kind) + }) { + self.parse_ident().unwrap() + } else { + return Ok(()); + }; + + let mut found_generics = false; + if self.check(&token::Lt) { + found_generics = true; + self.eat_to_tokens(&[&token::Gt]); + self.bump(); // `>` + } + + let err = if self.check(&token::OpenDelim(Delimiter::Brace)) { + // possible public struct definition where `struct` was forgotten + Some(errors::MissingKeywordForItemDefinition::Struct { span: sp, ident }) + } else if self.check(&token::OpenDelim(Delimiter::Parenthesis)) { + // possible public function or tuple struct definition where `fn`/`struct` was + // forgotten + self.bump(); // `(` + let is_method = self.recover_self_param(); + + self.consume_block(Delimiter::Parenthesis, ConsumeClosingDelim::Yes); + + let err = + if self.check(&token::RArrow) || self.check(&token::OpenDelim(Delimiter::Brace)) { + self.eat_to_tokens(&[&token::OpenDelim(Delimiter::Brace)]); + self.bump(); // `{` + self.consume_block(Delimiter::Brace, ConsumeClosingDelim::Yes); + if is_method { + errors::MissingKeywordForItemDefinition::Method { span: sp, ident } + } else { + errors::MissingKeywordForItemDefinition::Function { span: sp, ident } + } + } else if self.check(&token::Semi) { + errors::MissingKeywordForItemDefinition::Struct { span: sp, ident } + } else { + errors::MissingKeywordForItemDefinition::Ambiguous { + span: sp, + subdiag: if found_generics { + None + } else if let Ok(snippet) = self.span_to_snippet(ident_sp) { + Some(errors::AmbiguousMissingKwForItemSub::SuggestMacro { + span: full_sp, + snippet, + }) + } else { + Some(errors::AmbiguousMissingKwForItemSub::HelpMacro) + }, + } + }; + Some(err) + } else if found_generics { + Some(errors::MissingKeywordForItemDefinition::Ambiguous { span: sp, subdiag: None }) + } else { + None + }; + + if let Some(err) = err { Err(self.dcx().create_err(err)) } else { Ok(()) } + } + + fn parse_item_builtin(&mut self) -> PResult<'a, Option<ItemInfo>> { + // To be expanded + return Ok(None); + } + + /// Parses an item macro, e.g., `item!();`. + fn parse_item_macro(&mut self, vis: &Visibility) -> PResult<'a, MacCall> { + let path = self.parse_path(PathStyle::Mod)?; // `foo::bar` + self.expect(&token::Not)?; // `!` + match self.parse_delim_args() { + // `( .. )` or `[ .. ]` (followed by `;`), or `{ .. }`. + Ok(args) => { + self.eat_semi_for_macro_if_needed(&args); + self.complain_if_pub_macro(vis, false); + Ok(MacCall { path, args }) + } + + Err(mut err) => { + // Maybe the user misspelled `macro_rules` (issue #91227) + if self.token.is_ident() + && path.segments.len() == 1 + && edit_distance("macro_rules", &path.segments[0].ident.to_string(), 2) + .is_some() + { + err.span_suggestion( + path.span, + "perhaps you meant to define a macro", + "macro_rules", + Applicability::MachineApplicable, + ); + } + Err(err) + } + } + } + + /// Recover if we parsed attributes and expected an item but there was none. + fn recover_attrs_no_item(&mut self, attrs: &[Attribute]) -> PResult<'a, ()> { + let ([start @ end] | [start, .., end]) = attrs else { + return Ok(()); + }; + let msg = if end.is_doc_comment() { + "expected item after doc comment" + } else { + "expected item after attributes" + }; + let mut err = self.dcx().struct_span_err(end.span, msg); + if end.is_doc_comment() { + err.span_label(end.span, "this doc comment doesn't document anything"); + } else if self.token.kind == TokenKind::Semi { + err.span_suggestion_verbose( + self.token.span, + "consider removing this semicolon", + "", + Applicability::MaybeIncorrect, + ); + } + if let [.., penultimate, _] = attrs { + err.span_label(start.span.to(penultimate.span), "other attributes here"); + } + Err(err) + } + + fn is_async_fn(&self) -> bool { + self.token.is_keyword(kw::Async) && self.is_keyword_ahead(1, &[kw::Fn]) + } + + fn parse_polarity(&mut self) -> ast::ImplPolarity { + // Disambiguate `impl !Trait for Type { ... }` and `impl ! { ... }` for the never type. + if self.check(&token::Not) && self.look_ahead(1, |t| t.can_begin_type()) { + self.bump(); // `!` + ast::ImplPolarity::Negative(self.prev_token.span) + } else { + ast::ImplPolarity::Positive + } + } + + /// Parses an implementation item. + /// + /// ```ignore (illustrative) + /// impl<'a, T> TYPE { /* impl items */ } + /// impl<'a, T> TRAIT for TYPE { /* impl items */ } + /// impl<'a, T> !TRAIT for TYPE { /* impl items */ } + /// impl<'a, T> const TRAIT for TYPE { /* impl items */ } + /// ``` + /// + /// We actually parse slightly more relaxed grammar for better error reporting and recovery. + /// ```ebnf + /// "impl" GENERICS "const"? "!"? TYPE "for"? (TYPE | "..") ("where" PREDICATES)? "{" BODY "}" + /// "impl" GENERICS "const"? "!"? TYPE ("where" PREDICATES)? "{" BODY "}" + /// ``` + fn parse_item_impl( + &mut self, + attrs: &mut AttrVec, + defaultness: Defaultness, + ) -> PResult<'a, ItemInfo> { + let safety = self.parse_safety(Case::Sensitive); + self.expect_keyword(kw::Impl)?; + + // First, parse generic parameters if necessary. + let mut generics = if self.choose_generics_over_qpath(0) { + self.parse_generics()? + } else { + let mut generics = Generics::default(); + // impl A for B {} + // /\ this is where `generics.span` should point when there are no type params. + generics.span = self.prev_token.span.shrink_to_hi(); + generics + }; + + let constness = self.parse_constness(Case::Sensitive); + if let Const::Yes(span) = constness { + self.psess.gated_spans.gate(sym::const_trait_impl, span); + } + + // Parse stray `impl async Trait` + if (self.token.uninterpolated_span().at_least_rust_2018() + && self.token.is_keyword(kw::Async)) + || self.is_kw_followed_by_ident(kw::Async) + { + self.bump(); + self.dcx().emit_err(errors::AsyncImpl { span: self.prev_token.span }); + } + + let polarity = self.parse_polarity(); + + // Parse both types and traits as a type, then reinterpret if necessary. + let err_path = |span| ast::Path::from_ident(Ident::new(kw::Empty, span)); + let ty_first = if self.token.is_keyword(kw::For) && self.look_ahead(1, |t| t != &token::Lt) + { + let span = self.prev_token.span.between(self.token.span); + self.dcx().emit_err(errors::MissingTraitInTraitImpl { + span, + for_span: span.to(self.token.span), + }); + + P(Ty { + kind: TyKind::Path(None, err_path(span)), + span, + id: DUMMY_NODE_ID, + tokens: None, + }) + } else { + self.parse_ty_with_generics_recovery(&generics)? + }; + + // If `for` is missing we try to recover. + let has_for = self.eat_keyword(kw::For); + let missing_for_span = self.prev_token.span.between(self.token.span); + + let ty_second = if self.token == token::DotDot { + // We need to report this error after `cfg` expansion for compatibility reasons + self.bump(); // `..`, do not add it to expected tokens + + // AST validation later detects this `TyKind::Dummy` and emits an + // error. (#121072 will hopefully remove all this special handling + // of the obsolete `impl Trait for ..` and then this can go away.) + Some(self.mk_ty(self.prev_token.span, TyKind::Dummy)) + } else if has_for || self.token.can_begin_type() { + Some(self.parse_ty()?) + } else { + None + }; + + generics.where_clause = self.parse_where_clause()?; + + let impl_items = self.parse_item_list(attrs, |p| p.parse_impl_item(ForceCollect::No))?; + + let item_kind = match ty_second { + Some(ty_second) => { + // impl Trait for Type + if !has_for { + self.dcx().emit_err(errors::MissingForInTraitImpl { span: missing_for_span }); + } + + let ty_first = ty_first.into_inner(); + let path = match ty_first.kind { + // This notably includes paths passed through `ty` macro fragments (#46438). + TyKind::Path(None, path) => path, + other => { + if let TyKind::ImplTrait(_, bounds, None) = other + && let [bound] = bounds.as_slice() + { + // Suggest removing extra `impl` keyword: + // `impl<T: Default> impl Default for Wrapper<T>` + // ^^^^^ + let extra_impl_kw = ty_first.span.until(bound.span()); + self.dcx().emit_err(errors::ExtraImplKeywordInTraitImpl { + extra_impl_kw, + impl_trait_span: ty_first.span, + }); + } else { + self.dcx().emit_err(errors::ExpectedTraitInTraitImplFoundType { + span: ty_first.span, + }); + } + err_path(ty_first.span) + } + }; + let trait_ref = TraitRef { path, ref_id: ty_first.id }; + + ItemKind::Impl(Box::new(Impl { + safety, + polarity, + defaultness, + constness, + generics, + of_trait: Some(trait_ref), + self_ty: ty_second, + items: impl_items, + })) + } + None => { + // impl Type + ItemKind::Impl(Box::new(Impl { + safety, + polarity, + defaultness, + constness, + generics, + of_trait: None, + self_ty: ty_first, + items: impl_items, + })) + } + }; + + Ok((Ident::empty(), item_kind)) + } + + fn parse_item_delegation(&mut self) -> PResult<'a, ItemInfo> { + let span = self.token.span; + self.expect_keyword(kw::Reuse)?; + + let (qself, path) = if self.eat_lt() { + let (qself, path) = self.parse_qpath(PathStyle::Expr)?; + (Some(qself), path) + } else { + (None, self.parse_path(PathStyle::Expr)?) + }; + + let rename = |this: &mut Self| { + Ok(if this.eat_keyword(kw::As) { Some(this.parse_ident()?) } else { None }) + }; + let body = |this: &mut Self| { + Ok(if this.check(&token::OpenDelim(Delimiter::Brace)) { + Some(this.parse_block()?) + } else { + this.expect(&token::Semi)?; + None + }) + }; + + let (ident, item_kind) = if self.eat(&token::PathSep) { + let (suffixes, _) = self.parse_delim_comma_seq(Delimiter::Brace, |p| { + Ok((p.parse_path_segment_ident()?, rename(p)?)) + })?; + let deleg = DelegationMac { qself, prefix: path, suffixes, body: body(self)? }; + (Ident::empty(), ItemKind::DelegationMac(Box::new(deleg))) + } else { + let rename = rename(self)?; + let ident = rename.unwrap_or_else(|| path.segments.last().unwrap().ident); + let deleg = Delegation { id: DUMMY_NODE_ID, qself, path, rename, body: body(self)? }; + (ident, ItemKind::Delegation(Box::new(deleg))) + }; + + let span = span.to(self.prev_token.span); + self.psess.gated_spans.gate(sym::fn_delegation, span); + + Ok((ident, item_kind)) + } + + fn parse_item_list<T>( + &mut self, + attrs: &mut AttrVec, + mut parse_item: impl FnMut(&mut Parser<'a>) -> PResult<'a, Option<Option<T>>>, + ) -> PResult<'a, ThinVec<T>> { + let open_brace_span = self.token.span; + + // Recover `impl Ty;` instead of `impl Ty {}` + if self.token == TokenKind::Semi { + self.dcx().emit_err(errors::UseEmptyBlockNotSemi { span: self.token.span }); + self.bump(); + return Ok(ThinVec::new()); + } + + self.expect(&token::OpenDelim(Delimiter::Brace))?; + attrs.extend(self.parse_inner_attributes()?); + + let mut items = ThinVec::new(); + while !self.eat(&token::CloseDelim(Delimiter::Brace)) { + if self.recover_doc_comment_before_brace() { + continue; + } + self.recover_vcs_conflict_marker(); + match parse_item(self) { + Ok(None) => { + let mut is_unnecessary_semicolon = !items.is_empty() + // When the close delim is `)` in a case like the following, `token.kind` is expected to be `token::CloseDelim(Delimiter::Parenthesis)`, + // but the actual `token.kind` is `token::CloseDelim(Delimiter::Brace)`. + // This is because the `token.kind` of the close delim is treated as the same as + // that of the open delim in `TokenTreesReader::parse_token_tree`, even if the delimiters of them are different. + // Therefore, `token.kind` should not be compared here. + // + // issue-60075.rs + // ``` + // trait T { + // fn qux() -> Option<usize> { + // let _ = if true { + // }); + // ^ this close delim + // Some(4) + // } + // ``` + && self + .span_to_snippet(self.prev_token.span) + .is_ok_and(|snippet| snippet == "}") + && self.token.kind == token::Semi; + let mut semicolon_span = self.token.span; + if !is_unnecessary_semicolon { + // #105369, Detect spurious `;` before assoc fn body + is_unnecessary_semicolon = self.token == token::OpenDelim(Delimiter::Brace) + && self.prev_token.kind == token::Semi; + semicolon_span = self.prev_token.span; + } + // We have to bail or we'll potentially never make progress. + let non_item_span = self.token.span; + let is_let = self.token.is_keyword(kw::Let); + + let mut err = + self.dcx().struct_span_err(non_item_span, "non-item in item list"); + self.consume_block(Delimiter::Brace, ConsumeClosingDelim::Yes); + if is_let { + err.span_suggestion( + non_item_span, + "consider using `const` instead of `let` for associated const", + "const", + Applicability::MachineApplicable, + ); + } else { + err.span_label(open_brace_span, "item list starts here") + .span_label(non_item_span, "non-item starts here") + .span_label(self.prev_token.span, "item list ends here"); + } + if is_unnecessary_semicolon { + err.span_suggestion( + semicolon_span, + "consider removing this semicolon", + "", + Applicability::MaybeIncorrect, + ); + } + err.emit(); + break; + } + Ok(Some(item)) => items.extend(item), + Err(err) => { + self.consume_block(Delimiter::Brace, ConsumeClosingDelim::Yes); + err.with_span_label( + open_brace_span, + "while parsing this item list starting here", + ) + .with_span_label(self.prev_token.span, "the item list ends here") + .emit(); + break; + } + } + } + Ok(items) + } + + /// Recover on a doc comment before `}`. + fn recover_doc_comment_before_brace(&mut self) -> bool { + if let token::DocComment(..) = self.token.kind { + if self.look_ahead(1, |tok| tok == &token::CloseDelim(Delimiter::Brace)) { + // FIXME: merge with `DocCommentDoesNotDocumentAnything` (E0585) + struct_span_code_err!( + self.dcx(), + self.token.span, + E0584, + "found a documentation comment that doesn't document anything", + ) + .with_span_label(self.token.span, "this doc comment doesn't document anything") + .with_help( + "doc comments must come before what they document, if a comment was \ + intended use `//`", + ) + .emit(); + self.bump(); + return true; + } + } + false + } + + /// Parses defaultness (i.e., `default` or nothing). + fn parse_defaultness(&mut self) -> Defaultness { + // We are interested in `default` followed by another identifier. + // However, we must avoid keywords that occur as binary operators. + // Currently, the only applicable keyword is `as` (`default as Ty`). + if self.check_keyword(kw::Default) + && self.look_ahead(1, |t| t.is_non_raw_ident_where(|i| i.name != kw::As)) + { + self.bump(); // `default` + Defaultness::Default(self.prev_token.uninterpolated_span()) + } else { + Defaultness::Final + } + } + + /// Is this an `(unsafe auto? | auto) trait` item? + fn check_auto_or_unsafe_trait_item(&mut self) -> bool { + // auto trait + self.check_keyword(kw::Auto) && self.is_keyword_ahead(1, &[kw::Trait]) + // unsafe auto trait + || self.check_keyword(kw::Unsafe) && self.is_keyword_ahead(1, &[kw::Trait, kw::Auto]) + } + + /// Parses `unsafe? auto? trait Foo { ... }` or `trait Foo = Bar;`. + fn parse_item_trait(&mut self, attrs: &mut AttrVec, lo: Span) -> PResult<'a, ItemInfo> { + let safety = self.parse_safety(Case::Sensitive); + // Parse optional `auto` prefix. + let is_auto = if self.eat_keyword(kw::Auto) { + self.psess.gated_spans.gate(sym::auto_traits, self.prev_token.span); + IsAuto::Yes + } else { + IsAuto::No + }; + + self.expect_keyword(kw::Trait)?; + let ident = self.parse_ident()?; + let mut generics = self.parse_generics()?; + + // Parse optional colon and supertrait bounds. + let had_colon = self.eat(&token::Colon); + let span_at_colon = self.prev_token.span; + let bounds = if had_colon { self.parse_generic_bounds()? } else { Vec::new() }; + + let span_before_eq = self.prev_token.span; + if self.eat(&token::Eq) { + // It's a trait alias. + if had_colon { + let span = span_at_colon.to(span_before_eq); + self.dcx().emit_err(errors::BoundsNotAllowedOnTraitAliases { span }); + } + + let bounds = self.parse_generic_bounds()?; + generics.where_clause = self.parse_where_clause()?; + self.expect_semi()?; + + let whole_span = lo.to(self.prev_token.span); + if is_auto == IsAuto::Yes { + self.dcx().emit_err(errors::TraitAliasCannotBeAuto { span: whole_span }); + } + if let Safety::Unsafe(_) = safety { + self.dcx().emit_err(errors::TraitAliasCannotBeUnsafe { span: whole_span }); + } + + self.psess.gated_spans.gate(sym::trait_alias, whole_span); + + Ok((ident, ItemKind::TraitAlias(generics, bounds))) + } else { + // It's a normal trait. + generics.where_clause = self.parse_where_clause()?; + let items = self.parse_item_list(attrs, |p| p.parse_trait_item(ForceCollect::No))?; + Ok(( + ident, + ItemKind::Trait(Box::new(Trait { is_auto, safety, generics, bounds, items })), + )) + } + } + + pub fn parse_impl_item( + &mut self, + force_collect: ForceCollect, + ) -> PResult<'a, Option<Option<P<AssocItem>>>> { + let fn_parse_mode = FnParseMode { req_name: |_| true, req_body: true }; + self.parse_assoc_item(fn_parse_mode, force_collect) + } + + pub fn parse_trait_item( + &mut self, + force_collect: ForceCollect, + ) -> PResult<'a, Option<Option<P<AssocItem>>>> { + let fn_parse_mode = + FnParseMode { req_name: |edition| edition >= Edition::Edition2018, req_body: false }; + self.parse_assoc_item(fn_parse_mode, force_collect) + } + + /// Parses associated items. + fn parse_assoc_item( + &mut self, + fn_parse_mode: FnParseMode, + force_collect: ForceCollect, + ) -> PResult<'a, Option<Option<P<AssocItem>>>> { + Ok(self.parse_item_(fn_parse_mode, force_collect)?.map( + |Item { attrs, id, span, vis, ident, kind, tokens }| { + let kind = match AssocItemKind::try_from(kind) { + Ok(kind) => kind, + Err(kind) => match kind { + ItemKind::Static(box StaticItem { ty, mutability: _, expr }) => { + self.dcx().emit_err(errors::AssociatedStaticItemNotAllowed { span }); + AssocItemKind::Const(Box::new(ConstItem { + defaultness: Defaultness::Final, + generics: Generics::default(), + ty, + expr, + })) + } + _ => return self.error_bad_item_kind(span, &kind, "`trait`s or `impl`s"), + }, + }; + Some(P(Item { attrs, id, span, vis, ident, kind, tokens })) + }, + )) + } + + /// Parses a `type` alias with the following grammar: + /// ```ebnf + /// TypeAlias = "type" Ident Generics (":" GenericBounds)? WhereClause ("=" Ty)? WhereClause ";" ; + /// ``` + /// The `"type"` has already been eaten. + fn parse_type_alias(&mut self, defaultness: Defaultness) -> PResult<'a, ItemInfo> { + let ident = self.parse_ident()?; + let mut generics = self.parse_generics()?; + + // Parse optional colon and param bounds. + let bounds = + if self.eat(&token::Colon) { self.parse_generic_bounds()? } else { Vec::new() }; + let before_where_clause = self.parse_where_clause()?; + + let ty = if self.eat(&token::Eq) { Some(self.parse_ty()?) } else { None }; + + let after_where_clause = self.parse_where_clause()?; + + let where_clauses = TyAliasWhereClauses { + before: TyAliasWhereClause { + has_where_token: before_where_clause.has_where_token, + span: before_where_clause.span, + }, + after: TyAliasWhereClause { + has_where_token: after_where_clause.has_where_token, + span: after_where_clause.span, + }, + split: before_where_clause.predicates.len(), + }; + let mut predicates = before_where_clause.predicates; + predicates.extend(after_where_clause.predicates); + let where_clause = WhereClause { + has_where_token: before_where_clause.has_where_token + || after_where_clause.has_where_token, + predicates, + span: DUMMY_SP, + }; + generics.where_clause = where_clause; + + self.expect_semi()?; + + Ok(( + ident, + ItemKind::TyAlias(Box::new(TyAlias { + defaultness, + generics, + where_clauses, + bounds, + ty, + })), + )) + } + + /// Parses a `UseTree`. + /// + /// ```text + /// USE_TREE = [`::`] `*` | + /// [`::`] `{` USE_TREE_LIST `}` | + /// PATH `::` `*` | + /// PATH `::` `{` USE_TREE_LIST `}` | + /// PATH [`as` IDENT] + /// ``` + fn parse_use_tree(&mut self) -> PResult<'a, UseTree> { + let lo = self.token.span; + + let mut prefix = + ast::Path { segments: ThinVec::new(), span: lo.shrink_to_lo(), tokens: None }; + let kind = if self.check(&token::OpenDelim(Delimiter::Brace)) + || self.check(&token::BinOp(token::Star)) + || self.is_import_coupler() + { + // `use *;` or `use ::*;` or `use {...};` or `use ::{...};` + let mod_sep_ctxt = self.token.span.ctxt(); + if self.eat(&token::PathSep) { + prefix + .segments + .push(PathSegment::path_root(lo.shrink_to_lo().with_ctxt(mod_sep_ctxt))); + } + + self.parse_use_tree_glob_or_nested()? + } else { + // `use path::*;` or `use path::{...};` or `use path;` or `use path as bar;` + prefix = self.parse_path(PathStyle::Mod)?; + + if self.eat(&token::PathSep) { + self.parse_use_tree_glob_or_nested()? + } else { + // Recover from using a colon as path separator. + while self.eat_noexpect(&token::Colon) { + self.dcx() + .emit_err(errors::SingleColonImportPath { span: self.prev_token.span }); + + // We parse the rest of the path and append it to the original prefix. + self.parse_path_segments(&mut prefix.segments, PathStyle::Mod, None)?; + prefix.span = lo.to(self.prev_token.span); + } + + UseTreeKind::Simple(self.parse_rename()?) + } + }; + + Ok(UseTree { prefix, kind, span: lo.to(self.prev_token.span) }) + } + + /// Parses `*` or `{...}`. + fn parse_use_tree_glob_or_nested(&mut self) -> PResult<'a, UseTreeKind> { + Ok(if self.eat(&token::BinOp(token::Star)) { + UseTreeKind::Glob + } else { + let lo = self.token.span; + UseTreeKind::Nested { + items: self.parse_use_tree_list()?, + span: lo.to(self.prev_token.span), + } + }) + } + + /// Parses a `UseTreeKind::Nested(list)`. + /// + /// ```text + /// USE_TREE_LIST = ∅ | (USE_TREE `,`)* USE_TREE [`,`] + /// ``` + fn parse_use_tree_list(&mut self) -> PResult<'a, ThinVec<(UseTree, ast::NodeId)>> { + self.parse_delim_comma_seq(Delimiter::Brace, |p| { + p.recover_vcs_conflict_marker(); + Ok((p.parse_use_tree()?, DUMMY_NODE_ID)) + }) + .map(|(r, _)| r) + } + + fn parse_rename(&mut self) -> PResult<'a, Option<Ident>> { + if self.eat_keyword(kw::As) { self.parse_ident_or_underscore().map(Some) } else { Ok(None) } + } + + fn parse_ident_or_underscore(&mut self) -> PResult<'a, Ident> { + match self.token.ident() { + Some((ident @ Ident { name: kw::Underscore, .. }, IdentIsRaw::No)) => { + self.bump(); + Ok(ident) + } + _ => self.parse_ident(), + } + } + + /// Parses `extern crate` links. + /// + /// # Examples + /// + /// ```ignore (illustrative) + /// extern crate foo; + /// extern crate bar as foo; + /// ``` + fn parse_item_extern_crate(&mut self) -> PResult<'a, ItemInfo> { + // Accept `extern crate name-like-this` for better diagnostics + let orig_name = self.parse_crate_name_with_dashes()?; + let (item_name, orig_name) = if let Some(rename) = self.parse_rename()? { + (rename, Some(orig_name.name)) + } else { + (orig_name, None) + }; + self.expect_semi()?; + Ok((item_name, ItemKind::ExternCrate(orig_name))) + } + + fn parse_crate_name_with_dashes(&mut self) -> PResult<'a, Ident> { + let ident = if self.token.is_keyword(kw::SelfLower) { + self.parse_path_segment_ident() + } else { + self.parse_ident() + }?; + + let dash = token::BinOp(token::BinOpToken::Minus); + if self.token != dash { + return Ok(ident); + } + + // Accept `extern crate name-like-this` for better diagnostics. + let mut dashes = vec![]; + let mut idents = vec![]; + while self.eat(&dash) { + dashes.push(self.prev_token.span); + idents.push(self.parse_ident()?); + } + + let fixed_name_sp = ident.span.to(idents.last().unwrap().span); + let mut fixed_name = ident.name.to_string(); + for part in idents { + write!(fixed_name, "_{}", part.name).unwrap(); + } + + self.dcx().emit_err(errors::ExternCrateNameWithDashes { + span: fixed_name_sp, + sugg: errors::ExternCrateNameWithDashesSugg { dashes }, + }); + + Ok(Ident::from_str_and_span(&fixed_name, fixed_name_sp)) + } + + /// Parses `extern` for foreign ABIs modules. + /// + /// `extern` is expected to have been consumed before calling this method. + /// + /// # Examples + /// + /// ```ignore (only-for-syntax-highlight) + /// extern "C" {} + /// extern {} + /// ``` + fn parse_item_foreign_mod( + &mut self, + attrs: &mut AttrVec, + mut safety: Safety, + ) -> PResult<'a, ItemInfo> { + let abi = self.parse_abi(); // ABI? + if safety == Safety::Default + && self.token.is_keyword(kw::Unsafe) + && self.look_ahead(1, |t| t.kind == token::OpenDelim(Delimiter::Brace)) + { + self.expect(&token::OpenDelim(Delimiter::Brace)).unwrap_err().emit(); + safety = Safety::Unsafe(self.token.span); + self.eat_keyword(kw::Unsafe); + } + let module = ast::ForeignMod { + safety, + abi, + items: self.parse_item_list(attrs, |p| p.parse_foreign_item(ForceCollect::No))?, + }; + Ok((Ident::empty(), ItemKind::ForeignMod(module))) + } + + /// Parses a foreign item (one in an `extern { ... }` block). + pub fn parse_foreign_item( + &mut self, + force_collect: ForceCollect, + ) -> PResult<'a, Option<Option<P<ForeignItem>>>> { + let fn_parse_mode = FnParseMode { req_name: |_| true, req_body: false }; + Ok(self.parse_item_(fn_parse_mode, force_collect)?.map( + |Item { attrs, id, span, vis, ident, kind, tokens }| { + let kind = match ForeignItemKind::try_from(kind) { + Ok(kind) => kind, + Err(kind) => match kind { + ItemKind::Const(box ConstItem { ty, expr, .. }) => { + let const_span = Some(span.with_hi(ident.span.lo())) + .filter(|span| span.can_be_used_for_suggestions()); + self.dcx().emit_err(errors::ExternItemCannotBeConst { + ident_span: ident.span, + const_span, + }); + ForeignItemKind::Static(Box::new(StaticForeignItem { + ty, + mutability: Mutability::Not, + expr, + })) + } + _ => return self.error_bad_item_kind(span, &kind, "`extern` blocks"), + }, + }; + Some(P(Item { attrs, id, span, vis, ident, kind, tokens })) + }, + )) + } + + fn error_bad_item_kind<T>(&self, span: Span, kind: &ItemKind, ctx: &'static str) -> Option<T> { + // FIXME(#100717): needs variant for each `ItemKind` (instead of using `ItemKind::descr()`) + let span = self.psess.source_map().guess_head_span(span); + let descr = kind.descr(); + self.dcx().emit_err(errors::BadItemKind { span, descr, ctx }); + None + } + + fn is_unsafe_foreign_mod(&self) -> bool { + self.token.is_keyword(kw::Unsafe) + && self.is_keyword_ahead(1, &[kw::Extern]) + && self.look_ahead( + 2 + self.look_ahead(2, |t| t.can_begin_literal_maybe_minus() as usize), + |t| t.kind == token::OpenDelim(Delimiter::Brace), + ) + } + + fn is_static_global(&mut self) -> bool { + if self.check_keyword(kw::Static) { + // Check if this could be a closure. + !self.look_ahead(1, |token| { + if token.is_keyword(kw::Move) { + return true; + } + matches!(token.kind, token::BinOp(token::Or) | token::OrOr) + }) + } else { + false + } + } + + /// Recover on `const mut` with `const` already eaten. + fn recover_const_mut(&mut self, const_span: Span) { + if self.eat_keyword(kw::Mut) { + let span = self.prev_token.span; + self.dcx() + .emit_err(errors::ConstGlobalCannotBeMutable { ident_span: span, const_span }); + } else if self.eat_keyword(kw::Let) { + let span = self.prev_token.span; + self.dcx().emit_err(errors::ConstLetMutuallyExclusive { span: const_span.to(span) }); + } + } + + /// Recover on `const impl` with `const` already eaten. + fn recover_const_impl( + &mut self, + const_span: Span, + attrs: &mut AttrVec, + defaultness: Defaultness, + ) -> PResult<'a, ItemInfo> { + let impl_span = self.token.span; + let err = self.expected_ident_found_err(); + + // Only try to recover if this is implementing a trait for a type + let mut impl_info = match self.parse_item_impl(attrs, defaultness) { + Ok(impl_info) => impl_info, + Err(recovery_error) => { + // Recovery failed, raise the "expected identifier" error + recovery_error.cancel(); + return Err(err); + } + }; + + match &mut impl_info.1 { + ItemKind::Impl(box Impl { of_trait: Some(trai), constness, .. }) => { + *constness = Const::Yes(const_span); + + let before_trait = trai.path.span.shrink_to_lo(); + let const_up_to_impl = const_span.with_hi(impl_span.lo()); + err.with_multipart_suggestion( + "you might have meant to write a const trait impl", + vec![(const_up_to_impl, "".to_owned()), (before_trait, "const ".to_owned())], + Applicability::MaybeIncorrect, + ) + .emit(); + } + ItemKind::Impl { .. } => return Err(err), + _ => unreachable!(), + } + + Ok(impl_info) + } + + /// Parse a static item with the prefix `"static" "mut"?` already parsed and stored in `mutability`. + /// + /// ```ebnf + /// Static = "static" "mut"? $ident ":" $ty (= $expr)? ";" ; + /// ``` + fn parse_static_item(&mut self, mutability: Mutability) -> PResult<'a, (Ident, StaticItem)> { + let ident = self.parse_ident()?; + + if self.token.kind == TokenKind::Lt && self.may_recover() { + let generics = self.parse_generics()?; + self.dcx().emit_err(errors::StaticWithGenerics { span: generics.span }); + } + + // Parse the type of a static item. That is, the `":" $ty` fragment. + // FIXME: This could maybe benefit from `.may_recover()`? + let ty = match (self.eat(&token::Colon), self.check(&token::Eq) | self.check(&token::Semi)) + { + (true, false) => self.parse_ty()?, + // If there wasn't a `:` or the colon was followed by a `=` or `;`, recover a missing type. + (colon, _) => self.recover_missing_global_item_type(colon, Some(mutability)), + }; + + let expr = if self.eat(&token::Eq) { Some(self.parse_expr()?) } else { None }; + + self.expect_semi()?; + + Ok((ident, StaticItem { ty, mutability, expr })) + } + + /// Parse a constant item with the prefix `"const"` already parsed. + /// + /// ```ebnf + /// Const = "const" ($ident | "_") Generics ":" $ty (= $expr)? WhereClause ";" ; + /// ``` + fn parse_const_item(&mut self) -> PResult<'a, (Ident, Generics, P<Ty>, Option<P<ast::Expr>>)> { + let ident = self.parse_ident_or_underscore()?; + + let mut generics = self.parse_generics()?; + + // Check the span for emptiness instead of the list of parameters in order to correctly + // recognize and subsequently flag empty parameter lists (`<>`) as unstable. + if !generics.span.is_empty() { + self.psess.gated_spans.gate(sym::generic_const_items, generics.span); + } + + // Parse the type of a constant item. That is, the `":" $ty` fragment. + // FIXME: This could maybe benefit from `.may_recover()`? + let ty = match ( + self.eat(&token::Colon), + self.check(&token::Eq) | self.check(&token::Semi) | self.check_keyword(kw::Where), + ) { + (true, false) => self.parse_ty()?, + // If there wasn't a `:` or the colon was followed by a `=`, `;` or `where`, recover a missing type. + (colon, _) => self.recover_missing_global_item_type(colon, None), + }; + + // Proactively parse a where-clause to be able to provide a good error message in case we + // encounter the item body following it. + let before_where_clause = + if self.may_recover() { self.parse_where_clause()? } else { WhereClause::default() }; + + let expr = if self.eat(&token::Eq) { Some(self.parse_expr()?) } else { None }; + + let after_where_clause = self.parse_where_clause()?; + + // Provide a nice error message if the user placed a where-clause before the item body. + // Users may be tempted to write such code if they are still used to the deprecated + // where-clause location on type aliases and associated types. See also #89122. + if before_where_clause.has_where_token + && let Some(expr) = &expr + { + self.dcx().emit_err(errors::WhereClauseBeforeConstBody { + span: before_where_clause.span, + name: ident.span, + body: expr.span, + sugg: if !after_where_clause.has_where_token { + self.psess.source_map().span_to_snippet(expr.span).ok().map(|body| { + errors::WhereClauseBeforeConstBodySugg { + left: before_where_clause.span.shrink_to_lo(), + snippet: body, + right: before_where_clause.span.shrink_to_hi().to(expr.span), + } + }) + } else { + // FIXME(generic_const_items): Provide a structured suggestion to merge the first + // where-clause into the second one. + None + }, + }); + } + + // Merge the predicates of both where-clauses since either one can be relevant. + // If we didn't parse a body (which is valid for associated consts in traits) and we were + // allowed to recover, `before_where_clause` contains the predicates, otherwise they are + // in `after_where_clause`. Further, both of them might contain predicates iff two + // where-clauses were provided which is syntactically ill-formed but we want to recover from + // it and treat them as one large where-clause. + let mut predicates = before_where_clause.predicates; + predicates.extend(after_where_clause.predicates); + let where_clause = WhereClause { + has_where_token: before_where_clause.has_where_token + || after_where_clause.has_where_token, + predicates, + span: if after_where_clause.has_where_token { + after_where_clause.span + } else { + before_where_clause.span + }, + }; + + if where_clause.has_where_token { + self.psess.gated_spans.gate(sym::generic_const_items, where_clause.span); + } + + generics.where_clause = where_clause; + + self.expect_semi()?; + + Ok((ident, generics, ty, expr)) + } + + /// We were supposed to parse `":" $ty` but the `:` or the type was missing. + /// This means that the type is missing. + fn recover_missing_global_item_type( + &mut self, + colon_present: bool, + m: Option<Mutability>, + ) -> P<Ty> { + // Construct the error and stash it away with the hope + // that typeck will later enrich the error with a type. + let kind = match m { + Some(Mutability::Mut) => "static mut", + Some(Mutability::Not) => "static", + None => "const", + }; + + let colon = match colon_present { + true => "", + false => ":", + }; + + let span = self.prev_token.span.shrink_to_hi(); + let err = self.dcx().create_err(errors::MissingConstType { span, colon, kind }); + err.stash(span, StashKey::ItemNoType); + + // The user intended that the type be inferred, + // so treat this as if the user wrote e.g. `const A: _ = expr;`. + P(Ty { kind: TyKind::Infer, span, id: ast::DUMMY_NODE_ID, tokens: None }) + } + + /// Parses an enum declaration. + fn parse_item_enum(&mut self) -> PResult<'a, ItemInfo> { + if self.token.is_keyword(kw::Struct) { + let span = self.prev_token.span.to(self.token.span); + let err = errors::EnumStructMutuallyExclusive { span }; + if self.look_ahead(1, |t| t.is_ident()) { + self.bump(); + self.dcx().emit_err(err); + } else { + return Err(self.dcx().create_err(err)); + } + } + + let prev_span = self.prev_token.span; + let id = self.parse_ident()?; + let mut generics = self.parse_generics()?; + generics.where_clause = self.parse_where_clause()?; + + // Possibly recover `enum Foo;` instead of `enum Foo {}` + let (variants, _) = if self.token == TokenKind::Semi { + self.dcx().emit_err(errors::UseEmptyBlockNotSemi { span: self.token.span }); + self.bump(); + (thin_vec![], Trailing::No) + } else { + self.parse_delim_comma_seq(Delimiter::Brace, |p| p.parse_enum_variant(id.span)) + .map_err(|mut err| { + err.span_label(id.span, "while parsing this enum"); + if self.token == token::Colon { + let snapshot = self.create_snapshot_for_diagnostic(); + self.bump(); + match self.parse_ty() { + Ok(_) => { + err.span_suggestion_verbose( + prev_span, + "perhaps you meant to use `struct` here", + "struct", + Applicability::MaybeIncorrect, + ); + } + Err(e) => { + e.cancel(); + } + } + self.restore_snapshot(snapshot); + } + self.eat_to_tokens(&[&token::CloseDelim(Delimiter::Brace)]); + self.bump(); // } + err + })? + }; + + let enum_definition = EnumDef { variants: variants.into_iter().flatten().collect() }; + Ok((id, ItemKind::Enum(enum_definition, generics))) + } + + fn parse_enum_variant(&mut self, span: Span) -> PResult<'a, Option<Variant>> { + self.recover_vcs_conflict_marker(); + let variant_attrs = self.parse_outer_attributes()?; + self.recover_vcs_conflict_marker(); + let help = "enum variants can be `Variant`, `Variant = <integer>`, \ + `Variant(Type, ..., TypeN)` or `Variant { fields: Types }`"; + self.collect_tokens_trailing_token( + variant_attrs, + ForceCollect::No, + |this, variant_attrs| { + let vlo = this.token.span; + + let vis = this.parse_visibility(FollowedByType::No)?; + if !this.recover_nested_adt_item(kw::Enum)? { + return Ok((None, TrailingToken::None)); + } + let ident = this.parse_field_ident("enum", vlo)?; + + if this.token == token::Not { + if let Err(err) = this.unexpected() { + err.with_note(fluent::parse_macro_expands_to_enum_variant).emit(); + } + + this.bump(); + this.parse_delim_args()?; + + return Ok((None, TrailingToken::MaybeComma)); + } + + let struct_def = if this.check(&token::OpenDelim(Delimiter::Brace)) { + // Parse a struct variant. + let (fields, recovered) = + match this.parse_record_struct_body("struct", ident.span, false) { + Ok((fields, recovered)) => (fields, recovered), + Err(mut err) => { + if this.token == token::Colon { + // We handle `enum` to `struct` suggestion in the caller. + return Err(err); + } + this.eat_to_tokens(&[&token::CloseDelim(Delimiter::Brace)]); + this.bump(); // } + err.span_label(span, "while parsing this enum"); + err.help(help); + let guar = err.emit(); + (thin_vec![], Recovered::Yes(guar)) + } + }; + VariantData::Struct { fields, recovered: recovered.into() } + } else if this.check(&token::OpenDelim(Delimiter::Parenthesis)) { + let body = match this.parse_tuple_struct_body() { + Ok(body) => body, + Err(mut err) => { + if this.token == token::Colon { + // We handle `enum` to `struct` suggestion in the caller. + return Err(err); + } + this.eat_to_tokens(&[&token::CloseDelim(Delimiter::Parenthesis)]); + this.bump(); // ) + err.span_label(span, "while parsing this enum"); + err.help(help); + err.emit(); + thin_vec![] + } + }; + VariantData::Tuple(body, DUMMY_NODE_ID) + } else { + VariantData::Unit(DUMMY_NODE_ID) + }; + + let disr_expr = + if this.eat(&token::Eq) { Some(this.parse_expr_anon_const()?) } else { None }; + + let vr = ast::Variant { + ident, + vis, + id: DUMMY_NODE_ID, + attrs: variant_attrs, + data: struct_def, + disr_expr, + span: vlo.to(this.prev_token.span), + is_placeholder: false, + }; + + Ok((Some(vr), TrailingToken::MaybeComma)) + }, + ) + .map_err(|mut err| { + err.help(help); + err + }) + } + + /// Parses `struct Foo { ... }`. + fn parse_item_struct(&mut self) -> PResult<'a, ItemInfo> { + let class_name = self.parse_ident()?; + + let mut generics = self.parse_generics()?; + + // There is a special case worth noting here, as reported in issue #17904. + // If we are parsing a tuple struct it is the case that the where clause + // should follow the field list. Like so: + // + // struct Foo<T>(T) where T: Copy; + // + // If we are parsing a normal record-style struct it is the case + // that the where clause comes before the body, and after the generics. + // So if we look ahead and see a brace or a where-clause we begin + // parsing a record style struct. + // + // Otherwise if we look ahead and see a paren we parse a tuple-style + // struct. + + let vdata = if self.token.is_keyword(kw::Where) { + let tuple_struct_body; + (generics.where_clause, tuple_struct_body) = + self.parse_struct_where_clause(class_name, generics.span)?; + + if let Some(body) = tuple_struct_body { + // If we see a misplaced tuple struct body: `struct Foo<T> where T: Copy, (T);` + let body = VariantData::Tuple(body, DUMMY_NODE_ID); + self.expect_semi()?; + body + } else if self.eat(&token::Semi) { + // If we see a: `struct Foo<T> where T: Copy;` style decl. + VariantData::Unit(DUMMY_NODE_ID) + } else { + // If we see: `struct Foo<T> where T: Copy { ... }` + let (fields, recovered) = self.parse_record_struct_body( + "struct", + class_name.span, + generics.where_clause.has_where_token, + )?; + VariantData::Struct { fields, recovered: recovered.into() } + } + // No `where` so: `struct Foo<T>;` + } else if self.eat(&token::Semi) { + VariantData::Unit(DUMMY_NODE_ID) + // Record-style struct definition + } else if self.token == token::OpenDelim(Delimiter::Brace) { + let (fields, recovered) = self.parse_record_struct_body( + "struct", + class_name.span, + generics.where_clause.has_where_token, + )?; + VariantData::Struct { fields, recovered: recovered.into() } + // Tuple-style struct definition with optional where-clause. + } else if self.token == token::OpenDelim(Delimiter::Parenthesis) { + let body = VariantData::Tuple(self.parse_tuple_struct_body()?, DUMMY_NODE_ID); + generics.where_clause = self.parse_where_clause()?; + self.expect_semi()?; + body + } else { + let err = + errors::UnexpectedTokenAfterStructName::new(self.token.span, self.token.clone()); + return Err(self.dcx().create_err(err)); + }; + + Ok((class_name, ItemKind::Struct(vdata, generics))) + } + + /// Parses `union Foo { ... }`. + fn parse_item_union(&mut self) -> PResult<'a, ItemInfo> { + let class_name = self.parse_ident()?; + + let mut generics = self.parse_generics()?; + + let vdata = if self.token.is_keyword(kw::Where) { + generics.where_clause = self.parse_where_clause()?; + let (fields, recovered) = self.parse_record_struct_body( + "union", + class_name.span, + generics.where_clause.has_where_token, + )?; + VariantData::Struct { fields, recovered: recovered.into() } + } else if self.token == token::OpenDelim(Delimiter::Brace) { + let (fields, recovered) = self.parse_record_struct_body( + "union", + class_name.span, + generics.where_clause.has_where_token, + )?; + VariantData::Struct { fields, recovered: recovered.into() } + } else { + let token_str = super::token_descr(&self.token); + let msg = format!