diff options
| author | Mazdak Farrokhzad <twingoow@gmail.com> | 2019-10-15 22:48:13 +0200 |
|---|---|---|
| committer | Mazdak Farrokhzad <twingoow@gmail.com> | 2019-11-10 03:57:18 +0100 |
| commit | 4ae2728fa8052915414127dce28245eb8f70842a (patch) | |
| tree | 27cc54d90904091e4dc9bf7ae5fa3b41be4b6187 /src/librustc_parse/parser | |
| parent | be023ebe850261c6bb202a02a686827d821c3697 (diff) | |
| download | rust-4ae2728fa8052915414127dce28245eb8f70842a.tar.gz rust-4ae2728fa8052915414127dce28245eb8f70842a.zip | |
move syntax::parse -> librustc_parse
also move MACRO_ARGUMENTS -> librustc_parse
Diffstat (limited to 'src/librustc_parse/parser')
| -rw-r--r-- | src/librustc_parse/parser/attr.rs | 351 | ||||
| -rw-r--r-- | src/librustc_parse/parser/diagnostics.rs | 1549 | ||||
| -rw-r--r-- | src/librustc_parse/parser/expr.rs | 1963 | ||||
| -rw-r--r-- | src/librustc_parse/parser/generics.rs | 308 | ||||
| -rw-r--r-- | src/librustc_parse/parser/item.rs | 2238 | ||||
| -rw-r--r-- | src/librustc_parse/parser/mod.rs | 1393 | ||||
| -rw-r--r-- | src/librustc_parse/parser/module.rs | 316 | ||||
| -rw-r--r-- | src/librustc_parse/parser/pat.rs | 1015 | ||||
| -rw-r--r-- | src/librustc_parse/parser/path.rs | 497 | ||||
| -rw-r--r-- | src/librustc_parse/parser/stmt.rs | 482 | ||||
| -rw-r--r-- | src/librustc_parse/parser/ty.rs | 460 |
11 files changed, 10572 insertions, 0 deletions
diff --git a/src/librustc_parse/parser/attr.rs b/src/librustc_parse/parser/attr.rs new file mode 100644 index 00000000000..524b551e54c --- /dev/null +++ b/src/librustc_parse/parser/attr.rs @@ -0,0 +1,351 @@ +use super::{SeqSep, Parser, TokenType, PathStyle}; +use syntax::attr; +use syntax::ast; +use syntax::util::comments; +use syntax::token::{self, Nonterminal, DelimToken}; +use syntax::tokenstream::{TokenStream, TokenTree}; +use syntax_pos::{Span, Symbol}; +use errors::PResult; + +use log::debug; + +#[derive(Debug)] +enum InnerAttributeParsePolicy<'a> { + Permitted, + NotPermitted { reason: &'a str, saw_doc_comment: bool, prev_attr_sp: Option<Span> }, +} + +const DEFAULT_UNEXPECTED_INNER_ATTR_ERR_MSG: &str = "an inner attribute is not \ + permitted in this context"; + +impl<'a> Parser<'a> { + /// Parses attributes that appear before an item. + pub(super) fn parse_outer_attributes(&mut self) -> PResult<'a, Vec<ast::Attribute>> { + let mut attrs: Vec<ast::Attribute> = Vec::new(); + let mut just_parsed_doc_comment = false; + loop { + debug!("parse_outer_attributes: self.token={:?}", self.token); + match self.token.kind { + token::Pound => { + let inner_error_reason = if just_parsed_doc_comment { + "an inner attribute is not permitted following an outer doc comment" + } else if !attrs.is_empty() { + "an inner attribute is not permitted following an outer attribute" + } else { + DEFAULT_UNEXPECTED_INNER_ATTR_ERR_MSG + }; + let inner_parse_policy = + InnerAttributeParsePolicy::NotPermitted { + reason: inner_error_reason, + saw_doc_comment: just_parsed_doc_comment, + prev_attr_sp: attrs.last().and_then(|a| Some(a.span)) + }; + let attr = self.parse_attribute_with_inner_parse_policy(inner_parse_policy)?; + attrs.push(attr); + just_parsed_doc_comment = false; + } + token::DocComment(s) => { + let attr = self.mk_doc_comment(s); + if attr.style != ast::AttrStyle::Outer { + let mut err = self.fatal("expected outer doc comment"); + err.note("inner doc comments like this (starting with \ + `//!` or `/*!