("expected `where` or `{{` after union name, found {token_str}"); + let mut err = self.dcx().struct_span_err(self.token.span, msg); + err.span_label(self.token.span, "expected `where` or `{` after union name"); + return Err(err); + }; + + Ok((class_name, ItemKind::Union(vdata, generics))) + } + + /// This function parses the fields of record structs: + /// + /// - `struct S { ... }` + /// - `enum E { Variant { ... } }` + pub(crate) fn parse_record_struct_body( + &mut self, + adt_ty: &str, + ident_span: Span, + parsed_where: bool, + ) -> PResult<'a, (ThinVec<FieldDef>, Recovered)> { + let mut fields = ThinVec::new(); + let mut recovered = Recovered::No; + if self.eat(&token::OpenDelim(Delimiter::Brace)) { + while self.token != token::CloseDelim(Delimiter::Brace) { + match self.parse_field_def(adt_ty) { + Ok(field) => { + fields.push(field); + } + Err(mut err) => { + self.consume_block(Delimiter::Brace, ConsumeClosingDelim::No); + err.span_label(ident_span, format!("while parsing this {adt_ty}")); + let guar = err.emit(); + recovered = Recovered::Yes(guar); + break; + } + } + } + self.eat(&token::CloseDelim(Delimiter::Brace)); + } else { + let token_str = super::token_descr(&self.token); + let where_str = if parsed_where { "" } else { "`where`, or " }; + let msg = format!("expected {where_str}`{{` after struct name, found {token_str}"); + let mut err = self.dcx().struct_span_err(self.token.span, msg); + err.span_label(self.token.span, format!("expected {where_str}`{{` after struct name",)); + return Err(err); + } + + Ok((fields, recovered)) + } + + pub(super) fn parse_tuple_struct_body(&mut self) -> PResult<'a, ThinVec<FieldDef>> { + // This is the case where we find `struct Foo<T>(T) where T: Copy;` + // Unit like structs are handled in parse_item_struct function + self.parse_paren_comma_seq(|p| { + let attrs = p.parse_outer_attributes()?; + p.collect_tokens_trailing_token(attrs, ForceCollect::No, |p, attrs| { + let mut snapshot = None; + if p.is_vcs_conflict_marker(&TokenKind::BinOp(token::Shl), &TokenKind::Lt) { + // Account for `<<<<<<<` diff markers. We can't proactively error here because + // that can be a valid type start, so we snapshot and reparse only we've + // encountered another parse error. + snapshot = Some(p.create_snapshot_for_diagnostic()); + } + let lo = p.token.span; + let vis = match p.parse_visibility(FollowedByType::Yes) { + Ok(vis) => vis, + Err(err) => { + if let Some(ref mut snapshot) = snapshot { + snapshot.recover_vcs_conflict_marker(); + } + return Err(err); + } + }; + let ty = match p.parse_ty() { + Ok(ty) => ty, + Err(err) => { + if let Some(ref mut snapshot) = snapshot { + snapshot.recover_vcs_conflict_marker(); + } + return Err(err); + } + }; + + Ok(( + FieldDef { + span: lo.to(ty.span), + vis, + ident: None, + id: DUMMY_NODE_ID, + ty, + attrs, + is_placeholder: false, + }, + TrailingToken::MaybeComma, + )) + }) + }) + .map(|(r, _)| r) + } + + /// Parses an element of a struct declaration. + fn parse_field_def(&mut self, adt_ty: &str) -> PResult<'a, FieldDef> { + self.recover_vcs_conflict_marker(); + let attrs = self.parse_outer_attributes()?; + self.recover_vcs_conflict_marker(); + self.collect_tokens_trailing_token(attrs, ForceCollect::No, |this, attrs| { + let lo = this.token.span; + let vis = this.parse_visibility(FollowedByType::No)?; + this.parse_single_struct_field(adt_ty, lo, vis, attrs) + .map(|field| (field, TrailingToken::None)) + }) + } + + /// Parses a structure field declaration. + fn parse_single_struct_field( + &mut self, + adt_ty: &str, + lo: Span, + vis: Visibility, + attrs: AttrVec, + ) -> PResult<'a, FieldDef> { + let mut seen_comma: bool = false; + let a_var = self.parse_name_and_ty(adt_ty, lo, vis, attrs)?; + if self.token == token::Comma { + seen_comma = true; + } + if self.eat(&token::Semi) { + let sp = self.prev_token.span; + let mut err = + self.dcx().struct_span_err(sp, format!("{adt_ty} fields are separated by `,`")); + err.span_suggestion_short( + sp, + "replace `;` with `,`", + ",", + Applicability::MachineApplicable, + ); + return Err(err); + } + match self.token.kind { + token::Comma => { + self.bump(); + } + token::CloseDelim(Delimiter::Brace) => {} + token::DocComment(..) => { + let previous_span = self.prev_token.span; + let mut err = errors::DocCommentDoesNotDocumentAnything { + span: self.token.span, + missing_comma: None, + }; + self.bump(); // consume the doc comment + let comma_after_doc_seen = self.eat(&token::Comma); + // `seen_comma` is always false, because we are inside doc block + // condition is here to make code more readable + if !seen_comma && comma_after_doc_seen { + seen_comma = true; + } + if comma_after_doc_seen || self.token == token::CloseDelim(Delimiter::Brace) { + self.dcx().emit_err(err); + } else { + if !seen_comma { + let sp = previous_span.shrink_to_hi(); + err.missing_comma = Some(sp); + } + return Err(self.dcx().create_err(err)); + } + } + _ => { + let sp = self.prev_token.span.shrink_to_hi(); + let msg = + format!("expected `,`, or `}}`, found {}", super::token_descr(&self.token)); + + // Try to recover extra trailing angle brackets + if let TyKind::Path(_, Path { segments, .. }) = &a_var.ty.kind { + if let Some(last_segment) = segments.last() { + let guar = self.check_trailing_angle_brackets( + last_segment, + &[&token::Comma, &token::CloseDelim(Delimiter::Brace)], + ); + if let Some(_guar) = guar { + // Handle a case like `Vec<u8>>,` where we can continue parsing fields + // after the comma + self.eat(&token::Comma); + + // `check_trailing_angle_brackets` already emitted a nicer error, as + // proven by the presence of `_guar`. We can continue parsing. + return Ok(a_var); + } + } + } + + let mut err = self.dcx().struct_span_err(sp, msg); + + if self.token.is_ident() + || (self.token.kind == TokenKind::Pound + && (self.look_ahead(1, |t| t == &token::OpenDelim(Delimiter::Bracket)))) + { + // This is likely another field, TokenKind::Pound is used for `#[..]` + // attribute for next field. Emit the diagnostic and continue parsing. + err.span_suggestion( + sp, + "try adding a comma", + ",", + Applicability::MachineApplicable, + ); + err.emit(); + } else { + return Err(err); + } + } + } + Ok(a_var) + } + + fn expect_field_ty_separator(&mut self) -> PResult<'a, ()> { + if let Err(err) = self.expect(&token::Colon) { + let sm = self.psess.source_map(); + let eq_typo = self.token.kind == token::Eq && self.look_ahead(1, |t| t.is_path_start()); + let semi_typo = self.token.kind == token::Semi + && self.look_ahead(1, |t| { + t.is_path_start() + // We check that we are in a situation like `foo; bar` to avoid bad suggestions + // when there's no type and `;` was used instead of a comma. + && match (sm.lookup_line(self.token.span.hi()), sm.lookup_line(t.span.lo())) { + (Ok(l), Ok(r)) => l.line == r.line, + _ => true, + } + }); + if eq_typo || semi_typo { + self.bump(); + // Gracefully handle small typos. + err.with_span_suggestion_short( + self.prev_token.span, + "field names and their types are separated with `:`", + ":", + Applicability::MachineApplicable, + ) + .emit(); + } else { + return Err(err); + } + } + Ok(()) + } + + /// Parses a structure field. + fn parse_name_and_ty( + &mut self, + adt_ty: &str, + lo: Span, + vis: Visibility, + attrs: AttrVec, + ) -> PResult<'a, FieldDef> { + let name = self.parse_field_ident(adt_ty, lo)?; + if self.token.kind == token::Not { + if let Err(mut err) = self.unexpected() { + // Encounter the macro invocation + err.subdiagnostic(self.dcx(), MacroExpandsToAdtField { adt_ty }); + return Err(err); + } + } + self.expect_field_ty_separator()?; + let ty = self.parse_ty_for_field_def()?; + if self.token.kind == token::Colon && self.look_ahead(1, |tok| tok.kind != token::Colon) { + self.dcx().emit_err(errors::SingleColonStructType { span: self.token.span }); + } + if self.token.kind == token::Eq { + self.bump(); + let const_expr = self.parse_expr_anon_const()?; + let sp = ty.span.shrink_to_hi().to(const_expr.value.span); + self.dcx().emit_err(errors::EqualsStructDefault { span: sp }); + } + Ok(FieldDef { + span: lo.to(self.prev_token.span), + ident: Some(name), + vis, + id: DUMMY_NODE_ID, + ty, + attrs, + is_placeholder: false, + }) + } + + /// Parses a field identifier. Specialized version of `parse_ident_common` + /// for better diagnostics and suggestions. + fn parse_field_ident(&mut self, adt_ty: &str, lo: Span) -> PResult<'a, Ident> { + let (ident, is_raw) = self.ident_or_err(true)?; + if ident.name == kw::Underscore { + self.psess.gated_spans.gate(sym::unnamed_fields, lo); + } else if matches!(is_raw, IdentIsRaw::No) && ident.is_reserved() { + let snapshot = self.create_snapshot_for_diagnostic(); + let err = if self.check_fn_front_matter(false, Case::Sensitive) { + let inherited_vis = + Visibility { span: DUMMY_SP, kind: VisibilityKind::Inherited, tokens: None }; + // We use `parse_fn` to get a span for the function + let fn_parse_mode = FnParseMode { req_name: |_| true, req_body: true }; + match self.parse_fn( + &mut AttrVec::new(), + fn_parse_mode, + lo, + &inherited_vis, + Case::Insensitive, + ) { + Ok(_) => { + self.dcx().struct_span_err( + lo.to(self.prev_token.span), + format!("functions are not allowed in {adt_ty} definitions"), + ) + .with_help( + "unlike in C++, Java, and C#, functions are declared in `impl` blocks", + ) + .with_help("see https://doc.rust-lang.org/book/ch05-03-method-syntax.html for more information") + } + Err(err) => { + err.cancel(); + self.restore_snapshot(snapshot); + self.expected_ident_found_err() + } + } + } else if self.eat_keyword(kw::Struct) { + match self.parse_item_struct() { + Ok((ident, _)) => self + .dcx() + .struct_span_err( + lo.with_hi(ident.span.hi()), + format!("structs are not allowed in {adt_ty} definitions"), + ) + .with_help( + "consider creating a new `struct` definition instead of nesting", + ), + Err(err) => { + err.cancel(); + self.restore_snapshot(snapshot); + self.expected_ident_found_err() + } + } + } else { + let mut err = self.expected_ident_found_err(); + if self.eat_keyword_noexpect(kw::Let) + && let removal_span = self.prev_token.span.until(self.token.span) + && let Ok(ident) = self + .parse_ident_common(false) + // Cancel this error, we don't need it. + .map_err(|err| err.cancel()) + && self.token.kind == TokenKind::Colon + { + err.span_suggestion( + removal_span, + "remove this `let` keyword", + String::new(), + Applicability::MachineApplicable, + ); + err.note("the `let` keyword is not allowed in `struct` fields"); + err.note("see <https://doc.rust-lang.org/book/ch05-01-defining-structs.html> for more information"); + err.emit(); + return Ok(ident); + } else { + self.restore_snapshot(snapshot); + } + err + }; + return Err(err); + } + self.bump(); + Ok(ident) + } + + /// Parses a declarative macro 2.0 definition. + /// The `macro` keyword has already been parsed. + /// ```ebnf + /// MacBody = "{" TOKEN_STREAM "}" ; + /// MacParams = "(" TOKEN_STREAM ")" ; + /// DeclMac = "macro" Ident MacParams? MacBody ; + /// ``` + fn parse_item_decl_macro(&mut self, lo: Span) -> PResult<'a, ItemInfo> { + let ident = self.parse_ident()?; + let body = if self.check(&token::OpenDelim(Delimiter::Brace)) { + self.parse_delim_args()? // `MacBody` + } else if self.check(&token::OpenDelim(Delimiter::Parenthesis)) { + let params = self.parse_token_tree(); // `MacParams` + let pspan = params.span(); + if !self.check(&token::OpenDelim(Delimiter::Brace)) { + self.unexpected()?; + } + let body = self.parse_token_tree(); // `MacBody` + // Convert `MacParams MacBody` into `{ MacParams => MacBody }`. + let bspan = body.span(); + let arrow = TokenTree::token_alone(token::FatArrow, pspan.between(bspan)); // `=>` + let tokens = TokenStream::new(vec![params, arrow, body]); + let dspan = DelimSpan::from_pair(pspan.shrink_to_lo(), bspan.shrink_to_hi()); + P(DelimArgs { dspan, delim: Delimiter::Brace, tokens }) + } else { + self.unexpected_any()? + }; + + self.psess.gated_spans.gate(sym::decl_macro, lo.to(self.prev_token.span)); + Ok((ident, ItemKind::MacroDef(ast::MacroDef { body, macro_rules: false }))) + } + + /// Is this a possibly malformed start of a `macro_rules! foo` item definition? + fn is_macro_rules_item(&mut self) -> IsMacroRulesItem { + if self.check_keyword(kw::MacroRules) { + let macro_rules_span = self.token.span; + + if self.look_ahead(1, |t| *t == token::Not) && self.look_ahead(2, |t| t.is_ident()) { + return IsMacroRulesItem::Yes { has_bang: true }; + } else if self.look_ahead(1, |t| (t.is_ident())) { + // macro_rules foo + self.dcx().emit_err(errors::MacroRulesMissingBang { + span: macro_rules_span, + hi: macro_rules_span.shrink_to_hi(), + }); + + return IsMacroRulesItem::Yes { has_bang: false }; + } + } + + IsMacroRulesItem::No + } + + /// Parses a `macro_rules! foo { ... }` declarative macro. + fn parse_item_macro_rules( + &mut self, + vis: &Visibility, + has_bang: bool, + ) -> PResult<'a, ItemInfo> { + self.expect_keyword(kw::MacroRules)?; // `macro_rules` + + if has_bang { + self.expect(&token::Not)?; // `!` + } + let ident = self.parse_ident()?; + + if self.eat(&token::Not) { + // Handle macro_rules! foo! + let span = self.prev_token.span; + self.dcx().emit_err(errors::MacroNameRemoveBang { span }); + } + + let body = self.parse_delim_args()?; + self.eat_semi_for_macro_if_needed(&body); + self.complain_if_pub_macro(vis, true); + + Ok((ident, ItemKind::MacroDef(ast::MacroDef { body, macro_rules: true }))) + } + + /// Item macro invocations or `macro_rules!` definitions need inherited visibility. + /// If that's not the case, emit an error. + fn complain_if_pub_macro(&self, vis: &Visibility, macro_rules: bool) { + if let VisibilityKind::Inherited = vis.kind { + return; + } + + let vstr = pprust::vis_to_string(vis); + let vstr = vstr.trim_end(); + if macro_rules { + self.dcx().emit_err(errors::MacroRulesVisibility { span: vis.span, vis: vstr }); + } else { + self.dcx().emit_err(errors::MacroInvocationVisibility { span: vis.span, vis: vstr }); + } + } + + fn eat_semi_for_macro_if_needed(&mut self, args: &DelimArgs) { + if args.need_semicolon() && !self.eat(&token::Semi) { + self.report_invalid_macro_expansion_item(args); + } + } + + fn report_invalid_macro_expansion_item(&self, args: &DelimArgs) { + let span = args.dspan.entire(); + let mut err = self.dcx().struct_span_err( + span, + "macros that expand to items must be delimited with braces or followed by a semicolon", + ); + // FIXME: This will make us not emit the help even for declarative + // macros within the same crate (that we can fix), which is sad. + if !span.from_expansion() { + let DelimSpan { open, close } = args.dspan; + err.multipart_suggestion( + "change the delimiters to curly braces", + vec![(open, "{".to_string()), (close, '}'.to_string())], + Applicability::MaybeIncorrect, + ); + err.span_suggestion( + span.with_neighbor(self.token.span).shrink_to_hi(), + "add a semicolon", + ';', + Applicability::MaybeIncorrect, + ); + } + err.emit(); + } + + /// Checks if current token is one of tokens which cannot be nested like `kw::Enum`. In case + /// it is, we try to parse the item and report error about nested types. + fn recover_nested_adt_item(&mut self, keyword: Symbol) -> PResult<'a, bool> { + if (self.token.is_keyword(kw::Enum) + || self.token.is_keyword(kw::Struct) + || self.token.is_keyword(kw::Union)) + && self.look_ahead(1, |t| t.is_ident()) + { + let kw_token = self.token.clone(); + let kw_str = pprust::token_to_string(&kw_token); + let item = self.parse_item(ForceCollect::No)?; + self.dcx().emit_err(errors::NestedAdt { + span: kw_token.span, + item: item.unwrap().span, + kw_str, + keyword: keyword.as_str(), + }); + // We successfully parsed the item but we must inform the caller about nested problem. + return Ok(false); + } + Ok(true) + } +} + +/// The parsing configuration used to parse a parameter list (see `parse_fn_params`). +/// +/// The function decides if, per-parameter `p`, `p` must have a pattern or just a type. +/// +/// This function pointer accepts an edition, because in edition 2015, trait declarations +/// were allowed to omit parameter names. In 2018, they became required. +type ReqName = fn(Edition) -> bool; + +/// Parsing configuration for functions. +/// +/// The syntax of function items is slightly different within trait definitions, +/// impl blocks, and modules. It is still parsed using the same code, just with +/// different flags set, so that even when the input is wrong and produces a parse +/// error, it still gets into the AST and the rest of the parser and +/// type checker can run. +#[derive(Clone, Copy)] +pub(crate) struct FnParseMode { + /// A function pointer that decides if, per-parameter `p`, `p` must have a + /// pattern or just a type. This field affects parsing of the parameters list. + /// + /// ```text + /// fn foo(alef: A) -> X { X::new() } + /// -----^^ affects parsing this part of the function signature + /// | + /// if req_name returns false, then this name is optional + /// + /// fn bar(A) -> X; + /// ^ + /// | + /// if req_name returns true, this is an error + /// ``` + /// + /// Calling this function pointer should only return false if: + /// + /// * The item is being parsed inside of a trait definition. + /// Within an impl block or a module, it should always evaluate + /// to true. + /// * The span is from Edition 2015. In particular, you can get a + /// 2015 span inside a 2021 crate using macros. + pub req_name: ReqName, + /// If this flag is set to `true`, then plain, semicolon-terminated function + /// prototypes are not allowed here. + /// + /// ```text + /// fn foo(alef: A) -> X { X::new() } + /// ^^^^^^^^^^^^ + /// | + /// this is always allowed + /// + /// fn bar(alef: A, bet: B) -> X; + /// ^ + /// | + /// if req_body is set to true, this is an error + /// ``` + /// + /// This field should only be set to false if the item is inside of a trait + /// definition or extern block. Within an impl block or a module, it should + /// always be set to true. + pub req_body: bool, +} + +/// Parsing of functions and methods. +impl<'a> Parser<'a> { + /// Parse a function starting from the front matter (`const ...`) to the body `{ ... }` or `;`. + fn parse_fn( + &mut self, + attrs: &mut AttrVec, + fn_parse_mode: FnParseMode, + sig_lo: Span, + vis: &Visibility, + case: Case, + ) -> PResult<'a, (Ident, FnSig, Generics, Option<P<Block>>)> { + let fn_span = self.token.span; + let header = self.parse_fn_front_matter(vis, case)?; // `const ... fn` + let ident = self.parse_ident()?; // `foo` + let mut generics = self.parse_generics()?; // `<'a, T, ...>` + let decl = match self.parse_fn_decl( + fn_parse_mode.req_name, + AllowPlus::Yes, + RecoverReturnSign::Yes, + ) { + Ok(decl) => decl, + Err(old_err) => { + // If we see `for Ty ...` then user probably meant `impl` item. + if self.token.is_keyword(kw::For) { + old_err.cancel(); + return Err(self.dcx().create_err(errors::FnTypoWithImpl { fn_span })); + } else { + return Err(old_err); + } + } + }; + generics.where_clause = self.parse_where_clause()?; // `where T: Ord` + + let mut sig_hi = self.prev_token.span; + let body = self.parse_fn_body(attrs, &ident, &mut sig_hi, fn_parse_mode.req_body)?; // `;` or `{ ... }`. + let fn_sig_span = sig_lo.to(sig_hi); + Ok((ident, FnSig { header, decl, span: fn_sig_span }, generics, body)) + } + + /// Parse the "body" of a function. + /// This can either be `;` when there's no body, + /// or e.g. a block when the function is a provided one. + fn parse_fn_body( + &mut self, + attrs: &mut AttrVec, + ident: &Ident, + sig_hi: &mut Span, + req_body: bool, + ) -> PResult<'a, Option<P<Block>>> { + let has_semi = if req_body { + self.token.kind == TokenKind::Semi + } else { + // Only include `;` in list of expected tokens if body is not required + self.check(&TokenKind::Semi) + }; + let (inner_attrs, body) = if has_semi { + // Include the trailing semicolon in the span of the signature + self.expect_semi()?; + *sig_hi = self.prev_token.span; + (AttrVec::new(), None) + } else if self.check(&token::OpenDelim(Delimiter::Brace)) || self.token.is_whole_block() { + self.parse_block_common(self.token.span, BlockCheckMode::Default, false) + .map(|(attrs, body)| (attrs, Some(body)))? + } else if self.token.kind == token::Eq { + // Recover `fn foo() = $expr;`. + self.bump(); // `=` + let eq_sp = self.prev_token.span; + let _ = self.parse_expr()?; + self.expect_semi()?; // `;` + let span = eq_sp.to(self.prev_token.span); + let guar = self.dcx().emit_err(errors::FunctionBodyEqualsExpr { + span, + sugg: errors::FunctionBodyEqualsExprSugg { eq: eq_sp, semi: self.prev_token.span }, + }); + (AttrVec::new(), Some(self.mk_block_err(span, guar))) + } else { + let expected = if req_body { + &[token::OpenDelim(Delimiter::Brace)][..] + } else { + &[token::Semi, token::OpenDelim(Delimiter::Brace)] + }; + if let Err(mut err) = self.expected_one_of_not_found(&[], expected) { + if self.token.kind == token::CloseDelim(Delimiter::Brace) { + // The enclosing `mod`, `trait` or `impl` is being closed, so keep the `fn` in + // the AST for typechecking. + err.span_label(ident.span, "while parsing this `fn`"); + err.emit(); + } else { + // check for typo'd Fn* trait bounds such as + // fn foo<F>() where F: FnOnce -> () {} + if self.token.kind == token::RArrow { + let machine_applicable = [sym::FnOnce, sym::FnMut, sym::Fn] + .into_iter() + .any(|s| self.prev_token.is_ident_named(s)); + + err.subdiagnostic( + self.dcx(), + errors::FnTraitMissingParen { + span: self.prev_token.span, + machine_applicable, + }, + ); + } + return Err(err); + } + } + (AttrVec::new(), None) + }; + attrs.extend(inner_attrs); + Ok(body) + } + + /// Is the current token the start of an `FnHeader` / not a valid parse? + /// + /// `check_pub` adds additional `pub` to the checks in case users place it + /// wrongly, can be used to ensure `pub` never comes after `default`. + pub(super) fn check_fn_front_matter(&mut self, check_pub: bool, case: Case) -> bool { + // We use an over-approximation here. + // `const const`, `fn const` won't parse, but we're not stepping over other syntax either. + // `pub` is added in case users got confused with the ordering like `async pub fn`, + // only if it wasn't preceded by `default` as `default pub` is invalid. + let quals: &[Symbol] = if check_pub { + &[kw::Pub, kw::Gen, kw::Const, kw::Async, kw::Unsafe, kw::Extern] + } else { + &[kw::Gen, kw::Const, kw::Async, kw::Unsafe, kw::Extern] + }; + self.check_keyword_case(kw::Fn, case) // Definitely an `fn`. + // `$qual fn` or `$qual $qual`: + || quals.iter().any(|&kw| self.check_keyword_case(kw, case)) + && self.look_ahead(1, |t| { + // `$qual fn`, e.g. `const fn` or `async fn`. + t.is_keyword_case(kw::Fn, case) + // Two qualifiers `$qual $qual` is enough, e.g. `async unsafe`. + || ( + ( + t.is_non_raw_ident_where(|i| + quals.contains(&i.name) + // Rule out 2015 `const async: T = val`. + && i.is_reserved() + ) + || case == Case::Insensitive + && t.is_non_raw_ident_where(|i| quals.iter().any(|qual| qual.as_str() == i.name.as_str().to_lowercase())) + ) + // Rule out `unsafe extern {`. + && !self.is_unsafe_foreign_mod() + // Rule out `async gen {` and `async gen move {` + && !self.is_async_gen_block()) + }) + // `extern ABI fn` + || self.check_keyword_case(kw::Extern, case) + && self.look_ahead(1, |t| t.can_begin_literal_maybe_minus()) + && (self.look_ahead(2, |t| t.is_keyword_case(kw::Fn, case)) || + // this branch is only for better diagnostic in later, `pub` is not allowed here + (self.may_recover() + && self.look_ahead(2, |t| t.is_keyword(kw::Pub)) + && self.look_ahead(3, |t| t.is_keyword_case(kw::Fn, case)))) + } + + /// Parses all the "front matter" (or "qualifiers") for a `fn` declaration, + /// up to and including the `fn` keyword. The formal grammar is: + /// + /// ```text + /// Extern = "extern" StringLit? ; + /// FnQual = "const"? "async"? "unsafe"? Extern? ; + /// FnFrontMatter = FnQual "fn" ; + /// ``` + /// + /// `vis` represents the visibility that was already parsed, if any. Use + /// `Visibility::Inherited` when no visibility is known. + pub(super) fn parse_fn_front_matter( + &mut self, + orig_vis: &Visibility, + case: Case, + ) -> PResult<'a, FnHeader> { + let sp_start = self.token.span; + let constness = self.parse_constness(case); + + let async_start_sp = self.token.span; + let coroutine_kind = self.parse_coroutine_kind(case); + + let unsafe_start_sp = self.token.span; + let safety = self.parse_safety(case); + + let ext_start_sp = self.token.span; + let ext = self.parse_extern(case); + + if let Some(CoroutineKind::Async { span, .. }) = coroutine_kind { + if span.is_rust_2015() { + self.dcx().emit_err(errors::AsyncFnIn2015 { + span, + help: errors::HelpUseLatestEdition::new(), + }); + } + } + + match coroutine_kind { + Some(CoroutineKind::Gen { span, .. }) | Some(CoroutineKind::AsyncGen { span, .. }) => { + self.psess.gated_spans.gate(sym::gen_blocks, span); + } + Some(CoroutineKind::Async { .. }) | None => {} + } + + if !self.eat_keyword_case(kw::Fn, case) { + // It is possible for `expect_one_of` to recover given the contents of + // `self.expected_tokens`, therefore, do not use `self.unexpected()` which doesn't + // account for this. + match self.expect_one_of(&[], &[]) { + Ok(Recovered::Yes(_)) => {} + Ok(Recovered::No) => unreachable!(), + Err(mut err) => { + // Qualifier keywords ordering check + enum WrongKw { + Duplicated(Span), + Misplaced(Span), + } + + // We may be able to recover + let mut recover_constness = constness; + let mut recover_coroutine_kind = coroutine_kind; + let mut recover_safety = safety; + // This will allow the machine fix to directly place the keyword in the correct place or to indicate + // that the keyword is already present and the second instance should be removed. + let wrong_kw = if self.check_keyword(kw::Const) { + match constness { + Const::Yes(sp) => Some(WrongKw::Duplicated(sp)), + Const::No => { + recover_constness = Const::Yes(self.token.span); + Some(WrongKw::Misplaced(async_start_sp)) + } + } + } else if self.check_keyword(kw::Async) { + match coroutine_kind { + Some(CoroutineKind::Async { span, .. }) => { + Some(WrongKw::Duplicated(span)) + } + Some(CoroutineKind::AsyncGen { span, .. }) => { + Some(WrongKw::Duplicated(span)) + } + Some(CoroutineKind::Gen { .. }) => { + recover_coroutine_kind = Some(CoroutineKind::AsyncGen { + span: self.token.span, + closure_id: DUMMY_NODE_ID, + return_impl_trait_id: DUMMY_NODE_ID, + }); + // FIXME(gen_blocks): This span is wrong, didn't want to think about it. + Some(WrongKw::Misplaced(unsafe_start_sp)) + } + None => { + recover_coroutine_kind = Some(CoroutineKind::Async { + span: self.token.span, + closure_id: DUMMY_NODE_ID, + return_impl_trait_id: DUMMY_NODE_ID, + }); + Some(WrongKw::Misplaced(unsafe_start_sp)) + } + } + } else if self.check_keyword(kw::Unsafe) { + match safety { + Safety::Unsafe(sp) => Some(WrongKw::Duplicated(sp)), + Safety::Default => { + recover_safety = Safety::Unsafe(self.token.span); + Some(WrongKw::Misplaced(ext_start_sp)) + } + } + } else { + None + }; + + // The keyword is already present, suggest removal of the second instance + if let Some(WrongKw::Duplicated(original_sp)) = wrong_kw { + let original_kw = self + .span_to_snippet(original_sp) + .expect("Span extracted directly from keyword should always work"); + + err.span_suggestion( + self.token.uninterpolated_span(), + format!("`{original_kw}` already used earlier, remove this one"), + "", + Applicability::MachineApplicable, + ) + .span_note(original_sp, format!("`{original_kw}` first seen here")); + } + // The keyword has not been seen yet, suggest correct placement in the function front matter + else if let Some(WrongKw::Misplaced(correct_pos_sp)) = wrong_kw { + let correct_pos_sp = correct_pos_sp.to(self.prev_token.span); + if let Ok(current_qual) = self.span_to_snippet(correct_pos_sp) { + let misplaced_qual_sp = self.token.uninterpolated_span(); + let misplaced_qual = self.span_to_snippet(misplaced_qual_sp).unwrap(); + + err.span_suggestion( + correct_pos_sp.to(misplaced_qual_sp), + format!("`{misplaced_qual}` must come before `{current_qual}`"), + format!("{misplaced_qual} {current_qual}"), + Applicability::MachineApplicable, + ).note("keyword order for functions declaration is `pub`, `default`, `const`, `async`, `unsafe`, `extern`"); + } + } + // Recover incorrect visibility order such as `async pub` + else if self.check_keyword(kw::Pub) { + let sp = sp_start.to(self.prev_token.span); + if let Ok(snippet) = self.span_to_snippet(sp) { + let current_vis = match self.parse_visibility(FollowedByType::No) { + Ok(v) => v, + Err(d) => { + d.cancel(); + return Err(err); + } + }; + let vs = pprust::vis_to_string(¤t_vis); + let vs = vs.trim_end(); + + // There was no explicit visibility + if matches!(orig_vis.kind, VisibilityKind::Inherited) { + err.span_suggestion( + sp_start.to(self.prev_token.span), + format!("visibility `{vs}` must come before `{snippet}`"), + format!("{vs} {snippet}"), + Applicability::MachineApplicable, + ); + } + // There was an explicit visibility + else { + err.span_suggestion( + current_vis.span, + "there is already a visibility modifier, remove one", + "", + Applicability::MachineApplicable, + ) + .span_note(orig_vis.span, "explicit visibility first seen here"); + } + } + } + + // FIXME(gen_blocks): add keyword recovery logic for genness + + if wrong_kw.is_some() + && self.may_recover() + && self.look_ahead(1, |tok| tok.is_keyword_case(kw::Fn, case)) + { + // Advance past the misplaced keyword and `fn` + self.bump(); + self.bump(); + err.emit(); + return Ok(FnHeader { + constness: recover_constness, + safety: recover_safety, + coroutine_kind: recover_coroutine_kind, + ext, + }); + } + + return Err(err); + } + } + } + + Ok(FnHeader { constness, safety, coroutine_kind, ext }) + } + + /// Parses the parameter list and result type of a function declaration. + pub(super) fn parse_fn_decl( + &mut self, + req_name: ReqName, + ret_allow_plus: AllowPlus, + recover_return_sign: RecoverReturnSign, + ) -> PResult<'a, P<FnDecl>> { + Ok(P(FnDecl { + inputs: self.parse_fn_params(req_name)?, + output: self.parse_ret_ty(ret_allow_plus, RecoverQPath::Yes, recover_return_sign)?, + })) + } + + /// Parses the parameter list of a function, including the `(` and `)` delimiters. + pub(super) fn parse_fn_params(&mut self, req_name: ReqName) -> PResult<'a, ThinVec<Param>> { + let mut first_param = true; + // Parse the arguments, starting out with `self` being allowed... + if self.token.kind != TokenKind::OpenDelim(Delimiter::Parenthesis) + // might be typo'd trait impl, handled elsewhere + && !self.token.is_keyword(kw::For) + { + // recover from missing argument list, e.g. `fn main -> () {}` + self.dcx() + .emit_err(errors::MissingFnParams { span: self.prev_token.span.shrink_to_hi() }); + return Ok(ThinVec::new()); + } + + let (mut params, _) = self.parse_paren_comma_seq(|p| { + p.recover_vcs_conflict_marker(); + let snapshot = p.create_snapshot_for_diagnostic(); + let param = p.parse_param_general(req_name, first_param).or_else(|e| { + let guar = e.emit(); + let lo = p.prev_token.span; + p.restore_snapshot(snapshot); + // Skip every token until next possible arg or end. + p.eat_to_tokens(&[&token::Comma, &token::CloseDelim(Delimiter::Parenthesis)]); + // Create a placeholder argument for proper arg count (issue #34264). + Ok(dummy_arg(Ident::new(kw::Empty, lo.to(p.prev_token.span)), guar)) + }); + // ...now that we've parsed the first argument, `self` is no longer allowed. + first_param = false; + param + })?; + // Replace duplicated recovered params with `_` pattern to avoid unnecessary errors. + self.deduplicate_recovered_params_names(&mut params); + Ok(params) + } + + /// Parses a single function parameter. + /// + /// - `self` is syntactically allowed when `first_param` holds. + fn parse_param_general(&mut self, req_name: ReqName, first_param: bool) -> PResult<'a, Param> { + let lo = self.token.span; + let attrs = self.parse_outer_attributes()?; + self.collect_tokens_trailing_token(attrs, ForceCollect::No, |this, attrs| { + // Possibly parse `self`. Recover if we parsed it and it wasn't allowed here. + if let Some(mut param) = this.parse_self_param()? { + param.attrs = attrs; + let res = if first_param { Ok(param) } else { this.recover_bad_self_param(param) }; + return Ok((res?, TrailingToken::None)); + } + + let is_name_required = match this.token.kind { + token::DotDotDot => false, + _ => req_name(this.token.span.with_neighbor(this.prev_token.span).edition()), + }; + let (pat, ty) = if is_name_required || this.is_named_param() { + debug!("parse_param_general parse_pat (is_name_required:{})", is_name_required); + let (pat, colon) = this.parse_fn_param_pat_colon()?; + if !colon { + let mut err = this.unexpected().unwrap_err(); + return if let Some(ident) = + this.parameter_without_type(&mut err, pat, is_name_required, first_param) + { + let guar = err.emit(); + Ok((dummy_arg(ident, guar), TrailingToken::None)) + } else { + Err(err) + }; + } + + this.eat_incorrect_doc_comment_for_param_type(); + (pat, this.parse_ty_for_param()?) + } else { + debug!("parse_param_general ident_to_pat"); + let parser_snapshot_before_ty = this.create_snapshot_for_diagnostic(); + this.eat_incorrect_doc_comment_for_param_type(); + let mut ty = this.parse_ty_for_param(); + if ty.is_ok() + && this.token != token::Comma + && this.token != token::CloseDelim(Delimiter::Parenthesis) + { + // This wasn't actually a type, but a pattern looking like a type, + // so we are going to rollback and re-parse for recovery. + ty = this.unexpected_any(); + } + match ty { + Ok(ty) => { + let ident = Ident::new(kw::Empty, this.prev_token.span); + let bm = BindingMode::NONE; + let pat = this.mk_pat_ident(ty.span, bm, ident); + (pat, ty) + } + // If this is a C-variadic argument and we hit an error, return the error. + Err(err) if this.token == token::DotDotDot => return Err(err), + // Recover from attempting to parse the argument as a type without pattern. + Err(err) => { + err.cancel(); + this.restore_snapshot(parser_snapshot_before_ty); + this.recover_arg_parse()? + } + } + }; + + let span = lo.to(this.prev_token.span); + + Ok(( + Param { attrs, id: ast::DUMMY_NODE_ID, is_placeholder: false, pat, span, ty }, + TrailingToken::None, + )) + }) + } + + /// Returns the parsed optional self parameter and whether a self shortcut was used. + fn parse_self_param(&mut self) -> PResult<'a, Option<Param>> { + // Extract an identifier *after* having confirmed that the token is one. + let expect_self_ident = |this: &mut Self| match this.token.ident() { + Some((ident, IdentIsRaw::No)) => { + this.bump(); + ident + } + _ => unreachable!(), + }; + // Is `self` `n` tokens ahead? + let is_isolated_self = |this: &Self, n| { + this.is_keyword_ahead(n, &[kw::SelfLower]) + && this.look_ahead(n + 1, |t| t != &token::PathSep) + }; + // Is `mut self` `n` tokens ahead? + let is_isolated_mut_self = + |this: &Self, n| this.is_keyword_ahead(n, &[kw::Mut]) && is_isolated_self(this, n + 1); + // Parse `self` or `self: TYPE`. We already know the current token is `self`. + let parse_self_possibly_typed = |this: &mut Self, m| { + let eself_ident = expect_self_ident(this); + let eself_hi = this.prev_token.span; + let eself = if this.eat(&token::Colon) { + SelfKind::Explicit(this.parse_ty()?, m) + } else { + SelfKind::Value(m) + }; + Ok((eself, eself_ident, eself_hi)) + }; + // Recover for the grammar `*self`, `*const self`, and `*mut self`. + let recover_self_ptr = |this: &mut Self| { + this.dcx().emit_err(errors::SelfArgumentPointer { span: this.token.span }); + + Ok((SelfKind::Value(Mutability::Not), expect_self_ident(this), this.prev_token.span)) + }; + + // Parse optional `self` parameter of a method. + // Only a limited set of initial token sequences is considered `self` parameters; anything + // else is parsed as a normal function parameter list, so some lookahead is required. + let eself_lo = self.token.span; + let (eself, eself_ident, eself_hi) = match self.token.uninterpolate().kind { + token::BinOp(token::And) => { + let eself = if is_isolated_self(self, 1) { + // `&self` + self.bump(); + SelfKind::Region(None, Mutability::Not) + } else if is_isolated_mut_self(self, 1) { + // `&mut self` + self.bump(); + self.bump(); + SelfKind::Region(None, Mutability::Mut) + } else if self.look_ahead(1, |t| t.is_lifetime()) && is_isolated_self(self, 2) { + // `&'lt self` + self.bump(); + let lt = self.expect_lifetime(); + SelfKind::Region(Some(lt), Mutability::Not) + } else if self.look_ahead(1, |t| t.is_lifetime()) && is_isolated_mut_self(self, 2) { + // `&'lt mut self` + self.bump(); + let lt = self.expect_lifetime(); + self.bump(); + SelfKind::Region(Some(lt), Mutability::Mut) + } else { + // `¬_self` + return Ok(None); + }; + (eself, expect_self_ident(self), self.prev_token.span) + } + // `*self` + token::BinOp(token::Star) if is_isolated_self(self, 1) => { + self.bump(); + recover_self_ptr(self)? + } + // `*mut self` and `*const self` + token::BinOp(token::Star) + if self.look_ahead(1, |t| t.is_mutability()) && is_isolated_self(self, 2) => + { + self.bump(); + self.bump(); + recover_self_ptr(self)? + } + // `self` and `self: TYPE` + token::Ident(..) if is_isolated_self(self, 0) => { + parse_self_possibly_typed(self, Mutability::Not)? + } + // `mut self` and `mut self: TYPE` + token::Ident(..) if is_isolated_mut_self(self, 0) => { + self.bump(); + parse_self_possibly_typed(self, Mutability::Mut)? + } + _ => return Ok(None), + }; + + let eself = source_map::respan(eself_lo.to(eself_hi), eself); + Ok(Some(Param::from_self(AttrVec::default(), eself, eself_ident))) + } + + fn is_named_param(&self) -> bool { + let offset = match &self.token.kind { + token::Interpolated(nt) => match &**nt { + token::NtPat(..) => return self.look_ahead(1, |t| t == &token::Colon), + _ => 0, + }, + token::BinOp(token::And) | token::AndAnd => 1, + _ if self.token.is_keyword(kw::Mut) => 1, + _ => 0, + }; + + self.look_ahead(offset, |t| t.is_ident()) + && self.look_ahead(offset + 1, |t| t == &token::Colon) + } + + fn recover_self_param(&mut self) -> bool { + matches!( + self.parse_outer_attributes() + .and_then(|_| self.parse_self_param()) + .map_err(|e| e.cancel()), + Ok(Some(_)) + ) + } +} + +enum IsMacroRulesItem { + Yes { has_bang: bool }, + No, +} diff --git a/compiler/rustc_parse/src/parser/mod.rs b/compiler/rustc_parse/src/parser/mod.rs new file mode 100644 index 00000000000..c2183258eef --- /dev/null +++ b/compiler/rustc_parse/src/parser/mod.rs @@ -0,0 +1,1643 @@ +pub mod attr; +mod attr_wrapper; +mod diagnostics; +mod expr; +mod generics; +mod item; +mod nonterminal; +mod pat; +mod path; +mod stmt; +mod ty; + +use crate::lexer::UnmatchedDelim; +pub use attr_wrapper::AttrWrapper; +pub use diagnostics::AttemptLocalParseRecovery; +pub(crate) use expr::ForbiddenLetReason; +pub(crate) use item::FnParseMode; +pub use pat::{CommaRecoveryMode, RecoverColon, RecoverComma}; +pub use path::PathStyle; + +use core::fmt; +use rustc_ast::ptr::P; +use rustc_ast::token::{self, Delimiter, IdentIsRaw, Nonterminal, Token, TokenKind}; +use rustc_ast::tokenstream::{AttributesData, DelimSpacing, DelimSpan, Spacing}; +use rustc_ast::tokenstream::{TokenStream, TokenTree, TokenTreeCursor}; +use rustc_ast::util::case::Case; +use rustc_ast::{ + self as ast, AnonConst, AttrArgs, AttrArgsEq, AttrId, ByRef, Const, CoroutineKind, DelimArgs, + Expr, ExprKind, Extern, HasAttrs, HasTokens, Mutability, Recovered, Safety, StrLit, Visibility, + VisibilityKind, DUMMY_NODE_ID, +}; +use rustc_ast_pretty::pprust; +use rustc_data_structures::fx::FxHashMap; +use rustc_data_structures::sync::Lrc; +use rustc_errors::{Applicability, Diag, FatalError, MultiSpan, PResult}; +use rustc_session::parse::ParseSess; +use rustc_span::symbol::{kw, sym, Ident, Symbol}; +use rustc_span::{Span, DUMMY_SP}; +use std::ops::Range; +use std::{mem, slice}; +use thin_vec::ThinVec; +use tracing::debug; + +use crate::errors::{ + self, IncorrectVisibilityRestriction, MismatchedClosingDelimiter, NonStringAbiLiteral, +}; + +#[cfg(test)] +mod tests; + +// Ideally, these tests would be in `rustc_ast`. But they depend on having a +// parser, so they are here. +#[cfg(test)] +mod tokenstream { + mod tests; +} +#[cfg(test)] +mod mut_visit { + mod tests; +} + +bitflags::bitflags! { + #[derive(Clone, Copy, Debug)] + struct Restrictions: u8 { + const STMT_EXPR = 1 << 0; + const NO_STRUCT_LITERAL = 1 << 1; + const CONST_EXPR = 1 << 2; + const ALLOW_LET = 1 << 3; + const IN_IF_GUARD = 1 << 4; + const IS_PAT = 1 << 5; + } +} + +#[derive(Clone, Copy, PartialEq, Debug)] +enum SemiColonMode { + Break, + Ignore, + Comma, +} + +#[derive(Clone, Copy, PartialEq, Debug)] +enum BlockMode { + Break, + Ignore, +} + +/// Whether or not we should force collection of tokens for an AST node, +/// regardless of whether or not it has attributes +#[derive(Clone, Copy, Debug, PartialEq)] +pub enum ForceCollect { + Yes, + No, +} + +#[derive(Debug, Eq, PartialEq)] +pub enum TrailingToken { + None, + Semi, + Gt, + /// If the trailing token is a comma, then capture it + /// Otherwise, ignore the trailing token + MaybeComma, +} + +/// Like `maybe_whole_expr`, but for things other than expressions. +#[macro_export] +macro_rules! maybe_whole { + ($p:expr, $constructor:ident, |$x:ident| $e:expr) => { + if let token::Interpolated(nt) = &$p.token.kind + && let token::$constructor(x) = &**nt + { + #[allow(unused_mut)] + let mut $x = x.clone(); + $p.bump(); + return Ok($e); + } + }; +} + +/// If the next tokens are ill-formed `$ty::` recover them as `<$ty>::`. +#[macro_export] +macro_rules! maybe_recover_from_interpolated_ty_qpath { + ($self: expr, $allow_qpath_recovery: expr) => { + if $allow_qpath_recovery + && $self.may_recover() + && $self.look_ahead(1, |t| t == &token::PathSep) + && let token::Interpolated(nt) = &$self.token.kind + && let token::NtTy(ty) = &**nt + { + let ty = ty.clone(); + $self.bump(); + return $self.maybe_recover_from_bad_qpath_stage_2($self.prev_token.span, ty); + } + }; +} + +#[derive(Clone, Copy, Debug)] +pub enum Recovery { + Allowed, + Forbidden, +} + +#[derive(Clone)] +pub struct Parser<'a> { + pub psess: &'a ParseSess, + /// The current token. + pub token: Token, + /// The spacing for the current token. + pub token_spacing: Spacing, + /// The previous token. + pub prev_token: Token, + pub capture_cfg: bool, + restrictions: Restrictions, + expected_tokens: Vec<TokenType>, + token_cursor: TokenCursor, + // The number of calls to `bump`, i.e. the position in the token stream. + num_bump_calls: usize, + // During parsing we may sometimes need to 'unglue' a glued token into two + // component tokens (e.g. '>>' into '>' and '>), so the parser can consume + // them one at a time. This process bypasses the normal capturing mechanism + // (e.g. `num_bump_calls` will not be incremented), since the 'unglued' + // tokens due not exist in the original `TokenStream`. + // + // If we end up consuming both unglued tokens, this is not an issue. We'll + // end up capturing the single 'glued' token. + // + // However, sometimes we may want to capture just the first 'unglued' + // token. For example, capturing the `Vec<u8>` in `Option<Vec<u8>>` + // requires us to unglue the trailing `>>` token. The `break_last_token` + // field is used to track this token. It gets appended to the captured + // stream when we evaluate a `LazyAttrTokenStream`. + break_last_token: bool, + /// This field is used to keep track of how many left angle brackets we have seen. This is + /// required in order to detect extra leading left angle brackets (`<` characters) and error + /// appropriately. + /// + /// See the comments in the `parse_path_segment` function for more details. + unmatched_angle_bracket_count: u16, + angle_bracket_nesting: u16, + + last_unexpected_token_span: Option<Span>, + /// If present, this `Parser` is not parsing Rust code but rather a macro call. + subparser_name: Option<&'static str>, + capture_state: CaptureState, + /// This allows us to recover when the user forget to add braces around + /// multiple statements in the closure body. + current_closure: Option<ClosureSpans>, + /// Whether the parser is allowed to do recovery. + /// This is disabled when parsing macro arguments, see #103534 + pub recovery: Recovery, +} + +// This type is used a lot, e.g. it's cloned when matching many declarative macro rules with nonterminals. Make sure +// it doesn't unintentionally get bigger. +#[cfg(target_pointer_width = "64")] +rustc_data_structures::static_assert_size!(Parser<'_>, 264); + +/// Stores span information about a closure. +#[derive(Clone, Debug)] +pub struct ClosureSpans { + pub whole_closure: Span, + pub closing_pipe: Span, + pub body: Span, +} + +/// Indicates a range of tokens that should be replaced by +/// the tokens in the provided vector. This is used in two +/// places during token collection: +/// +/// 1. During the parsing of an AST node that may have a `#[derive]` +/// attribute, we parse a nested AST node that has `#[cfg]` or `#[cfg_attr]` +/// In this case, we use a `ReplaceRange` to replace the entire inner AST node +/// with `FlatToken::AttrTarget`, allowing us to perform eager cfg-expansion +/// on an `AttrTokenStream`. +/// +/// 2. When we parse an inner attribute while collecting tokens. We +/// remove inner attributes from the token stream entirely, and +/// instead track them through the `attrs` field on the AST node. +/// This allows us to easily manipulate them (for example, removing +/// the first macro inner attribute to invoke a proc-macro). +/// When create a `TokenStream`, the inner attributes get inserted +/// into the proper place in the token stream. +pub type ReplaceRange = (Range<u32>, Vec<(FlatToken, Spacing)>); + +/// Controls how we capture tokens. Capturing can be expensive, +/// so we try to avoid performing capturing in cases where +/// we will never need an `AttrTokenStream`. +#[derive(Copy, Clone, Debug)] +pub enum Capturing { + /// We aren't performing any capturing - this is the default mode. + No, + /// We are capturing tokens + Yes, +} + +#[derive(Clone, Debug)] +struct CaptureState { + capturing: Capturing, + replace_ranges: Vec<ReplaceRange>, + inner_attr_ranges: FxHashMap<AttrId, ReplaceRange>, +} + +/// Iterator over a `TokenStream` that produces `Token`s. It's a bit odd that +/// we (a) lex tokens into a nice tree structure (`TokenStream`), and then (b) +/// use this type to emit them as a linear sequence. But a linear sequence is +/// what the parser expects, for the most part. +#[derive(Clone, Debug)] +struct TokenCursor { + // Cursor for the current (innermost) token stream. The delimiters for this + // token stream are found in `self.stack.last()`; when that is `None` then + // we are in the outermost token stream which never has delimiters. + tree_cursor: TokenTreeCursor, + + // Token streams surrounding the current one. The delimiters for stack[n]'s + // tokens are in `stack[n-1]`. `stack[0]` (when present) has no delimiters + // because it's the outermost token stream which never has delimiters. + stack: Vec<(TokenTreeCursor, DelimSpan, DelimSpacing, Delimiter)>, +} + +impl TokenCursor { + fn next(&mut self) -> (Token, Spacing) { + self.inlined_next() + } + + /// This always-inlined version should only be used on hot code paths. + #[inline(always)] + fn inlined_next(&mut self) -> (Token, Spacing) { + loop { + // FIXME: we currently don't return `Delimiter::Invisible` open/close delims. To fix + // #67062 we will need to, whereupon the `delim != Delimiter::Invisible` conditions + // below can be removed. + if let Some(tree) = self.tree_cursor.next_ref() { + match tree { + &TokenTree::Token(ref token, spacing) => { + debug_assert!(!matches!( + token.kind, + token::OpenDelim(_) | token::CloseDelim(_) + )); + return (token.clone(), spacing); + } + &TokenTree::Delimited(sp, spacing, delim, ref tts) => { + let trees = tts.clone().into_trees(); + self.stack.push(( + mem::replace(&mut self.tree_cursor, trees), + sp, + spacing, + delim, + )); + if delim != Delimiter::Invisible { + return (Token::new(token::OpenDelim(delim), sp.open), spacing.open); + } + // No open delimiter to return; continue on to the next iteration. + } + }; + } else if let Some((tree_cursor, span, spacing, delim)) = self.stack.pop() { + // We have exhausted this token stream. Move back to its parent token stream. + self.tree_cursor = tree_cursor; + if delim != Delimiter::Invisible { + return (Token::new(token::CloseDelim(delim), span.close), spacing.close); + } + // No close delimiter to return; continue on to the next iteration. + } else { + // We have exhausted the outermost token stream. The use of + // `Spacing::Alone` is arbitrary and immaterial, because the + // `Eof` token's spacing is never used. + return (Token::new(token::Eof, DUMMY_SP), Spacing::Alone); + } + } + } +} + +#[derive(Debug, Clone, PartialEq)] +enum TokenType { + Token(TokenKind), + Keyword(Symbol), + Operator, + Lifetime, + Ident, + Path, + Type, + Const, +} + +impl TokenType { + fn to_string(&self) -> String { + match self { + TokenType::Token(t) => format!("`{}`", pprust::token_kind_to_string(t)), + TokenType::Keyword(kw) => format!("`{kw}`"), + TokenType::Operator => "an operator".to_string(), + TokenType::Lifetime => "lifetime".to_string(), + TokenType::Ident => "identifier".to_string(), + TokenType::Path => "path".to_string(), + TokenType::Type => "type".to_string(), + TokenType::Const => "a const expression".to_string(), + } + } +} + +/// Used by [`Parser::expect_any_with_type`]. +#[derive(Copy, Clone, Debug)] +enum TokenExpectType { + /// Unencountered tokens are inserted into [`Parser::expected_tokens`]. + /// See [`Parser::check`]. + Expect, + + /// Unencountered tokens are not inserted into [`Parser::expected_tokens`]. + /// See [`Parser::check_noexpect`]. + NoExpect, +} + +/// A sequence separator. +#[derive(Debug)] +struct SeqSep { + /// The separator token. + sep: Option<TokenKind>, + /// `true` if a trailing separator is allowed. + trailing_sep_allowed: bool, +} + +impl SeqSep { + fn trailing_allowed(t: TokenKind) -> SeqSep { + SeqSep { sep: Some(t), trailing_sep_allowed: true } + } + + fn none() -> SeqSep { + SeqSep { sep: None, trailing_sep_allowed: false } + } +} + +#[derive(Debug)] +pub enum FollowedByType { + Yes, + No, +} + +#[derive(Copy, Clone, Debug)] +pub enum Trailing { + No, + Yes, +} + +#[derive(Clone, Copy, Debug, PartialEq, Eq)] +pub enum TokenDescription { + ReservedIdentifier, + Keyword, + ReservedKeyword, + DocComment, +} + +impl TokenDescription { + pub fn from_token(token: &Token) -> Option<Self> { + match token.kind { + _ if token.is_special_ident() => Some(TokenDescription::ReservedIdentifier), + _ if token.is_used_keyword() => Some(TokenDescription::Keyword), + _ if token.is_unused_keyword() => Some(TokenDescription::ReservedKeyword), + token::DocComment(..) => Some(TokenDescription::DocComment), + _ => None, + } + } +} + +pub(super) fn token_descr(token: &Token) -> String { + let name = pprust::token_to_string(token).to_string(); + + let kind = match (TokenDescription::from_token(token), &token.kind) { + (Some(TokenDescription::ReservedIdentifier), _) => Some("reserved identifier"), + (Some(TokenDescription::Keyword), _) => Some("keyword"), + (Some(TokenDescription::ReservedKeyword), _) => Some("reserved keyword"), + (Some(TokenDescription::DocComment), _) => Some("doc comment"), + (None, TokenKind::NtIdent(..)) => Some("identifier"), + (None, TokenKind::NtLifetime(..)) => Some("lifetime"), + (None, TokenKind::Interpolated(node)) => Some(node.descr()), + (None, _) => None, + }; + + if let Some(kind) = kind { format!("{kind} `{name}`") } else { format!("`{name}`") } +} + +impl<'a> Parser<'a> { + pub fn new( + psess: &'a ParseSess, + stream: TokenStream, + subparser_name: Option<&'static str>, + ) -> Self { + let mut parser = Parser { + psess, + token: Token::dummy(), + token_spacing: Spacing::Alone, + prev_token: Token::dummy(), + capture_cfg: false, + restrictions: Restrictions::empty(), + expected_tokens: Vec::new(), + token_cursor: TokenCursor { tree_cursor: stream.into_trees(), stack: Vec::new() }, + num_bump_calls: 0, + break_last_token: false, + unmatched_angle_bracket_count: 0, + angle_bracket_nesting: 0, + last_unexpected_token_span: None, + subparser_name, + capture_state: CaptureState { + capturing: Capturing::No, + replace_ranges: Vec::new(), + inner_attr_ranges: Default::default(), + }, + current_closure: None, + recovery: Recovery::Allowed, + }; + + // Make parser point to the first token. + parser.bump(); + + parser + } + + #[inline] + pub fn recovery(mut self, recovery: Recovery) -> Self { + self.recovery = recovery; + self + } + + /// Whether the parser is allowed to recover from broken code. + /// + /// If this returns false, recovering broken code into valid code (especially if this recovery does lookahead) + /// is not allowed. All recovery done by the parser must be gated behind this check. + /// + /// Technically, this only needs to restrict eager recovery by doing lookahead at more tokens. + /// But making the distinction is very subtle, and simply forbidding all recovery is a lot simpler to uphold. + #[inline] + fn may_recover(&self) -> bool { + matches!(self.recovery, Recovery::Allowed) + } + + /// Version of [`unexpected`](Parser::unexpected) that "returns" any type in the `Ok` + /// (both those functions never return "Ok", and so can lie like that in the type). + pub fn unexpected_any<T>(&mut self) -> PResult<'a, T> { + match self.expect_one_of(&[], &[]) { + Err(e) => Err(e), + // We can get `Ok(true)` from `recover_closing_delimiter` + // which is called in `expected_one_of_not_found`. + Ok(_) => FatalError.raise(), + } + } + + pub fn unexpected(&mut self) -> PResult<'a, ()> { + self.unexpected_any() + } + + /// Expects and consumes the token `t`. Signals an error if the next token is not `t`. + pub fn expect(&mut self, t: &TokenKind) -> PResult<'a, Recovered> { + if self.expected_tokens.is_empty() { + if self.token == *t { + self.bump(); + Ok(Recovered::No) + } else { + self.unexpected_try_recover(t) + } + } else { + self.expect_one_of(slice::from_ref(t), &[]) + } + } + + /// Expect next token to be edible or inedible token. If edible, + /// then consume it; if inedible, then return without consuming + /// anything. Signal a fatal error if next token is unexpected. + pub fn expect_one_of( + &mut self, + edible: &[TokenKind], + inedible: &[TokenKind], + ) -> PResult<'a, Recovered> { + if edible.contains(&self.token.kind) { + self.bump(); + Ok(Recovered::No) + } else if inedible.contains(&self.token.kind) { + // leave it in the input + Ok(Recovered::No) + } else if self.token.kind != token::Eof + && self.last_unexpected_token_span == Some(self.token.span) + { + FatalError.raise(); + } else { + self.expected_one_of_not_found(edible, inedible) + } + } + + // Public for rustfmt usage. + pub fn parse_ident(&mut self) -> PResult<'a, Ident> { + self.parse_ident_common(true) + } + + fn parse_ident_common(&mut self, recover: bool) -> PResult<'a, Ident> { + let (ident, is_raw) = self.ident_or_err(recover)?; + + if matches!(is_raw, IdentIsRaw::No) && ident.is_reserved() { + let err = self.expected_ident_found_err(); + if recover { + err.emit(); + } else { + return Err(err); + } + } + self.bump(); + Ok(ident) + } + + fn ident_or_err(&mut self, recover: bool) -> PResult<'a, (Ident, IdentIsRaw)> { + match self.token.ident() { + Some(ident) => Ok(ident), + None => self.expected_ident_found(recover), + } + } + + /// Checks if the next token is `tok`, and returns `true` if so. + /// + /// This method will automatically add `tok` to `expected_tokens` if `tok` is not + /// encountered. + #[inline] + fn check(&mut self, tok: &TokenKind) -> bool { + let is_present = self.token == *tok; + if !is_present { + self.expected_tokens.push(TokenType::Token(tok.clone())); + } + is_present + } + + #[inline] + fn check_noexpect(&self, tok: &TokenKind) -> bool { + self.token == *tok + } + + /// Consumes a token 'tok' if it exists. Returns whether the given token was present. + /// + /// the main purpose of this function is to reduce the cluttering of the suggestions list + /// which using the normal eat method could introduce in some cases. + #[inline] + pub fn eat_noexpect(&mut self, tok: &TokenKind) -> bool { + let is_present = self.check_noexpect(tok); + if is_present { + self.bump() + } + is_present + } + + /// Consumes a token 'tok' if it exists. Returns whether the given token was present. + #[inline] + pub fn eat(&mut self, tok: &TokenKind) -> bool { + let is_present = self.check(tok); + if is_present { + self.bump() + } + is_present + } + + /// If the next token is the given keyword, returns `true` without eating it. + /// An expectation is also added for diagnostics purposes. + #[inline] + fn check_keyword(&mut self, kw: Symbol) -> bool { + self.expected_tokens.push(TokenType::Keyword(kw)); + self.token.is_keyword(kw) + } + + #[inline] + fn check_keyword_case(&mut self, kw: Symbol, case: Case) -> bool { + if self.check_keyword(kw) { + return true; + } + + if case == Case::Insensitive + && let Some((ident, IdentIsRaw::No)) = self.token.ident() + && ident.as_str().to_lowercase() == kw.as_str().to_lowercase() + { + true + } else { + false + } + } + + /// If the next token is the given keyword, eats it and returns `true`. + /// Otherwise, returns `false`. An expectation is also added for diagnostics purposes. + // Public for rustfmt usage. + #[inline] + pub fn eat_keyword(&mut self, kw: Symbol) -> bool { + if self.check_keyword(kw) { + self.bump(); + true + } else { + false + } + } + + /// Eats a keyword, optionally ignoring the case. + /// If the case differs (and is ignored) an error is issued. + /// This is useful for recovery. + #[inline] + fn eat_keyword_case(&mut self, kw: Symbol, case: Case) -> bool { + if self.eat_keyword(kw) { + return true; + } + + if case == Case::Insensitive + && let Some((ident, IdentIsRaw::No)) = self.token.ident() + && ident.as_str().to_lowercase() == kw.as_str().to_lowercase() + { + self.dcx().emit_err(errors::KwBadCase { span: ident.span, kw: kw.as_str() }); + self.bump(); + return true; + } + + false + } + + #[inline] + fn eat_keyword_noexpect(&mut self, kw: Symbol) -> bool { + if self.token.is_keyword(kw) { + self.bump(); + true + } else { + false + } + } + + /// If the given word is not a keyword, signals an error. + /// If the next token is not the given word, signals an error. + /// Otherwise, eats it. + fn expect_keyword(&mut self, kw: Symbol) -> PResult<'a, ()> { + if !self.eat_keyword(kw) { self.unexpected() } else { Ok(()) } + } + + /// Is the given keyword `kw` followed by a non-reserved identifier? + fn is_kw_followed_by_ident(&self, kw: Symbol) -> bool { + self.token.is_keyword(kw) && self.look_ahead(1, |t| t.is_ident() && !t.is_reserved_ident()) + } + + #[inline] + fn check_or_expected(&mut self, ok: bool, typ: TokenType) -> bool { + if ok { + true + } else { + self.expected_tokens.push(typ); + false + } + } + + fn check_ident(&mut self) -> bool { + self.check_or_expected(self.token.is_ident(), TokenType::Ident) + } + + fn check_path(&mut self) -> bool { + self.check_or_expected(self.token.is_path_start(), TokenType::Path) + } + + fn check_type(&mut self) -> bool { + self.check_or_expected(self.token.can_begin_type(), TokenType::Type) + } + + fn check_const_arg(&mut self) -> bool { + self.check_or_expected(self.token.can_begin_const_arg(), TokenType::Const) + } + + fn check_const_closure(&self) -> bool { + self.is_keyword_ahead(0, &[kw::Const]) + && self.look_ahead(1, |t| match &t.kind { + // async closures do not work with const closures, so we do not parse that here. + token::Ident(kw::Move | kw::Static, _) | token::OrOr | token::BinOp(token::Or) => { + true + } + _ => false, + }) + } + + fn check_inline_const(&self, dist: usize) -> bool { + self.is_keyword_ahead(dist, &[kw::Const]) + && self.look_ahead(dist + 1, |t| match &t.kind { + token::Interpolated(nt) => matches!(&**nt, token::NtBlock(..)), + token::OpenDelim(Delimiter::Brace) => true, + _ => false, + }) + } + + /// Checks to see if the next token is either `+` or `+=`. + /// Otherwise returns `false`. + #[inline] + fn check_plus(&mut self) -> bool { + self.check_or_expected( + self.token.is_like_plus(), + TokenType::Token(token::BinOp(token::Plus)), + ) + } + + /// Eats the expected token if it's present possibly breaking + /// compound tokens like multi-character operators in process. + /// Returns `true` if the token was eaten. + fn break_and_eat(&mut self, expected: TokenKind) -> bool { + if self.token.kind == expected { + self.bump(); + return true; + } + match self.token.kind.break_two_token_op() { + Some((first, second)) if first == expected => { + let first_span = self.psess.source_map().start_point(self.token.span); + let second_span = self.token.span.with_lo(first_span.hi()); + self.token = Token::new(first, first_span); + // Keep track of this token - if we end token capturing now, + // we'll want to append this token to the captured stream. + // + // If we consume any additional tokens, then this token + // is not needed (we'll capture the entire 'glued' token), + // and `bump` will set this field to `None` + self.break_last_token = true; + // Use the spacing of the glued token as the spacing of the + // unglued second token. + self.bump_with((Token::new(second, second_span), self.token_spacing)); + true + } + _ => { + self.expected_tokens.push(TokenType::Token(expected)); + false + } + } + } + + /// Eats `+` possibly breaking tokens like `+=` in process. + fn eat_plus(&mut self) -> bool { + self.break_and_eat(token::BinOp(token::Plus)) + } + + /// Eats `&` possibly breaking tokens like `&&` in process. + /// Signals an error if `&` is not eaten. + fn expect_and(&mut self) -> PResult<'a, ()> { + if self.break_and_eat(token::BinOp(token::And)) { Ok(()) } else { self.unexpected() } + } + + /// Eats `|` possibly breaking tokens like `||` in process. + /// Signals an error if `|` was not eaten. + fn expect_or(&mut self) -> PResult<'a, ()> { + if self.break_and_eat(token::BinOp(token::Or)) { Ok(()) } else { self.unexpected() } + } + + /// Eats `<` possibly breaking tokens like `<<` in process. + fn eat_lt(&mut self) -> bool { + let ate = self.break_and_eat(token::Lt); + if ate { + // See doc comment for `unmatched_angle_bracket_count`. + self.unmatched_angle_bracket_count += 1; + debug!("eat_lt: (increment) count={:?}", self.unmatched_angle_bracket_count); + } + ate + } + + /// Eats `<` possibly breaking tokens like `<<` in process. + /// Signals an error if `<` was not eaten. + fn expect_lt(&mut self) -> PResult<'a, ()> { + if self.eat_lt() { Ok(()) } else { self.unexpected() } + } + + /// Eats `>` possibly breaking tokens like `>>` in process. + /// Signals an error if `>` was not eaten. + fn expect_gt(&mut self) -> PResult<'a, ()> { + if self.break_and_eat(token::Gt) { + // See doc comment for `unmatched_angle_bracket_count`. + if self.unmatched_angle_bracket_count > 0 { + self.unmatched_angle_bracket_count -= 1; + debug!("expect_gt: (decrement) count={:?}", self.unmatched_angle_bracket_count); + } + Ok(()) + } else { + self.unexpected() + } + } + + /// Checks if the next token is contained within `kets`, and returns `true` if so. + fn expect_any_with_type(&mut self, kets: &[&TokenKind], expect: TokenExpectType) -> bool { + kets.iter().any(|k| match expect { + TokenExpectType::Expect => self.check(k), + TokenExpectType::NoExpect => self.check_noexpect(k), + }) + } + + /// Parses a sequence until the specified delimiters. The function + /// `f` must consume tokens until reaching the next separator or + /// closing bracket. + fn parse_seq_to_before_tokens<T>( + &mut self, + kets: &[&TokenKind], + sep: SeqSep, + expect: TokenExpectType, + mut f: impl FnMut(&mut Parser<'a>) -> PResult<'a, T>, + ) -> PResult<'a, (ThinVec<T>, Trailing, Recovered)> { + let mut first = true; + let mut recovered = Recovered::No; + let mut trailing = Trailing::No; + let mut v = ThinVec::new(); + + while !self.expect_any_with_type(kets, expect) { + if let token::CloseDelim(..) | token::Eof = self.token.kind { + break; + } + if let Some(t) = &sep.sep { + if first { + // no separator for the first element + first = false; + } else { + // check for separator + match self.expect(t) { + Ok(Recovered::No) => { + self.current_closure.take(); + } + Ok(Recovered::Yes(guar)) => { + self.current_closure.take(); + recovered = Recovered::Yes(guar); + break; + } + Err(mut expect_err) => { + let sp = self.prev_token.span.shrink_to_hi(); + let token_str = pprust::token_kind_to_string(t); + + match self.current_closure.take() { + Some(closure_spans) if self.token.kind == TokenKind::Semi => { + // Finding a semicolon instead of a comma + // after a closure body indicates that the + // closure body may be a block but the user + // forgot to put braces around its + // statements. + + self.recover_missing_braces_around_closure_body( + closure_spans, + expect_err, + )?; + + continue; + } + + _ => { + // Attempt to keep parsing if it was a similar separator. + if let Some(tokens) = t.similar_tokens() { + if tokens.contains(&self.token.kind) { + self.bump(); + } + } + } + } + + // If this was a missing `@` in a binding pattern + // bail with a suggestion + // https://github.com/rust-lang/rust/issues/72373 + if self.prev_token.is_ident() && self.token.kind == token::DotDot { + let msg = format!( + "if you meant to bind the contents of the rest of the array \ + pattern into `{}`, use `@`", + pprust::token_to_string(&self.prev_token) + ); + expect_err + .with_span_suggestion_verbose( + self.prev_token.span.shrink_to_hi().until(self.token.span), + msg, + " @ ", + Applicability::MaybeIncorrect, + ) + .emit(); + break; + } + + // Attempt to keep parsing if it was an omitted separator. + self.last_unexpected_token_span = None; + match f(self) { + Ok(t) => { + // Parsed successfully, therefore most probably the code only + // misses a separator. + expect_err + .with_span_suggestion_short( + sp, + format!("missing `{token_str}`"), + token_str, + Applicability::MaybeIncorrect, + ) + .emit(); + + v.push(t); + continue; + } + Err(e) => { + // Parsing failed, therefore it must be something more serious + // than just a missing separator. + for xx in &e.children { + // propagate the help message from sub error 'e' to main error 'expect_err; + expect_err.children.push(xx.clone()); + } + e.cancel(); + if self.token == token::Colon { + // we will try to recover in `maybe_recover_struct_lit_bad_delims` + return Err(expect_err); + } else if let [token::CloseDelim(Delimiter::Parenthesis)] = kets + { + return Err(expect_err); + } else { + expect_err.emit(); + break; + } + } + } + } + } + } + } + if sep.trailing_sep_allowed && self.expect_any_with_type(kets, expect) { + trailing = Trailing::Yes; + break; + } + + let t = f(self)?; + v.push(t); + } + + Ok((v, trailing, recovered)) + } + + fn recover_missing_braces_around_closure_body( + &mut self, + closure_spans: ClosureSpans, + mut expect_err: Diag<'_>, + ) -> PResult<'a, ()> { + let initial_semicolon = self.token.span; + + while self.eat(&TokenKind::Semi) { + let _ = + self.parse_stmt_without_recovery(false, ForceCollect::Yes).unwrap_or_else(|e| { + e.cancel(); + None + }); + } + + expect_err + .primary_message("closure bodies that contain statements must be surrounded by braces"); + + let preceding_pipe_span = closure_spans.closing_pipe; + let following_token_span = self.token.span; + + let mut first_note = MultiSpan::from(vec![initial_semicolon]); + first_note.push_span_label( + initial_semicolon, + "this `;` turns the preceding closure into a statement", + ); + first_note.push_span_label( + closure_spans.body, + "this expression is a statement because of the trailing semicolon", + ); + expect_err.span_note(first_note, "statement found outside of a block"); + + let mut second_note = MultiSpan::from(vec![closure_spans.whole_closure]); + second_note.push_span_label(closure_spans.whole_closure, "this is the parsed closure..."); + second_note.push_span_label( + following_token_span, + "...but likely you meant the closure to end here", + ); + expect_err.span_note(second_note, "the closure body may be incorrectly delimited"); + + expect_err.span(vec![preceding_pipe_span, following_token_span]); + + let opening_suggestion_str = " {".to_string(); + let closing_suggestion_str = "}".to_string(); + + expect_err.multipart_suggestion( + "try adding braces", + vec![ + (preceding_pipe_span.shrink_to_hi(), opening_suggestion_str), + (following_token_span.shrink_to_lo(), closing_suggestion_str), + ], + Applicability::MaybeIncorrect, + ); + + expect_err.emit(); + + Ok(()) + } + + /// Parses a sequence, not including the delimiters. The function + /// `f` must consume tokens until reaching the next separator or + /// closing bracket. + fn parse_seq_to_before_end<T>( + &mut self, + ket: &TokenKind, + sep: SeqSep, + f: impl FnMut(&mut Parser<'a>) -> PResult<'a, T>, + ) -> PResult<'a, (ThinVec<T>, Trailing, Recovered)> { + self.parse_seq_to_before_tokens(&[ket], sep, TokenExpectType::Expect, f) + } + + /// Parses a sequence, including only the closing delimiter. The function + /// `f` must consume tokens until reaching the next separator or + /// closing bracket. + fn parse_seq_to_end<T>( + &mut self, + ket: &TokenKind, + sep: SeqSep, + f: impl FnMut(&mut Parser<'a>) -> PResult<'a, T>, + ) -> PResult<'a, (ThinVec<T>, Trailing)> { + let (val, trailing, recovered) = self.parse_seq_to_before_end(ket, sep, f)?; + if matches!(recovered, Recovered::No) { + self.eat(ket); + } + Ok((val, trailing)) + } + + /// Parses a sequence, including both delimiters. The function + /// `f` must consume tokens until reaching the next separator or + /// closing bracket. + fn parse_unspanned_seq<T>( + &mut self, + bra: &TokenKind, + ket: &TokenKind, + sep: SeqSep, + f: impl FnMut(&mut Parser<'a>) -> PResult<'a, T>, + ) -> PResult<'a, (ThinVec<T>, Trailing)> { + self.expect(bra)?; + self.parse_seq_to_end(ket, sep, f) + } + + /// Parses a comma-separated sequence, including both delimiters. + /// The function `f` must consume tokens until reaching the next separator or + /// closing bracket. + fn parse_delim_comma_seq<T>( + &mut self, + delim: Delimiter, + f: impl FnMut(&mut Parser<'a>) -> PResult<'a, T>, + ) -> PResult<'a, (ThinVec<T>, Trailing)> { + self.parse_unspanned_seq( + &token::OpenDelim(delim), + &token::CloseDelim(delim), + SeqSep::trailing_allowed(token::Comma), + f, + ) + } + + /// Parses a comma-separated sequence delimited by parentheses (e.g. `(x, y)`). + /// The function `f` must consume tokens until reaching the next separator or + /// closing bracket. + fn parse_paren_comma_seq<T>( + &mut self, + f: impl FnMut(&mut Parser<'a>) -> PResult<'a, T>, + ) -> PResult<'a, (ThinVec<T>, Trailing)> { + self.parse_delim_comma_seq(Delimiter::Parenthesis, f) + } + + /// Advance the parser by one token using provided token as the next one. + fn bump_with(&mut self, next: (Token, Spacing)) { + self.inlined_bump_with(next) + } + + /// This always-inlined version should only be used on hot code paths. + #[inline(always)] + fn inlined_bump_with(&mut self, (next_token, next_spacing): (Token, Spacing)) { + // Update the current and previous tokens. + self.prev_token = mem::replace(&mut self.token, next_token); + self.token_spacing = next_spacing; + + // Diagnostics. + self.expected_tokens.clear(); + } + + /// Advance the parser by one token. + pub fn bump(&mut self) { + // Note: destructuring here would give nicer code, but it was found in #96210 to be slower + // than `.0`/`.1` access. + let mut next = self.token_cursor.inlined_next(); + self.num_bump_calls += 1; + // We've retrieved an token from the underlying + // cursor, so we no longer need to worry about + // an unglued token. See `break_and_eat` for more details + self.break_last_token = false; + if next.0.span.is_dummy() { + // Tweak the location for better diagnostics, but keep syntactic context intact. + let fallback_span = self.token.span; + next.0.span = fallback_span.with_ctxt(next.0.span.ctxt()); + } + debug_assert!(!matches!( + next.0.kind, + token::OpenDelim(Delimiter::Invisible) | token::CloseDelim(Delimiter::Invisible) + )); + self.inlined_bump_with(next) + } + + /// Look-ahead `dist` tokens of `self.token` and get access to that token there. + /// When `dist == 0` then the current token is looked at. `Eof` will be + /// returned if the look-ahead is any distance past the end of the tokens. + pub fn look_ahead<R>(&self, dist: usize, looker: impl FnOnce(&Token) -> R) -> R { + if dist == 0 { + return looker(&self.token); + } + + if let Some(&(_, span, _, delim)) = self.token_cursor.stack.last() + && delim != Delimiter::Invisible + { + // We are not in the outermost token stream, and the token stream + // we are in has non-skipped delimiters. Look for skipped + // delimiters in the lookahead range. + let tree_cursor = &self.token_cursor.tree_cursor; + let all_normal = (0..dist).all(|i| { + let token = tree_cursor.look_ahead(i); + !matches!(token, Some(TokenTree::Delimited(.., Delimiter::Invisible, _))) + }); + if all_normal { + // There were no skipped delimiters. Do lookahead by plain indexing. + return match tree_cursor.look_ahead(dist - 1) { + Some(tree) => { + // Indexing stayed within the current token stream. + match tree { + TokenTree::Token(token, _) => looker(token), + TokenTree::Delimited(dspan, _, delim, _) => { + looker(&Token::new(token::OpenDelim(*delim), dspan.open)) + } + } + } + None => { + // Indexing went past the end of the current token + // stream. Use the close delimiter, no matter how far + // ahead `dist` went. + looker(&Token::new(token::CloseDelim(delim), span.close)) + } + }; + } + } + + // We are in a more complex case. Just clone the token cursor and use + // `next`, skipping delimiters as necessary. Slow but simple. + let mut cursor = self.token_cursor.clone(); + let mut i = 0; + let mut token = Token::dummy(); + while i < dist { + token = cursor.next().0; + if matches!( + token.kind, + token::OpenDelim(Delimiter::Invisible) | token::CloseDelim(Delimiter::Invisible) + ) { + continue; + } + i += 1; + } + looker(&token) + } + + /// Returns whether any of the given keywords are `dist` tokens ahead of the current one. + pub(crate) fn is_keyword_ahead(&self, dist: usize, kws: &[Symbol]) -> bool { + self.look_ahead(dist, |t| kws.iter().any(|&kw| t.is_keyword(kw))) + } + + /// Parses asyncness: `async` or nothing. + fn parse_coroutine_kind(&mut self, case: Case) -> Option<CoroutineKind> { + let span = self.token.uninterpolated_span(); + if self.eat_keyword_case(kw::Async, case) { + // FIXME(gen_blocks): Do we want to unconditionally parse `gen` and then + // error if edition <= 2024, like we do with async and edition <= 2018? + if self.token.uninterpolated_span().at_least_rust_2024() + && self.eat_keyword_case(kw::Gen, case) + { + let gen_span = self.prev_token.uninterpolated_span(); + Some(CoroutineKind::AsyncGen { + span: span.to(gen_span), + closure_id: DUMMY_NODE_ID, + return_impl_trait_id: DUMMY_NODE_ID, + }) + } else { + Some(CoroutineKind::Async { + span, + closure_id: DUMMY_NODE_ID, + return_impl_trait_id: DUMMY_NODE_ID, + }) + } + } else if self.token.uninterpolated_span().at_least_rust_2024() + && self.eat_keyword_case(kw::Gen, case) + { + Some(CoroutineKind::Gen { + span, + closure_id: DUMMY_NODE_ID, + return_impl_trait_id: DUMMY_NODE_ID, + }) + } else { + None + } + } + + /// Parses fn unsafety: `unsafe`, `safe` or nothing. + fn parse_safety(&mut self, case: Case) -> Safety { + if self.eat_keyword_case(kw::Unsafe, case) { + Safety::Unsafe(self.prev_token.uninterpolated_span()) + } else { + Safety::Default + } + } + + /// Parses constness: `const` or nothing. + fn parse_constness(&mut self, case: Case) -> Const { + self.parse_constness_(case, false) + } + + /// Parses constness for closures (case sensitive, feature-gated) + fn parse_closure_constness(&mut self) -> Const { + let constness = self.parse_constness_(Case::Sensitive, true); + if let Const::Yes(span) = constness { + self.psess.gated_spans.gate(sym::const_closures, span); + } + constness + } + + fn parse_constness_(&mut self, case: Case, is_closure: bool) -> Const { + // Avoid const blocks and const closures to be parsed as const items + if (self.check_const_closure() == is_closure) + && !self + .look_ahead(1, |t| *t == token::OpenDelim(Delimiter::Brace) || t.is_whole_block()) + && self.eat_keyword_case(kw::Const, case) + { + Const::Yes(self.prev_token.uninterpolated_span()) + } else { + Const::No + } + } + + /// Parses inline const expressions. + fn parse_const_block(&mut self, span: Span, pat: bool) -> PResult<'a, P<Expr>> { + if pat { + self.psess.gated_spans.gate(sym::inline_const_pat, span); + } + self.eat_keyword(kw::Const); + let (attrs, blk) = self.parse_inner_attrs_and_block()?; + let anon_const = AnonConst { + id: DUMMY_NODE_ID, + value: self.mk_expr(blk.span, ExprKind::Block(blk, None)), + }; + let blk_span = anon_const.value.span; + Ok(self.mk_expr_with_attrs(span.to(blk_span), ExprKind::ConstBlock(anon_const), attrs)) + } + + /// Parses mutability (`mut` or nothing). + fn parse_mutability(&mut self) -> Mutability { + if self.eat_keyword(kw::Mut) { Mutability::Mut } else { Mutability::Not } + } + + /// Parses reference binding mode (`ref`, `ref mut`, or nothing). + fn parse_byref(&mut self) -> ByRef { + if self.eat_keyword(kw::Ref) { ByRef::Yes(self.parse_mutability()) } else { ByRef::No } + } + + /// Possibly parses mutability (`const` or `mut`). + fn parse_const_or_mut(&mut self) -> Option<Mutability> { + if self.eat_keyword(kw::Mut) { + Some(Mutability::Mut) + } else if self.eat_keyword(kw::Const) { + Some(Mutability::Not) + } else { + None + } + } + + fn parse_field_name(&mut self) -> PResult<'a, Ident> { + if let token::Literal(token::Lit { kind: token::Integer, symbol, suffix }) = self.token.kind + { + if let Some(suffix) = suffix { + self.expect_no_tuple_index_suffix(self.token.span, suffix); + } + self.bump(); + Ok(Ident::new(symbol, self.prev_token.span)) + } else { + self.parse_ident_common(true) + } + } + + fn parse_delim_args(&mut self) -> PResult<'a, P<DelimArgs>> { + if let Some(args) = self.parse_delim_args_inner() { + Ok(P(args)) + } else { + self.unexpected_any() + } + } + + fn parse_attr_args(&mut self) -> PResult<'a, AttrArgs> { + Ok(if let Some(args) = self.parse_delim_args_inner() { + AttrArgs::Delimited(args) + } else { + if self.eat(&token::Eq) { + let eq_span = self.prev_token.span; + AttrArgs::Eq(eq_span, AttrArgsEq::Ast(self.parse_expr_force_collect()?)) + } else { + AttrArgs::Empty + } + }) + } + + fn parse_delim_args_inner(&mut self) -> Option<DelimArgs> { + let delimited = self.check(&token::OpenDelim(Delimiter::Parenthesis)) + || self.check(&token::OpenDelim(Delimiter::Bracket)) + || self.check(&token::OpenDelim(Delimiter::Brace)); + + delimited.then(|| { + let TokenTree::Delimited(dspan, _, delim, tokens) = self.parse_token_tree() else { + unreachable!