`) can only appear before items"); + return Err(err); + } + attrs.push(attr); + self.bump(); + just_parsed_doc_comment = true; + } + _ => break, + } + } + Ok(attrs) + } + + fn mk_doc_comment(&self, s: Symbol) -> ast::Attribute { + let style = comments::doc_comment_style(&s.as_str()); + attr::mk_doc_comment(style, s, self.token.span) + } + + /// Matches `attribute = # ! [ meta_item ]`. + /// + /// If `permit_inner` is `true`, then a leading `!` indicates an inner + /// attribute. + pub fn parse_attribute(&mut self, permit_inner: bool) -> PResult<'a, ast::Attribute> { + debug!("parse_attribute: permit_inner={:?} self.token={:?}", + permit_inner, + self.token); + let inner_parse_policy = if permit_inner { + InnerAttributeParsePolicy::Permitted + } else { + InnerAttributeParsePolicy::NotPermitted { + reason: DEFAULT_UNEXPECTED_INNER_ATTR_ERR_MSG, + saw_doc_comment: false, + prev_attr_sp: None + } + }; + self.parse_attribute_with_inner_parse_policy(inner_parse_policy) + } + + /// The same as `parse_attribute`, except it takes in an `InnerAttributeParsePolicy` + /// that prescribes how to handle inner attributes. + fn parse_attribute_with_inner_parse_policy( + &mut self, + inner_parse_policy: InnerAttributeParsePolicy<'_> + ) -> PResult<'a, ast::Attribute> { + debug!("parse_attribute_with_inner_parse_policy: inner_parse_policy={:?} self.token={:?}", + inner_parse_policy, + self.token); + let (span, item, style) = match self.token.kind { + token::Pound => { + let lo = self.token.span; + self.bump(); + + if let InnerAttributeParsePolicy::Permitted = inner_parse_policy { + self.expected_tokens.push(TokenType::Token(token::Not)); + } + + let style = if self.token == token::Not { + self.bump(); + ast::AttrStyle::Inner + } else { + ast::AttrStyle::Outer + }; + + self.expect(&token::OpenDelim(token::Bracket))?; + let item = self.parse_attr_item()?; + self.expect(&token::CloseDelim(token::Bracket))?; + let hi = self.prev_span; + + let attr_sp = lo.to(hi); + + // Emit error if inner attribute is encountered and not permitted + if style == ast::AttrStyle::Inner { + if let InnerAttributeParsePolicy::NotPermitted { reason, + saw_doc_comment, prev_attr_sp } = inner_parse_policy { + let prev_attr_note = if saw_doc_comment { + "previous doc comment" + } else { + "previous outer attribute" + }; + + let mut diagnostic = self + .diagnostic() + .struct_span_err(attr_sp, reason); + + if let Some(prev_attr_sp) = prev_attr_sp { + diagnostic + .span_label(attr_sp, "not permitted following an outer attibute") + .span_label(prev_attr_sp, prev_attr_note); + } + + diagnostic + .note("inner attributes, like `#![no_std]`, annotate the item \ + enclosing them, and are usually found at the beginning of \ + source files. Outer attributes, like `#[test]`, annotate the \ + item following them.") + .emit() + } + } + + (attr_sp, item, style) + } + _ => { + let token_str = self.this_token_to_string(); + return Err(self.fatal(&format!