() + }; + DelimArgs { dspan, delim, tokens } + }) + } + + fn parse_or_use_outer_attributes( + &mut self, + already_parsed_attrs: Option<AttrWrapper>, + ) -> PResult<'a, AttrWrapper> { + if let Some(attrs) = already_parsed_attrs { + Ok(attrs) + } else { + self.parse_outer_attributes() + } + } + + /// Parses a single token tree from the input. + pub fn parse_token_tree(&mut self) -> TokenTree { + match self.token.kind { + token::OpenDelim(..) => { + // Grab the tokens within the delimiters. + let stream = self.token_cursor.tree_cursor.stream.clone(); + let (_, span, spacing, delim) = *self.token_cursor.stack.last().unwrap(); + + // Advance the token cursor through the entire delimited + // sequence. After getting the `OpenDelim` we are *within* the + // delimited sequence, i.e. at depth `d`. After getting the + // matching `CloseDelim` we are *after* the delimited sequence, + // i.e. at depth `d - 1`. + let target_depth = self.token_cursor.stack.len() - 1; + loop { + // Advance one token at a time, so `TokenCursor::next()` + // can capture these tokens if necessary. + self.bump(); + if self.token_cursor.stack.len() == target_depth { + debug_assert!(matches!(self.token.kind, token::CloseDelim(_))); + break; + } + } + + // Consume close delimiter + self.bump(); + TokenTree::Delimited(span, spacing, delim, stream) + } + token::CloseDelim(_) | token::Eof => unreachable!(), + _ => { + let prev_spacing = self.token_spacing; + self.bump(); + TokenTree::Token(self.prev_token.clone(), prev_spacing) + } + } + } + + pub fn parse_tokens(&mut self) -> TokenStream { + let mut result = Vec::new(); + loop { + match self.token.kind { + token::Eof | token::CloseDelim(..) => break, + _ => result.push(self.parse_token_tree()), + } + } + TokenStream::new(result) + } + + /// Evaluates the closure with restrictions in place. + /// + /// Afters the closure is evaluated, restrictions are reset. + fn with_res<T>(&mut self, res: Restrictions, f: impl FnOnce(&mut Self) -> T) -> T { + let old = self.restrictions; + self.restrictions = res; + let res = f(self); + self.restrictions = old; + res + } + + /// Parses `pub` and `pub(in path)` plus shortcuts `pub(crate)` for `pub(in crate)`, `pub(self)` + /// for `pub(in self)` and `pub(super)` for `pub(in super)`. + /// If the following element can't be a tuple (i.e., it's a function definition), then + /// it's not a tuple struct field), and the contents within the parentheses aren't valid, + /// so emit a proper diagnostic. + // Public for rustfmt usage. + pub fn parse_visibility(&mut self, fbt: FollowedByType) -> PResult<'a, Visibility> { + maybe_whole!(self, NtVis, |vis| vis.into_inner()); + + if !self.eat_keyword(kw::Pub) { + // We need a span for our `Spanned<VisibilityKind>`, but there's inherently no + // keyword to grab a span from for inherited visibility; an empty span at the + // beginning of the current token would seem to be the "Schelling span". + return Ok(Visibility { + span: self.token.span.shrink_to_lo(), + kind: VisibilityKind::Inherited, + tokens: None, + }); + } + let lo = self.prev_token.span; + + if self.check(&token::OpenDelim(Delimiter::Parenthesis)) { + // We don't `self.bump()` the `(` yet because this might be a struct definition where + // `()` or a tuple might be allowed. For example, `struct Struct(pub (), pub (usize));`. + // Because of this, we only `bump` the `(` if we're assured it is appropriate to do so + // by the following tokens. + if self.is_keyword_ahead(1, &[kw::In]) { + // Parse `pub(in path)`. + self.bump(); // `(` + self.bump(); // `in` + let path = self.parse_path(PathStyle::Mod)?; // `path` + self.expect(&token::CloseDelim(Delimiter::Parenthesis))?; // `)` + let vis = VisibilityKind::Restricted { + path: P(path), + id: ast::DUMMY_NODE_ID, + shorthand: false, + }; + return Ok(Visibility { + span: lo.to(self.prev_token.span), + kind: vis, + tokens: None, + }); + } else if self.look_ahead(2, |t| t == &token::CloseDelim(Delimiter::Parenthesis)) + && self.is_keyword_ahead(1, &[kw::Crate, kw::Super, kw::SelfLower]) + { + // Parse `pub(crate)`, `pub(self)`, or `pub(super)`. + self.bump(); // `(` + let path = self.parse_path(PathStyle::Mod)?; // `crate`/`super`/`self` + self.expect(&token::CloseDelim(Delimiter::Parenthesis))?; // `)` + let vis = VisibilityKind::Restricted { + path: P(path), + id: ast::DUMMY_NODE_ID, + shorthand: true, + }; + return Ok(Visibility { + span: lo.to(self.prev_token.span), + kind: vis, + tokens: None, + }); + } else if let FollowedByType::No = fbt { + // Provide this diagnostic if a type cannot follow; + // in particular, if this is not a tuple struct. + self.recover_incorrect_vis_restriction()?; + // Emit diagnostic, but continue with public visibility. + } + } + + Ok(Visibility { span: lo, kind: VisibilityKind::Public, tokens: None }) + } + + /// Recovery for e.g. `pub(something) fn ...` or `struct X { pub(something) y: Z }` + fn recover_incorrect_vis_restriction(&mut self) -> PResult<'a, ()> { + self.bump(); // `(` + let path = self.parse_path(PathStyle::Mod)?; + self.expect(&token::CloseDelim(Delimiter::Parenthesis))?; // `)` + + let path_str = pprust::path_to_string(&path); + self.dcx() + .emit_err(IncorrectVisibilityRestriction { span: path.span, inner_str: path_str }); + + Ok(()) + } + + /// Parses `extern string_literal?`. + fn parse_extern(&mut self, case: Case) -> Extern { + if self.eat_keyword_case(kw::Extern, case) { + let mut extern_span = self.prev_token.span; + let abi = self.parse_abi(); + if let Some(abi) = abi { + extern_span = extern_span.to(abi.span); + } + Extern::from_abi(abi, extern_span) + } else { + Extern::None + } + } + + /// Parses a string literal as an ABI spec. + fn parse_abi(&mut self) -> Option<StrLit> { + match self.parse_str_lit() { + Ok(str_lit) => Some(str_lit), + Err(Some(lit)) => match lit.kind { + ast::LitKind::Err(_) => None, + _ => { + self.dcx().emit_err(NonStringAbiLiteral { span: lit.span }); + None + } + }, + Err(None) => None, + } + } + + pub fn collect_tokens_no_attrs<R: HasAttrs + HasTokens>( + &mut self, + f: impl FnOnce(&mut Self) -> PResult<'a, R>, + ) -> PResult<'a, R> { + // The only reason to call `collect_tokens_no_attrs` is if you want tokens, so use + // `ForceCollect::Yes` + self.collect_tokens_trailing_token( + AttrWrapper::empty(), + ForceCollect::Yes, + |this, _attrs| Ok((f(this)?, TrailingToken::None)), + ) + } + + /// `::{` or `::*` + fn is_import_coupler(&mut self) -> bool { + self.check(&token::PathSep) + && self.look_ahead(1, |t| { + *t == token::OpenDelim(Delimiter::Brace) || *t == token::BinOp(token::Star) + }) + } + + // debug view of the parser's token stream, up to `{lookahead}` tokens + pub fn debug_lookahead(&self, lookahead: usize) -> impl fmt::Debug + '_ { + struct DebugParser<'dbg> { + parser: &'dbg Parser<'dbg>, + lookahead: usize, + } + + impl fmt::Debug for DebugParser<'_> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + let Self { parser, lookahead } = self; + let mut dbg_fmt = f.debug_struct("Parser"); // or at least, one view of + + // we don't need N spans, but we want at least one, so print all of prev_token + dbg_fmt.field("prev_token", &parser.prev_token); + // make it easier to peek farther ahead by taking TokenKinds only until EOF + let tokens = (0..*lookahead) + .map(|i| parser.look_ahead(i, |tok| tok.kind.clone())) + .scan(parser.prev_token == TokenKind::Eof, |eof, tok| { + let current = eof.then_some(tok.clone()); // include a trailing EOF token + *eof |= &tok == &TokenKind::Eof; + current + }); + dbg_fmt.field_with("tokens", |field| field.debug_list().entries(tokens).finish()); + dbg_fmt.field("approx_token_stream_pos", &parser.num_bump_calls); + + // some fields are interesting for certain values, as they relate to macro parsing + if let Some(subparser) = parser.subparser_name { + dbg_fmt.field("subparser_name", &subparser); + } + if let Recovery::Forbidden = parser.recovery { + dbg_fmt.field("recovery", &parser.recovery); + } + + // imply there's "more to know" than this view + dbg_fmt.finish_non_exhaustive() + } + } + + DebugParser { parser: self, lookahead } + } + + pub fn clear_expected_tokens(&mut self) { + self.expected_tokens.clear(); + } + + pub fn approx_token_stream_pos(&self) -> usize { + self.num_bump_calls + } +} + +pub(crate) fn make_unclosed_delims_error( + unmatched: UnmatchedDelim, + psess: &ParseSess, +) -> Option<Diag<'_>> { + // `None` here means an `Eof` was found. We already emit those errors elsewhere, we add them to + // `unmatched_delims` only for error recovery in the `Parser`. + let found_delim = unmatched.found_delim?; + let mut spans = vec![unmatched.found_span]; + if let Some(sp) = unmatched.unclosed_span { + spans.push(sp); + }; + let err = psess.dcx.create_err(MismatchedClosingDelimiter { + spans, + delimiter: pprust::token_kind_to_string(&token::CloseDelim(found_delim)).to_string(), + unmatched: unmatched.found_span, + opening_candidate: unmatched.candidate_span, + unclosed: unmatched.unclosed_span, + }); + Some(err) +} + +/// A helper struct used when building an `AttrTokenStream` from +/// a `LazyAttrTokenStream`. Both delimiter and non-delimited tokens +/// are stored as `FlatToken::Token`. A vector of `FlatToken`s +/// is then 'parsed' to build up an `AttrTokenStream` with nested +/// `AttrTokenTree::Delimited` tokens. +#[derive(Debug, Clone)] +pub enum FlatToken { + /// A token - this holds both delimiter (e.g. '{' and '}') + /// and non-delimiter tokens + Token(Token), + /// Holds the `AttributesData` for an AST node. The + /// `AttributesData` is inserted directly into the + /// constructed `AttrTokenStream` as + /// an `AttrTokenTree::Attributes`. + AttrTarget(AttributesData), + /// A special 'empty' token that is ignored during the conversion + /// to an `AttrTokenStream`. This is used to simplify the + /// handling of replace ranges. + Empty, +} + +// Metavar captures of various kinds. +#[derive(Clone, Debug)] +pub enum ParseNtResult { + Tt(TokenTree), + Ident(Ident, IdentIsRaw), + Lifetime(Ident), + + /// This case will eventually be removed, along with `Token::Interpolate`. + Nt(Lrc<Nonterminal>), +} diff --git a/compiler/rustc_parse/src/parser/mut_visit/tests.rs b/compiler/rustc_parse/src/parser/mut_visit/tests.rs new file mode 100644 index 00000000000..b3cb28af657 --- /dev/null +++ b/compiler/rustc_parse/src/parser/mut_visit/tests.rs @@ -0,0 +1,71 @@ +use crate::parser::tests::{matches_codepattern, string_to_crate}; +use rustc_ast as ast; +use rustc_ast::mut_visit::MutVisitor; +use rustc_ast_pretty::pprust; +use rustc_span::create_default_session_globals_then; +use rustc_span::symbol::Ident; + +// This version doesn't care about getting comments or doc-strings in. +fn print_crate_items(krate: &ast::Crate) -> String { + krate.items.iter().map(|i| pprust::item_to_string(i)).collect::<Vec<_>>().join(" ") +} + +// Change every identifier to "zz". +struct ToZzIdentMutVisitor; + +impl MutVisitor for ToZzIdentMutVisitor { + const VISIT_TOKENS: bool = true; + + fn visit_ident(&mut self, ident: &mut Ident) { + *ident = Ident::from_str("zz"); + } +} + +// Maybe add to `expand.rs`. +macro_rules! assert_pred { + ($pred:expr, $predname:expr, $a:expr , $b:expr) => {{ + let pred_val = $pred; + let a_val = $a; + let b_val = $b; + if !(pred_val(&a_val, &b_val)) { + panic!("expected args satisfying {}, got {} and {}", $predname, a_val, b_val); + } + }}; +} + +// Make sure idents get transformed everywhere. +#[test] +fn ident_transformation() { + create_default_session_globals_then(|| { + let mut zz_visitor = ToZzIdentMutVisitor; + let mut krate = + string_to_crate("#[a] mod b {fn c (d : e, f : g) {h!(i,j,k);l;m}}".to_string()); + zz_visitor.visit_crate(&mut krate); + assert_pred!( + matches_codepattern, + "matches_codepattern", + print_crate_items(&krate), + "#[zz]mod zz{fn zz(zz:zz,zz:zz){zz!(zz,zz,zz);zz;zz}}".to_string() + ); + }) +} + +// Make sure idents get transformed even inside macro defs. +#[test] +fn ident_transformation_in_defs() { + create_default_session_globals_then(|| { + let mut zz_visitor = ToZzIdentMutVisitor; + let mut krate = string_to_crate( + "macro_rules! a {(b $c:expr $(d $e:token)f+ => \ + (g $(d $d $e)+))} " + .to_string(), + ); + zz_visitor.visit_crate(&mut krate); + assert_pred!( + matches_codepattern, + "matches_codepattern", + print_crate_items(&krate), + "macro_rules! zz{(zz$zz:zz$(zz $zz:zz)zz+=>(zz$(zz$zz$zz)+))}".to_string() + ); + }) +} diff --git a/compiler/rustc_parse/src/parser/nonterminal.rs b/compiler/rustc_parse/src/parser/nonterminal.rs new file mode 100644 index 00000000000..619c4c63e51 --- /dev/null +++ b/compiler/rustc_parse/src/parser/nonterminal.rs @@ -0,0 +1,214 @@ +use rustc_ast::ptr::P; +use rustc_ast::token::{self, Delimiter, Nonterminal::*, NonterminalKind, Token}; +use rustc_ast::HasTokens; +use rustc_ast_pretty::pprust; +use rustc_data_structures::sync::Lrc; +use rustc_errors::PResult; +use rustc_span::symbol::{kw, Ident}; + +use crate::errors::UnexpectedNonterminal; +use crate::parser::pat::{CommaRecoveryMode, RecoverColon, RecoverComma}; +use crate::parser::{FollowedByType, ForceCollect, ParseNtResult, Parser, PathStyle}; + +impl<'a> Parser<'a> { + /// Checks whether a non-terminal may begin with a particular token. + /// + /// Returning `false` is a *stability guarantee* that such a matcher will *never* begin with + /// that token. Be conservative (return true) if not sure. Inlined because it has a single call + /// site. + #[inline] + pub fn nonterminal_may_begin_with(kind: NonterminalKind, token: &Token) -> bool { + /// Checks whether the non-terminal may contain a single (non-keyword) identifier. + fn may_be_ident(nt: &token::Nonterminal) -> bool { + match nt { + NtStmt(_) + | NtPat(_) + | NtExpr(_) + | NtTy(_) + | NtLiteral(_) // `true`, `false` + | NtMeta(_) + | NtPath(_) => true, + + NtItem(_) + | NtBlock(_) + | NtVis(_) => false, + } + } + + match kind { + NonterminalKind::Expr2021 => { + token.can_begin_expr() + // This exception is here for backwards compatibility. + && !token.is_keyword(kw::Let) + // This exception is here for backwards compatibility. + && !token.is_keyword(kw::Const) + } + NonterminalKind::Expr => { + token.can_begin_expr() + // This exception is here for backwards compatibility. + && !token.is_keyword(kw::Let) + && (token.span.edition().at_least_rust_2024() || !token.is_keyword(kw::Const)) + } + NonterminalKind::Ty => token.can_begin_type(), + NonterminalKind::Ident => get_macro_ident(token).is_some(), + NonterminalKind::Literal => token.can_begin_literal_maybe_minus(), + NonterminalKind::Vis => match token.kind { + // The follow-set of :vis + "priv" keyword + interpolated + token::Comma + | token::Ident(..) + | token::NtIdent(..) + | token::NtLifetime(..) + | token::Interpolated(_) => true, + _ => token.can_begin_type(), + }, + NonterminalKind::Block => match &token.kind { + token::OpenDelim(Delimiter::Brace) => true, + token::NtLifetime(..) => true, + token::Interpolated(nt) => match &**nt { + NtBlock(_) | NtStmt(_) | NtExpr(_) | NtLiteral(_) => true, + NtItem(_) | NtPat(_) | NtTy(_) | NtMeta(_) | NtPath(_) | NtVis(_) => false, + }, + _ => false, + }, + NonterminalKind::Path | NonterminalKind::Meta => match &token.kind { + token::PathSep | token::Ident(..) | token::NtIdent(..) => true, + token::Interpolated(nt) => may_be_ident(nt), + _ => false, + }, + NonterminalKind::PatParam { .. } | NonterminalKind::PatWithOr => match &token.kind { + // box, ref, mut, and other identifiers (can stricten) + token::Ident(..) | token::NtIdent(..) | + token::OpenDelim(Delimiter::Parenthesis) | // tuple pattern + token::OpenDelim(Delimiter::Bracket) | // slice pattern + token::BinOp(token::And) | // reference + token::BinOp(token::Minus) | // negative literal + token::AndAnd | // double reference + token::Literal(_) | // literal + token::DotDot | // range pattern (future compat) + token::DotDotDot | // range pattern (future compat) + token::PathSep | // path + token::Lt | // path (UFCS constant) + token::BinOp(token::Shl) => true, // path (double UFCS) + // leading vert `|` or-pattern + token::BinOp(token::Or) => matches!(kind, NonterminalKind::PatWithOr), + token::Interpolated(nt) => may_be_ident(nt), + _ => false, + }, + NonterminalKind::Lifetime => match &token.kind { + token::Lifetime(_) | token::NtLifetime(..) => true, + _ => false, + }, + NonterminalKind::TT | NonterminalKind::Item | NonterminalKind::Stmt => { + !matches!(token.kind, token::CloseDelim(_)) + } + } + } + + /// Parse a non-terminal (e.g. MBE `:pat` or `:ident`). Inlined because there is only one call + /// site. + #[inline] + pub fn parse_nonterminal(&mut self, kind: NonterminalKind) -> PResult<'a, ParseNtResult> { + // A `macro_rules!` invocation may pass a captured item/expr to a proc-macro, + // which requires having captured tokens available. Since we cannot determine + // in advance whether or not a proc-macro will be (transitively) invoked, + // we always capture tokens for any `Nonterminal` which needs them. + let mut nt = match kind { + // Note that TT is treated differently to all the others. + NonterminalKind::TT => return Ok(ParseNtResult::Tt(self.parse_token_tree())), + NonterminalKind::Item => match self.parse_item(ForceCollect::Yes)? { + Some(item) => NtItem(item), + None => { + return Err(self + .dcx() + .create_err(UnexpectedNonterminal::Item(self.token.span))); + } + }, + NonterminalKind::Block => { + // While a block *expression* may have attributes (e.g. `#[my_attr] { ... }`), + // the ':block' matcher does not support them + NtBlock(self.collect_tokens_no_attrs(|this| this.parse_block())?) + } + NonterminalKind::Stmt => match self.parse_stmt(ForceCollect::Yes)? { + Some(s) => NtStmt(P(s)), + None => { + return Err(self + .dcx() + .create_err(UnexpectedNonterminal::Statement(self.token.span))); + } + }, + NonterminalKind::PatParam { .. } | NonterminalKind::PatWithOr => { + NtPat(self.collect_tokens_no_attrs(|this| match kind { + NonterminalKind::PatParam { .. } => this.parse_pat_no_top_alt(None, None), + NonterminalKind::PatWithOr => this.parse_pat_allow_top_alt( + None, + RecoverComma::No, + RecoverColon::No, + CommaRecoveryMode::EitherTupleOrPipe, + ), + _ => unreachable!(), + })?) + } + + NonterminalKind::Expr | NonterminalKind::Expr2021 => { + NtExpr(self.parse_expr_force_collect()?) + } + NonterminalKind::Literal => { + // The `:literal` matcher does not support attributes + NtLiteral(self.collect_tokens_no_attrs(|this| this.parse_literal_maybe_minus())?) + } + + NonterminalKind::Ty => { + NtTy(self.collect_tokens_no_attrs(|this| this.parse_ty_no_question_mark_recover())?) + } + + // this could be handled like a token, since it is one + NonterminalKind::Ident => { + return if let Some((ident, is_raw)) = get_macro_ident(&self.token) { + self.bump(); + Ok(ParseNtResult::Ident(ident, is_raw)) + } else { + Err(self.dcx().create_err(UnexpectedNonterminal::Ident { + span: self.token.span, + token: self.token.clone(), + })) + }; + } + NonterminalKind::Path => { + NtPath(P(self.collect_tokens_no_attrs(|this| this.parse_path(PathStyle::Type))?)) + } + NonterminalKind::Meta => NtMeta(P(self.parse_attr_item(true)?)), + NonterminalKind::Vis => { + NtVis(P(self + .collect_tokens_no_attrs(|this| this.parse_visibility(FollowedByType::Yes))?)) + } + NonterminalKind::Lifetime => { + return if self.check_lifetime() { + Ok(ParseNtResult::Lifetime(self.expect_lifetime().ident)) + } else { + Err(self.dcx().create_err(UnexpectedNonterminal::Lifetime { + span: self.token.span, + token: self.token.clone(), + })) + }; + } + }; + + // If tokens are supported at all, they should be collected. + if matches!(nt.tokens_mut(), Some(None)) { + panic!( + "Missing tokens for nt {:?} at {:?}: {:?}", + nt, + nt.use_span(), + pprust::nonterminal_to_string(&nt) + ); + } + + Ok(ParseNtResult::Nt(Lrc::new(nt))) + } +} + +/// The token is an identifier, but not `_`. +/// We prohibit passing `_` to macros expecting `ident` for now. +fn get_macro_ident(token: &Token) -> Option<(Ident, token::IdentIsRaw)> { + token.ident().filter(|(ident, _)| ident.name != kw::Underscore) +} diff --git a/compiler/rustc_parse/src/parser/pat.rs b/compiler/rustc_parse/src/parser/pat.rs new file mode 100644 index 00000000000..8af415f7c9d --- /dev/null +++ b/compiler/rustc_parse/src/parser/pat.rs @@ -0,0 +1,1426 @@ +use super::{ForceCollect, Parser, PathStyle, Restrictions, Trailing, TrailingToken}; +use crate::errors::{ + self, AmbiguousRangePattern, DotDotDotForRemainingFields, DotDotDotRangeToPatternNotAllowed, + DotDotDotRestPattern, EnumPatternInsteadOfIdentifier, ExpectedBindingLeftOfAt, + ExpectedCommaAfterPatternField, GenericArgsInPatRequireTurbofishSyntax, + InclusiveRangeExtraEquals, InclusiveRangeMatchArrow, InclusiveRangeNoEnd, InvalidMutInPattern, + PatternOnWrongSideOfAt, RemoveLet, RepeatedMutInPattern, SwitchRefBoxOrder, + TopLevelOrPatternNotAllowed, TopLevelOrPatternNotAllowedSugg, TrailingVertNotAllowed, + UnexpectedExpressionInPattern, UnexpectedLifetimeInPattern, UnexpectedParenInRangePat, + UnexpectedParenInRangePatSugg, UnexpectedVertVertBeforeFunctionParam, + UnexpectedVertVertInPattern, +}; +use crate::parser::expr::could_be_unclosed_char_literal; +use crate::{maybe_recover_from_interpolated_ty_qpath, maybe_whole}; +use rustc_ast::mut_visit::{noop_visit_pat, MutVisitor}; +use rustc_ast::ptr::P; +use rustc_ast::token::{self, BinOpToken, Delimiter, Token}; +use rustc_ast::{ + self as ast, AttrVec, BindingMode, ByRef, Expr, ExprKind, MacCall, Mutability, Pat, PatField, + PatFieldsRest, PatKind, Path, QSelf, RangeEnd, RangeSyntax, +}; +use rustc_ast_pretty::pprust; +use rustc_errors::{Applicability, Diag, PResult}; +use rustc_session::errors::ExprParenthesesNeeded; +use rustc_span::source_map::{respan, Spanned}; +use rustc_span::symbol::{kw, sym, Ident}; +use rustc_span::{ErrorGuaranteed, Span}; +use thin_vec::{thin_vec, ThinVec}; + +#[derive(PartialEq, Copy, Clone)] +pub enum Expected { + ParameterName, + ArgumentName, + Identifier, + BindingPattern, +} + +impl Expected { + // FIXME(#100717): migrate users of this to proper localization + fn to_string_or_fallback(expected: Option<Expected>) -> &'static str { + match expected { + Some(Expected::ParameterName) => "parameter name", + Some(Expected::ArgumentName) => "argument name", + Some(Expected::Identifier) => "identifier", + Some(Expected::BindingPattern) => "binding pattern", + None => "pattern", + } + } +} + +const WHILE_PARSING_OR_MSG: &str = "while parsing this or-pattern starting here"; + +/// Whether or not to recover a `,` when parsing or-patterns. +#[derive(PartialEq, Copy, Clone)] +pub enum RecoverComma { + Yes, + No, +} + +/// Whether or not to recover a `:` when parsing patterns that were meant to be paths. +#[derive(PartialEq, Copy, Clone)] +pub enum RecoverColon { + Yes, + No, +} + +/// Whether or not to recover a `a, b` when parsing patterns as `(a, b)` or that *and* `a | b`. +#[derive(PartialEq, Copy, Clone)] +pub enum CommaRecoveryMode { + LikelyTuple, + EitherTupleOrPipe, +} + +/// The result of `eat_or_separator`. We want to distinguish which case we are in to avoid +/// emitting duplicate diagnostics. +#[derive(Debug, Clone, Copy)] +enum EatOrResult { + /// We recovered from a trailing vert. + TrailingVert, + /// We ate an `|` (or `||` and recovered). + AteOr, + /// We did not eat anything (i.e. the current token is not `|` or `||`). + None, +} + +/// The syntax location of a given pattern. Used for diagnostics. +#[derive(Clone, Copy)] +pub enum PatternLocation { + LetBinding, + FunctionParameter, +} + +impl<'a> Parser<'a> { + /// Parses a pattern. + /// + /// Corresponds to `pat<no_top_alt>` in RFC 2535 and does not admit or-patterns + /// at the top level. Used when parsing the parameters of lambda expressions, + /// functions, function pointers, and `pat` macro fragments. + pub fn parse_pat_no_top_alt( + &mut self, + expected: Option<Expected>, + syntax_loc: Option<PatternLocation>, + ) -> PResult<'a, P<Pat>> { + self.parse_pat_with_range_pat(true, expected, syntax_loc) + } + + /// Parses a pattern. + /// + /// Corresponds to `top_pat` in RFC 2535 and allows or-pattern at the top level. + /// Used for parsing patterns in all cases when `pat<no_top_alt>` is not used. + /// + /// Note that after the FCP in <https://github.com/rust-lang/rust/issues/81415>, + /// a leading vert is allowed in nested or-patterns, too. This allows us to + /// simplify the grammar somewhat. + pub fn parse_pat_allow_top_alt( + &mut self, + expected: Option<Expected>, + rc: RecoverComma, + ra: RecoverColon, + rt: CommaRecoveryMode, + ) -> PResult<'a, P<Pat>> { + self.parse_pat_allow_top_alt_inner(expected, rc, ra, rt, None).map(|(pat, _)| pat) + } + + /// Returns the pattern and a bool indicating whether we recovered from a trailing vert (true = + /// recovered). + fn parse_pat_allow_top_alt_inner( + &mut self, + expected: Option<Expected>, + rc: RecoverComma, + ra: RecoverColon, + rt: CommaRecoveryMode, + syntax_loc: Option<PatternLocation>, + ) -> PResult<'a, (P<Pat>, bool)> { + // Keep track of whether we recovered from a trailing vert so that we can avoid duplicated + // suggestions (which bothers rustfix). + // + // Allow a '|' before the pats (RFCs 1925, 2530, and 2535). + let (leading_vert_span, mut trailing_vert) = match self.eat_or_separator(None) { + EatOrResult::AteOr => (Some(self.prev_token.span), false), + EatOrResult::TrailingVert => (None, true), + EatOrResult::None => (None, false), + }; + + // Parse the first pattern (`p_0`). + let mut first_pat = match self.parse_pat_no_top_alt(expected, syntax_loc) { + Ok(pat) => pat, + Err(err) + if self.token.is_reserved_ident() + && !self.token.is_keyword(kw::In) + && !self.token.is_keyword(kw::If) => + { + err.emit(); + self.bump(); + self.mk_pat(self.token.span, PatKind::Wild) + } + Err(err) => return Err(err), + }; + if rc == RecoverComma::Yes && !first_pat.could_be_never_pattern() { + self.maybe_recover_unexpected_comma(first_pat.span, rt)?; + } + + // If the next token is not a `|`, + // this is not an or-pattern and we should exit here. + if !self.check(&token::BinOp(token::Or)) && self.token != token::OrOr { + // If we parsed a leading `|` which should be gated, + // then we should really gate the leading `|`. + // This complicated procedure is done purely for diagnostics UX. + + // Check if the user wrote `foo:bar` instead of `foo::bar`. + if ra == RecoverColon::Yes { + first_pat = self.maybe_recover_colon_colon_in_pat_typo(first_pat, expected); + } + + if let Some(leading_vert_span) = leading_vert_span { + // If there was a leading vert, treat this as an or-pattern. This improves + // diagnostics. + let span = leading_vert_span.to(self.prev_token.span); + return Ok((self.mk_pat(span, PatKind::Or(thin_vec![first_pat])), trailing_vert)); + } + + return Ok((first_pat, trailing_vert)); + } + + // Parse the patterns `p_1 | ... | p_n` where `n > 0`. + let lo = leading_vert_span.unwrap_or(first_pat.span); + let mut pats = thin_vec![first_pat]; + loop { + match self.eat_or_separator(Some(lo)) { + EatOrResult::AteOr => {} + EatOrResult::None => break, + EatOrResult::TrailingVert => { + trailing_vert = true; + break; + } + } + let pat = self.parse_pat_no_top_alt(expected, syntax_loc).map_err(|mut err| { + err.span_label(lo, WHILE_PARSING_OR_MSG); + err + })?; + if rc == RecoverComma::Yes && !pat.could_be_never_pattern() { + self.maybe_recover_unexpected_comma(pat.span, rt)?; + } + pats.push(pat); + } + let or_pattern_span = lo.to(self.prev_token.span); + + Ok((self.mk_pat(or_pattern_span, PatKind::Or(pats)), trailing_vert)) + } + + /// Parse a pattern and (maybe) a `Colon` in positions where a pattern may be followed by a + /// type annotation (e.g. for `let` bindings or `fn` params). + /// + /// Generally, this corresponds to `pat_no_top_alt` followed by an optional `Colon`. It will + /// eat the `Colon` token if one is present. + /// + /// The return value represents the parsed pattern and `true` if a `Colon` was parsed (`false` + /// otherwise). + pub(super) fn parse_pat_before_ty( + &mut self, + expected: Option<Expected>, + rc: RecoverComma, + syntax_loc: PatternLocation, + ) -> PResult<'a, (P<Pat>, bool)> { + // We use `parse_pat_allow_top_alt` regardless of whether we actually want top-level + // or-patterns so that we can detect when a user tries to use it. This allows us to print a + // better error message. + let (pat, trailing_vert) = self.parse_pat_allow_top_alt_inner( + expected, + rc, + RecoverColon::No, + CommaRecoveryMode::LikelyTuple, + Some(syntax_loc), + )?; + let colon = self.eat(&token::Colon); + + if let PatKind::Or(pats) = &pat.kind { + let span = pat.span; + let pat = pprust::pat_to_string(&pat); + let sub = if pats.len() == 1 { + Some(TopLevelOrPatternNotAllowedSugg::RemoveLeadingVert { span, pat }) + } else { + Some(TopLevelOrPatternNotAllowedSugg::WrapInParens { span, pat }) + }; + + let err = self.dcx().create_err(match syntax_loc { + PatternLocation::LetBinding => { + TopLevelOrPatternNotAllowed::LetBinding { span, sub } + } + PatternLocation::FunctionParameter => { + TopLevelOrPatternNotAllowed::FunctionParameter { span, sub } + } + }); + if trailing_vert { + err.delay_as_bug(); + } else { + err.emit(); + } + } + + Ok((pat, colon)) + } + + /// Parse the pattern for a function or function pointer parameter, followed by a colon. + /// + /// The return value represents the parsed pattern and `true` if a `Colon` was parsed (`false` + /// otherwise). + pub(super) fn parse_fn_param_pat_colon(&mut self) -> PResult<'a, (P<Pat>, bool)> { + // In order to get good UX, we first recover in the case of a leading vert for an illegal + // top-level or-pat. Normally, this means recovering both `|` and `||`, but in this case, + // a leading `||` probably doesn't indicate an or-pattern attempt, so we handle that + // separately. + if let token::OrOr = self.token.kind { + self.dcx().emit_err(UnexpectedVertVertBeforeFunctionParam { span: self.token.span }); + self.bump(); + } + + self.parse_pat_before_ty( + Some(Expected::ParameterName), + RecoverComma::No, + PatternLocation::FunctionParameter, + ) + } + + /// Eat the or-pattern `|` separator. + /// If instead a `||` token is encountered, recover and pretend we parsed `|`. + fn eat_or_separator(&mut self, lo: Option<Span>) -> EatOrResult { + if self.recover_trailing_vert(lo) { + EatOrResult::TrailingVert + } else if matches!(self.token.kind, token::OrOr) { + // Found `||`; Recover and pretend we parsed `|`. + self.dcx().emit_err(UnexpectedVertVertInPattern { span: self.token.span, start: lo }); + self.bump(); + EatOrResult::AteOr + } else if self.eat(&token::BinOp(token::Or)) { + EatOrResult::AteOr + } else { + EatOrResult::None + } + } + + /// Recover if `|` or `||` is the current token and we have one of the + /// tokens `=>`, `if`, `=`, `:`, `;`, `,`, `]`, `)`, or `}` ahead of us. + /// + /// These tokens all indicate that we reached the end of the or-pattern + /// list and can now reliably say that the `|` was an illegal trailing vert. + /// Note that there are more tokens such as `@` for which we know that the `|` + /// is an illegal parse. However, the user's intent is less clear in that case. + fn recover_trailing_vert(&mut self, lo: Option<Span>) -> bool { + let is_end_ahead = self.look_ahead(1, |token| { + matches!( + &token.uninterpolate().kind, + token::FatArrow // e.g. `a | => 0,`. + | token::Ident(kw::If, token::IdentIsRaw::No) // e.g. `a | if expr`. + | token::Eq // e.g. `let a | = 0`. + | token::Semi // e.g. `let a |;`. + | token::Colon // e.g. `let a | :`. + | token::Comma // e.g. `let (a |,)`. + | token::CloseDelim(Delimiter::Bracket) // e.g. `let [a | ]`. + | token::CloseDelim(Delimiter::Parenthesis) // e.g. `let (a | )`. + | token::CloseDelim(Delimiter::Brace) // e.g. `let A { f: a | }`. + ) + }); + match (is_end_ahead, &self.token.kind) { + (true, token::BinOp(token::Or) | token::OrOr) => { + // A `|` or possibly `||` token shouldn't be here. Ban it. + self.dcx().emit_err(TrailingVertNotAllowed { + span: self.token.span, + start: lo, + token: self.token.clone(), + note_double_vert: matches!(self.token.kind, token::OrOr).then_some(()), + }); + self.bump(); + true + } + _ => false, + } + } + + /// Ensures that the last parsed pattern (or pattern range bound) is not followed by a method call or an operator. + /// + /// `is_end_bound` indicates whether the last parsed thing was the end bound of a range pattern (see [`parse_pat_range_end`](Self::parse_pat_range_end)) + /// in order to say "expected a pattern range bound" instead of "expected a pattern"; + /// ```text + /// 0..=1 + 2 + /// ^^^^^ + /// ``` + /// Only the end bound is spanned, and this function have no idea if there were a `..=` before `pat_span`, hence the parameter. + #[must_use = "the pattern must be discarded as `PatKind::Err` if this function returns Some"] + fn maybe_recover_trailing_expr( + &mut self, + pat_span: Span, + is_end_bound: bool, + ) -> Option<ErrorGuaranteed> { + if self.prev_token.is_keyword(kw::Underscore) || !self.may_recover() { + // Don't recover anything after an `_` or if recovery is disabled. + return None; + } + + // Check for `.hello()`, but allow `.Hello()` to be recovered as `, Hello()` in `parse_seq_to_before_tokens()`. + let has_trailing_method = self.check_noexpect(&token::Dot) + && self.look_ahead(1, |tok| { + tok.ident() + .and_then(|(ident, _)| ident.name.as_str().chars().next()) + .is_some_and(char::is_lowercase) + }) + && self.look_ahead(2, |tok| tok.kind == token::OpenDelim(Delimiter::Parenthesis)); + + // Check for operators. + // `|` is excluded as it is used in pattern alternatives and lambdas, + // `?` is included for error propagation, + // `[` is included for indexing operations, + // `[]` is excluded as `a[]` isn't an expression and should be recovered as `a, []` (cf. `tests/ui/parser/pat-lt-bracket-7.rs`) + let has_trailing_operator = matches!(self.token.kind, token::BinOp(op) if op != BinOpToken::Or) + || self.token.kind == token::Question + || (self.token.kind == token::OpenDelim(Delimiter::Bracket) + && self.look_ahead(1, |tok| tok.kind != token::CloseDelim(Delimiter::Bracket))); + + if !has_trailing_method && !has_trailing_operator { + // Nothing to recover here. + return None; + } + + // Let's try to parse an expression to emit a better diagnostic. + let mut snapshot = self.create_snapshot_for_diagnostic(); + snapshot.restrictions.insert(Restrictions::IS_PAT); + + // Parse `?`, `.f`, `(arg0, arg1, ...)` or `[expr]` until they've all been eaten. + if let Ok(expr) = snapshot + .parse_expr_dot_or_call_with( + self.mk_expr(pat_span, ExprKind::Dummy), // equivalent to transforming the parsed pattern into an `Expr` + pat_span, + AttrVec::new(), + ) + .map_err(|err| err.cancel()) + { + let non_assoc_span = expr.span; + + // Parse an associative expression such as `+ expr`, `% expr`, ... + // Assignements, ranges and `|` are disabled by [`Restrictions::IS_PAT`]. + if let Ok(expr) = + snapshot.parse_expr_assoc_with(0, expr.into()).map_err(|err| err.cancel()) + { + // We got a valid expression. + self.restore_snapshot(snapshot); + self.restrictions.remove(Restrictions::IS_PAT); + + let is_bound = is_end_bound + // is_start_bound: either `..` or `)..` + || self.token.is_range_separator() + || self.token.kind == token::CloseDelim(Delimiter::Parenthesis) + && self.look_ahead(1, Token::is_range_separator); + + // Check that `parse_expr_assoc_with` didn't eat a rhs. + let is_method_call = has_trailing_method && non_assoc_span == expr.span; + + return Some(self.dcx().emit_err(UnexpectedExpressionInPattern { + span: expr.span, + is_bound, + is_method_call, + })); + } + } + + // We got a trailing method/operator, but we couldn't parse an expression. + None + } + + /// Parses a pattern, with a setting whether modern range patterns (e.g., `a..=b`, `a..b` are + /// allowed). + fn parse_pat_with_range_pat( + &mut self, + allow_range_pat: bool, + expected: Option<Expected>, + syntax_loc: Option<PatternLocation>, + ) -> PResult<'a, P<Pat>> { + maybe_recover_from_interpolated_ty_qpath!(self, true); + maybe_whole!(self, NtPat, |pat| pat); + + let mut lo = self.token.span; + + if self.token.is_keyword(kw::Let) && self.look_ahead(1, |tok| tok.can_begin_pattern()) { + self.bump(); + self.dcx().emit_err(RemoveLet { span: lo }); + lo = self.token.span; + } + + let pat = if self.check(&token::BinOp(token::And)) || self.token.kind == token::AndAnd { + self.parse_pat_deref(expected)? + } else if self.check(&token::OpenDelim(Delimiter::Parenthesis)) { + self.parse_pat_tuple_or_parens()? + } else if self.check(&token::OpenDelim(Delimiter::Bracket)) { + // Parse `[pat, pat,...]` as a slice pattern. + let (pats, _) = self.parse_delim_comma_seq(Delimiter::Bracket, |p| { + p.parse_pat_allow_top_alt( + None, + RecoverComma::No, + RecoverColon::No, + CommaRecoveryMode::EitherTupleOrPipe, + ) + })?; + PatKind::Slice(pats) + } else if self.check(&token::DotDot) && !self.is_pat_range_end_start(1) { + // A rest pattern `..`. + self.bump(); // `..` + PatKind::Rest + } else if self.check(&token::DotDotDot) && !self.is_pat_range_end_start(1) { + self.recover_dotdotdot_rest_pat(lo) + } else if let Some(form) = self.parse_range_end() { + self.parse_pat_range_to(form)? // `..=X`, `...X`, or `..X`. + } else if self.eat(&token::Not) { + // Parse `!` + self.psess.gated_spans.gate(sym::never_patterns, self.prev_token.span); + PatKind::Never + } else if self.eat_keyword(kw::Underscore) { + // Parse `_` + PatKind::Wild + } else if self.eat_keyword(kw::Mut) { + self.parse_pat_ident_mut()? + } else if self.eat_keyword(kw::Ref) { + if self.check_keyword(kw::Box) { + // Suggest `box ref`. + let span = self.prev_token.span.to(self.token.span); + self.bump(); + self.dcx().emit_err(SwitchRefBoxOrder { span }); + } + // Parse ref ident @ pat / ref mut ident @ pat + let mutbl = self.parse_mutability(); + self.parse_pat_ident(BindingMode(ByRef::Yes(mutbl), Mutability::Not), syntax_loc)? + } else if self.eat_keyword(kw::Box) { + self.parse_pat_box()? + } else if self.check_inline_const(0) { + // Parse `const pat` + let const_expr = self.parse_const_block(lo.to(self.token.span), true)?; + + if let Some(re) = self.parse_range_end() { + self.parse_pat_range_begin_with(const_expr, re)? + } else { + PatKind::Lit(const_expr) + } + } else if self.is_builtin() { + self.parse_pat_builtin()? + } + // Don't eagerly error on semantically invalid tokens when matching + // declarative macros, as the input to those doesn't have to be + // semantically valid. For attribute/derive proc macros this is not the + // case, so doing the recovery for them is fine. + else if self.can_be_ident_pat() + || (self.is_lit_bad_ident().is_some() && self.may_recover()) + { + // Parse `ident @ pat` + // This can give false positives and parse nullary enums, + // they are dealt with later in resolve. + self.parse_pat_ident(BindingMode::NONE, syntax_loc)? + } else if self.is_start_of_pat_with_path() { + // Parse pattern starting with a path + let (qself, path) = if self.eat_lt() { + // Parse a qualified path + let (qself, path) = self.parse_qpath(PathStyle::Pat)?; + (Some(qself), path) + } else { + // Parse an unqualified path + (None, self.parse_path(PathStyle::Pat)?) + }; + let span = lo.to(self.prev_token.span); + + if qself.is_none() && self.check(&token::Not) { + self.parse_pat_mac_invoc(path)? + } else if let Some(form) = self.parse_range_end() { + let begin = self.mk_expr(span, ExprKind::Path(qself, path)); + self.parse_pat_range_begin_with(begin, form)? + } else if self.check(&token::OpenDelim(Delimiter::Brace)) { + self.parse_pat_struct(qself, path)? + } else if self.check(&token::OpenDelim(Delimiter::Parenthesis)) { + self.parse_pat_tuple_struct(qself, path)? + } else { + match self.maybe_recover_trailing_expr(span, false) { + Some(guar) => PatKind::Err(guar), + None => PatKind::Path(qself, path), + } + } + } else if let token::Lifetime(lt) = self.token.kind + // In pattern position, we're totally fine with using "next token isn't colon" + // as a heuristic. We could probably just always try to recover if it's a lifetime, + // because we never have `'a: label {}` in a pattern position anyways, but it does + // keep us from suggesting something like `let 'a: Ty = ..` => `let 'a': Ty = ..` + && could_be_unclosed_char_literal(Ident::with_dummy_span(lt)) + && !self.look_ahead(1, |token| matches!(token.kind, token::Colon)) + { + // Recover a `'a` as a `'a'` literal + let lt = self.expect_lifetime(); + let (lit, _) = + self.recover_unclosed_char(lt.ident, Parser::mk_token_lit_char, |self_| { + let expected = Expected::to_string_or_fallback(expected); + let msg = format!( + "expected {}, found {}", + expected, + super::token_descr(&self_.token) + ); + + self_ + .dcx() + .struct_span_err(self_.token.span, msg) + .with_span_label(self_.token.span, format!("expected {expected}")) + }); + PatKind::Lit(self.mk_expr(lo, ExprKind::Lit(lit))) + } else { + // Try to parse everything else as literal with optional minus + match self.parse_literal_maybe_minus() { + Ok(begin) => { + let begin = match self.maybe_recover_trailing_expr(begin.span, false) { + Some(guar) => self.mk_expr_err(begin.span, guar), + None => begin, + }; + + match self.parse_range_end() { + Some(form) => self.parse_pat_range_begin_with(begin, form)?, + None => PatKind::Lit(begin), + } + } + Err(err) => return self.fatal_unexpected_non_pat(err, expected), + } + }; + + let pat = self.mk_pat(lo.to(self.prev_token.span), pat); + let pat = self.maybe_recover_from_bad_qpath(pat)?; + let pat = self.recover_intersection_pat(pat)?; + + if !allow_range_pat { + self.ban_pat_range_if_ambiguous(&pat) + } + + Ok(pat) + } + + /// Recover from a typoed `...` pattern that was encountered + /// Ref: Issue #70388 + fn recover_dotdotdot_rest_pat(&mut self, lo: Span) -> PatKind { + // A typoed rest pattern `...`. + self.bump(); // `...` + + // The user probably mistook `...` for a rest pattern `..`. + self.dcx().emit_err(DotDotDotRestPattern { span: lo }); + PatKind::Rest + } + + /// Try to recover the more general form `intersect ::= $pat_lhs @ $pat_rhs`. + /// + /// Allowed binding patterns generated by `binding ::= ref? mut? $ident @ $pat_rhs` + /// should already have been parsed by now at this point, + /// if the next token is `@` then we can try to parse the more general form. + /// + /// Consult `parse_pat_ident` for the `binding` grammar. + /// + /// The notion of intersection patterns are found in + /// e.g. [F#][and] where they are called AND-patterns. + /// + /// [and]: https://docs.microsoft.com/en-us/dotnet/fsharp/language-reference/pattern-matching + fn recover_intersection_pat(&mut self, lhs: P<Pat>) -> PResult<'a, P<Pat>> { + if self.token.kind != token::At { + // Next token is not `@` so it's not going to be an intersection pattern. + return Ok(lhs); + } + + // At this point we attempt to parse `@ $pat_rhs` and emit an error. + self.bump(); // `@` + let mut rhs = self.parse_pat_no_top_alt(None, None)?; + let whole_span = lhs.span.to(rhs.span); + + if let PatKind::Ident(_, _, sub @ None) = &mut rhs.kind { + // The user inverted the order, so help them fix that. + let lhs_span = lhs.span; + // Move the LHS into the RHS as a subpattern. + // The RHS is now the full pattern. + *sub = Some(lhs); + + self.dcx().emit_err(PatternOnWrongSideOfAt { + whole_span, + whole_pat: pprust::pat_to_string(&rhs), + pattern: lhs_span, + binding: rhs.span, + }); + } else { + // The special case above doesn't apply so we may have e.g. `A(x) @ B(y)`. + rhs.kind = PatKind::Wild; + self.dcx().emit_err(ExpectedBindingLeftOfAt { + whole_span, + lhs: lhs.span, + rhs: rhs.span, + }); + } + + rhs.span = whole_span; + Ok(rhs) + } + + /// Ban a range pattern if it has an ambiguous interpretation. + fn ban_pat_range_if_ambiguous(&self, pat: &Pat) { + match pat.kind { + PatKind::Range( + .., + Spanned { node: RangeEnd::Included(RangeSyntax::DotDotDot), .. }, + ) => return, + PatKind::Range(..) => {} + _ => return, + } + + self.dcx() + .emit_err(AmbiguousRangePattern { span: pat.span, pat: pprust::pat_to_string(pat) }); + } + + /// Parse `&pat` / `&mut pat`. + fn parse_pat_deref(&mut self, expected: Option<Expected>) -> PResult<'a, PatKind> { + self.expect_and()?; + if let token::Lifetime(name) = self.token.kind { + self.bump(); // `'a` + + self.dcx() + .emit_err(UnexpectedLifetimeInPattern { span: self.prev_token.span, symbol: name }); + } + + let mutbl = self.parse_mutability(); + let subpat = self.parse_pat_with_range_pat(false, expected, None)?; + Ok(PatKind::Ref(subpat, mutbl)) + } + + /// Parse a tuple or parenthesis pattern. + fn parse_pat_tuple_or_parens(&mut self) -> PResult<'a, PatKind> { + let open_paren = self.token.span; + + let (fields, trailing_comma) = self.parse_paren_comma_seq(|p| { + p.parse_pat_allow_top_alt( + None, + RecoverComma::No, + RecoverColon::No, + CommaRecoveryMode::LikelyTuple, + ) + })?; + + // Here, `(pat,)` is a tuple pattern. + // For backward compatibility, `(..)` is a tuple pattern as well. + let paren_pattern = + fields.len() == 1 && !(matches!(trailing_comma, Trailing::Yes) || fields[0].is_rest()); + if paren_pattern { + let pat = fields.into_iter().next().unwrap(); + let close_paren = self.prev_token.span; + + match &pat.kind { + // recover ranges with parentheses around the `(start)..` + PatKind::Lit(begin) + if self.may_recover() + && let Some(form) = self.parse_range_end() => + { + self.dcx().emit_err(UnexpectedParenInRangePat { + span: vec![open_paren, close_paren], + sugg: UnexpectedParenInRangePatSugg { + start_span: open_paren, + end_span: close_paren, + }, + }); + + self.parse_pat_range_begin_with(begin.clone(), form) + } + // recover ranges with parentheses around the `(start)..` + PatKind::Err(guar) + if self.may_recover() + && let Some(form) = self.parse_range_end() => + { + self.dcx().emit_err(UnexpectedParenInRangePat { + span: vec![open_paren, close_paren], + sugg: UnexpectedParenInRangePatSugg { + start_span: open_paren, + end_span: close_paren, + }, + }); + + self.parse_pat_range_begin_with(self.mk_expr_err(pat.span, *guar), form) + } + + // (pat) with optional parentheses + _ => Ok(PatKind::Paren(pat)), + } + } else { + Ok(PatKind::Tuple(fields)) + } + } + + /// Parse a mutable binding with the `mut` token already eaten. + fn parse_pat_ident_mut(&mut self) -> PResult<'a, PatKind> { + let mut_span = self.prev_token.span; + + self.recover_additional_muts(); + + let byref = self.parse_byref(); + + self.recover_additional_muts(); + + // Make sure we don't allow e.g. `let mut $p;` where `$p:pat`. + if let token::Interpolated(nt) = &self.token.kind { + if let token::NtPat(..) = &**nt { + self.expected_ident_found_err().emit(); + } + } + + // Parse the pattern we hope to be an identifier. + let mut pat = self.parse_pat_no_top_alt(Some(Expected::Identifier), None)?; + + // If we don't have `mut $ident (@ pat)?`, error. + if let PatKind::Ident(BindingMode(br @ ByRef::No, m @ Mutability::Not), ..) = &mut pat.kind + { + // Don't recurse into the subpattern. + // `mut` on the outer binding doesn't affect the inner bindings. + *br = byref; + *m = Mutability::Mut; + } else { + // Add `mut` to any binding in the parsed pattern. + let changed_any_binding = Self::make_all_value_bindings_mutable(&mut pat); + self.ban_mut_general_pat(mut_span, &pat, changed_any_binding); + } + + if matches!(pat.kind, PatKind::Ident(BindingMode(ByRef::Yes(_), Mutability::Mut), ..)) { + self.psess.gated_spans.gate(sym::mut_ref, pat.span); + } + Ok(pat.into_inner().kind) + } + + /// Turn all by-value immutable bindings in a pattern into mutable bindings. + /// Returns `true` if any change was made. + fn make_all_value_bindings_mutable(pat: &mut P<Pat>) -> bool { + struct AddMut(bool); + impl MutVisitor for AddMut { + fn visit_pat(&mut self, pat: &mut P<Pat>) { + if let PatKind::Ident(BindingMode(ByRef::No, m @ Mutability::Not), ..) = + &mut pat.kind + { + self.0 = true; + *m = Mutability::Mut; + } + noop_visit_pat(pat, self); + } + } + + let mut add_mut = AddMut(false); + add_mut.visit_pat(pat); + add_mut.0 + } + + /// Error on `mut $pat` where `$pat` is not an ident. + fn ban_mut_general_pat(&self, lo: Span, pat: &Pat, changed_any_binding: bool) { + self.dcx().emit_err(if changed_any_binding { + InvalidMutInPattern::NestedIdent { + span: lo.to(pat.span), + pat: pprust::pat_to_string(pat), + } + } else { + InvalidMutInPattern::NonIdent { span: lo.until(pat.span) } + }); + } + + /// Eat any extraneous `mut`s and error + recover if we ate any. + fn recover_additional_muts(&mut self) { + let lo = self.token.span; + while self.eat_keyword(kw::Mut) {} + if lo == self.token.span { + return; + } + + self.dcx().emit_err(RepeatedMutInPattern { span: lo.to(self.prev_token.span) }); + } + + /// Parse macro invocation + fn parse_pat_mac_invoc(&mut self, path: Path) -> PResult<'a, PatKind> { + self.bump(); + let args = self.parse_delim_args()?; + let mac = P(MacCall { path, args }); + Ok(PatKind::MacCall(mac)) + } + + fn fatal_unexpected_non_pat( + &mut self, + err: Diag<'a>, + expected: Option<Expected>, + ) -> PResult<'a, P<Pat>> { + err.cancel(); + + let expected = Expected::to_string_or_fallback(expected); + let msg = format!("expected {}, found {}", expected, super::token_descr(&self.token)); + + let mut err = self.dcx().struct_span_err(self.token.span, msg); + err.span_label(self.token.span, format!("expected {expected}")); + + let sp = self.psess.source_map().start_point(self.token.span); + if let Some(sp) = self.psess.ambiguous_block_expr_parse.borrow().get(&sp) { + err.subdiagnostic(self.dcx(), ExprParenthesesNeeded::surrounding(*sp)); + } + + Err(err) + } + + /// Parses the range pattern end form `".." | "..." | "..=" ;`. + fn parse_range_end(&mut self) -> Option<Spanned<RangeEnd>> { + let re = if self.eat(&token::DotDotDot) { + RangeEnd::Included(RangeSyntax::DotDotDot) + } else if self.eat(&token::DotDotEq) { + RangeEnd::Included(RangeSyntax::DotDotEq) + } else if self.eat(&token::DotDot) { + RangeEnd::Excluded + } else { + return None; + }; + Some(respan(self.prev_token.span, re)) + } + + /// Parse a range pattern `$begin $form $end?` where `$form = ".." | "..." | "..=" ;`. + /// `$begin $form` has already been parsed. + fn parse_pat_range_begin_with( + &mut self, + begin: P<Expr>, + re: Spanned<RangeEnd>, + ) -> PResult<'a, PatKind> { + let end = if self.is_pat_range_end_start(0) { + // Parsing e.g. `X..=Y`. + Some(self.parse_pat_range_end()?) + } else { + // Parsing e.g. `X..`. + if let RangeEnd::Included(_) = re.node { + // FIXME(Centril): Consider semantic errors instead in `ast_validation`. + self.inclusive_range_with_incorrect_end(); + } + None + }; + Ok(PatKind::Range(Some(begin), end, re)) + } + + pub(super) fn inclusive_range_with_incorrect_end(&mut self) -> ErrorGuaranteed { + let tok = &self.token; + let span = self.prev_token.span; + // If the user typed "..==" instead of "..=", we want to give them + // a specific error message telling them to use "..=". + // If they typed "..=>", suggest they use ".. =>". + // Otherwise, we assume that they meant to type a half open exclusive + // range and give them an error telling them to do that instead. + let no_space = tok.span.lo() == span.hi(); + match tok.kind { + token::Eq if no_space => { + let span_with_eq = span.to(tok.span); + + // Ensure the user doesn't receive unhelpful unexpected token errors + self.bump(); + if self.is_pat_range_end_start(0) { + let _ = self.parse_pat_range_end().map_err(|e| e.cancel()); + } + + self.dcx().emit_err(InclusiveRangeExtraEquals { span: span_with_eq }) + } + token::Gt if no_space => { + let after_pat = span.with_hi(span.hi() - rustc_span::BytePos(1)).shrink_to_hi(); + self.dcx().emit_err(InclusiveRangeMatchArrow { span, arrow: tok.span, after_pat }) + } + _ => self.dcx().emit_err(InclusiveRangeNoEnd { span }), + } + } + + /// Parse a range-to pattern, `..X` or `..=X` where `X` remains to be parsed. + /// + /// The form `...X` is prohibited to reduce confusion with the potential + /// expression syntax `...expr` for splatting in expressions. + fn parse_pat_range_to(&mut self, mut re: Spanned<RangeEnd>) -> PResult<'a, PatKind> { + let end = self.parse_pat_range_end()?; + if let RangeEnd::Included(syn @ RangeSyntax::DotDotDot) = &mut re.node { + *syn = RangeSyntax::DotDotEq; + self.dcx().emit_err(DotDotDotRangeToPatternNotAllowed { span: re.span }); + } + Ok(PatKind::Range(None, Some(end), re)) + } + + /// Is the token `dist` away from the current suitable as the start of a range patterns end? + fn is_pat_range_end_start(&self, dist: usize) -> bool { + self.check_inline_const(dist) + || self.look_ahead(dist, |t| { + t.is_path_start() // e.g. `MY_CONST`; + || t.kind == token::Dot // e.g. `.5` for recovery; + || t.can_begin_literal_maybe_minus() // e.g. `42`. + || t.is_whole_expr() + || t.is_lifetime() // recover `'a` instead of `'a'` + || (self.may_recover() // recover leading `(` + && t.kind == token::OpenDelim(Delimiter::Parenthesis) + && self.look_ahead(dist + 1, |t| t.kind != token::OpenDelim(Delimiter::Parenthesis)) + && self.is_pat_range_end_start(dist + 1)) + }) + } + + /// Parse a range pattern end bound + fn parse_pat_range_end(&mut self) -> PResult<'a, P<Expr>> { + // recover leading `(` + let open_paren = (self.may_recover() + && self.eat_noexpect(&token::OpenDelim(Delimiter::Parenthesis))) + .then_some(self.prev_token.span); + + let bound = if self.check_inline_const(0) { + self.parse_const_block(self.token.span, true) + } else if self.check_path() { + let lo = self.token.span; + let (qself, path) = if self.eat_lt() { + // Parse a qualified path + let (qself, path) = self.parse_qpath(PathStyle::Pat)?; + (Some(qself), path) + } else { + // Parse an unqualified path + (None, self.parse_path(PathStyle::Pat)?) + }; + let hi = self.prev_token.span; + Ok(self.mk_expr(lo.to(hi), ExprKind::Path(qself, path))) + } else { + self.parse_literal_maybe_minus() + }?; + + let recovered = self.maybe_recover_trailing_expr(bound.span, true); + + // recover trailing `)` + if let Some(open_paren) = open_paren { + self.expect(&token::CloseDelim(Delimiter::Parenthesis))?; + + self.dcx().emit_err(UnexpectedParenInRangePat { + span: vec![open_paren, self.prev_token.span], + sugg: UnexpectedParenInRangePatSugg { + start_span: open_paren, + end_span: self.prev_token.span, + }, + }); + } + + Ok(match recovered { + Some(guar) => self.mk_expr_err(bound.span, guar), + None => bound, + }) + } + + /// Is this the start of a pattern beginning with a path? + fn is_start_of_pat_with_path(&mut self) -> bool { + self.check_path() + // Just for recovery (see `can_be_ident`). + || self.token.is_ident() && !self.token.is_bool_lit() && !self.token.is_keyword(kw::In) + } + + /// Would `parse_pat_ident` be appropriate here? + fn can_be_ident_pat(&mut self) -> bool { + self.check_ident() + && !self.token.is_bool_lit() // Avoid `true` or `false` as a binding as it is a literal. + && !self.token.is_path_segment_keyword() // Avoid e.g. `Self` as it is a path. + // Avoid `in`. Due to recovery in the list parser this messes with `for ( $pat in $expr )`. + && !self.token.is_keyword(kw::In) + // Try to do something more complex? + && self.look_ahead(1, |t| !matches!(t.kind, token::OpenDelim(Delimiter::Parenthesis) // A tuple struct pattern. + | token::OpenDelim(Delimiter::Brace) // A struct pattern. + | token::DotDotDot | token::DotDotEq | token::DotDot // A range pattern. + | token::PathSep // A tuple / struct variant pattern. + | token::Not)) // A macro expanding to a pattern. + } + + /// Parses `ident` or `ident @ pat`. + /// Used by the copy foo and ref foo patterns to give a good + /// error message when parsing mistakes like `ref foo(a, b)`. + fn parse_pat_ident( + &mut self, + binding_annotation: BindingMode, + syntax_loc: Option<PatternLocation>, + ) -> PResult<'a, PatKind> { + let ident = self.parse_ident_common(false)?; + + if self.may_recover() + && !matches!(syntax_loc, Some(PatternLocation::FunctionParameter)) + && self.check_noexpect(&token::Lt) + && self.look_ahead(1, |t| t.can_begin_type()) + { + return Err(self.dcx().create_err(GenericArgsInPatRequireTurbofishSyntax { + span: self.token.span, + suggest_turbofish: self.token.span.shrink_to_lo(), + })); + } + + let sub = if self.eat(&token::At) { + Some(self.parse_pat_no_top_alt(Some(Expected::BindingPattern), None)?) + } else { + None + }; + + // Just to be friendly, if they write something like `ref Some(i)`, + // we end up here with `(` as the current token. + // This shortly leads to a parse error. Note that if there is no explicit + // binding mode then we do not end up here, because the lookahead + // will direct us over to `parse_enum_variant()`. + if self.token == token::OpenDelim(Delimiter::Parenthesis) { + return Err(self + .dcx() + .create_err(EnumPatternInsteadOfIdentifier { span: self.prev_token.span })); + } + + // Check for method calls after the `ident`, + // but not `ident @ subpat` as `subpat` was already checked and `ident` continues with `@`. + + let pat = if sub.is_none() + && let Some(guar) = self.maybe_recover_trailing_expr(ident.span, false) + { + PatKind::Err(guar) + } else { + PatKind::Ident(binding_annotation, ident, sub) + }; + Ok(pat) + } + + /// Parse a struct ("record") pattern (e.g. `Foo { ... }` or `Foo::Bar { ... }`). + fn parse_pat_struct(&mut self, qself: Option<P<QSelf>>, path: Path) -> PResult<'a, PatKind> { + if qself.is_some() { + // Feature gate the use of qualified paths in patterns + self.psess.gated_spans.gate(sym::more_qualified_paths, path.span); + } + self.bump(); + let (fields, etc) = self.parse_pat_fields().unwrap_or_else(|mut e| { + e.span_label(path.span, "while parsing the fields for this pattern"); + e.emit(); + self.recover_stmt(); + // When recovering, pretend we had `Foo { .. }`, to avoid cascading errors. + (ThinVec::new(), PatFieldsRest::Rest) + }); + self.bump(); + Ok(PatKind::Struct(qself, path, fields, etc)) + } + + /// Parse tuple struct or tuple variant pattern (e.g. `Foo(...)` or `Foo::Bar(...)`). + fn parse_pat_tuple_struct( + &mut self, + qself: Option<P<QSelf>>, + path: Path, + ) -> PResult<'a, PatKind> { + let (fields, _) = self.parse_paren_comma_seq(|p| { + p.parse_pat_allow_top_alt( + None, + RecoverComma::No, + RecoverColon::No, + CommaRecoveryMode::EitherTupleOrPipe, + ) + })?; + if qself.is_some() { + self.psess.gated_spans.gate(sym::more_qualified_paths, path.span); + } + Ok(PatKind::TupleStruct(qself, path, fields)) + } + + /// Are we sure this could not possibly be the start of a pattern? + /// + /// Currently, this only accounts for tokens that can follow identifiers + /// in patterns, but this can be extended as necessary. + fn isnt_pattern_start(&self) -> bool { + [ + token::Eq, + token::Colon, + token::Comma, + token::Semi, + token::At, + token::OpenDelim(Delimiter::Brace), + token::CloseDelim(Delimiter::Brace), + token::CloseDelim(Delimiter::Parenthesis), + ] + .contains(&self.token.kind) + } + + fn parse_pat_builtin(&mut self) -> PResult<'a, PatKind> { + self.parse_builtin(|self_, _lo, ident| { + Ok(match ident.name { + // builtin#deref(PAT) + sym::deref => Some(ast::PatKind::Deref(self_.parse_pat_allow_top_alt( + None, + RecoverComma::Yes, + RecoverColon::Yes, + CommaRecoveryMode::LikelyTuple, + )?)), + _ => None, + }) + }) + } + + /// Parses `box pat` + fn parse_pat_box(&mut self) -> PResult<'a, PatKind> { + let box_span = self.prev_token.span; + + if self.isnt_pattern_start() { + let descr = super::token_descr(&self.token); + self.dcx().emit_err(errors::BoxNotPat { + span: self.token.span, + kw: box_span, + lo: box_span.shrink_to_lo(), + descr, + }); + + // We cannot use `parse_pat_ident()` since it will complain `box` + // is not an identifier. + let sub = if self.eat(&token::At) { + Some(self.parse_pat_no_top_alt(Some(Expected::BindingPattern), None)?) + } else { + None + }; + + Ok(PatKind::Ident(BindingMode::NONE, Ident::new(kw::Box, box_span), sub)) + } else { + let pat = self.parse_pat_with_range_pat(false, None, None)?; + self.psess.gated_spans.gate(sym::box_patterns, box_span.to(self.prev_token.span)); + Ok(PatKind::Box(pat)) + } + } + + /// Parses the fields of a struct-like pattern. + fn parse_pat_fields(&mut self) -> PResult<'a, (ThinVec<PatField>, PatFieldsRest)> { + let mut fields = ThinVec::new(); + let mut etc = PatFieldsRest::None; + let mut ate_comma = true; + let mut delayed_err: Option<Diag<'a>> = None; + let mut first_etc_and_maybe_comma_span = None; + let mut last_non_comma_dotdot_span = None; + + while self.token != token::CloseDelim(Delimiter::Brace) { + let attrs = match self.parse_outer_attributes() { + Ok(attrs) => attrs, + Err(err) => { + if let Some(delayed) = delayed_err { + delayed.emit(); + } + return Err(err); + } + }; + let lo = self.token.span; + + // check that a comma comes after every field + if !ate_comma { + let mut err = + self.dcx().create_err(ExpectedCommaAfterPatternField { span: self.token.span }); + if let Some(delayed) = delayed_err { + delayed.emit(); + } + self.recover_misplaced_pattern_modifiers(&fields, &mut err); + return Err(err); + } + ate_comma = false; + + if self.check(&token::DotDot) + || self.check_noexpect(&token::DotDotDot) + || self.check_keyword(kw::Underscore) + { + etc = PatFieldsRest::Rest; + let mut etc_sp = self.token.span; + if first_etc_and_maybe_comma_span.is_none() { + if let Some(comma_tok) = self + .look_ahead(1, |t| if *t == token::Comma { Some(t.clone()) } else { None }) + { + let nw_span = self + .psess + .source_map() + .span_extend_to_line(comma_tok.span) + .trim_start(comma_tok.span.shrink_to_lo()) + .map(|s| self.psess.source_map().span_until_non_whitespace(s)); + first_etc_and_maybe_comma_span = nw_span.map(|s| etc_sp.to(s)); + } else { + first_etc_and_maybe_comma_span = + Some(self.psess.source_map().span_until_non_whitespace(etc_sp)); + } + } + + self.recover_bad_dot_dot(); + self.bump(); // `..` || `...` || `_` + + if self.token == token::CloseDelim(Delimiter::Brace) { + break; + } + let token_str = super::token_descr(&self.token); + let msg = format!("expected `}}`, found {token_str}"); + let mut err = self.dcx().struct_span_err(self.token.span, msg); + + err.span_label(self.token.span, "expected `}`"); + let mut comma_sp = None; + if self.token == token::Comma { + // Issue #49257 + let nw_span = + self.psess.source_map().span_until_non_whitespace(self.token.span); + etc_sp = etc_sp.to(nw_span); + err.span_label( + etc_sp, + "`..` must be at the end and cannot have a trailing comma", + ); + comma_sp = Some(self.token.span); + self.bump(); + ate_comma = true; + } + + if self.token == token::CloseDelim(Delimiter::Brace) { + // If the struct looks otherwise well formed, recover and continue. + if let Some(sp) = comma_sp { + err.span_suggestion_short( + sp, + "remove this comma", + "", + Applicability::MachineApplicable, + ); + } + err.emit(); + break; + } else if self.token.is_ident() && ate_comma { + // Accept fields coming after `..,`. + // This way we avoid "pattern missing fields" errors afterwards. + // We delay this error until the end in order to have a span for a + // suggested fix. + if let Some(delayed_err) = delayed_err { + delayed_err.emit(); + return Err(err); + } else { + delayed_err = Some(err); + } + } else { + if let Some(err) = delayed_err { + err.emit(); + } + return Err(err); + } + } + + let field = + self.collect_tokens_trailing_token(attrs, ForceCollect::No, |this, attrs| { + let field = match this.parse_pat_field(lo, attrs) { + Ok(field) => Ok(field), + Err(err) => { + if let Some(delayed_err) = delayed_err.take() { + delayed_err.emit(); + } + return Err(err); + } + }?; + ate_comma = this.eat(&token::Comma); + + last_non_comma_dotdot_span = Some(this.prev_token.span); + + // We just ate a comma, so there's no need to use + // `TrailingToken::Comma` + Ok((field, TrailingToken::None)) + })?; + + fields.push(field) + } + + if let Some(mut err) = delayed_err { + if let Some(first_etc_span) = first_etc_and_maybe_comma_span { + if self.prev_token == token::DotDot { + // We have `.., x, ..`. + err.multipart_suggestion( + "remove the starting `..`", + vec![(first_etc_span, String::new())], + Applicability::MachineApplicable, + ); + } else { + if let Some(last_non_comma_dotdot_span) = last_non_comma_dotdot_span { + // We have `.., x`. + err.multipart_suggestion( + "move the `..` to the end of the field list", + vec![ + (first_etc_span, String::new()), + ( + self.token.span.to(last_non_comma_dotdot_span.shrink_to_hi()), + format!("{} .. }}", if ate_comma { "" } else { "," }), + ), + ], + Applicability::MachineApplicable, + ); + } + } + } + err.emit(); + } + Ok((fields, etc)) + } + + /// If the user writes `S { ref field: name }` instead of `S { field: ref name }`, we suggest + /// the correct code. + fn recover_misplaced_pattern_modifiers(&self, fields: &ThinVec<PatField>, err: &mut Diag<'a>) { + if let Some(last) = fields.iter().last() + && last.is_shorthand + && let PatKind::Ident(binding, ident, None) = last.pat.kind + && binding != BindingMode::NONE + && self.token == token::Colon + // We found `ref mut? ident:`, try to parse a `name,` or `name }`. + && let Some(name_span) = self.look_ahead(1, |t| t.is_ident().then(|| t.span)) + && self.look_ahead(2, |t| { + t == &token::Comma || t == &token::CloseDelim(Delimiter::Brace) + }) + { + let span = last.pat.span.with_hi(ident.span.lo()); + // We have `S { ref field: name }` instead of `S { field: ref name }` + err.multipart_suggestion( + "the pattern modifiers belong after the `:`", + vec![ + (span, String::new()), + (name_span.shrink_to_lo(), binding.prefix_str().to_string()), + ], + Applicability::MachineApplicable, + ); + } + } + + /// Recover on `...` or `_` as if it were `..` to avoid further errors. + /// See issue #46718. + fn recover_bad_dot_dot(&self) { + if self.token == token::DotDot { + return; + } + + let token_str = pprust::token_to_string(&self.token); + self.dcx().emit_err(DotDotDotForRemainingFields { span: self.token.span, token_str }); + } + + fn parse_pat_field(&mut self, lo: Span, attrs: AttrVec) -> PResult<'a, PatField> { + // Check if a colon exists one ahead. This means we're parsing a fieldname. + let hi; + let (subpat, fieldname, is_shorthand) = if self.look_ahead(1, |t| t == &token::Colon) { + // Parsing a pattern of the form `fieldname: pat`. + let fieldname = self.parse_field_name()?; + self.bump(); + let pat = self.parse_pat_allow_top_alt( + None, + RecoverComma::No, + RecoverColon::No, + CommaRecoveryMode::EitherTupleOrPipe, + )?; + hi = pat.span; + (pat, fieldname, false) + } else { + // Parsing a pattern of the form `(box) (ref) (mut) fieldname`. + let is_box = self.eat_keyword(kw::Box); + let boxed_span = self.token.span; + let mutability = self.parse_mutability(); + let by_ref = self.parse_byref(); + + let fieldname = self.parse_field_name()?; + hi = self.prev_token.span; + let ann = BindingMode(by_ref, mutability); + let fieldpat = self.mk_pat_ident(boxed_span.to(hi), ann, fieldname); + let subpat = + if is_box { self.mk_pat(lo.to(hi), PatKind::Box(fieldpat)) } else { fieldpat }; + (subpat, fieldname, true) + }; + + Ok(PatField { + ident: fieldname, + pat: subpat, + is_shorthand, + attrs, + id: ast::DUMMY_NODE_ID, + span: lo.to(hi), + is_placeholder: false, + }) + } + + pub(super) fn mk_pat_ident(&self, span: Span, ann: BindingMode, ident: Ident) -> P<Pat> { + self.mk_pat(span, PatKind::Ident(ann, ident, None)) + } + + pub(super) fn mk_pat(&self, span: Span, kind: PatKind) -> P<Pat> { + P(Pat { kind, span, id: ast::DUMMY_NODE_ID, tokens: None }) + } +} diff --git a/compiler/rustc_parse/src/parser/path.rs b/compiler/rustc_parse/src/parser/path.rs new file mode 100644 index 00000000000..d845e8ab90d --- /dev/null +++ b/compiler/rustc_parse/src/parser/path.rs @@ -0,0 +1,967 @@ +use super::ty::{AllowPlus, RecoverQPath, RecoverReturnSign}; +use super::{Parser, Restrictions, TokenType}; +use crate::errors::PathSingleColon; +use crate::parser::{CommaRecoveryMode, RecoverColon, RecoverComma}; +use crate::{errors, maybe_whole}; +use ast::token::IdentIsRaw; +use rustc_ast::ptr::P; +use rustc_ast::token::{self, Delimiter, Token, TokenKind}; +use rustc_ast::{ + self as ast, AngleBracketedArg, AngleBracketedArgs, AnonConst, AssocConstraint, + AssocConstraintKind, BlockCheckMode, GenericArg, GenericArgs, Generics, ParenthesizedArgs, + Path, PathSegment, QSelf, +}; +use rustc_errors::{Applicability, Diag, PResult}; +use rustc_span::symbol::{kw, sym, Ident}; +use rustc_span::{BytePos, Span}; +use std::mem; +use thin_vec::ThinVec; +use tracing::debug; + +/// Specifies how to parse a path. +#[derive(Copy, Clone, PartialEq)] +pub enum PathStyle { + /// In some contexts, notably in expressions, paths with generic arguments are ambiguous + /// with something else. For example, in expressions `segment < ....` can be interpreted + /// as a comparison and `segment ( ....` can be interpreted as a function call. + /// In all such contexts the non-path interpretation is preferred by default for practical + /// reasons, but the path interpretation can be forced by the disambiguator `::`, e.g. + /// `x<y>` - comparisons, `x::<y>` - unambiguously a path. + /// + /// Also, a path may never be followed by a `:`. This means that we can eagerly recover if + /// we encounter it. + Expr, + /// The same as `Expr`, but may be followed by a `:`. + /// For example, this code: + /// ```rust + /// struct S; + /// + /// let S: S; + /// // ^ Followed by a `:` + /// ``` + Pat, + /// In other contexts, notably in types, no ambiguity exists and paths can be written + /// without the disambiguator, e.g., `x<y>` - unambiguously a path. + /// Paths with disambiguators are still accepted, `x::<Y>` - unambiguously a path too. + Type, + /// A path with generic arguments disallowed, e.g., `foo::bar::Baz`, used in imports, + /// visibilities or attributes. + /// Technically, this variant is unnecessary and e.g., `Expr` can be used instead + /// (paths in "mod" contexts have to be checked later for absence of generic arguments + /// anyway, due to macros), but it is used to avoid weird suggestions about expected + /// tokens when something goes wrong. + Mod, +} + +impl PathStyle { + fn has_generic_ambiguity(&self) -> bool { + matches!(self, Self::Expr | Self::Pat) + } +} + +impl<'a> Parser<'a> { + /// Parses a qualified path. + /// Assumes that the leading `<` has been parsed already. + /// + /// `qualified_path = <type [as trait_ref]>::path` + /// + /// # Examples + /// `<T>::default` + /// `<T as U>::a` + /// `<T as U>::F::a<S>` (without disambiguator) + /// `<T as U>::F::a::<S>` (with disambiguator) + pub(super) fn parse_qpath(&mut self, style: PathStyle) -> PResult<'a, (P<QSelf>, Path)> { + let lo = self.prev_token.span; + let ty = self.parse_ty()?; + + // `path` will contain the prefix of the path up to the `>`, + // if any (e.g., `U` in the `<T as U>::*` examples + // above). `path_span` has the span of that path, or an empty + // span in the case of something like `<T>::Bar`. + let (mut path, path_span); + if self.eat_keyword(kw::As) { + let path_lo = self.token.span; + path = self.parse_path(PathStyle::Type)?; + path_span = path_lo.to(self.prev_token.span); + } else { + path_span = self.token.span.to(self.token.span); + path = ast::Path { segments: ThinVec::new(), span: path_span, tokens: None }; + } + + // See doc comment for `unmatched_angle_bracket_count`. + self.expect(&token::Gt)?; + if self.unmatched_angle_bracket_count > 0 { + self.unmatched_angle_bracket_count -= 1; + debug!("parse_qpath: (decrement) count={:?}", self.unmatched_angle_bracket_count); + } + + let is_import_coupler = self.is_import_coupler(); + if !is_import_coupler && !self.recover_colon_before_qpath_proj() { + self.expect(&token::PathSep)?; + } + + let qself = P(QSelf { ty, path_span, position: path.segments.len() }); + if !is_import_coupler { + self.parse_path_segments(&mut path.segments, style, None)?; + } + + Ok(( + qself, + Path { segments: path.segments, span: lo.to(self.prev_token.span), tokens: None }, + )) + } + + /// Recover from an invalid single colon, when the user likely meant a qualified path. + /// We avoid emitting this if not followed by an identifier, as our assumption that the user + /// intended this to be a qualified path may not be correct. + /// + /// ```ignore (diagnostics) + /// <Bar as Baz<T>>:Qux + /// ^ help: use double colon + /// ``` + fn recover_colon_before_qpath_proj(&mut self) -> bool { + if !self.check_noexpect(&TokenKind::Colon) + || self.look_ahead(1, |t| !t.is_ident() || t.is_reserved_ident()) + { + return false; + } + + self.bump(); // colon + + self.dcx() + .struct_span_err( + self.prev_token.span, + "found single colon before projection in qualified path", + ) + .with_span_suggestion( + self.prev_token.span, + "use double colon", + "::", + Applicability::MachineApplicable, + ) + .emit(); + + true + } + + pub(super) fn parse_path(&mut self, style: PathStyle) -> PResult<'a, Path> { + self.parse_path_inner(style, None) + } + + /// Parses simple paths. + /// + /// `path = [::] segment+` + /// `segment = ident | ident[::]<args> | ident[::](args) [-> type]` + /// + /// # Examples + /// `a::b::C<D>` (without disambiguator) + /// `a::b::C::<D>` (with disambiguator) + /// `Fn(Args)` (without disambiguator) + /// `Fn::(Args)` (with disambiguator) + pub(super) fn parse_path_inner( + &mut self, + style: PathStyle, + ty_generics: Option<&Generics>, + ) -> PResult<'a, Path> { + let reject_generics_if_mod_style = |parser: &Parser<'_>, path: Path| { + // Ensure generic arguments don't end up in attribute paths, such as: + // + // macro_rules! m { + // ($p:path) => { #[$p] struct S; } + // } + // + // m!(inline<u8>); //~ ERROR: unexpected generic arguments in path + // + if style == PathStyle::Mod && path.segments.iter().any(|segment| segment.args.is_some()) + { + let span = path + .segments + .iter() + .filter_map(|segment| segment.args.as_ref()) + .map(|arg| arg.span()) + .collect::<Vec<_>>(); + parser.dcx().emit_err(errors::GenericsInPath { span }); + // Ignore these arguments to prevent unexpected behaviors. + let segments = path + .segments + .iter() + .map(|segment| PathSegment { ident: segment.ident, id: segment.id, args: None }) + .collect(); + Path { segments, ..path } + } else { + path + } + }; + + maybe_whole!(self, NtPath, |path| reject_generics_if_mod_style(self, path.into_inner())); + + if let token::Interpolated(nt) = &self.token.kind { + if let token::NtTy(ty) = &**nt { + if let ast::TyKind::Path(None, path) = &ty.kind { + let path = path.clone(); + self.bump(); + return Ok(reject_generics_if_mod_style(self, path)); + } + } + } + + let lo = self.token.span; + let mut segments = ThinVec::new(); + let mod_sep_ctxt = self.token.span.ctxt(); + if self.eat(&token::PathSep) { + segments.push(PathSegment::path_root(lo.shrink_to_lo().with_ctxt(mod_sep_ctxt))); + } + self.parse_path_segments(&mut segments, style, ty_generics)?; + Ok(Path { segments, span: lo.to(self.prev_token.span), tokens: None }) + } + + pub(super) fn parse_path_segments( + &mut self, + segments: &mut ThinVec<PathSegment>, + style: PathStyle, + ty_generics: Option<&Generics>, + ) -> PResult<'a, ()> { + loop { + let segment = self.parse_path_segment(style, ty_generics)?; + if style.has_generic_ambiguity() { + // In order to check for trailing angle brackets, we must have finished + // recursing (`parse_path_segment` can indirectly call this function), + // that is, the next token must be the highlighted part of the below example: + // + // `Foo::<Bar as Baz<T>>::Qux` + // ^ here + // + // As opposed to the below highlight (if we had only finished the first + // recursion): + // + // `Foo::<Bar as Baz<T>>::Qux` + // ^ here + // + // `PathStyle::Expr` is only provided at the root invocation and never in + // `parse_path_segment` to recurse and therefore can be checked to maintain + // this invariant. + self.check_trailing_angle_brackets(&segment, &[&token::PathSep]); + } + segments.push(segment); + + if self.is_import_coupler() || !self.eat(&token::PathSep) { + if style == PathStyle::Expr + && self.may_recover() + && self.token == token::Colon + && self.look_ahead(1, |token| token.is_ident() && !token.is_reserved_ident()) + { + // Emit a special error message for `a::b:c` to help users + // otherwise, `a: c` might have meant to introduce a new binding + if self.token.span.lo() == self.prev_token.span.hi() + && self.look_ahead(1, |token| self.token.span.hi() == token.span.lo()) + { + self.bump(); // bump past the colon + self.dcx().emit_err(PathSingleColon { + span: self.prev_token.span, + type_ascription: self + .psess + .unstable_features + .is_nightly_build() + .then_some(()), + }); + } + continue; + } + + return Ok(()); + } + } + } + + pub(super) fn parse_path_segment( + &mut self, + style: PathStyle, + ty_generics: Option<&Generics>, + ) -> PResult<'a, PathSegment> { + let ident = self.parse_path_segment_ident()?; + let is_args_start = |token: &Token| { + matches!( + token.kind, + token::Lt + | token::BinOp(token::Shl) + | token::OpenDelim(Delimiter::Parenthesis) + | token::LArrow + ) + }; + let check_args_start = |this: &mut Self| { + this.expected_tokens.extend_from_slice(&[ + TokenType::Token(token::Lt), + TokenType::Token(token::OpenDelim(Delimiter::Parenthesis)), + ]); + is_args_start(&this.token) + }; + + Ok( + if style == PathStyle::Type && check_args_start(self) + || style != PathStyle::Mod + && self.check(&token::PathSep) + && self.look_ahead(1, |t| is_args_start(t)) + { + // We use `style == PathStyle::Expr` to check if this is in a recursion or not. If + // it isn't, then we reset the unmatched angle bracket count as we're about to start + // parsing a new path. + if style == PathStyle::Expr { + self.unmatched_angle_bracket_count = 0; + } + + // Generic arguments are found - `<`, `(`, `::<` or `::(`. + self.eat(&token::PathSep); + let lo = self.token.span; + let args = if self.eat_lt() { + // `<'a, T, A = U>` + let args = self.parse_angle_args_with_leading_angle_bracket_recovery( + style, + lo, + ty_generics, + )?; + self.expect_gt().map_err(|mut err| { + // Try to recover a `:` into a `::` + if self.token == token::Colon + && self.look_ahead(1, |token| { + token.is_ident() && !token.is_reserved_ident() + }) + { + err.cancel(); + err = self.dcx().create_err(PathSingleColon { + span: self.token.span, + type_ascription: self + .psess + .unstable_features + .is_nightly_build() + .then_some(()), + }); + } + // Attempt to find places where a missing `>` might belong. + else if let Some(arg) = args + .iter() + .rev() + .find(|arg| !matches!(arg, AngleBracketedArg::Constraint(_))) + { + err.span_suggestion_verbose( + arg.span().shrink_to_hi(), + "you might have meant to end the type parameters here", + ">", + Applicability::MaybeIncorrect, + ); + } + err + })?; + let span = lo.to(self.prev_token.span); + AngleBracketedArgs { args, span }.into() + } else if self.may_recover() + && self.token.kind == token::OpenDelim(Delimiter::Parenthesis) + // FIXME(return_type_notation): Could also recover `...` here. + && self.look_ahead(1, |tok| tok.kind == token::DotDot) + { + self.bump(); + self.dcx() + .emit_err(errors::BadReturnTypeNotationDotDot { span: self.token.span }); + self.bump(); + self.expect(&token::CloseDelim(Delimiter::Parenthesis))?; + let span = lo.to(self.prev_token.span); + + if self.eat_noexpect(&token::RArrow) { + let lo = self.prev_token.span; + let ty = self.parse_ty()?; + self.dcx() + .emit_err(errors::BadReturnTypeNotationOutput { span: lo.to(ty.span) }); + } + + ParenthesizedArgs { + span, + inputs: ThinVec::new(), + inputs_span: span, + output: ast::FnRetTy::Default(self.prev_token.span.shrink_to_hi()), + } + .into() + } else { + // `(T, U) -> R` + + let prev_token_before_parsing = self.prev_token.clone(); + let token_before_parsing = self.token.clone(); + let mut snapshot = None; + if self.may_recover() + && prev_token_before_parsing.kind == token::PathSep + && (style == PathStyle::Expr && self.token.can_begin_expr() + || style == PathStyle::Pat && self.token.can_begin_pattern()) + { + snapshot = Some(self.create_snapshot_for_diagnostic()); + } + + let (inputs, _) = match self.parse_paren_comma_seq(|p| p.parse_ty()) { + Ok(output) => output, + Err(mut error) if prev_token_before_parsing.kind == token::PathSep => { + error.span_label( + prev_token_before_parsing.span.to(token_before_parsing.span), + "while parsing this parenthesized list of type arguments starting here", + ); + + if let Some(mut snapshot) = snapshot { + snapshot.recover_fn_call_leading_path_sep( + style, + prev_token_before_parsing, + &mut error, + ) + } + + return Err(error); + } + Err(error) => return Err(error), + }; + let inputs_span = lo.to(self.prev_token.span); + let output = + self.parse_ret_ty(AllowPlus::No, RecoverQPath::No, RecoverReturnSign::No)?; + let span = ident.span.to(self.prev_token.span); + ParenthesizedArgs { span, inputs, inputs_span, output }.into() + }; + + PathSegment { ident, args: Some(args), id: ast::DUMMY_NODE_ID } + } else { + // Generic arguments are not found. + PathSegment::from_ident(ident) + }, + ) + } + + pub(super) fn parse_path_segment_ident(&mut self) -> PResult<'a, Ident> { + match self.token.ident() { + Some((ident, IdentIsRaw::No)) if ident.is_path_segment_keyword() => { + self.bump(); + Ok(ident) + } + _ => self.parse_ident(), + } + } + + /// Recover `$path::(...)` as `$path(...)`. + /// + /// ```ignore (diagnostics) + /// foo::(420, "bar") + /// ^^ remove extra separator to make the function call + /// // or + /// match x { + /// Foo::(420, "bar") => { ... }, + /// ^^ remove extra separator to turn this into tuple struct pattern + /// _ => { ... }, + /// } + /// ``` + fn recover_fn_call_leading_path_sep( + &mut self, + style: PathStyle, + prev_token_before_parsing: Token, + error: &mut Diag<'_>, + ) { + match style { + PathStyle::Expr + if let Ok(_) = self + .parse_paren_comma_seq(|p| p.parse_expr()) + .map_err(|error| error.cancel()) => {} + PathStyle::Pat + if let Ok(_) = self + .parse_paren_comma_seq(|p| { + p.parse_pat_allow_top_alt( + None, + RecoverComma::No, + RecoverColon::No, + CommaRecoveryMode::LikelyTuple, + ) + }) + .map_err(|error| error.cancel()) => {} + _ => { + return; + } + } + + if let token::PathSep | token::RArrow = self.token.kind { + return; + } + + error.span_suggestion_verbose( + prev_token_before_parsing.span, + format!( + "consider removing the `::` here to {}", + match style { + PathStyle::Expr => "call the expression", + PathStyle::Pat => "turn this into a tuple struct pattern", + _ => { + return; + } + } + ), + "", + Applicability::MaybeIncorrect, + ); + } + + /// Parses generic args (within a path segment) with recovery for extra leading angle brackets. + /// For the purposes of understanding the parsing logic of generic arguments, this function + /// can be thought of being the same as just calling `self.parse_angle_args()` if the source + /// had the correct amount of leading angle brackets. + /// + /// ```ignore (diagnostics) + /// bar::<<<<T as Foo>::Output>(); + /// ^^ help: remove extra angle brackets + /// ``` + fn parse_angle_args_with_leading_angle_bracket_recovery( + &mut self, + style: PathStyle, + lo: Span, + ty_generics: Option<&Generics>, + ) -> PResult<'a, ThinVec<AngleBracketedArg>> { + // We need to detect whether there are extra leading left angle brackets and produce an + // appropriate error and suggestion. This cannot be implemented by looking ahead at + // upcoming tokens for a matching `>` character - if there are unmatched `<` tokens + // then there won't be matching `>` tokens to find. + // + // To explain how this detection works, consider the following example: + // + // ```ignore (diagnostics) + // bar::<<<<T as Foo>::Output>(); + // ^^ help: remove extra angle brackets + // ``` + // + // Parsing of the left angle brackets starts in this function. We start by parsing the + // `<` token (incrementing the counter of unmatched angle brackets on `Parser` via + // `eat_lt`): + // + // *Upcoming tokens:* `<<<<T as Foo>::Output>;` + // *Unmatched count:* 1 + // *`parse_path_segment` calls deep:* 0 + // + // This has the effect of recursing as this function is called if a `<` character + // is found within the expected generic arguments: + // + // *Upcoming tokens:* `<<<T as Foo>::Output>;` + // *Unmatched count:* 2 + // *`parse_path_segment` calls deep:* 1 + // + // Eventually we will have recursed until having consumed all of the `<` tokens and + // this will be reflected in the count: + // + // *Upcoming tokens:* `T as Foo>::Output>;` + // *Unmatched count:* 4 + // `parse_path_segment` calls deep:* 3 + // + // The parser will continue until reaching the first `>` - this will decrement the + // unmatched angle bracket count and return to the parent invocation of this function + // having succeeded in parsing: + // + // *Upcoming tokens:* `::Output>;` + // *Unmatched count:* 3 + // *`parse_path_segment` calls deep:* 2 + // + // This will continue until the next `>` character which will also return successfully + // to the parent invocation of this function and decrement the count: + // + // *Upcoming tokens:* `;` + // *Unmatched count:* 2 + // *`parse_path_segment` calls deep:* 1 + // + // At this point, this function will expect to find another matching `>` character but + // won't be able to and will return an error. This will continue all the way up the + // call stack until the first invocation: + // + // *Upcoming tokens:* `;` + // *Unmatched count:* 2 + // *`parse_path_segment` calls deep:* 0 + // + // In doing this, we have managed to work out how many unmatched leading left angle + // brackets there are, but we cannot recover as the unmatched angle brackets have + // already been consumed. To remedy this, we keep a snapshot of the parser state + // before we do the above. We can then inspect whether we ended up with a parsing error + // and unmatched left angle brackets and if so, restore the parser state before we + // consumed any `<` characters to emit an error and consume the erroneous tokens to + // recover by attempting to parse again. + // + // In practice, the recursion of this function is indirect and there will be other + // locations that consume some `<` characters - as long as we update the count when + // this happens, it isn't an issue. + + let is_first_invocation = style == PathStyle::Expr; + // Take a snapshot before attempting to parse - we can restore this later. + let snapshot = is_first_invocation.then(|| self.clone()); + + self.angle_bracket_nesting += 1; + debug!("parse_generic_args_with_leading_angle_bracket_recovery: (snapshotting)"); + match self.parse_angle_args(ty_generics) { + Ok(args) => { + self.angle_bracket_nesting -= 1; + Ok(args) + } + Err(e) if self.angle_bracket_nesting > 10 => { + self.angle_bracket_nesting -= 1; + // When encountering severely malformed code where there are several levels of + // nested unclosed angle args (`f::<f::<f::<f::<...`), we avoid severe O(n^2) + // behavior by bailing out earlier (#117080). + e.emit(); + rustc_errors::FatalError.raise(); + } + Err(e) if is_first_invocation && self.unmatched_angle_bracket_count > 0 => { + self.angle_bracket_nesting -= 1; + + // Swap `self` with our backup of the parser state before attempting to parse + // generic arguments. + let snapshot = mem::replace(self, snapshot.unwrap()); + + // Eat the unmatched angle brackets. + let all_angle_brackets = (0..snapshot.unmatched_angle_bracket_count) + .fold(true, |a, _| a && self.eat_lt()); + + if !all_angle_brackets { + // If there are other tokens in between the extraneous `<`s, we cannot simply + // suggest to remove them. This check also prevents us from accidentally ending + // up in the middle of a multibyte character (issue #84104). + let _ = mem::replace(self, snapshot); + Err(e) + } else { + // Cancel error from being unable to find `>`. We know the error + // must have been this due to a non-zero unmatched angle bracket + // count. + e.cancel(); + + debug!( + "parse_generic_args_with_leading_angle_bracket_recovery: (snapshot failure) \ + snapshot.count={:?}", + snapshot.unmatched_angle_bracket_count, + ); + + // Make a span over ${unmatched angle bracket count} characters. + // This is safe because `all_angle_brackets` ensures that there are only `<`s, + // i.e. no multibyte characters, in this range. + let span = lo + .with_hi(lo.lo() + BytePos(snapshot.unmatched_angle_bracket_count.into())); + self.dcx().emit_err(errors::UnmatchedAngle { + span, + plural: snapshot.unmatched_angle_bracket_count > 1, + }); + + // Try again without unmatched angle bracket characters. + self.parse_angle_args(ty_generics) + } + } + Err(e) => { + self.angle_bracket_nesting -= 1; + Err(e) + } + } + } + + /// Parses (possibly empty) list of generic arguments / associated item constraints, + /// possibly including trailing comma. + pub(super) fn parse_angle_args( + &mut self, + ty_generics: Option<&Generics>, + ) -> PResult<'a, ThinVec<AngleBracketedArg>> { + let mut args = ThinVec::new(); + while let Some(arg) = self.parse_angle_arg(ty_generics)? { + args.push(arg); + if !self.eat(&token::Comma) { + if self.check_noexpect(&TokenKind::Semi) + && self.look_ahead(1, |t| t.is_ident() || t.is_lifetime()) + { + // Add `>` to the list of expected tokens. + self.check(&token::Gt); + // Handle `,` to `;` substitution + let mut err = self.unexpected().unwrap_err(); + self.bump(); + err.span_suggestion_verbose( + self.prev_token.span.until(self.token.span), + "use a comma to separate type parameters", + ", ", + Applicability::MachineApplicable, + ); + err.emit(); + continue; + } + if !self.token.kind.should_end_const_arg() { + if self.handle_ambiguous_unbraced_const_arg(&mut args)? { + // We've managed to (partially) recover, so continue trying to parse + // arguments. + continue; + } + } + break; + } + } + Ok(args) + } + + /// Parses a single argument in the angle arguments `<...>` of a path segment. + fn parse_angle_arg( + &mut self, + ty_generics: Option<&Generics>, + ) -> PResult<'a, Option<AngleBracketedArg>> { + let lo = self.token.span; + let arg = self.parse_generic_arg(ty_generics)?; + match arg { + Some(arg) => { + // we are using noexpect here because we first want to find out if either `=` or `:` + // is present and then use that info to push the other token onto the tokens list + let separated = + self.check_noexpect(&token::Colon) || self.check_noexpect(&token::Eq); + if separated && (self.check(&token::Colon) | self.check(&token::Eq)) { + let arg_span = arg.span(); + let (binder, ident, gen_args) = match self.get_ident_from_generic_arg(&arg) { + Ok(ident_gen_args) => ident_gen_args, + Err(()) => return Ok(Some(AngleBracketedArg::Arg(arg))), + }; + if binder { + // FIXME(compiler-errors): this could be improved by suggesting lifting + // this up to the trait, at least before this becomes real syntax. + // e.g. `Trait<for<'a> Assoc = Ty>` -> `for<'a> Trait<Assoc = Ty>` + return Err(self.dcx().struct_span_err( + arg_span, + "`for<...>` is not allowed on associated type bounds", + )); + } + let kind = if self.eat(&token::Colon) { + // Parse associated type constraint bound. + + let bounds = self.parse_generic_bounds()?; + AssocConstraintKind::Bound { bounds } + } else if self.eat(&token::Eq) { + self.parse_assoc_equality_term( + ident, + gen_args.as_ref(), + self.prev_token.span, + )? + } else { + unreachable!(); + }; + + let span = lo.to(self.prev_token.span); + // Gate associated type bounds, e.g., `Iterator<Item: Ord>`. + if let AssocConstraintKind::Bound { .. } = kind { + if let Some(ast::GenericArgs::Parenthesized(args)) = &gen_args + && args.inputs.is_empty() + && matches!(args.output, ast::FnRetTy::Default(..)) + { + self.psess.gated_spans.gate(sym::return_type_notation, span); + } + } + let constraint = + AssocConstraint { id: ast::DUMMY_NODE_ID, ident, gen_args, kind, span }; + Ok(Some(AngleBracketedArg::Constraint(constraint))) + } else { + // we only want to suggest `:` and `=` in contexts where the previous token + // is an ident and the current token or the next token is an ident + if self.prev_token.is_ident() + && (self.token.is_ident() || self.look_ahead(1, |token| token.is_ident())) + { + self.check(&token::Colon); + self.check(&token::Eq); + } + Ok(Some(AngleBracketedArg::Arg(arg))) + } + } + _ => Ok(None), + } + } + + /// Parse the term to the right of an associated item equality constraint. + /// + /// That is, parse `$term` in `Item = $term` where `$term` is a type or + /// a const expression (wrapped in curly braces if complex). + fn parse_assoc_equality_term( + &mut self, + ident: Ident, + gen_args: Option<&GenericArgs>, + eq: Span, + ) -> PResult<'a, AssocConstraintKind> { + let arg = self.parse_generic_arg(None)?; + let span = ident.span.to(self.prev_token.span); + let term = match arg { + Some(GenericArg::Type(ty)) => ty.into(), + Some(GenericArg::Const(c)) => { + self.psess.gated_spans.gate(sym::associated_const_equality, span); + c.into() + } + Some(GenericArg::Lifetime(lt)) => { + let guar = self.dcx().emit_err(errors::LifetimeInEqConstraint { + span: lt.ident.span, + lifetime: lt.ident, + binding_label: span, + colon_sugg: gen_args + .map_or(ident.span, |args| args.span()) + .between(lt.ident.span), + }); + self.mk_ty(lt.ident.span, ast::TyKind::Err(guar)).into() + } + None => { + let after_eq = eq.shrink_to_hi(); + let before_next = self.token.span.shrink_to_lo(); + let mut err = self + .dcx() + .struct_span_err(after_eq.to(before_next), "missing type to the right of `=`"); + if matches!(self.token.kind, token::Comma | token::Gt) { + err.span_suggestion( + self.psess.source_map().next_point(eq).to(before_next), + "to constrain the associated type, add a type after `=`", + " TheType", + Applicability::HasPlaceholders, + ); + err.span_suggestion( + eq.to(before_next), + format!("remove the `=` if `{ident}` is a type"), + "", + Applicability::MaybeIncorrect, + ) + } else { + err.span_label( + self.token.span, + format!("expected type, found {}", super::token_descr(&self.token)), + ) + }; + return Err(err); + } + }; + Ok(AssocConstraintKind::Equality { term }) + } + + /// We do not permit arbitrary expressions as const arguments. They must be one of: + /// - An expression surrounded in `{}`. + /// - A literal. + /// - A numeric literal prefixed by `-`. + /// - A single-segment path. + pub(super) fn expr_is_valid_const_arg(&self, expr: &P<rustc_ast::Expr>) -> bool { + match &expr.kind { + ast::ExprKind::Block(_, _) + | ast::ExprKind::Lit(_) + | ast::ExprKind::IncludedBytes(..) => true, + ast::ExprKind::Unary(ast::UnOp::Neg, expr) => { + matches!(expr.kind, ast::ExprKind::Lit(_)) + } + // We can only resolve single-segment paths at the moment, because multi-segment paths + // require type-checking: see `visit_generic_arg` in `src/librustc_resolve/late.rs`. + ast::ExprKind::Path(None, path) + if path.segments.len() == 1 && path.segments[0].args.is_none() => + { + true + } + _ => false, + } + } + + /// Parse a const argument, e.g. `<3>`. It is assumed the angle brackets will be parsed by + /// the caller. + pub(super) fn parse_const_arg(&mut self) -> PResult<'a, AnonConst> { + // Parse const argument. + let value = if let token::OpenDelim(Delimiter::Brace) = self.token.kind { + self.parse_expr_block(None, self.token.span, BlockCheckMode::Default)? + } else { + self.handle_unambiguous_unbraced_const_arg()? + }; + Ok(AnonConst { id: ast::DUMMY_NODE_ID, value }) + } + + /// Parse a generic argument in a path segment. + /// This does not include constraints, e.g., `Item = u8`, which is handled in `parse_angle_arg`. + pub(super) fn parse_generic_arg( + &mut self, + ty_generics: Option<&Generics>, + ) -> PResult<'a, Option<GenericArg>> { + let start = self.token.span; + let arg = if self.check_lifetime() && self.look_ahead(1, |t| !t.is_like_plus()) { + // Parse lifetime argument. + GenericArg::Lifetime(self.expect_lifetime()) + } else if self.check_const_arg() { + // Parse const argument. + GenericArg::Const(self.parse_const_arg()?) + } else if self.check_type() { + // Parse type argument. + + // Proactively create a parser snapshot enabling us to rewind and try to reparse the + // input as a const expression in case we fail to parse a type. If we successfully + // do so, we will report an error that it needs to be wrapped in braces. + let mut snapshot = None; + if self.may_recover() && self.token.can_begin_expr() { + snapshot = Some(self.create_snapshot_for_diagnostic()); + } + + match self.parse_ty() { + Ok(ty) => { + // Since the type parser recovers from some malformed slice and array types and + // successfully returns a type, we need to look for `TyKind::Err`s in the + // type to determine if error recovery has occurred and if the input is not a + // syntactically valid type after all. + if let ast::TyKind::Slice(inner_ty) | ast::TyKind::Array(inner_ty, _) = &ty.kind + && let ast::TyKind::Err(_) = inner_ty.kind + && let Some(snapshot) = snapshot + && let Some(expr) = + self.recover_unbraced_const_arg_that_can_begin_ty(snapshot) + { + return Ok(Some( + self.dummy_const_arg_needs_braces( + self.dcx() + .struct_span_err(expr.span, "invalid const generic expression"), + expr.span, + ), + )); + } + + GenericArg::Type(ty) + } + Err(err) => { + if let Some(snapshot) = snapshot + && let Some(expr) = + self.recover_unbraced_const_arg_that_can_begin_ty(snapshot) + { + return Ok(Some(self.dummy_const_arg_needs_braces(err, expr.span))); + } + // Try to recover from possible `const` arg without braces. + return self.recover_const_arg(start, err).map(Some); + } + } + } else if self.token.is_keyword(kw::Const) { + return self.recover_const_param_declaration(ty_generics); + } else { + // Fall back by trying to parse a const-expr expression. If we successfully do so, + // then we should report an error that it needs to be wrapped in braces. + let snapshot = self.create_snapshot_for_diagnostic(); + match self.parse_expr_res(Restrictions::CONST_EXPR, None) { + Ok(expr) => { + return Ok(Some(self.dummy_const_arg_needs_braces( + self.dcx().struct_span_err(expr.span, "invalid const generic expression"), + expr.span, + ))); + } + Err(err) => { + self.restore_snapshot(snapshot); + err.cancel(); + return Ok(None); + } + } + }; + Ok(Some(arg)) + } + + /// Given a arg inside of generics, we try to destructure it as if it were the LHS in + /// `LHS = ...`, i.e. an associated type binding. + /// This returns a bool indicating if there are any `for<'a, 'b>` binder args, the + /// identifier, and any GAT arguments. + fn get_ident_from_generic_arg( + &self, + gen_arg: &GenericArg, + ) -> Result<(bool, Ident, Option<GenericArgs>), ()> { + if let GenericArg::Type(ty) = gen_arg { + if let ast::TyKind::Path(qself, path) = &ty.kind + && qself.is_none() + && let [seg] = path.segments.as_slice() + { + return Ok((false, seg.ident, seg.args.as_deref().cloned())); + } else if let ast::TyKind::TraitObject(bounds, ast::TraitObjectSyntax::None) = &ty.kind + && let [ast::GenericBound::Trait(trait_ref, ast::TraitBoundModifiers::NONE)] = + bounds.as_slice() + && let [seg] = trait_ref.trait_ref.path.segments.as_slice() + { + return Ok((true, seg.ident, seg.args.as_deref().cloned())); + } + } + Err(()) + } +} diff --git a/compiler/rustc_parse/src/parser/stmt.rs b/compiler/rustc_parse/src/parser/stmt.rs new file mode 100644 index 00000000000..7424fbea9b0 --- /dev/null +++ b/compiler/rustc_parse/src/parser/stmt.rs @@ -0,0 +1,853 @@ +use super::attr::InnerAttrForbiddenReason; +use super::diagnostics::AttemptLocalParseRecovery; +use super::expr::LhsExpr; +use super::pat::{PatternLocation, RecoverComma}; +use super::path::PathStyle; +use super::TrailingToken; +use super::{ + AttrWrapper, BlockMode, FnParseMode, ForceCollect, Parser, Restrictions, SemiColonMode, +}; +use crate::errors; +use crate::maybe_whole; + +use crate::errors::MalformedLoopLabel; +use ast::Label; +use rustc_ast as ast; +use rustc_ast::ptr::P; +use rustc_ast::token::{self, Delimiter, TokenKind}; +use rustc_ast::util::classify; +use rustc_ast::{AttrStyle, AttrVec, LocalKind, MacCall, MacCallStmt, MacStmtStyle}; +use rustc_ast::{Block, BlockCheckMode, Expr, ExprKind, HasAttrs, Local, Recovered, Stmt}; +use rustc_ast::{StmtKind, DUMMY_NODE_ID}; +use rustc_errors::{Applicability, Diag, PResult}; +use rustc_span::symbol::{kw, sym, Ident}; +use rustc_span::{BytePos, ErrorGuaranteed, Span}; + +use std::borrow::Cow; +use std::mem; +use thin_vec::{thin_vec, ThinVec}; + +impl<'a> Parser<'a> { + /// Parses a statement. This stops just before trailing semicolons on everything but items. + /// e.g., a `StmtKind::Semi` parses to a `StmtKind::Expr`, leaving the trailing `;` unconsumed. + // Public for rustfmt usage. + pub fn parse_stmt(&mut self, force_collect: ForceCollect) -> PResult<'a, Option<Stmt>> { + Ok(self.parse_stmt_without_recovery(false, force_collect).unwrap_or_else(|e| { + e.emit(); + self.recover_stmt_(SemiColonMode::Break, BlockMode::Ignore); + None + })) + } + + /// If `force_collect` is [`ForceCollect::Yes`], forces collection of tokens regardless of + /// whether or not we have attributes. + // Public for `cfg_eval` macro expansion. + pub fn parse_stmt_without_recovery( + &mut self, + capture_semi: bool, + force_collect: ForceCollect, + ) -> PResult<'a, Option<Stmt>> { + let attrs = self.parse_outer_attributes()?; + let lo = self.token.span; + + maybe_whole!(self, NtStmt, |stmt| { + stmt.visit_attrs(|stmt_attrs| { + attrs.prepend_to_nt_inner(stmt_attrs); + }); + Some(stmt.into_inner()) + }); + + if self.token.is_keyword(kw::Mut) && self.is_keyword_ahead(1, &[kw::Let]) { + self.bump(); + let mut_let_span = lo.to(self.token.span); + self.dcx().emit_err(errors::InvalidVariableDeclaration { + span: mut_let_span, + sub: errors::InvalidVariableDeclarationSub::SwitchMutLetOrder(mut_let_span), + }); + } + + Ok(Some(if self.token.is_keyword(kw::Let) { + self.parse_local_mk(lo, attrs, capture_semi, force_collect)? + } else if self.is_kw_followed_by_ident(kw::Mut) && self.may_recover() { + self.recover_stmt_local_after_let( + lo, + attrs, + errors::InvalidVariableDeclarationSub::MissingLet, + )? + } else if self.is_kw_followed_by_ident(kw::Auto) && self.may_recover() { + self.bump(); // `auto` + self.recover_stmt_local_after_let( + lo, + attrs, + errors::InvalidVariableDeclarationSub::UseLetNotAuto, + )? + } else if self.is_kw_followed_by_ident(sym::var) && self.may_recover() { + self.bump(); // `var` + self.recover_stmt_local_after_let( + lo, + attrs, + errors::InvalidVariableDeclarationSub::UseLetNotVar, + )? + } else if self.check_path() + && !self.token.is_qpath_start() + && !self.is_path_start_item() + && !self.is_builtin() + { + // We have avoided contextual keywords like `union`, items with `crate` visibility, + // or `auto trait` items. We aim to parse an arbitrary path `a::b` but not something + // that starts like a path (1 token), but it fact not a path. + // Also, we avoid stealing syntax from `parse_item_`. + match force_collect { + ForceCollect::Yes => { + self.collect_tokens_no_attrs(|this| this.parse_stmt_path_start(lo, attrs))? + } + ForceCollect::No => match self.parse_stmt_path_start(lo, attrs) { + Ok(stmt) => stmt, + Err(mut err) => { + self.suggest_add_missing_let_for_stmt(&mut err); + return Err(err); + } + }, + } + } else if let Some(item) = self.parse_item_common( + attrs.clone(), + false, + true, + FnParseMode { req_name: |_| true, req_body: true }, + force_collect, + )? { + // FIXME: Bad copy of attrs + self.mk_stmt(lo.to(item.span), StmtKind::Item(P(item))) + } else if self.eat(&token::Semi) { + // Do not attempt to parse an expression if we're done here. + self.error_outer_attrs(attrs); + self.mk_stmt(lo, StmtKind::Empty) + } else if self.token != token::CloseDelim(Delimiter::Brace) { + // Remainder are line-expr stmts. + let e = match force_collect { + ForceCollect::Yes => self.collect_tokens_no_attrs(|this| { + this.parse_expr_res(Restrictions::STMT_EXPR, Some(attrs)) + })?, + ForceCollect::No => self.parse_expr_res(Restrictions::STMT_EXPR, Some(attrs))?, + }; + if matches!(e.kind, ExprKind::Assign(..)) && self.eat_keyword(kw::Else) { + let bl = self.parse_block()?; + // Destructuring assignment ... else. + // This is not allowed, but point it out in a nice way. + self.dcx().emit_err(errors::AssignmentElseNotAllowed { span: e.span.to(bl.span) }); + } + self.mk_stmt(lo.to(e.span), StmtKind::Expr(e)) + } else { + self.error_outer_attrs(attrs); + return Ok(None); + })) + } + + fn parse_stmt_path_start(&mut self, lo: Span, attrs: AttrWrapper) -> PResult<'a, Stmt> { + let stmt = self.collect_tokens_trailing_token(attrs, ForceCollect::No, |this, attrs| { + let path = this.parse_path(PathStyle::Expr)?; + + if this.eat(&token::Not) { + let stmt_mac = this.parse_stmt_mac(lo, attrs, path)?; + if this.token == token::Semi { + return Ok((stmt_mac, TrailingToken::Semi)); + } else { + return Ok((stmt_mac, TrailingToken::None)); + } + } + + let expr = if this.eat(&token::OpenDelim(Delimiter::Brace)) { + this.parse_expr_struct(None, path, true)? + } else { + let hi = this.prev_token.span; + this.mk_expr(lo.to(hi), ExprKind::Path(None, path)) + }; + + let expr = this.with_res(Restrictions::STMT_EXPR, |this| { + this.parse_expr_dot_or_call_with(expr, lo, attrs) + })?; + // `DUMMY_SP` will get overwritten later in this function + Ok((this.mk_stmt(rustc_span::DUMMY_SP, StmtKind::Expr(expr)), TrailingToken::None)) + })?; + + if let StmtKind::Expr(expr) = stmt.kind { + // Perform this outside of the `collect_tokens_trailing_token` closure, + // since our outer attributes do not apply to this part of the expression + let expr = self.with_res(Restrictions::STMT_EXPR, |this| { + this.parse_expr_assoc_with( + 0, + LhsExpr::AlreadyParsed { expr, starts_statement: true }, + ) + })?; + Ok(self.mk_stmt(lo.to(self.prev_token.span), StmtKind::Expr(expr))) + } else { + Ok(stmt) + } + } + + /// Parses a statement macro `mac!(args)` provided a `path` representing `mac`. + /// At this point, the `!` token after the path has already been eaten. + fn parse_stmt_mac(&mut self, lo: Span, attrs: AttrVec, path: ast::Path) -> PResult<'a, Stmt> { + let args = self.parse_delim_args()?; + let hi = self.prev_token.span; + + let style = match args.delim { + Delimiter::Brace => MacStmtStyle::Braces, + _ => MacStmtStyle::NoBraces, + }; + + let mac = P(MacCall { path, args }); + + let kind = if (style == MacStmtStyle::Braces + && self.token != token::Dot + && self.token != token::Question) + || self.token == token::Semi + || self.token == token::Eof + { + StmtKind::MacCall(P(MacCallStmt { mac, style, attrs, tokens: None })) + } else { + // Since none of the above applied, this is an expression statement macro. + let e = self.mk_expr(lo.to(hi), ExprKind::MacCall(mac)); + let e = self.maybe_recover_from_bad_qpath(e)?; + let e = self.parse_expr_dot_or_call_with(e, lo, attrs)?; + let e = self.parse_expr_assoc_with( + 0, + LhsExpr::AlreadyParsed { expr: e, starts_statement: false }, + )?; + StmtKind::Expr(e) + }; + Ok(self.mk_stmt(lo.to(hi), kind)) + } + + /// Error on outer attributes in this context. + /// Also error if the previous token was a doc comment. + fn error_outer_attrs(&self, attrs: AttrWrapper) { + if !attrs.is_empty() + && let attrs @ [.., last] = &*attrs.take_for_recovery(self.psess) + { + if last.is_doc_comment() { + self.dcx().emit_err(errors::DocCommentDoesNotDocumentAnything { + span: last.span, + missing_comma: None, + }); + } else if attrs.iter().any(|a| a.style == AttrStyle::Outer) { + self.dcx().emit_err(errors::ExpectedStatementAfterOuterAttr { span: last.span }); + } + } + } + + fn recover_stmt_local_after_let( + &mut self, + lo: Span, + attrs: AttrWrapper, + subdiagnostic: fn(Span) -> errors::InvalidVariableDeclarationSub, + ) -> PResult<'a, Stmt> { + let stmt = + self.collect_tokens_trailing_token(attrs, ForceCollect::Yes, |this, attrs| { + let local = this.parse_local(attrs)?; + // FIXME - maybe capture semicolon in recovery? + Ok(( + this.mk_stmt(lo.to(this.prev_token.span), StmtKind::Let(local)), + TrailingToken::None, + )) + })?; + self.dcx() + .emit_err(errors::InvalidVariableDeclaration { span: lo, sub: subdiagnostic(lo) }); + Ok(stmt) + } + + fn parse_local_mk( + &mut self, + lo: Span, + attrs: AttrWrapper, + capture_semi: bool, + force_collect: ForceCollect, + ) -> PResult<'a, Stmt> { + self.collect_tokens_trailing_token(attrs, force_collect, |this, attrs| { + this.expect_keyword(kw::Let)?; + let local = this.parse_local(attrs)?; + let trailing = if capture_semi && this.token.kind == token::Semi { + TrailingToken::Semi + } else { + TrailingToken::None + }; + Ok((this.mk_stmt(lo.to(this.prev_token.span), StmtKind::Let(local)), trailing)) + }) + } + + /// Parses a local variable declaration. + fn parse_local(&mut self, attrs: AttrVec) -> PResult<'a, P<Local>> { + let lo = self.prev_token.span; + + if self.token.is_keyword(kw::Const) && self.look_ahead(1, |t| t.is_ident()) { + self.dcx().emit_err(errors::ConstLetMutuallyExclusive { span: lo.to(self.token.span) }); + self.bump(); + } + + let (pat, colon) = + self.parse_pat_before_ty(None, RecoverComma::Yes, PatternLocation::LetBinding)?; + + let (err, ty, colon_sp) = if colon { + // Save the state of the parser before parsing type normally, in case there is a `:` + // instead of an `=` typo. + let parser_snapshot_before_type = self.clone(); + let colon_sp = self.prev_token.span; + match self.parse_ty() { + Ok(ty) => (None, Some(ty), Some(colon_sp)), + Err(mut err) => { + err.span_label( + colon_sp, + format!( + "while parsing the type for {}", + pat.descr() + .map_or_else(|| "the binding".to_string(), |n| format!("`{n}`")) + ), + ); + // we use noexpect here because we don't actually expect Eq to be here + // but we are still checking for it in order to be able to handle it if + // it is there + let err = if self.check_noexpect(&token::Eq) { + err.emit(); + None + } else { + // Rewind to before attempting to parse the type and continue parsing. + let parser_snapshot_after_type = + mem::replace(self, parser_snapshot_before_type); + Some((parser_snapshot_after_type, colon_sp, err)) + }; + (err, None, Some(colon_sp)) + } + } + } else { + (None, None, None) + }; + let init = match (self.parse_initializer(err.is_some()), err) { + (Ok(init), None) => { + // init parsed, ty parsed + init + } + (Ok(init), Some((_, colon_sp, mut err))) => { + // init parsed, ty error + // Could parse the type as if it were the initializer, it is likely there was a + // typo in the code: `:` instead of `=`. Add suggestion and emit the error. + err.span_suggestion_short( + colon_sp, + "use `=` if you meant to assign", + " =", + Applicability::MachineApplicable, + ); + err.emit(); + // As this was parsed successfully, continue as if the code has been fixed for the + // rest of the file. It will still fail due to the emitted error, but we avoid + // extra noise. + init + } + (Err(init_err), Some((snapshot, _, ty_err))) => { + // init error, ty error + init_err.cancel(); + // Couldn't parse the type nor the initializer, only raise the type error and + // return to the parser state before parsing the type as the initializer. + // let x: <parse_error>; + *self = snapshot; + return Err(ty_err); + } + (Err(err), None) => { + // init error, ty parsed + // Couldn't parse the initializer and we're not attempting to recover a failed + // parse of the type, return the error. + return Err(err); + } + }; + let kind = match init { + None => LocalKind::Decl, + Some(init) => { + if self.eat_keyword(kw::Else) { + if self.token.is_keyword(kw::If) { + // `let...else if`. Emit the same error that `parse_block()` would, + // but explicitly point out that this pattern is not allowed. + let msg = "conditional `else if` is not supported for `let...else`"; + return Err(self.error_block_no_opening_brace_msg(Cow::from(msg))); + } + let els = self.parse_block()?; + self.check_let_else_init_bool_expr(&init); + self.check_let_else_init_trailing_brace(&init); + LocalKind::InitElse(init, els) + } else { + LocalKind::Init(init) + } + } + }; + let hi = if self.token == token::Semi { self.token.span } else { self.prev_token.span }; + Ok(P(ast::Local { + ty, + pat, + kind, + id: DUMMY_NODE_ID, + span: lo.to(hi), + colon_sp, + attrs, + tokens: None, + })) + } + + fn check_let_else_init_bool_expr(&self, init: &ast::Expr) { + if let ast::ExprKind::Binary(op, ..) = init.kind { + if op.node.is_lazy() { + self.dcx().emit_err(errors::InvalidExpressionInLetElse { + span: init.span, + operator: op.node.as_str(), + sugg: errors::WrapInParentheses::Expression { + left: init.span.shrink_to_lo(), + right: init.span.shrink_to_hi(), + }, + }); + } + } + } + + fn check_let_else_init_trailing_brace(&self, init: &ast::Expr) { + if let Some(trailing) = classify::expr_trailing_brace(init) { + let sugg = match &trailing.kind { + ExprKind::MacCall(mac) => errors::WrapInParentheses::MacroArgs { + left: mac.args.dspan.open, + right: mac.args.dspan.close, + }, + _ => errors::WrapInParentheses::Expression { + left: trailing.span.shrink_to_lo(), + right: trailing.span.shrink_to_hi(), + }, + }; + self.dcx().emit_err(errors::InvalidCurlyInLetElse { + span: trailing.span.with_lo(trailing.span.hi() - BytePos(1)), + sugg, + }); + } + } + + /// Parses the RHS of a local variable declaration (e.g., `= 14;`). + fn parse_initializer(&mut self, eq_optional: bool) -> PResult<'a, Option<P<Expr>>> { + let eq_consumed = match self.token.kind { + token::BinOpEq(..) => { + // Recover `let x <op>= 1` as `let x = 1` + self.dcx() + .emit_err(errors::CompoundAssignmentExpressionInLet { span: self.token.span }); + self.bump(); + true + } + _ => self.eat(&token::Eq), + }; + + Ok(if eq_consumed || eq_optional { Some(self.parse_expr()?) } else { None }) + } + + /// Parses a block. No inner attributes are allowed. + pub fn parse_block(&mut self) -> PResult<'a, P<Block>> { + let (attrs, block) = self.parse_inner_attrs_and_block()?; + if let [.., last] = &*attrs { + self.error_on_forbidden_inner_attr( + last.span, + super::attr::InnerAttrPolicy::Forbidden(Some( + InnerAttrForbiddenReason::InCodeBlock, + )), + ); + } + Ok(block) + } + + fn error_block_no_opening_brace_msg(&mut self, msg: Cow<'static, str>) -> Diag<'a> { + let sp = self.token.span; + let mut e = self.dcx().struct_span_err(sp, msg); + let do_not_suggest_help = self.token.is_keyword(kw::In) || self.token == token::Colon; + + // Check to see if the user has written something like + // + // if (cond) + // bar; + // + // which is valid in other languages, but not Rust. + match self.parse_stmt_without_recovery(false, ForceCollect::No) { + // If the next token is an open brace, e.g., we have: + // + // if expr other_expr { + // ^ ^ ^- lookahead(1) is a brace + // | |- current token is not "else" + // |- (statement we just parsed) + // + // the place-inside-a-block suggestion would be more likely wrong than right. + // + // FIXME(compiler-errors): this should probably parse an arbitrary expr and not + // just lookahead one token, so we can see if there's a brace after _that_, + // since we want to protect against: + // `if 1 1 + 1 {` being suggested as `if { 1 } 1 + 1 {` + // + + + Ok(Some(_)) + if (!self.token.is_keyword(kw::Else) + && self.look_ahead(1, |t| t == &token::OpenDelim(Delimiter::Brace))) + || do_not_suggest_help => {} + // Do not suggest `if foo println!("") {;}` (as would be seen in test for #46836). + Ok(Some(Stmt { kind: StmtKind::Empty, .. })) => {} + Ok(Some(stmt)) => { + let stmt_own_line = self.psess.source_map().is_line_before_span_empty(sp); + let stmt_span = if stmt_own_line && self.eat(&token::Semi) { + // Expand the span to include the semicolon. + stmt.span.with_hi(self.prev_token.span.hi()) + } else { + stmt.span + }; + e.multipart_suggestion( + "try placing this code inside a block", + vec![ + (stmt_span.shrink_to_lo(), "{ ".to_string()), + (stmt_span.shrink_to_hi(), " }".to_string()), + ], + // Speculative; has been misleading in the past (#46836). + Applicability::MaybeIncorrect, + ); + } + Err(e) => { + self.recover_stmt_(SemiColonMode::Break, BlockMode::Ignore); + e.cancel(); + } + _ => {} + } + e.span_label(sp, "expected `{`"); + e + } + + fn error_block_no_opening_brace<T>(&mut self) -> PResult<'a, T> { + let tok = super::token_descr(&self.token); + let msg = format!("expected `{{`, found {tok}"); + Err(self.error_block_no_opening_brace_msg(Cow::from(msg))) + } + + /// Parses a block. Inner attributes are allowed. + pub(super) fn parse_inner_attrs_and_block(&mut self) -> PResult<'a, (AttrVec, P<Block>)> { + self.parse_block_common(self.token.span, BlockCheckMode::Default, true) + } + + /// Parses a block. Inner attributes are allowed. + pub(super) fn parse_block_common( + &mut self, + lo: Span, + blk_mode: BlockCheckMode, + can_be_struct_literal: bool, + ) -> PResult<'a, (AttrVec, P<Block>)> { + maybe_whole!(self, NtBlock, |block| (AttrVec::new(), block)); + + let maybe_ident = self.prev_token.clone(); + self.maybe_recover_unexpected_block_label(); + if !self.eat(&token::OpenDelim(Delimiter::Brace)) { + return self.error_block_no_opening_brace(); + } + + let attrs = self.parse_inner_attributes()?; + let tail = match self.maybe_suggest_struct_literal( + lo, + blk_mode, + maybe_ident, + can_be_struct_literal, + ) { + Some(tail) => tail?, + None => self.parse_block_tail(lo, blk_mode, AttemptLocalParseRecovery::Yes)?, + }; + Ok((attrs, tail)) + } + + /// Parses the rest of a block expression or function body. + /// Precondition: already parsed the '{'. + pub(crate) fn parse_block_tail( + &mut self, + lo: Span, + s: BlockCheckMode, + recover: AttemptLocalParseRecovery, + ) -> PResult<'a, P<Block>> { + let mut stmts = ThinVec::new(); + let mut snapshot = None; + while !self.eat(&token::CloseDelim(Delimiter::Brace)) { + if self.token == token::Eof { + break; + } + if self.is_vcs_conflict_marker(&TokenKind::BinOp(token::Shl), &TokenKind::Lt) { + // Account for `<<<<<<<` diff markers. We can't proactively error here because + // that can be a valid path start, so we snapshot and reparse only we've + // encountered another parse error. + snapshot = Some(self.create_snapshot_for_diagnostic()); + } + let stmt = match self.parse_full_stmt(recover) { + Err(mut err) if recover.yes() => { + if let Some(ref mut snapshot) = snapshot { + snapshot.recover_vcs_conflict_marker(); + } + if self.token == token::Colon { + // if a previous and next token of the current one is + // integer literal (e.g. `1:42`), it's likely a range + // expression for Pythonistas and we can suggest so. + if self.prev_token.is_integer_lit() + && self.may_recover() + && self.look_ahead(1, |token| token.is_integer_lit()) + { + // FIXME(hkmatsumoto): Might be better to trigger + // this only when parsing an index expression. + err.span_suggestion_verbose( + self.token.span, + "you might have meant a range expression", + "..", + Applicability::MaybeIncorrect, + ); + } else { + // if next token is following a colon, it's likely a path + // and we can suggest a path separator + self.bump(); + if self.token.span.lo() == self.prev_token.span.hi() { + err.span_suggestion_verbose( + self.prev_token.span, + "maybe write a path separator here", + "::", + Applicability::MaybeIncorrect, + ); + } + if self.psess.unstable_features.is_nightly_build() { + // FIXME(Nilstrieb): Remove this again after a few months. + err.note("type ascription syntax has been removed, see issue #101728 <https://github.com/rust-lang/rust/issues/101728>"); + } + } + } + + let guar = err.emit(); + self.recover_stmt_(SemiColonMode::Ignore, BlockMode::Ignore); + Some(self.mk_stmt_err(self.token.span, guar)) + } + Ok(stmt) => stmt, + Err(err) => return Err(err), + }; + if let Some(stmt) = stmt { + stmts.push(stmt); + } else { + // Found only `;` or `}`. + continue; + }; + } + Ok(self.mk_block(stmts, s, lo.to(self.prev_token.span))) + } + + /// Parses a statement, including the trailing semicolon. + pub fn parse_full_stmt( + &mut self, + recover: AttemptLocalParseRecovery, + ) -> PResult<'a, Option<Stmt>> { + // Skip looking for a trailing semicolon when we have an interpolated statement. + maybe_whole!(self, NtStmt, |stmt| Some(stmt.into_inner())); + + let Some(mut stmt) = self.parse_stmt_without_recovery(true, ForceCollect::No)? else { + return Ok(None); + }; + + let mut eat_semi = true; + let mut add_semi_to_stmt = false; + + match &mut stmt.kind { + // Expression without semicolon. + StmtKind::Expr(expr) + if classify::expr_requires_semi_to_be_stmt(expr) + && !expr.attrs.is_empty() + && ![token::Eof, token::Semi, token::CloseDelim(Delimiter::Brace)] + .contains(&self.token.kind) => + { + // The user has written `#[attr] expr` which is unsupported. (#106020) + let guar = self.attr_on_non_tail_expr(&expr); + // We already emitted an error, so don't emit another type error + let sp = expr.span.to(self.prev_token.span); + *expr = self.mk_expr_err(sp, guar); + } + + // Expression without semicolon. + StmtKind::Expr(expr) + if self.token != token::Eof && classify::expr_requires_semi_to_be_stmt(expr) => + { + // Just check for errors and recover; do not eat semicolon yet. + + let expect_result = + self.expect_one_of(&[], &[token::Semi, token::CloseDelim(Delimiter::Brace)]); + + // Try to both emit a better diagnostic, and avoid further errors by replacing + // the `expr` with `ExprKind::Err`. + let replace_with_err = 'break_recover: { + match expect_result { + Ok(Recovered::No) => None, + Ok(Recovered::Yes(guar)) => { + // Skip type error to avoid extra errors. + Some(guar) + } + Err(e) => { + if self.recover_colon_as_semi() { + // recover_colon_as_semi has already emitted a nicer error. + e.delay_as_bug(); + add_semi_to_stmt = true; + eat_semi = false; + + break 'break_recover None; + } + + match &expr.kind { + ExprKind::Path(None, ast::Path { segments, .. }) + if segments.len() == 1 => + { + if self.token == token::Colon + && self.look_ahead(1, |token| { + token.is_whole_block() + || matches!( + token.kind, + token::Ident( + kw::For | kw::Loop | kw::While, + token::IdentIsRaw::No + ) | token::OpenDelim(Delimiter::Brace) + ) + }) + { + let snapshot = self.create_snapshot_for_diagnostic(); + let label = Label { + ident: Ident::from_str_and_span( + &format!("'{}", segments[0].ident), + segments[0].ident.span, + ), + }; + match self.parse_expr_labeled(label, false) { + Ok(labeled_expr) => { + e.cancel(); + self.dcx().emit_err(MalformedLoopLabel { + span: label.ident.span, + correct_label: label.ident, + }); + *expr = labeled_expr; + break 'break_recover None; + } + Err(err) => { + err.cancel(); + self.restore_snapshot(snapshot); + } + } + } + } + _ => {} + } + + let res = + self.check_mistyped_turbofish_with_multiple_type_params(e, expr); + + Some(if recover.no() { + res? + } else { + res.unwrap_or_else(|e| { + let guar = e.emit(); + self.recover_stmt(); + guar + }) + }) + } + } + }; + + if let Some(guar) = replace_with_err { + // We already emitted an error, so don't emit another type error + let sp = expr.span.to(self.prev_token.span); + *expr = self.mk_expr_err(sp, guar); + } + } + StmtKind::Expr(_) | StmtKind::MacCall(_) => {} + StmtKind::Let(local) if let Err(mut e) = self.expect_semi() => { + // We might be at the `,` in `let x = foo<bar, baz>;`. Try to recover. + match &mut local.kind { + LocalKind::Init(expr) | LocalKind::InitElse(expr, _) => { + self.check_mistyped_turbofish_with_multiple_type_params(e, expr)?; + // We found `foo<bar, baz>`, have we fully recovered? + self.expect_semi()?; + } + LocalKind::Decl => { + if let Some(colon_sp) = local.colon_sp { + e.span_label( + colon_sp, + format!( + "while parsing the type for {}", + local.pat.descr().map_or_else( + || "the binding".to_string(), + |n| format!("`{n}`") + ) + ), + ); + let suggest_eq = if self.token.kind == token::Dot + && let _ = self.bump() + && let mut snapshot = self.create_snapshot_for_diagnostic() + && let Ok(_) = snapshot + .parse_dot_suffix_expr( + colon_sp, + self.mk_expr_err( + colon_sp, + self.dcx() + .delayed_bug("error during `:` -> `=` recovery"), + ), + ) + .map_err(Diag::cancel) + { + true + } else if let Some(op) = self.check_assoc_op() + && op.node.can_continue_expr_unambiguously() + { + true + } else { + false + }; + if suggest_eq { + e.span_suggestion_short( + colon_sp, + "use `=` if you meant to assign", + "=", + Applicability::MaybeIncorrect, + ); + } + } + return Err(e); + } + } + eat_semi = false; + } + StmtKind::Empty | StmtKind::Item(_) | StmtKind::Let(_) | StmtKind::Semi(_) => { + eat_semi = false + } + } + + if add_semi_to_stmt || (eat_semi && self.eat(&token::Semi)) { + stmt = stmt.add_trailing_semicolon(); + } + + stmt.span = stmt.span.to(self.prev_token.span); + Ok(Some(stmt)) + } + + pub(super) fn mk_block( + &self, + stmts: ThinVec<Stmt>, + rules: BlockCheckMode, + span: Span, + ) -> P<Block> { + P(Block { + stmts, + id: DUMMY_NODE_ID, + rules, + span, + tokens: None, + could_be_bare_literal: false, + }) + } + + pub(super) fn mk_stmt(&self, span: Span, kind: StmtKind) -> Stmt { + Stmt { id: DUMMY_NODE_ID, kind, span } + } + + pub(super) fn mk_stmt_err(&self, span: Span, guar: ErrorGuaranteed) -> Stmt { + self.mk_stmt(span, StmtKind::Expr(self.mk_expr_err(span, guar))) + } + + pub(super) fn mk_block_err(&self, span: Span, guar: ErrorGuaranteed) -> P<Block> { + self.mk_block(thin_vec![self.mk_stmt_err(span, guar)], BlockCheckMode::Default, span) + } +} diff --git a/compiler/rustc_parse/src/parser/tests.rs b/compiler/rustc_parse/src/parser/tests.rs new file mode 100644 index 00000000000..a31e350541a --- /dev/null +++ b/compiler/rustc_parse/src/parser/tests.rs @@ -0,0 +1,1422 @@ +use crate::parser::ForceCollect; +use crate::{new_parser_from_source_str, parser::Parser, source_file_to_stream}; +use ast::token::IdentIsRaw; +use rustc_ast::ptr::P; +use rustc_ast::token::{self, Delimiter, Token}; +use rustc_ast::tokenstream::{DelimSpacing, DelimSpan, Spacing, TokenStream, TokenTree}; +use rustc_ast::visit; +use rustc_ast::{self as ast, PatKind}; +use rustc_ast_pretty::pprust::item_to_string; +use rustc_data_structures::sync::Lrc; +use rustc_errors::emitter::HumanEmitter; +use rustc_errors::{DiagCtxt, MultiSpan, PResult}; +use rustc_session::parse::ParseSess; +use rustc_span::create_default_session_globals_then; +use rustc_span::source_map::{FilePathMapping, SourceMap}; +use rustc_span::symbol::{kw, sym, Symbol}; +use rustc_span::{BytePos, FileName, Pos, Span}; +use std::io; +use std::io::prelude::*; +use std::iter::Peekable; +use std::path::{Path, PathBuf}; +use std::str; +use std::sync::{Arc, Mutex}; +use termcolor::WriteColor; + +fn psess() -> ParseSess { + ParseSess::new(vec![crate::DEFAULT_LOCALE_RESOURCE, crate::DEFAULT_LOCALE_RESOURCE]) +} + +/// Map string to parser (via tts). +fn string_to_parser(psess: &ParseSess, source_str: String) -> Parser<'_> { + new_parser_from_source_str(psess, PathBuf::from("bogofile").into(), source_str) +} + +fn create_test_handler() -> (DiagCtxt, Lrc<SourceMap>, Arc<Mutex<Vec<u8>>>) { + let output = Arc::new(Mutex::new(Vec::new())); + let source_map = Lrc::new(SourceMap::new(FilePathMapping::empty())); + let fallback_bundle = rustc_errors::fallback_fluent_bundle( + vec![crate::DEFAULT_LOCALE_RESOURCE, crate::DEFAULT_LOCALE_RESOURCE], + false, + ); + let emitter = HumanEmitter::new(Box::new(Shared { data: output.clone() }), fallback_bundle) + .sm(Some(source_map.clone())) + .diagnostic_width(Some(140)); + let dcx = DiagCtxt::new(Box::new(emitter)); + (dcx, source_map, output) +} + +/// Returns the result of parsing the given string via the given callback. +/// +/// If there are any errors, this will panic. +fn with_error_checking_parse<'a, T, F>(s: String, psess: &'a ParseSess, f: F) -> T +where + F: FnOnce(&mut Parser<'a>) -> PResult<'a, T>, +{ + let mut p = string_to_parser(&psess, s); + let x = f(&mut p).unwrap(); + p.psess.dcx.abort_if_errors(); + x +} + +/// Verifies that parsing the given string using the given callback will +/// generate an error that contains the given text. +fn with_expected_parse_error<T, F>(source_str: &str, expected_output: &str, f: F) +where + F: for<'a> FnOnce(&mut Parser<'a>) -> PResult<'a, T>, +{ + let (handler, source_map, output) = create_test_handler(); + let psess = ParseSess::with_dcx(handler, source_map); + let mut p = string_to_parser(&psess, source_str.to_string()); + let result = f(&mut p); + assert!(result.is_ok()); + + let bytes = output.lock().unwrap(); + let actual_output = str::from_utf8(&bytes).unwrap(); + println!("expected output:\n------\n{}------", expected_output); + println!("actual output:\n------\n{}------", actual_output); + + assert!(actual_output.contains(expected_output)) +} + +/// Maps a string to tts, using a made-up filename. +pub(crate) fn string_to_stream(source_str: String) -> TokenStream { + let psess = psess(); + source_file_to_stream( + &psess, + psess.source_map().new_source_file(PathBuf::from("bogofile").into(), source_str), + None, + ) +} + +/// Parses a string, returns a crate. +pub(crate) fn string_to_crate(source_str: String) -> ast::Crate { + let psess = psess(); + with_error_checking_parse(source_str, &psess, |p| p.parse_crate_mod()) +} + +/// Does the given string match the pattern? whitespace in the first string +/// may be deleted or replaced with other whitespace to match the pattern. +/// This function is relatively Unicode-ignorant; fortunately, the careful design +/// of UTF-8 mitigates this ignorance. It doesn't do NKF-normalization(?). +pub(crate) fn matches_codepattern(a: &str, b: &str) -> bool { + let mut a_iter = a.chars().peekable(); + let mut b_iter = b.chars().peekable(); + + loop { + let (a, b) = match (a_iter.peek(), b_iter.peek()) { + (None, None) => return true, + (None, _) => return false, + (Some(&a), None) => { + if rustc_lexer::is_whitespace(a) { + break; // Trailing whitespace check is out of loop for borrowck. + } else { + return false; + } + } + (Some(&a), Some(&b)) => (a, b), + }; + + if rustc_lexer::is_whitespace(a) && rustc_lexer::is_whitespace(b) { + // Skip whitespace for `a` and `b`. + scan_for_non_ws_or_end(&mut a_iter); + scan_for_non_ws_or_end(&mut b_iter); + } else if rustc_lexer::is_whitespace(a) { + // Skip whitespace for `a`. + scan_for_non_ws_or_end(&mut a_iter); + } else if a == b { + a_iter.next(); + b_iter.next(); + } else { + return false; + } + } + + // Check if a has *only* trailing whitespace. + a_iter.all(rustc_lexer::is_whitespace) +} + +/// Advances the given peekable `Iterator` until it reaches a non-whitespace character. +fn scan_for_non_ws_or_end<I: Iterator<Item = char>>(iter: &mut Peekable<I>) { + while iter.peek().copied().is_some_and(rustc_lexer::is_whitespace) { + iter.next(); + } +} + +/// Identifies a position in the text by the n'th occurrence of a string. +struct Position { + string: &'static str, + count: usize, +} + +struct SpanLabel { + start: Position, + end: Position, + label: &'static str, +} + +struct Shared<T: Write> { + data: Arc<Mutex<T>>, +} + +impl<T: Write> WriteColor for Shared<T> { + fn supports_color(&self) -> bool { + false + } + + fn set_color(&mut self, _spec: &termcolor::ColorSpec) -> io::Result<()> { + Ok(()) + } + + fn reset(&mut self) -> io::Result<()> { + Ok(()) + } +} + +impl<T: Write> Write for Shared<T> { + fn write(&mut self, buf: &[u8]) -> io::Result<usize> { + self.data.lock().unwrap().write(buf) + } + + fn flush(&mut self) -> io::Result<()> { + self.data.lock().unwrap().flush() + } +} + +#[allow(rustc::untranslatable_diagnostic)] // no translation needed for tests +fn test_harness(file_text: &str, span_labels: Vec<SpanLabel>, expected_output: &str) { + create_default_session_globals_then(|| { + let (handler, source_map, output) = create_test_handler(); + source_map.new_source_file(Path::new("test.rs").to_owned().into(), file_text.to_owned()); + + let primary_span = make_span(&file_text, &span_labels[0].start, &span_labels[0].end); + let mut msp = MultiSpan::from_span(primary_span); + for span_label in span_labels { + let span = make_span(&file_text, &span_label.start, &span_label.end); + msp.push_span_label(span, span_label.label); + println!("span: {:?} label: {:?}", span, span_label.label); + println!("text: {:?}", source_map.span_to_snippet(span)); + } + + handler.span_err(msp, "foo"); + + assert!( + expected_output.chars().next() == Some('\n'), + "expected output should begin with newline" + ); + let expected_output = &expected_output[1..]; + + let bytes = output.lock().unwrap(); + let actual_output = str::from_utf8(&bytes).unwrap(); + println!("expected output:\n------\n{}------", expected_output); + println!("actual output:\n------\n{}------", actual_output); + + assert!(expected_output == actual_output) + }) +} + +fn make_span(file_text: &str, start: &Position, end: &Position) -> Span { + let start = make_pos(file_text, start); + let end = make_pos(file_text, end) + end.string.len(); // just after matching thing ends + assert!(start <= end); + Span::with_root_ctxt(BytePos(start as u32), BytePos(end as u32)) +} + +fn make_pos(file_text: &str, pos: &Position) -> usize { + let mut remainder = file_text; + let mut offset = 0; + for _ in 0..pos.count { + if let Some(n) = remainder.find(&pos.string) { + offset += n; + remainder = &remainder[n + 1..]; + } else { + panic!("failed to find {} instances of {:?} in {:?}", pos.count, pos.string, file_text); + } + } + offset +} + +#[test] +fn ends_on_col0() { + test_harness( + r#" +fn foo() { +} +"#, + vec![SpanLabel { + start: Position { string: "{", count: 1 }, + end: Position { string: "}", count: 1 }, + label: "test", + }], + r#" +error: foo + --> test.rs:2:10 + | +2 | fn foo() { + | __________^ +3 | | } + | |_^ test + +"#, + ); +} + +#[test] +fn ends_on_col2() { + test_harness( + r#" +fn foo() { + + + } +"#, + vec![SpanLabel { + start: Position { string: "{", count: 1 }, + end: Position { string: "}", count: 1 }, + label: "test", + }], + r#" +error: foo + --> test.rs:2:10 + | +2 | fn foo() { + | __________^ +... | +5 | | } + | |___^ test + +"#, + ); +} +#[test] +fn non_nested() { + test_harness( + r#" +fn foo() { + X0 Y0 + X1 Y1 + X2 Y2 +} +"#, + vec![ + SpanLabel { + start: Position { string: "X0", count: 1 }, + end: Position { string: "X2", count: 1 }, + label: "`X` is a good letter", + }, + SpanLabel { + start: Position { string: "Y0", count: 1 }, + end: Position { string: "Y2", count: 1 }, + label: "`Y` is a good letter too", + }, + ], + r#" +error: foo + --> test.rs:3:3 + | +3 | X0 Y0 + | ___^__- + | |___| + | || +4 | || X1 Y1 +5 | || X2 Y2 + | ||____^__- `Y` is a good letter too + | |_____| + | `X` is a good letter + +"#, + ); +} + +#[test] +fn nested() { + test_harness( + r#" +fn foo() { + X0 Y0 + Y1 X1 +} +"#, + vec![ + SpanLabel { + start: Position { string: "X0", count: 1 }, + end: Position { string: "X1", count: 1 }, + label: "`X` is a good letter", + }, + SpanLabel { + start: Position { string: "Y0", count: 1 }, + end: Position { string: "Y1", count: 1 }, + label: "`Y` is a good letter too", + }, + ], + r#" +error: foo + --> test.rs:3:3 + | +3 | X0 Y0 + | ___^__- + | |___| + | || +4 | || Y1 X1 + | ||____-__^ `X` is a good letter + | |____| + | `Y` is a good letter too + +"#, + ); +} + +#[test] +fn different_overlap() { + test_harness( + r#" +fn foo() { + X0 Y0 Z0 + X1 Y1 Z1 + X2 Y2 Z2 + X3 Y3 Z3 +} +"#, + vec![ + SpanLabel { + start: Position { string: "Y0", count: 1 }, + end: Position { string: "X2", count: 1 }, + label: "`X` is a good letter", + }, + SpanLabel { + start: Position { string: "Z1", count: 1 }, + end: Position { string: "X3", count: 1 }, + label: "`Y` is a good letter too", + }, + ], + r#" +error: foo + --> test.rs:3:6 + | +3 | X0 Y0 Z0 + | _______^ +4 | | X1 Y1 Z1 + | | _________- +5 | || X2 Y2 Z2 + | ||____^ `X` is a good letter +6 | | X3 Y3 Z3 + | |____- `Y` is a good letter too + +"#, + ); +} + +#[test] +fn triple_overlap() { + test_harness( + r#" +fn foo() { + X0 Y0 Z0 + X1 Y1 Z1 + X2 Y2 Z2 +} +"#, + vec![ + SpanLabel { + start: Position { string: "X0", count: 1 }, + end: Position { string: "X2", count: 1 }, + label: "`X` is a good letter", + }, + SpanLabel { + start: Position { string: "Y0", count: 1 }, + end: Position { string: "Y2", count: 1 }, + label: "`Y` is a good letter too", + }, + SpanLabel { + start: Position { string: "Z0", count: 1 }, + end: Position { string: "Z2", count: 1 }, + label: "`Z` label", + }, + ], + r#" +error: foo + --> test.rs:3:3 + | +3 | X0 Y0 Z0 + | ___^__-__- + | |___|__| + | ||___| + | ||| +4 | ||| X1 Y1 Z1 +5 | ||| X2 Y2 Z2 + | |||____^__-__- `Z` label + | ||_____|__| + | |______| `Y` is a good letter too + | `X` is a good letter + +"#, + ); +} + +#[test] +fn triple_exact_overlap() { + test_harness( + r#" +fn foo() { + X0 Y0 Z0 + X1 Y1 Z1 + X2 Y2 Z2 +} +"#, + vec![ + SpanLabel { + start: Position { string: "X0", count: 1 }, + end: Position { string: "X2", count: 1 }, + label: "`X` is a good letter", + }, + SpanLabel { + start: Position { string: "X0", count: 1 }, + end: Position { string: "X2", count: 1 }, + label: "`Y` is a good letter too", + }, + SpanLabel { + start: Position { string: "X0", count: 1 }, + end: Position { string: "X2", count: 1 }, + label: "`Z` label", + }, + ], + r#" +error: foo + --> test.rs:3:3 + | +3 | / X0 Y0 Z0 +4 | | X1 Y1 Z1 +5 | | X2 Y2 Z2 + | | ^ + | | | + | | `X` is a good letter + | |____`Y` is a good letter too + | `Z` label + +"#, + ); +} + +#[test] +fn minimum_depth() { + test_harness( + r#" +fn foo() { + X0 Y0 Z0 + X1 Y1 Z1 + X2 Y2 Z2 + X3 Y3 Z3 +} +"#, + vec![ + SpanLabel { + start: Position { string: "Y0", count: 1 }, + end: Position { string: "X1", count: 1 }, + label: "`X` is a good letter", + }, + SpanLabel { + start: Position { string: "Y1", count: 1 }, + end: Position { string: "Z2", count: 1 }, + label: "`Y` is a good letter too", + }, + SpanLabel { + start: Position { string: "X2", count: 1 }, + end: Position { string: "Y3", count: 1 }, + label: "`Z`", + }, + ], + r#" +error: foo + --> test.rs:3:6 + | +3 | X0 Y0 Z0 + | _______^ +4 | | X1 Y1 Z1 + | | ____^_- + | ||____| + | | `X` is a good letter +5 | | X2 Y2 Z2 + | |___-______- `Y` is a good letter too + | ___| + | | +6 | | X3 Y3 Z3 + | |_______- `Z` + +"#, + ); +} + +#[test] +fn non_overlapping() { + test_harness( + r#" +fn foo() { + X0 Y0 Z0 + X1 Y1 Z1 + X2 Y2 Z2 + X3 Y3 Z3 +} +"#, + vec![ + SpanLabel { + start: Position { string: "X0", count: 1 }, + end: Position { string: "X1", count: 1 }, + label: "`X` is a good letter", + }, + SpanLabel { + start: Position { string: "Y2", count: 1 }, + end: Position { string: "Z3", count: 1 }, + label: "`Y` is a good letter too", + }, + ], + r#" +error: foo + --> test.rs:3:3 + | +3 | / X0 Y0 Z0 +4 | | X1 Y1 Z1 + | |____^ `X` is a good letter +5 | X2 Y2 Z2 + | ______- +6 | | X3 Y3 Z3 + | |__________- `Y` is a good letter too + +"#, + ); +} + +#[test] +fn overlapping_start_and_end() { + test_harness( + r#" +fn foo() { + X0 Y0 Z0 + X1 Y1 Z1 + X2 Y2 Z2 + X3 Y3 Z3 +} +"#, + vec![ + SpanLabel { + start: Position { string: "Y0", count: 1 }, + end: Position { string: "X1", count: 1 }, + label: "`X` is a good letter", + }, + SpanLabel { + start: Position { string: "Z1", count: 1 }, + end: Position { string: "Z3", count: 1 }, + label: "`Y` is a good letter too", + }, + ], + r#" +error: foo + --> test.rs:3:6 + | +3 | X0 Y0 Z0 + | _______^ +4 | | X1 Y1 Z1 + | | ____^____- + | ||____| + | | `X` is a good letter +5 | | X2 Y2 Z2 +6 | | X3 Y3 Z3 + | |__________- `Y` is a good letter too + +"#, + ); +} + +#[test] +fn multiple_labels_primary_without_message() { + test_harness( + r#" +fn foo() { + a { b { c } d } +} +"#, + vec![ + SpanLabel { + start: Position { string: "b", count: 1 }, + end: Position { string: "}", count: 1 }, + label: "", + }, + SpanLabel { + start: Position { string: "a", count: 1 }, + end: Position { string: "d", count: 1 }, + label: "`a` is a good letter", + }, + SpanLabel { + start: Position { string: "c", count: 1 }, + end: Position { string: "c", count: 1 }, + label: "", + }, + ], + r#" +error: foo + --> test.rs:3:7 + | +3 | a { b { c } d } + | ----^^^^-^^-- `a` is a good letter + +"#, + ); +} + +#[test] +fn multiple_labels_secondary_without_message() { + test_harness( + r#" +fn foo() { + a { b { c } d } +} +"#, + vec![ + SpanLabel { + start: Position { string: "a", count: 1 }, + end: Position { string: "d", count: 1 }, + label: "`a` is a good letter", + }, + SpanLabel { + start: Position { string: "b", count: 1 }, + end: Position { string: "}", count: 1 }, + label: "", + }, + ], + r#" +error: foo + --> test.rs:3:3 + | +3 | a { b { c } d } + | ^^^^-------^^ `a` is a good letter + +"#, + ); +} + +#[test] +fn multiple_labels_primary_without_message_2() { + test_harness( + r#" +fn foo() { + a { b { c } d } +} +"#, + vec![ + SpanLabel { + start: Position { string: "b", count: 1 }, + end: Position { string: "}", count: 1 }, + label: "`b` is a good letter", + }, + SpanLabel { + start: Position { string: "a", count: 1 }, + end: Position { string: "d", count: 1 }, + label: "", + }, + SpanLabel { + start: Position { string: "c", count: 1 }, + end: Position { string: "c", count: 1 }, + label: "", + }, + ], + r#" +error: foo + --> test.rs:3:7 + | +3 | a { b { c } d } + | ----^^^^-^^-- + | | + | `b` is a good letter + +"#, + ); +} + +#[test] +fn multiple_labels_secondary_without_message_2() { + test_harness( + r#" +fn foo() { + a { b { c } d } +} +"#, + vec![ + SpanLabel { + start: Position { string: "a", count: 1 }, + end: Position { string: "d", count: 1 }, + label: "", + }, + SpanLabel { + start: Position { string: "b", count: 1 }, + end: Position { string: "}", count: 1 }, + label: "`b` is a good letter", + }, + ], + r#" +error: foo + --> test.rs:3:3 + | +3 | a { b { c } d } + | ^^^^-------^^ + | | + | `b` is a good letter + +"#, + ); +} + +#[test] +fn multiple_labels_secondary_without_message_3() { + test_harness( + r#" +fn foo() { + a bc d +} +"#, + vec![ + SpanLabel { + start: Position { string: "a", count: 1 }, + end: Position { string: "b", count: 1 }, + label: "`a` is a good letter", + }, + SpanLabel { + start: Position { string: "c", count: 1 }, + end: Position { string: "d", count: 1 }, + label: "", + }, + ], + r#" +error: foo + --> test.rs:3:3 + | +3 | a bc d + | ^^^^---- + | | + | `a` is a good letter + +"#, + ); +} + +#[test] +fn multiple_labels_without_message() { + test_harness( + r#" +fn foo() { + a { b { c } d } +} +"#, + vec![ + SpanLabel { + start: Position { string: "a", count: 1 }, + end: Position { string: "d", count: 1 }, + label: "", + }, + SpanLabel { + start: Position { string: "b", count: 1 }, + end: Position { string: "}", count: 1 }, + label: "", + }, + ], + r#" +error: foo + --> test.rs:3:3 + | +3 | a { b { c } d } + | ^^^^-------^^ + +"#, + ); +} + +#[test] +fn multiple_labels_without_message_2() { + test_harness( + r#" +fn foo() { + a { b { c } d } +} +"#, + vec![ + SpanLabel { + start: Position { string: "b", count: 1 }, + end: Position { string: "}", count: 1 }, + label: "", + }, + SpanLabel { + start: Position { string: "a", count: 1 }, + end: Position { string: "d", count: 1 }, + label: "", + }, + SpanLabel { + start: Position { string: "c", count: 1 }, + end: Position { string: "c", count: 1 }, + label: "", + }, + ], + r#" +error: foo + --> test.rs:3:7 + | +3 | a { b { c } d } + | ----^^^^-^^-- + +"#, + ); +} + +#[test] +fn multiple_labels_with_message() { + test_harness( + r#" +fn foo() { + a { b { c } d } +} +"#, + vec![ + SpanLabel { + start: Position { string: "a", count: 1 }, + end: Position { string: "d", count: 1 }, + label: "`a` is a good letter", + }, + SpanLabel { + start: Position { string: "b", count: 1 }, + end: Position { string: "}", count: 1 }, + label: "`b` is a good letter", + }, + ], + r#" +error: foo + --> test.rs:3:3 + | +3 | a { b { c } d } + | ^^^^-------^^ + | | | + | | `b` is a good letter + | `a` is a good letter + +"#, + ); +} + +#[test] +fn single_label_with_message() { + test_harness( + r#" +fn foo() { + a { b { c } d } +} +"#, + vec![SpanLabel { + start: Position { string: "a", count: 1 }, + end: Position { string: "d", count: 1 }, + label: "`a` is a good letter", + }], + r#" +error: foo + --> test.rs:3:3 + | +3 | a { b { c } d } + | ^^^^^^^^^^^^^ `a` is a good letter + +"#, + ); +} + +#[test] +fn single_label_without_message() { + test_harness( + r#" +fn foo() { + a { b { c } d } +} +"#, + vec![SpanLabel { + start: Position { string: "a", count: 1 }, + end: Position { string: "d", count: 1 }, + label: "", + }], + r#" +error: foo + --> test.rs:3:3 + | +3 | a { b { c } d } + | ^^^^^^^^^^^^^ + +"#, + ); +} + +#[test] +fn long_snippet() { + test_harness( + r#" +fn foo() { + X0 Y0 Z0 + X1 Y1 Z1 +1 +2 +3 +4 +5 +6 +7 +8 +9 +10 + X2 Y2 Z2 + X3 Y3 Z3 +} +"#, + vec![ + SpanLabel { + start: Position { string: "Y0", count: 1 }, + end: Position { string: "X1", count: 1 }, + label: "`X` is a good letter", + }, + SpanLabel { + start: Position { string: "Z1", count: 1 }, + end: Position { string: "Z3", count: 1 }, + label: "`Y` is a good letter too", + }, + ], + r#" +error: foo + --> test.rs:3:6 + | +3 | X0 Y0 Z0 + | _______^ +4 | | X1 Y1 Z1 + | | ____^____- + | ||____| + | | `X` is a good letter +5 | | 1 +6 | | 2 +7 | | 3 +... | +15 | | X2 Y2 Z2 +16 | | X3 Y3 Z3 + | |__________- `Y` is a good letter too + +"#, + ); +} + +#[test] +fn long_snippet_multiple_spans() { + test_harness( + r#" +fn foo() { + X0 Y0 Z0 +1 +2 +3 + X1 Y1 Z1 +4 +5 +6 + X2 Y2 Z2 +7 +8 +9 +10 + X3 Y3 Z3 +} +"#, + vec![ + SpanLabel { + start: Position { string: "Y0", count: 1 }, + end: Position { string: "Y3", count: 1 }, + label: "`Y` is a good letter", + }, + SpanLabel { + start: Position { string: "Z1", count: 1 }, + end: Position { string: "Z2", count: 1 }, + label: "`Z` is a good letter too", + }, + ], + r#" +error: foo + --> test.rs:3:6 + | +3 | X0 Y0 Z0 + | _______^ +4 | | 1 +5 | | 2 +6 | | 3 +7 | | X1 Y1 Z1 + | | _________- +8 | || 4 +9 | || 5 +10 | || 6 +11 | || X2 Y2 Z2 + | ||__________- `Z` is a good letter too +... | +15 | | 10 +16 | | X3 Y3 Z3 + | |________^ `Y` is a good letter + +"#, + ); +} + +/// Parses an item. +/// +/// Returns `Ok(Some(item))` when successful, `Ok(None)` when no item was found, and `Err` +/// when a syntax error occurred. +fn parse_item_from_source_str( + name: FileName, + source: String, + psess: &ParseSess, +) -> PResult<'_, Option<P<ast::Item>>> { + new_parser_from_source_str(psess, name, source).parse_item(ForceCollect::No) +} + +// Produces a `rustc_span::span`. +fn sp(a: u32, b: u32) -> Span { + Span::with_root_ctxt(BytePos(a), BytePos(b)) +} + +/// Parses a string, return an expression. +fn string_to_expr(source_str: String) -> P<ast::Expr> { + with_error_checking_parse(source_str, &psess(), |p| p.parse_expr()) +} + +/// Parses a string, returns an item. +fn string_to_item(source_str: String) -> Option<P<ast::Item>> { + with_error_checking_parse(source_str, &psess(), |p| p.parse_item(ForceCollect::No)) +} + +#[test] +fn bad_path_expr_1() { + // This should trigger error: expected identifier, found keyword `return` + create_default_session_globals_then(|| { + with_expected_parse_error( + "::abc::def::return", + "expected identifier, found keyword `return`", + |p| p.parse_expr(), + ); + }) +} + +// Checks the token-tree-ization of macros. +#[test] +fn string_to_tts_macro() { + create_default_session_globals_then(|| { + let stream = string_to_stream("macro_rules! zip (($a)=>($a))".to_string()); + let tts = &stream.trees().collect::<Vec<_>>()[..]; + + match tts { + [ + TokenTree::Token( + Token { kind: token::Ident(name_macro_rules, IdentIsRaw::No), .. }, + _, + ), + TokenTree::Token(Token { kind: token::Not, .. }, _), + TokenTree::Token(Token { kind: token::Ident(name_zip, IdentIsRaw::No), .. }, _), + TokenTree::Delimited(.., macro_delim, macro_tts), + ] if name_macro_rules == &kw::MacroRules && name_zip.as_str() == "zip" => { + let tts = ¯o_tts.trees().collect::<Vec<_>>(); + match &tts[..] { + [ + TokenTree::Delimited(.., first_delim, first_tts), + TokenTree::Token(Token { kind: token::FatArrow, .. }, _), + TokenTree::Delimited(.., second_delim, second_tts), + ] if macro_delim == &Delimiter::Parenthesis => { + let tts = &first_tts.trees().collect::<Vec<_>>(); + match &tts[..] { + [ + TokenTree::Token(Token { kind: token::Dollar, .. }, _), + TokenTree::Token( + Token { kind: token::Ident(name, IdentIsRaw::No), .. }, + _, + ), + ] if first_delim == &Delimiter::Parenthesis && name.as_str() == "a" => { + } + _ => panic!("value 3: {:?} {:?}", first_delim, first_tts), + } + let tts = &second_tts.trees().collect::<Vec<_>>(); + match &tts[..] { + [ + TokenTree::Token(Token { kind: token::Dollar, .. }, _), + TokenTree::Token( + Token { kind: token::Ident(name, IdentIsRaw::No), .. }, + _, + ), + ] if second_delim == &Delimiter::Parenthesis + && name.as_str() == "a" => {} + _ => panic!("value 4: {:?} {:?}", second_delim, second_tts), + } + } + _ => panic!("value 2: {:?} {:?}", macro_delim, macro_tts), + } + } + _ => panic!("value: {:?}", tts), + } + }) +} + +#[test] +fn string_to_tts_1() { + create_default_session_globals_then(|| { + let tts = string_to_stream("fn a(b: i32) { b; }".to_string()); + + let expected = TokenStream::new(vec![ + TokenTree::token_alone(token::Ident(kw::Fn, IdentIsRaw::No), sp(0, 2)), + TokenTree::token_joint_hidden( + token::Ident(Symbol::intern("a"), IdentIsRaw::No), + sp(3, 4), + ), + TokenTree::Delimited( + DelimSpan::from_pair(sp(4, 5), sp(11, 12)), + // `JointHidden` because the `(` is followed immediately by + // `b`, `Alone` because the `)` is followed by whitespace. + DelimSpacing::new(Spacing::JointHidden, Spacing::Alone), + Delimiter::Parenthesis, + TokenStream::new(vec![ + TokenTree::token_joint( + token::Ident(Symbol::intern("b"), IdentIsRaw::No), + sp(5, 6), + ), + TokenTree::token_alone(token::Colon, sp(6, 7)), + // `JointHidden` because the `i32` is immediately followed by the `)`. + TokenTree::token_joint_hidden( + token::Ident(sym::i32, IdentIsRaw::No), + sp(8, 11), + ), + ]) + .into(), + ), + TokenTree::Delimited( + DelimSpan::from_pair(sp(13, 14), sp(18, 19)), + // First `Alone` because the `{` is followed by whitespace, + // second `Alone` because the `}` is followed immediately by + // EOF. + DelimSpacing::new(Spacing::Alone, Spacing::Alone), + Delimiter::Brace, + TokenStream::new(vec![ + TokenTree::token_joint( + token::Ident(Symbol::intern("b"), IdentIsRaw::No), + sp(15, 16), + ), + // `Alone` because the `;` is followed by whitespace. + TokenTree::token_alone(token::Semi, sp(16, 17)), + ]) + .into(), + ), + ]); + + assert_eq!(tts, expected); + }) +} + +#[test] +fn parse_use() { + create_default_session_globals_then(|| { + let use_s = "use foo::bar::baz;"; + let vitem = string_to_item(use_s.to_string()).unwrap(); + let vitem_s = item_to_string(&vitem); + assert_eq!(&vitem_s[..], use_s); + + let use_s = "use foo::bar as baz;"; + let vitem = string_to_item(use_s.to_string()).unwrap(); + let vitem_s = item_to_string(&vitem); + assert_eq!(&vitem_s[..], use_s); + }) +} + +#[test] +fn parse_extern_crate() { + create_default_session_globals_then(|| { + let ex_s = "extern crate foo;"; + let vitem = string_to_item(ex_s.to_string()).unwrap(); + let vitem_s = item_to_string(&vitem); + assert_eq!(&vitem_s[..], ex_s); + + let ex_s = "extern crate foo as bar;"; + let vitem = string_to_item(ex_s.to_string()).unwrap(); + let vitem_s = item_to_string(&vitem); + assert_eq!(&vitem_s[..], ex_s); + }) +} + +fn get_spans_of_pat_idents(src: &str) -> Vec<Span> { + let item = string_to_item(src.to_string()).unwrap(); + + struct PatIdentVisitor { + spans: Vec<Span>, + } + impl<'a> visit::Visitor<'a> for PatIdentVisitor { + fn visit_pat(&mut self, p: &'a ast::Pat) { + match &p.kind { + PatKind::Ident(_, ident, _) => { + self.spans.push(ident.span); + } + _ => { + visit::walk_pat(self, p); + } + } + } + } + let mut v = PatIdentVisitor { spans: Vec::new() }; + visit::walk_item(&mut v, &item); + return v.spans; +} + +#[test] +fn span_of_self_arg_pat_idents_are_correct() { + create_default_session_globals_then(|| { + let srcs = [ + "impl z { fn a (&self, &myarg: i32) {} }", + "impl z { fn a (&mut self, &myarg: i32) {} }", + "impl z { fn a (&'a self, &myarg: i32) {} }", + "impl z { fn a (self, &myarg: i32) {} }", + "impl z { fn a (self: Foo, &myarg: i32) {} }", + ]; + + for src in srcs { + let spans = get_spans_of_pat_idents(src); + let (lo, hi) = (spans[0].lo(), spans[0].hi()); + assert!( + "self" == &src[lo.to_usize()..hi.to_usize()], + "\"{}\" != \"self\". src=\"{}\"", + &src[lo.to_usize()..hi.to_usize()], + src + ) + } + }) +} + +#[test] +fn parse_exprs() { + create_default_session_globals_then(|| { + // just make sure that they parse.... + string_to_expr("3 + 4".to_string()); + string_to_expr("a::z.froob(b,&(987+3))".to_string()); + }) +} + +#[test] +fn attrs_fix_bug() { + create_default_session_globals_then(|| { + string_to_item( + "pub fn mk_file_writer(path: &Path, flags: &[FileFlag]) + -> Result<Box<Writer>, String> { +#[cfg(windows)] +fn wb() -> c_int { + (O_WRONLY | libc::consts::os::extra::O_BINARY) as c_int +} + +#[cfg(unix)] +fn wb() -> c_int { O_WRONLY as c_int } + +let mut fflags: c_int = wb(); +}" + .to_string(), + ); + }) +} + +#[test] +fn crlf_doc_comments() { + create_default_session_globals_then(|| { + let psess = psess(); + + let name_1 = FileName::Custom("crlf_source_1".to_string()); + let source = "/// doc comment\r\nfn foo() {}".to_string(); + let item = parse_item_from_source_str(name_1, source, &psess).unwrap().unwrap(); + let doc = item.attrs.iter().filter_map(|at| at.doc_str()).next().unwrap(); + assert_eq!(doc.as_str(), " doc comment"); + + let name_2 = FileName::Custom("crlf_source_2".to_string()); + let source = "/// doc comment\r\n/// line 2\r\nfn foo() {}".to_string(); + let item = parse_item_from_source_str(name_2, source, &psess).unwrap().unwrap(); + let docs = item.attrs.iter().filter_map(|at| at.doc_str()).collect::<Vec<_>>(); + let b: &[_] = &[Symbol::intern(" doc comment"), Symbol::intern(" line 2")]; + assert_eq!(&docs[..], b); + + let name_3 = FileName::Custom("clrf_source_3".to_string()); + let source = "/** doc comment\r\n * with CRLF */\r\nfn foo() {}".to_string(); + let item = parse_item_from_source_str(name_3, source, &psess).unwrap().unwrap(); + let doc = item.attrs.iter().filter_map(|at| at.doc_str()).next().unwrap(); + assert_eq!(doc.as_str(), " doc comment\n * with CRLF "); + }); +} + +#[test] +fn ttdelim_span() { + fn parse_expr_from_source_str( + name: FileName, + source: String, + psess: &ParseSess, + ) -> PResult<'_, P<ast::Expr>> { + new_parser_from_source_str(psess, name, source).parse_expr() + } + + create_default_session_globals_then(|| { + let psess = psess(); + let expr = parse_expr_from_source_str( + PathBuf::from("foo").into(), + "foo!( fn main() { body } )".to_string(), + &psess, + ) + .unwrap(); + + let ast::ExprKind::MacCall(mac) = &expr.kind else { panic!("not a macro") }; + let span = mac.args.tokens.trees().last().unwrap().span(); + + match psess.source_map().span_to_snippet(span) { + Ok(s) => assert_eq!(&s[..], "{ body }"), + Err(_) => panic!("could not get snippet"), + } + }); +} + +// This tests that when parsing a string (rather than a file) we don't try +// and read in a file for a module declaration and just parse a stub. +// See `recurse_into_file_modules` in the parser. +#[test] +fn out_of_line_mod() { + create_default_session_globals_then(|| { + let item = parse_item_from_source_str( + PathBuf::from("foo").into(), + "mod foo { struct S; mod this_does_not_exist; }".to_owned(), + &psess(), + ) + .unwrap() + .unwrap(); + + let ast::ItemKind::Mod(_, mod_kind) = &item.kind else { panic!() }; + assert!(matches!(mod_kind, ast::ModKind::Loaded(items, ..) if items.len() == 2)); + }); +} + +#[test] +fn eqmodws() { + assert_eq!(matches_codepattern("", ""), true); + assert_eq!(matches_codepattern("", "a"), false); + assert_eq!(matches_codepattern("a", ""), false); + assert_eq!(matches_codepattern("a", "a"), true); + assert_eq!(matches_codepattern("a b", "a \n\t\r b"), true); + assert_eq!(matches_codepattern("a b ", "a \n\t\r b"), true); + assert_eq!(matches_codepattern("a b", "a \n\t\r b "), false); + assert_eq!(matches_codepattern("a b", "a b"), true); + assert_eq!(matches_codepattern("ab", "a b"), false); + assert_eq!(matches_codepattern("a b", "ab"), true); + assert_eq!(matches_codepattern(" a b", "ab"), true); +} + +#[test] +fn pattern_whitespace() { + assert_eq!(matches_codepattern("", "\x0C"), false); + assert_eq!(matches_codepattern("a b ", "a \u{0085}\n\t\r b"), true); + assert_eq!(matches_codepattern("a b", "a \u{0085}\n\t\r b "), false); +} + +#[test] +fn non_pattern_whitespace() { + // These have the property 'White_Space' but not 'Pattern_White_Space' + assert_eq!(matches_codepattern("a b", "a\u{2002}b"), false); + assert_eq!(matches_codepattern("a b", "a\u{2002}b"), false); + assert_eq!(matches_codepattern("\u{205F}a b", "ab"), false); + assert_eq!(matches_codepattern("a \u{3000}b", "ab"), false); +} diff --git a/compiler/rustc_parse/src/parser/tokenstream/tests.rs b/compiler/rustc_parse/src/parser/tokenstream/tests.rs new file mode 100644 index 00000000000..9be00a14791 --- /dev/null +++ b/compiler/rustc_parse/src/parser/tokenstream/tests.rs @@ -0,0 +1,108 @@ +use crate::parser::tests::string_to_stream; +use rustc_ast::token::{self, IdentIsRaw}; +use rustc_ast::tokenstream::{TokenStream, TokenTree}; +use rustc_span::create_default_session_globals_then; +use rustc_span::{BytePos, Span, Symbol}; + +fn string_to_ts(string: &str) -> TokenStream { + string_to_stream(string.to_owned()) +} + +fn sp(a: u32, b: u32) -> Span { + Span::with_root_ctxt(BytePos(a), BytePos(b)) +} + +#[test] +fn test_concat() { + create_default_session_globals_then(|| { + let test_res = string_to_ts("foo::bar::baz"); + let test_fst = string_to_ts("foo::bar"); + let test_snd = string_to_ts("::baz"); + let mut eq_res = TokenStream::default(); + eq_res.push_stream(test_fst); + eq_res.push_stream(test_snd); + assert_eq!(test_res.trees().count(), 5); + assert_eq!(eq_res.trees().count(), 5); + assert_eq!(test_res.eq_unspanned(&eq_res), true); + }) +} + +#[test] +fn test_to_from_bijection() { + create_default_session_globals_then(|| { + let test_start = string_to_ts("foo::bar(baz)"); + let test_end = test_start.trees().cloned().collect(); + assert_eq!(test_start, test_end) + }) +} + +#[test] +fn test_eq_0() { + create_default_session_globals_then(|| { + let test_res = string_to_ts("foo"); + let test_eqs = string_to_ts("foo"); + assert_eq!(test_res, test_eqs) + }) +} + +#[test] +fn test_eq_1() { + create_default_session_globals_then(|| { + let test_res = string_to_ts("::bar::baz"); + let test_eqs = string_to_ts("::bar::baz"); + assert_eq!(test_res, test_eqs) + }) +} + +#[test] +fn test_eq_3() { + create_default_session_globals_then(|| { + let test_res = string_to_ts(""); + let test_eqs = string_to_ts(""); + assert_eq!(test_res, test_eqs) + }) +} + +#[test] +fn test_diseq_0() { + create_default_session_globals_then(|| { + let test_res = string_to_ts("::bar::baz"); + let test_eqs = string_to_ts("bar::baz"); + assert_eq!(test_res == test_eqs, false) + }) +} + +#[test] +fn test_diseq_1() { + create_default_session_globals_then(|| { + let test_res = string_to_ts("(bar,baz)"); + let test_eqs = string_to_ts("bar,baz"); + assert_eq!(test_res == test_eqs, false) + }) +} + +#[test] +fn test_is_empty() { + create_default_session_globals_then(|| { + let test0 = TokenStream::default(); + let test1 = + TokenStream::token_alone(token::Ident(Symbol::intern("a"), IdentIsRaw::No), sp(0, 1)); + let test2 = string_to_ts("foo(bar::baz)"); + + assert_eq!(test0.is_empty(), true); + assert_eq!(test1.is_empty(), false); + assert_eq!(test2.is_empty(), false); + }) +} + +#[test] +fn test_dotdotdot() { + create_default_session_globals_then(|| { + let mut stream = TokenStream::default(); + stream.push_tree(TokenTree::token_joint(token::Dot, sp(0, 1))); + stream.push_tree(TokenTree::token_joint(token::Dot, sp(1, 2))); + stream.push_tree(TokenTree::token_alone(token::Dot, sp(2, 3))); + assert!(stream.eq_unspanned(&string_to_ts("..."))); + assert_eq!(stream.trees().count(), 1); + }) +} diff --git a/compiler/rustc_parse/src/parser/ty.rs b/compiler/rustc_parse/src/parser/ty.rs new file mode 100644 index 00000000000..2df8f58507b --- /dev/null +++ b/compiler/rustc_parse/src/parser/ty.rs @@ -0,0 +1,1214 @@ +use super::{Parser, PathStyle, SeqSep, TokenType, Trailing}; + +use crate::errors::{ + self, DynAfterMut, ExpectedFnPathFoundFnKeyword, ExpectedMutOrConstInRawPointerType, + FnPointerCannotBeAsync, FnPointerCannotBeConst, FnPtrWithGenerics, FnPtrWithGenericsSugg, + HelpUseLatestEdition, InvalidDynKeyword, LifetimeAfterMut, NeedPlusAfterTraitObjectLifetime, + NestedCVariadicType, ReturnTypesUseThinArrow, +}; +use crate::{maybe_recover_from_interpolated_ty_qpath, maybe_whole}; + +use rustc_ast::ptr::P; +use rustc_ast::token::{self, Delimiter, Token, TokenKind}; +use rustc_ast::util::case::Case; +use rustc_ast::{ + self as ast, BareFnTy, BoundAsyncness, BoundConstness, BoundPolarity, FnRetTy, GenericBound, + GenericBounds, GenericParam, Generics, Lifetime, MacCall, MutTy, Mutability, PolyTraitRef, + PreciseCapturingArg, TraitBoundModifiers, TraitObjectSyntax, Ty, TyKind, DUMMY_NODE_ID, +}; +use rustc_errors::{Applicability, PResult}; +use rustc_span::symbol::{kw, sym, Ident}; +use rustc_span::{Span, Symbol}; +use thin_vec::{thin_vec, ThinVec}; + +#[derive(Copy, Clone, PartialEq)] +pub(super) enum AllowPlus { + Yes, + No, +} + +#[derive(PartialEq)] +pub(super) enum RecoverQPath { + Yes, + No, +} + +pub(super) enum RecoverQuestionMark { + Yes, + No, +} + +/// Signals whether parsing a type should recover `->`. +/// +/// More specifically, when parsing a function like: +/// ```compile_fail +/// fn foo() => u8 { 0 } +/// fn bar(): u8 { 0 } +/// ``` +/// The compiler will try to recover interpreting `foo() => u8` as `foo() -> u8` when calling +/// `parse_ty` with anything except `RecoverReturnSign::No`, and it will try to recover `bar(): u8` +/// as `bar() -> u8` when passing `RecoverReturnSign::Yes` to `parse_ty` +#[derive(Copy, Clone, PartialEq)] +pub(super) enum RecoverReturnSign { + Yes, + OnlyFatArrow, + No, +} + +impl RecoverReturnSign { + /// [RecoverReturnSign::Yes] allows for recovering `fn foo() => u8` and `fn foo(): u8`, + /// [RecoverReturnSign::OnlyFatArrow] allows for recovering only `fn foo() => u8` (recovering + /// colons can cause problems when parsing where clauses), and + /// [RecoverReturnSign::No] doesn't allow for any recovery of the return type arrow + fn can_recover(self, token: &TokenKind) -> bool { + match self { + Self::Yes => matches!(token, token::FatArrow | token::Colon), + Self::OnlyFatArrow => matches!(token, token::FatArrow), + Self::No => false, + } + } +} + +// Is `...` (`CVarArgs`) legal at this level of type parsing? +#[derive(PartialEq)] +enum AllowCVariadic { + Yes, + No, +} + +/// Returns `true` if `IDENT t` can start a type -- `IDENT::a::b`, `IDENT<u8, u8>`, +/// `IDENT<<u8 as Trait>::AssocTy>`. +/// +/// Types can also be of the form `IDENT(u8, u8) -> u8`, however this assumes +/// that `IDENT` is not the ident of a fn trait. +fn can_continue_type_after_non_fn_ident(t: &Token) -> bool { + t == &token::PathSep || t == &token::Lt || t == &token::BinOp(token::Shl) +} + +fn can_begin_dyn_bound_in_edition_2015(t: &Token) -> bool { + // `Not`, `Tilde` & `Const` are deliberately not part of this list to + // contain the number of potential regressions esp. in MBE code. + // `Const` would regress `rfc-2632-const-trait-impl/mbe-dyn-const-2015.rs`. + // `Not` would regress `dyn!(...)` macro calls in Rust 2015. + t.is_path_start() + || t.is_lifetime() + || t == &TokenKind::Question + || t.is_keyword(kw::For) + || t == &TokenKind::OpenDelim(Delimiter::Parenthesis) +} + +impl<'a> Parser<'a> { + /// Parses a type. + pub fn parse_ty(&mut self) -> PResult<'a, P<Ty>> { + self.parse_ty_common( + AllowPlus::Yes, + AllowCVariadic::No, + RecoverQPath::Yes, + RecoverReturnSign::Yes, + None, + RecoverQuestionMark::Yes, + ) + } + + pub(super) fn parse_ty_with_generics_recovery( + &mut self, + ty_params: &Generics, + ) -> PResult<'a, P<Ty>> { + self.parse_ty_common( + AllowPlus::Yes, + AllowCVariadic::No, + RecoverQPath::Yes, + RecoverReturnSign::Yes, + Some(ty_params), + RecoverQuestionMark::Yes, + ) + } + + /// Parse a type suitable for a field definition. + /// The difference from `parse_ty` is that this version + /// allows anonymous structs and unions. + pub fn parse_ty_for_field_def(&mut self) -> PResult<'a, P<Ty>> { + if self.can_begin_anon_struct_or_union() { + self.parse_anon_struct_or_union() + } else { + self.parse_ty() + } + } + + /// Parse a type suitable for a function or function pointer parameter. + /// The difference from `parse_ty` is that this version allows `...` + /// (`CVarArgs`) at the top level of the type. + pub(super) fn parse_ty_for_param(&mut self) -> PResult<'a, P<Ty>> { + self.parse_ty_common( + AllowPlus::Yes, + AllowCVariadic::Yes, + RecoverQPath::Yes, + RecoverReturnSign::Yes, + None, + RecoverQuestionMark::Yes, + ) + } + + /// Parses a type in restricted contexts where `+` is not permitted. + /// + /// Example 1: `&'a TYPE` + /// `+` is prohibited to maintain operator priority (P(+) < P(&)). + /// Example 2: `value1 as TYPE + value2` + /// `+` is prohibited to avoid interactions with expression grammar. + pub(super) fn parse_ty_no_plus(&mut self) -> PResult<'a, P<Ty>> { + self.parse_ty_common( + AllowPlus::No, + AllowCVariadic::No, + RecoverQPath::Yes, + RecoverReturnSign::Yes, + None, + RecoverQuestionMark::Yes, + ) + } + + /// Parses a type following an `as` cast. Similar to `parse_ty_no_plus`, but signaling origin + /// for better diagnostics involving `?`. + pub(super) fn parse_as_cast_ty(&mut self) -> PResult<'a, P<Ty>> { + self.parse_ty_common( + AllowPlus::No, + AllowCVariadic::No, + RecoverQPath::Yes, + RecoverReturnSign::Yes, + None, + RecoverQuestionMark::No, + ) + } + + pub(super) fn parse_ty_no_question_mark_recover(&mut self) -> PResult<'a, P<Ty>> { + self.parse_ty_common( + AllowPlus::Yes, + AllowCVariadic::No, + RecoverQPath::Yes, + RecoverReturnSign::Yes, + None, + RecoverQuestionMark::No, + ) + } + + /// Parse a type without recovering `:` as `->` to avoid breaking code such as `where fn() : for<'a>` + pub(super) fn parse_ty_for_where_clause(&mut self) -> PResult<'a, P<Ty>> { + self.parse_ty_common( + AllowPlus::Yes, + AllowCVariadic::Yes, + RecoverQPath::Yes, + RecoverReturnSign::OnlyFatArrow, + None, + RecoverQuestionMark::Yes, + ) + } + + /// Parses an optional return type `[ -> TY ]` in a function declaration. + pub(super) fn parse_ret_ty( + &mut self, + allow_plus: AllowPlus, + recover_qpath: RecoverQPath, + recover_return_sign: RecoverReturnSign, + ) -> PResult<'a, FnRetTy> { + Ok(if self.eat(&token::RArrow) { + // FIXME(Centril): Can we unconditionally `allow_plus`? + let ty = self.parse_ty_common( + allow_plus, + AllowCVariadic::No, + recover_qpath, + recover_return_sign, + None, + RecoverQuestionMark::Yes, + )?; + FnRetTy::Ty(ty) + } else if recover_return_sign.can_recover(&self.token.kind) { + // Don't `eat` to prevent `=>` from being added as an expected token which isn't + // actually expected and could only confuse users + self.bump(); + self.dcx().emit_err(ReturnTypesUseThinArrow { span: self.prev_token.span }); + let ty = self.parse_ty_common( + allow_plus, + AllowCVariadic::No, + recover_qpath, + recover_return_sign, + None, + RecoverQuestionMark::Yes, + )?; + FnRetTy::Ty(ty) + } else { + FnRetTy::Default(self.prev_token.span.shrink_to_hi()) + }) + } + + fn parse_ty_common( + &mut self, + allow_plus: AllowPlus, + allow_c_variadic: AllowCVariadic, + recover_qpath: RecoverQPath, + recover_return_sign: RecoverReturnSign, + ty_generics: Option<&Generics>, + recover_question_mark: RecoverQuestionMark, + ) -> PResult<'a, P<Ty>> { + let allow_qpath_recovery = recover_qpath == RecoverQPath::Yes; + maybe_recover_from_interpolated_ty_qpath!(self, allow_qpath_recovery); + maybe_whole!(self, NtTy, |ty| ty); + + let lo = self.token.span; + let mut impl_dyn_multi = false; + let kind = if self.check(&token::OpenDelim(Delimiter::Parenthesis)) { + self.parse_ty_tuple_or_parens(lo, allow_plus)? + } else if self.eat(&token::Not) { + // Never type `!` + TyKind::Never + } else if self.eat(&token::BinOp(token::Star)) { + self.parse_ty_ptr()? + } else if self.eat(&token::OpenDelim(Delimiter::Bracket)) { + self.parse_array_or_slice_ty()? + } else if self.check(&token::BinOp(token::And)) || self.check(&token::AndAnd) { + // Reference + self.expect_and()?; + self.parse_borrowed_pointee()? + } else if self.eat_keyword_noexpect(kw::Typeof) { + self.parse_typeof_ty()? + } else if self.eat_keyword(kw::Underscore) { + // A type to be inferred `_` + TyKind::Infer + } else if self.check_fn_front_matter(false, Case::Sensitive) { + // Function pointer type + self.parse_ty_bare_fn(lo, ThinVec::new(), None, recover_return_sign)? + } else if self.check_keyword(kw::For) { + let for_span = self.token.span; + // Function pointer type or bound list (trait object type) starting with a poly-trait. + // `for<'lt> [unsafe] [extern "ABI"] fn (&'lt S) -> T` + // `for<'lt> Trait1<'lt> + Trait2 + 'a` + let lifetime_defs = self.parse_late_bound_lifetime_defs()?; + if self.check_fn_front_matter(false, Case::Sensitive) { + self.parse_ty_bare_fn( + lo, + lifetime_defs, + Some(self.prev_token.span.shrink_to_lo()), + recover_return_sign, + )? + } else { + // Try to recover `for<'a> dyn Trait` or `for<'a> impl Trait`. + if self.may_recover() + && (self.eat_keyword_noexpect(kw::Impl) || self.eat_keyword_noexpect(kw::Dyn)) + { + let kw = self.prev_token.ident().unwrap().0; + let removal_span = kw.span.with_hi(self.token.span.lo()); + let path = self.parse_path(PathStyle::Type)?; + let parse_plus = allow_plus == AllowPlus::Yes && self.check_plus(); + let kind = + self.parse_remaining_bounds_path(lifetime_defs, path, lo, parse_plus)?; + let err = self.dcx().create_err(errors::TransposeDynOrImpl { + span: kw.span, + kw: kw.name.as_str(), + sugg: errors::TransposeDynOrImplSugg { + removal_span, + insertion_span: for_span.shrink_to_lo(), + kw: kw.name.as_str(), + }, + }); + + // Take the parsed bare trait object and turn it either + // into a `dyn` object or an `impl Trait`. + let kind = match (kind, kw.name) { + (TyKind::TraitObject(bounds, _), kw::Dyn) => { + TyKind::TraitObject(bounds, TraitObjectSyntax::Dyn) + } + (TyKind::TraitObject(bounds, _), kw::Impl) => { + TyKind::ImplTrait(ast::DUMMY_NODE_ID, bounds, None) + } + _ => return Err(err), + }; + err.emit(); + kind + } else { + let path = self.parse_path(PathStyle::Type)?; + let parse_plus = allow_plus == AllowPlus::Yes && self.check_plus(); + self.parse_remaining_bounds_path(lifetime_defs, path, lo, parse_plus)? + } + } + } else if self.eat_keyword(kw::Impl) { + self.parse_impl_ty(&mut impl_dyn_multi)? + } else if self.is_explicit_dyn_type() { + self.parse_dyn_ty(&mut impl_dyn_multi)? + } else if self.eat_lt() { + // Qualified path + let (qself, path) = self.parse_qpath(PathStyle::Type)?; + TyKind::Path(Some(qself), path) + } else if self.check_path() { + self.parse_path_start_ty(lo, allow_plus, ty_generics)? + } else if self.can_begin_bound() { + self.parse_bare_trait_object(lo, allow_plus)? + } else if self.eat(&token::DotDotDot) { + match allow_c_variadic { + AllowCVariadic::Yes => TyKind::CVarArgs, + AllowCVariadic::No => { + // FIXME(Centril): Should we just allow `...` syntactically + // anywhere in a type and use semantic restrictions instead? + let guar = self + .dcx() + .emit_err(NestedCVariadicType { span: lo.to(self.prev_token.span) }); + TyKind::Err(guar) + } + } + } else { + let msg = format!("expected type, found {}", super::token_descr(&self.token)); + let mut err = self.dcx().struct_span_err(self.token.span, msg); + err.span_label(self.token.span, "expected type"); + return Err(err); + }; + + let span = lo.to(self.prev_token.span); + let mut ty = self.mk_ty(span, kind); + + // Try to recover from use of `+` with incorrect priority. + match allow_plus { + AllowPlus::Yes => self.maybe_recover_from_bad_type_plus(&ty)?, + AllowPlus::No => self.maybe_report_ambiguous_plus(impl_dyn_multi, &ty), + } + if let RecoverQuestionMark::Yes = recover_question_mark { + ty = self.maybe_recover_from_question_mark(ty); + } + if allow_qpath_recovery { self.maybe_recover_from_bad_qpath(ty) } else { Ok(ty) } + } + + /// Parse an anonymous struct or union (only for field definitions): + /// ```ignore (feature-not-ready) + /// #[repr(C)] + /// struct Foo { + /// _: struct { // anonymous struct + /// x: u32, + /// y: f64, + /// } + /// _: union { // anonymous union + /// z: u32, + /// w: f64, + /// } + /// } + /// ``` + fn parse_anon_struct_or_union(&mut self) -> PResult<'a, P<Ty>> { + assert!(self.token.is_keyword(kw::Union) || self.token.is_keyword(kw::Struct)); + let is_union = self.token.is_keyword(kw::Union); + + let lo = self.token.span; + self.bump(); + + let (fields, _recovered) = + self.parse_record_struct_body(if is_union { "union" } else { "struct" }, lo, false)?; + let span = lo.to(self.prev_token.span); + self.psess.gated_spans.gate(sym::unnamed_fields, span); + let id = ast::DUMMY_NODE_ID; + let kind = + if is_union { TyKind::AnonUnion(id, fields) } else { TyKind::AnonStruct(id, fields) }; + Ok(self.mk_ty(span, kind)) + } + + /// Parses either: + /// - `(TYPE)`, a parenthesized type. + /// - `(TYPE,)`, a tuple with a single field of type TYPE. + fn parse_ty_tuple_or_parens(&mut self, lo: Span, allow_plus: AllowPlus) -> PResult<'a, TyKind> { + let mut trailing_plus = false; + let (ts, trailing) = self.parse_paren_comma_seq(|p| { + let ty = p.parse_ty()?; + trailing_plus = p.prev_token.kind == TokenKind::BinOp(token::Plus); + Ok(ty) + })?; + + if ts.len() == 1 && matches!(trailing, Trailing::No) { + let ty = ts.into_iter().next().unwrap().into_inner(); + let maybe_bounds = allow_plus == AllowPlus::Yes && self.token.is_like_plus(); + match ty.kind { + // `(TY_BOUND_NOPAREN) + BOUND + ...`. + TyKind::Path(None, path) if maybe_bounds => { + self.parse_remaining_bounds_path(ThinVec::new(), path, lo, true) + } + TyKind::TraitObject(bounds, TraitObjectSyntax::None) + if maybe_bounds && bounds.len() == 1 && !trailing_plus => + { + self.parse_remaining_bounds(bounds, true) + } + // `(TYPE)` + _ => Ok(TyKind::Paren(P(ty))), + } + } else { + Ok(TyKind::Tup(ts)) + } + } + + fn parse_bare_trait_object(&mut self, lo: Span, allow_plus: AllowPlus) -> PResult<'a, TyKind> { + let lt_no_plus = self.check_lifetime() && !self.look_ahead(1, |t| t.is_like_plus()); + let bounds = self.parse_generic_bounds_common(allow_plus)?; + if lt_no_plus { + self.dcx().emit_err(NeedPlusAfterTraitObjectLifetime { span: lo }); + } + Ok(TyKind::TraitObject(bounds, TraitObjectSyntax::None)) + } + + fn parse_remaining_bounds_path( + &mut self, + generic_params: ThinVec<GenericParam>, + path: ast::Path, + lo: Span, + parse_plus: bool, + ) -> PResult<'a, TyKind> { + let poly_trait_ref = PolyTraitRef::new(generic_params, path, lo.to(self.prev_token.span)); + let bounds = vec![GenericBound::Trait(poly_trait_ref, TraitBoundModifiers::NONE)]; + self.parse_remaining_bounds(bounds, parse_plus) + } + + /// Parse the remainder of a bare trait object type given an already parsed list. + fn parse_remaining_bounds( + &mut self, + mut bounds: GenericBounds, + plus: bool, + ) -> PResult<'a, TyKind> { + if plus { + self.eat_plus(); // `+`, or `+=` gets split and `+` is discarded + bounds.append(&mut self.parse_generic_bounds()?); + } + Ok(TyKind::TraitObject(bounds, TraitObjectSyntax::None)) + } + + /// Parses a raw pointer type: `*[const | mut] $type`. + fn parse_ty_ptr(&mut self) -> PResult<'a, TyKind> { + let mutbl = self.parse_const_or_mut().unwrap_or_else(|| { + let span = self.prev_token.span; + self.dcx().emit_err(ExpectedMutOrConstInRawPointerType { + span, + after_asterisk: span.shrink_to_hi(), + }); + Mutability::Not + }); + let ty = self.parse_ty_no_plus()?; + Ok(TyKind::Ptr(MutTy { ty, mutbl })) + } + + /// Parses an array (`[TYPE; EXPR]`) or slice (`[TYPE]`) type. + /// The opening `[` bracket is already eaten. + fn parse_array_or_slice_ty(&mut self) -> PResult<'a, TyKind> { + let elt_ty = match self.parse_ty() { + Ok(ty) => ty, + Err(err) + if self.look_ahead(1, |t| t.kind == token::CloseDelim(Delimiter::Bracket)) + | self.look_ahead(1, |t| t.kind == token::Semi) => + { + // Recover from `[LIT; EXPR]` and `[LIT]` + self.bump(); + let guar = err.emit(); + self.mk_ty(self.prev_token.span, TyKind::Err(guar)) + } + Err(err) => return Err(err), + }; + + let ty = if self.eat(&token::Semi) { + let mut length = self.parse_expr_anon_const()?; + if let Err(e) = self.expect(&token::CloseDelim(Delimiter::Bracket)) { + // Try to recover from `X<Y, ...>` when `X::<Y, ...>` works + self.check_mistyped_turbofish_with_multiple_type_params(e, &mut length.value)?; + self.expect(&token::CloseDelim(Delimiter::Bracket))?; + } + TyKind::Array(elt_ty, length) + } else { + self.expect(&token::CloseDelim(Delimiter::Bracket))?; + TyKind::Slice(elt_ty) + }; + + Ok(ty) + } + + fn parse_borrowed_pointee(&mut self) -> PResult<'a, TyKind> { + let and_span = self.prev_token.span; + let mut opt_lifetime = self.check_lifetime().then(|| self.expect_lifetime()); + let mut mutbl = self.parse_mutability(); + if self.token.is_lifetime() && mutbl == Mutability::Mut && opt_lifetime.is_none() { + // A lifetime is invalid here: it would be part of a bare trait bound, which requires + // it to be followed by a plus, but we disallow plus in the pointee type. + // So we can handle this case as an error here, and suggest `'a mut`. + // If there *is* a plus next though, handling the error later provides better suggestions + // (like adding parentheses) + if !self.look_ahead(1, |t| t.is_like_plus()) { + let lifetime_span = self.token.span; + let span = and_span.to(lifetime_span); + + let (suggest_lifetime, snippet) = + if let Ok(lifetime_src) = self.span_to_snippet(lifetime_span) { + (Some(span), lifetime_src) + } else { + (None, String::new()) + }; + self.dcx().emit_err(LifetimeAfterMut { span, suggest_lifetime, snippet }); + + opt_lifetime = Some(self.expect_lifetime()); + } + } else if self.token.is_keyword(kw::Dyn) + && mutbl == Mutability::Not + && self.look_ahead(1, |t| t.is_keyword(kw::Mut)) + { + // We have `&dyn mut ...`, which is invalid and should be `&mut dyn ...`. + let span = and_span.to(self.look_ahead(1, |t| t.span)); + self.dcx().emit_err(DynAfterMut { span }); + + // Recovery + mutbl = Mutability::Mut; + let (dyn_tok, dyn_tok_sp) = (self.token.clone(), self.token_spacing); + self.bump(); + self.bump_with((dyn_tok, dyn_tok_sp)); + } + let ty = self.parse_ty_no_plus()?; + Ok(TyKind::Ref(opt_lifetime, MutTy { ty, mutbl })) + } + + // Parses the `typeof(EXPR)`. + // To avoid ambiguity, the type is surrounded by parentheses. + fn parse_typeof_ty(&mut self) -> PResult<'a, TyKind> { + self.expect(&token::OpenDelim(Delimiter::Parenthesis))?; + let expr = self.parse_expr_anon_const()?; + self.expect(&token::CloseDelim(Delimiter::Parenthesis))?; + Ok(TyKind::Typeof(expr)) + } + + /// Parses a function pointer type (`TyKind::BareFn`). + /// ```ignore (illustrative) + /// [unsafe] [extern "ABI"] fn (S) -> T + /// // ^~~~~^ ^~~~^ ^~^ ^ + /// // | | | | + /// // | | | Return type + /// // Function Style ABI Parameter types + /// ``` + /// We actually parse `FnHeader FnDecl`, but we error on `const` and `async` qualifiers. + fn parse_ty_bare_fn( + &mut self, + lo: Span, + mut params: ThinVec<GenericParam>, + param_insertion_point: Option<Span>, + recover_return_sign: RecoverReturnSign, + ) -> PResult<'a, TyKind> { + let inherited_vis = rustc_ast::Visibility { + span: rustc_span::DUMMY_SP, + kind: rustc_ast::VisibilityKind::Inherited, + tokens: None, + }; + let span_start = self.token.span; + let ast::FnHeader { ext, safety, constness, coroutine_kind } = + self.parse_fn_front_matter(&inherited_vis, Case::Sensitive)?; + if self.may_recover() && self.token.kind == TokenKind::Lt { + self.recover_fn_ptr_with_generics(lo, &mut params, param_insertion_point)?; + } + let decl = self.parse_fn_decl(|_| false, AllowPlus::No, recover_return_sign)?; + let whole_span = lo.to(self.prev_token.span); + if let ast::Const::Yes(span) = constness { + // If we ever start to allow `const fn()`, then update + // feature gating for `#![feature(const_extern_fn)]` to + // cover it. + self.dcx().emit_err(FnPointerCannotBeConst { span: whole_span, qualifier: span }); + } + if let Some(ast::CoroutineKind::Async { span, .. }) = coroutine_kind { + self.dcx().emit_err(FnPointerCannotBeAsync { span: whole_span, qualifier: span }); + } + // FIXME(gen_blocks): emit a similar error for `gen fn()` + let decl_span = span_start.to(self.token.span); + Ok(TyKind::BareFn(P(BareFnTy { ext, safety, generic_params: params, decl, decl_span }))) + } + + /// Recover from function pointer types with a generic parameter list (e.g. `fn<'a>(&'a str)`). + fn recover_fn_ptr_with_generics( + &mut self, + lo: Span, + params: &mut ThinVec<GenericParam>, + param_insertion_point: Option<Span>, + ) -> PResult<'a, ()> { + let generics = self.parse_generics()?; + let arity = generics.params.len(); + + let mut lifetimes: ThinVec<_> = generics + .params + .into_iter() + .filter(|param| matches!(param.kind, ast::GenericParamKind::Lifetime)) + .collect(); + + let sugg = if !lifetimes.is_empty() { + let snippet = + lifetimes.iter().map(|param| param.ident.as_str()).intersperse(", ").collect(); + + let (left, snippet) = if let Some(span) = param_insertion_point { + (span, if params.is_empty() { snippet } else { format!(", {snippet}") }) + } else { + (lo.shrink_to_lo(), format!("for<{snippet}> ")) + }; + + Some(FnPtrWithGenericsSugg { + left, + snippet, + right: generics.span, + arity, + for_param_list_exists: param_insertion_point.is_some(), + }) + } else { + None + }; + + self.dcx().emit_err(FnPtrWithGenerics { span: generics.span, sugg }); + params.append(&mut lifetimes); + Ok(()) + } + + /// Parses an `impl B0 + ... + Bn` type. + fn parse_impl_ty(&mut self, impl_dyn_multi: &mut bool) -> PResult<'a, TyKind> { + if self.token.is_lifetime() { + self.look_ahead(1, |t| { + if let token::Ident(sym, _) = t.kind { + // parse pattern with "'a Sized" we're supposed to give suggestion like + // "'a + Sized" + self.dcx().emit_err(errors::MissingPlusBounds { + span: self.token.span, + hi: self.token.span.shrink_to_hi(), + sym, + }); + } + }) + } + + // parse precise captures, if any. This is `use<'lt, 'lt, P, P>`; a list of + // lifetimes and ident params (including SelfUpper). These are validated later + // for order, duplication, and whether they actually reference params. + let precise_capturing = if self.eat_keyword(kw::Use) { + let use_span = self.prev_token.span; + self.psess.gated_spans.gate(sym::precise_capturing, use_span); + let (args, args_span) = self.parse_precise_capturing_args()?; + Some(P((args, use_span.to(args_span)))) + } else { + None + }; + + // Always parse bounds greedily for better error recovery. + let bounds = self.parse_generic_bounds()?; + + *impl_dyn_multi = bounds.len() > 1 || self.prev_token.kind == TokenKind::BinOp(token::Plus); + + Ok(TyKind::ImplTrait(ast::DUMMY_NODE_ID, bounds, precise_capturing)) + } + + fn parse_precise_capturing_args( + &mut self, + ) -> PResult<'a, (ThinVec<PreciseCapturingArg>, Span)> { + let lo = self.token.span; + let (args, _) = self.parse_unspanned_seq( + &TokenKind::Lt, + &TokenKind::Gt, + SeqSep::trailing_allowed(token::Comma), + |self_| { + if self_.check_keyword(kw::SelfUpper) { + self_.bump(); + Ok(PreciseCapturingArg::Arg( + ast::Path::from_ident(self_.prev_token.ident().unwrap().0), + DUMMY_NODE_ID, + )) + } else if self_.check_ident() { + Ok(PreciseCapturingArg::Arg( + ast::Path::from_ident(self_.parse_ident()?), + DUMMY_NODE_ID, + )) + } else if self_.check_lifetime() { + Ok(PreciseCapturingArg::Lifetime(self_.expect_lifetime())) + } else { + self_.unexpected_any() + } + }, + )?; + Ok((args, lo.to(self.prev_token.span))) + } + + /// Is a `dyn B0 + ... + Bn` type allowed here? + fn is_explicit_dyn_type(&mut self) -> bool { + self.check_keyword(kw::Dyn) + && (self.token.uninterpolated_span().at_least_rust_2018() + || self.look_ahead(1, |t| { + (can_begin_dyn_bound_in_edition_2015(t) + || t.kind == TokenKind::BinOp(token::Star)) + && !can_continue_type_after_non_fn_ident(t) + })) + } + + /// Parses a `dyn B0 + ... + Bn` type. + /// + /// Note that this does *not* parse bare trait objects. + fn parse_dyn_ty(&mut self, impl_dyn_multi: &mut bool) -> PResult<'a, TyKind> { + let lo = self.token.span; + self.bump(); // `dyn` + + // parse dyn* types + let syntax = if self.eat(&TokenKind::BinOp(token::Star)) { + self.psess.gated_spans.gate(sym::dyn_star, lo.to(self.prev_token.span)); + TraitObjectSyntax::DynStar + } else { + TraitObjectSyntax::Dyn + }; + + // Always parse bounds greedily for better error recovery. + let bounds = self.parse_generic_bounds()?; + *impl_dyn_multi = bounds.len() > 1 || self.prev_token.kind == TokenKind::BinOp(token::Plus); + Ok(TyKind::TraitObject(bounds, syntax)) + } + + /// Parses a type starting with a path. + /// + /// This can be: + /// 1. a type macro, `mac!(...)`, + /// 2. a bare trait object, `B0 + ... + Bn`, + /// 3. or a path, `path::to::MyType`. + fn parse_path_start_ty( + &mut self, + lo: Span, + allow_plus: AllowPlus, + ty_generics: Option<&Generics>, + ) -> PResult<'a, TyKind> { + // Simple path + let path = self.parse_path_inner(PathStyle::Type, ty_generics)?; + if self.eat(&token::Not) { + // Macro invocation in type position + Ok(TyKind::MacCall(P(MacCall { path, args: self.parse_delim_args()? }))) + } else if allow_plus == AllowPlus::Yes && self.check_plus() { + // `Trait1 + Trait2 + 'a` + self.parse_remaining_bounds_path(ThinVec::new(), path, lo, true) + } else { + // Just a type path. + Ok(TyKind::Path(None, path)) + } + } + + pub(super) fn parse_generic_bounds(&mut self) -> PResult<'a, GenericBounds> { + self.parse_generic_bounds_common(AllowPlus::Yes) + } + + /// Parses bounds of a type parameter `BOUND + BOUND + ...`, possibly with trailing `+`. + /// + /// See `parse_generic_bound` for the `BOUND` grammar. + fn parse_generic_bounds_common(&mut self, allow_plus: AllowPlus) -> PResult<'a, GenericBounds> { + let mut bounds = Vec::new(); + + // In addition to looping while we find generic bounds: + // We continue even if we find a keyword. This is necessary for error recovery on, + // for example, `impl fn()`. The only keyword that can go after generic bounds is + // `where`, so stop if it's it. + // We also continue if we find types (not traits), again for error recovery. + while self.can_begin_bound() + || (self.may_recover() + && (self.token.can_begin_type() + || (self.token.is_reserved_ident() && !self.token.is_keyword(kw::Where)))) + { + if self.token.is_keyword(kw::Dyn) { + // Account for `&dyn Trait + dyn Other`. + self.dcx().emit_err(InvalidDynKeyword { span: self.token.span }); + self.bump(); + } + bounds.push(self.parse_generic_bound()?); + if allow_plus == AllowPlus::No || !self.eat_plus() { + break; + } + } + + Ok(bounds) + } + + pub(super) fn can_begin_anon_struct_or_union(&mut self) -> bool { + (self.token.is_keyword(kw::Struct) || self.token.is_keyword(kw::Union)) + && self.look_ahead(1, |t| t == &token::OpenDelim(Delimiter::Brace)) + } + + /// Can the current token begin a bound? + fn can_begin_bound(&mut self) -> bool { + self.check_path() + || self.check_lifetime() + || self.check(&token::Not) + || self.check(&token::Question) + || self.check(&token::Tilde) + || self.check_keyword(kw::For) + || self.check(&token::OpenDelim(Delimiter::Parenthesis)) + || self.check_keyword(kw::Const) + || self.check_keyword(kw::Async) + } + + /// Parses a bound according to the grammar: + /// ```ebnf + /// BOUND = TY_BOUND | LT_BOUND + /// ``` + fn parse_generic_bound(&mut self) -> PResult<'a, GenericBound> { + let lo = self.token.span; + let leading_token = self.prev_token.clone(); + let has_parens = self.eat(&token::OpenDelim(Delimiter::Parenthesis)); + let inner_lo = self.token.span; + + let modifiers = self.parse_trait_bound_modifiers()?; + let bound = if self.token.is_lifetime() { + self.error_lt_bound_with_modifiers(modifiers); + self.parse_generic_lt_bound(lo, inner_lo, has_parens)? + } else { + self.parse_generic_ty_bound(lo, has_parens, modifiers, &leading_token)? + }; + + Ok(bound) + } + + /// Parses a lifetime ("outlives") bound, e.g. `'a`, according to: + /// ```ebnf + /// LT_BOUND = LIFETIME + /// ``` + fn parse_generic_lt_bound( + &mut self, + lo: Span, + inner_lo: Span, + has_parens: bool, + ) -> PResult<'a, GenericBound> { + let bound = GenericBound::Outlives(self.expect_lifetime()); + if has_parens { + // FIXME(Centril): Consider not erroring here and accepting `('lt)` instead, + // possibly introducing `GenericBound::Paren(P<GenericBound>)`? + self.recover_paren_lifetime(lo, inner_lo)?; + } + Ok(bound) + } + + /// Emits an error if any trait bound modifiers were present. + fn error_lt_bound_with_modifiers(&self, modifiers: TraitBoundModifiers) { + match modifiers.constness { + BoundConstness::Never => {} + BoundConstness::Always(span) | BoundConstness::Maybe(span) => { + self.dcx().emit_err(errors::ModifierLifetime { + span, + modifier: modifiers.constness.as_str(), + }); + } + } + + match modifiers.polarity { + BoundPolarity::Positive => {} + BoundPolarity::Negative(span) | BoundPolarity::Maybe(span) => { + self.dcx().emit_err(errors::ModifierLifetime { + span, + modifier: modifiers.polarity.as_str(), + }); + } + } + } + + /// Recover on `('lifetime)` with `(` already eaten. + fn recover_paren_lifetime(&mut self, lo: Span, inner_lo: Span) -> PResult<'a, ()> { + let inner_span = inner_lo.to(self.prev_token.span); + self.expect(&token::CloseDelim(Delimiter::Parenthesis))?; + let span = lo.to(self.prev_token.span); + let (sugg, snippet) = if let Ok(snippet) = self.span_to_snippet(inner_span) { + (Some(span), snippet) + } else { + (None, String::new()) + }; + + self.dcx().emit_err(errors::ParenthesizedLifetime { span, sugg, snippet }); + Ok(()) + } + + /// Parses the modifiers that may precede a trait in a bound, e.g. `?Trait` or `~const Trait`. + /// + /// If no modifiers are present, this does not consume any tokens. + /// + /// ```ebnf + /// TRAIT_BOUND_MODIFIERS = [["~"] "const"] ["?" | "!"] + /// ``` + fn parse_trait_bound_modifiers(&mut self) -> PResult<'a, TraitBoundModifiers> { + let constness = if self.eat(&token::Tilde) { + let tilde = self.prev_token.span; + self.expect_keyword(kw::Const)?; + let span = tilde.to(self.prev_token.span); + self.psess.gated_spans.gate(sym::const_trait_impl, span); + BoundConstness::Maybe(span) + } else if self.eat_keyword(kw::Const) { + self.psess.gated_spans.gate(sym::const_trait_impl, self.prev_token.span); + BoundConstness::Always(self.prev_token.span) + } else { + BoundConstness::Never + }; + + let asyncness = if self.token.uninterpolated_span().at_least_rust_2018() + && self.eat_keyword(kw::Async) + { + self.psess.gated_spans.gate(sym::async_closure, self.prev_token.span); + BoundAsyncness::Async(self.prev_token.span) + } else if self.may_recover() + && self.token.uninterpolated_span().is_rust_2015() + && self.is_kw_followed_by_ident(kw::Async) + { + self.bump(); // eat `async` + self.dcx().emit_err(errors::AsyncBoundModifierIn2015 { + span: self.prev_token.span, + help: HelpUseLatestEdition::new(), + }); + self.psess.gated_spans.gate(sym::async_closure, self.prev_token.span); + BoundAsyncness::Async(self.prev_token.span) + } else { + BoundAsyncness::Normal + }; + + let polarity = if self.eat(&token::Question) { + BoundPolarity::Maybe(self.prev_token.span) + } else if self.eat(&token::Not) { + self.psess.gated_spans.gate(sym::negative_bounds, self.prev_token.span); + BoundPolarity::Negative(self.prev_token.span) + } else { + BoundPolarity::Positive + }; + + Ok(TraitBoundModifiers { constness, asyncness, polarity }) + } + + /// Parses a type bound according to: + /// ```ebnf + /// TY_BOUND = TY_BOUND_NOPAREN | (TY_BOUND_NOPAREN) + /// TY_BOUND_NOPAREN = [TRAIT_BOUND_MODIFIERS] [for<LT_PARAM_DEFS>] SIMPLE_PATH + /// ``` + /// + /// For example, this grammar accepts `~const ?for<'a: 'b> m::Trait<'a>`. + fn parse_generic_ty_bound( + &mut self, + lo: Span, + has_parens: bool, + modifiers: TraitBoundModifiers, + leading_token: &Token, + ) -> PResult<'a, GenericBound> { + let mut lifetime_defs = self.parse_late_bound_lifetime_defs()?; + let mut path = if self.token.is_keyword(kw::Fn) + && self.look_ahead(1, |tok| tok.kind == TokenKind::OpenDelim(Delimiter::Parenthesis)) + && let Some(path) = self.recover_path_from_fn() + { + path + } else if !self.token.is_path_start() && self.token.can_begin_type() { + let ty = self.parse_ty_no_plus()?; + // Instead of finding a path (a trait), we found a type. + let mut err = self.dcx().struct_span_err(ty.span, "expected a trait, found type"); + + // If we can recover, try to extract a path from the type. Note + // that we do not use the try operator when parsing the type because + // if it fails then we get a parser error which we don't want (we're trying + // to recover from errors, not make more). + let path = if self.may_recover() { + let (span, message, sugg, path, applicability) = match &ty.kind { + TyKind::Ptr(..) | TyKind::Ref(..) + if let TyKind::Path(_, path) = &ty.peel_refs().kind => + { + ( + ty.span.until(path.span), + "consider removing the indirection", + "", + path, + Applicability::MaybeIncorrect, + ) + } + TyKind::ImplTrait(_, bounds, None) + if let [GenericBound::Trait(tr, ..), ..] = bounds.as_slice() => + { + ( + ty.span.until(tr.span), + "use the trait bounds directly", + "", + &tr.trait_ref.path, + Applicability::MachineApplicable, + ) + } + _ => return Err(err), + }; + + err.span_suggestion_verbose(span, message, sugg, applicability); + + path.clone() + } else { + return Err(err); + }; + + err.emit(); + + path + } else { + self.parse_path(PathStyle::Type)? + }; + + if self.may_recover() && self.token == TokenKind::OpenDelim(Delimiter::Parenthesis) { + self.recover_fn_trait_with_lifetime_params(&mut path, &mut lifetime_defs)?; + } + + if has_parens { + // Someone has written something like `&dyn (Trait + Other)`. The correct code + // would be `&(dyn Trait + Other)` + if self.token.is_like_plus() && leading_token.is_keyword(kw::Dyn) { + let bounds = vec![]; + self.parse_remaining_bounds(bounds, true)?; + self.expect(&token::CloseDelim(Delimiter::Parenthesis))?; + self.dcx().emit_err(errors::IncorrectParensTraitBounds { + span: vec![lo, self.prev_token.span], + sugg: errors::IncorrectParensTraitBoundsSugg { + wrong_span: leading_token.span.shrink_to_hi().to(lo), + new_span: leading_token.span.shrink_to_lo(), + }, + }); + } else { + self.expect(&token::CloseDelim(Delimiter::Parenthesis))?; + } + } + + let poly_trait = PolyTraitRef::new(lifetime_defs, path, lo.to(self.prev_token.span)); + Ok(GenericBound::Trait(poly_trait, modifiers)) + } + + // recovers a `Fn(..)` parenthesized-style path from `fn(..)` + fn recover_path_from_fn(&mut self) -> Option<ast::Path> { + let fn_token_span = self.token.span; + self.bump(); + let args_lo = self.token.span; + let snapshot = self.create_snapshot_for_diagnostic(); + match self.parse_fn_decl(|_| false, AllowPlus::No, RecoverReturnSign::OnlyFatArrow) { + Ok(decl) => { + self.dcx().emit_err(ExpectedFnPathFoundFnKeyword { fn_token_span }); + Some(ast::Path { + span: fn_token_span.to(self.prev_token.span), + segments: thin_vec![ast::PathSegment { + ident: Ident::new(Symbol::intern("Fn"), fn_token_span), + id: DUMMY_NODE_ID, + args: Some(P(ast::GenericArgs::Parenthesized(ast::ParenthesizedArgs { + span: args_lo.to(self.prev_token.span), + inputs: decl.inputs.iter().map(|a| a.ty.clone()).collect(), + inputs_span: args_lo.until(decl.output.span()), + output: decl.output.clone(), + }))), + }], + tokens: None, + }) + } + Err(diag) => { + diag.cancel(); + self.restore_snapshot(snapshot); + None + } + } + } + + /// Optionally parses `for<$generic_params>`. + pub(super) fn parse_late_bound_lifetime_defs(&mut self) -> PResult<'a, ThinVec<GenericParam>> { + if self.eat_keyword(kw::For) { + self.expect_lt()?; + let params = self.parse_generic_params()?; + self.expect_gt()?; + // We rely on AST validation to rule out invalid cases: There must not be type + // parameters, and the lifetime parameters must not have bounds. + Ok(params) + } else { + Ok(ThinVec::new()) + } + } + + /// Recover from `Fn`-family traits (Fn, FnMut, FnOnce) with lifetime arguments + /// (e.g. `FnOnce<'a>(&'a str) -> bool`). Up to generic arguments have already + /// been eaten. + fn recover_fn_trait_with_lifetime_params( + &mut self, + fn_path: &mut ast::Path, + lifetime_defs: &mut ThinVec<GenericParam>, + ) -> PResult<'a, ()> { + let fn_path_segment = fn_path.segments.last_mut().unwrap(); + let generic_args = if let Some(p_args) = &fn_path_segment.args { + p_args.clone().into_inner() + } else { + // Normally it wouldn't come here because the upstream should have parsed + // generic parameters (otherwise it's impossible to call this function). + return Ok(()); + }; + let lifetimes = + if let ast::GenericArgs::AngleBracketed(ast::AngleBracketedArgs { span: _, args }) = + &generic_args + { + args.into_iter() + .filter_map(|arg| { + if let ast::AngleBracketedArg::Arg(generic_arg) = arg + && let ast::GenericArg::Lifetime(lifetime) = generic_arg + { + Some(lifetime) + } else { + None + } + }) + .collect() + } else { + Vec::new() + }; + // Only try to recover if the trait has lifetime params. + if lifetimes.is_empty() { + return Ok(()); + } + + // Parse `(T, U) -> R`. + let inputs_lo = self.token.span; + let inputs: ThinVec<_> = + self.parse_fn_params(|_| false)?.into_iter().map(|input| input.ty).collect(); + let inputs_span = inputs_lo.to(self.prev_token.span); + let output = self.parse_ret_ty(AllowPlus::No, RecoverQPath::No, RecoverReturnSign::No)?; + let args = ast::ParenthesizedArgs { + span: fn_path_segment.span().to(self.prev_token.span), + inputs, + inputs_span, + output, + } + .into(); + *fn_path_segment = ast::PathSegment { + ident: fn_path_segment.ident, + args: Some(args), + id: ast::DUMMY_NODE_ID, + }; + + // Convert parsed `<'a>` in `Fn<'a>` into `for<'a>`. + let mut generic_params = lifetimes + .iter() + .map(|lt| GenericParam { + id: lt.id, + ident: lt.ident, + attrs: ast::AttrVec::new(), + bounds: Vec::new(), + is_placeholder: false, + kind: ast::GenericParamKind::Lifetime, + colon_span: None, + }) + .collect::<ThinVec<GenericParam>>(); + lifetime_defs.append(&mut generic_params); + + let generic_args_span = generic_args.span(); + let snippet = format!( + "for<{}> ", + lifetimes.iter().map(|lt| lt.ident.as_str()).intersperse(", ").collect::<String>(), + ); + let before_fn_path = fn_path.span.shrink_to_lo(); + self.dcx() + .struct_span_err(generic_args_span, "`Fn` traits cannot take lifetime parameters") + .with_multipart_suggestion( + "consider using a higher-ranked trait bound instead", + vec![(generic_args_span, "".to_owned()), (before_fn_path, snippet)], + Applicability::MaybeIncorrect, + ) + .emit(); + Ok(()) + } + + pub(super) fn check_lifetime(&mut self) -> bool { + self.expected_tokens.push(TokenType::Lifetime); + self.token.is_lifetime() + } + + /// Parses a single lifetime `'a` or panics. + pub(super) fn expect_lifetime(&mut self) -> Lifetime { + if let Some(ident) = self.token.lifetime() { + self.bump(); + Lifetime { ident, id: ast::DUMMY_NODE_ID } + } else { + self.dcx().span_bug(self.token.span, "not a lifetime") + } + } + + pub(super) fn mk_ty(&self, span: Span, kind: TyKind) -> P<Ty> { + P(Ty { kind, span, id: ast::DUMMY_NODE_ID, tokens: None }) + } +} |