("expected `#`, found `{}`", token_str))); + } + }; + + Ok(attr::mk_attr_from_item(style, item, span)) + } + + /// 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) -> PResult<'a, ast::AttrItem> { + let item = match self.token.kind { + token::Interpolated(ref nt) => match **nt { + Nonterminal::NtMeta(ref item) => Some(item.clone()), + _ => None, + }, + _ => None, + }; + Ok(if let Some(item) = item { + self.bump(); + item + } else { + let path = self.parse_path(PathStyle::Mod)?; + let tokens = if self.check(&token::OpenDelim(DelimToken::Paren)) || + self.check(&token::OpenDelim(DelimToken::Bracket)) || + self.check(&token::OpenDelim(DelimToken::Brace)) { + self.parse_token_tree().into() + } else if self.eat(&token::Eq) { + let eq = TokenTree::token(token::Eq, self.prev_span); + let mut is_interpolated_expr = false; + if let token::Interpolated(nt) = &self.token.kind { + if let token::NtExpr(..) = **nt { + is_interpolated_expr = true; + } + } + let token_tree = if is_interpolated_expr { + // We need to accept arbitrary interpolated expressions to continue + // supporting things like `doc = $expr` that work on stable. + // Non-literal interpolated expressions are rejected after expansion. + self.parse_token_tree() + } else { + self.parse_unsuffixed_lit()?.token_tree() + }; + TokenStream::new(vec![eq.into(), token_tree.into()]) + } else { + TokenStream::default() + }; + ast::AttrItem { path, tokens } + }) + } + + /// 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*`. + crate fn parse_inner_attributes(&mut self) -> PResult<'a, Vec<ast::Attribute>> { + let mut attrs: Vec<ast::Attribute> = vec![]; + loop { + match self.token.kind { + token::Pound => { + // Don't even try to parse if it's not an inner attribute. + if !self.look_ahead(1, |t| t == &token::Not) { + break; + } + + let attr = self.parse_attribute(true)?; + assert_eq!(attr.style, ast::AttrStyle::Inner); + attrs.push(attr); + } + token::DocComment(s) => { + // We need to get the position of this token before we bump. + let attr = self.mk_doc_comment(s); + if attr.style == ast::AttrStyle::Inner { + attrs.push(attr); + self.bump(); + } else { + break; + } + } + _ => break, + } + } + Ok(attrs) + } + + fn parse_unsuffixed_lit(&mut self) -> PResult<'a, ast::Lit> { + let lit = self.parse_lit()?; + debug!("checking if {:?} is unusuffixed", lit); + + if !lit.kind.is_unsuffixed() { + let msg = "suffixed literals are not allowed in attributes"; + self.diagnostic().struct_span_err(lit.span, msg) + .help("instead of using a suffixed literal \ + (1u8, 1.0f32, etc.), use an unsuffixed version \ + (1, 1.0, etc.).") + .emit() + } + + 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)>)> { + self.expect(&token::OpenDelim(token::Paren))?; + + 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.check(&token::CloseDelim(token::Paren)) { + let lo = self.token.span.lo(); + let item = self.parse_attr_item()?; + expanded_attrs.push((item, self.prev_span.with_lo(lo))); + self.expect_one_of(&[token::Comma], &[token::CloseDelim(token::Paren)])?; + } + + self.expect(&token::CloseDelim(token::Paren))?; + Ok((cfg_predicate, expanded_attrs)) + } + + /// Matches the following grammar (per RFC 1559). + /// + /// meta_item : PATH ( '=' UNSUFFIXED_LIT | '(' meta_item_inner? ')' )? ; + /// meta_item_inner : (meta_item | UNSUFFIXED_LIT) (',' meta_item_inner)? ; + pub fn parse_meta_item(&mut self) -> PResult<'a, ast::MetaItem> { + let nt_meta = match self.token.kind { + token::Interpolated(ref nt) => match **nt { + token::NtMeta(ref e) => Some(e.clone()), + _ => None, + }, + _ => None, + }; + + if let Some(item) = nt_meta { + return match item.meta(item.path.span) { + Some(meta) => { + self.bump(); + 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_span); + Ok(ast::MetaItem { path, kind, span }) + } + + crate fn parse_meta_item_kind(&mut self) -> PResult<'a, ast::MetaItemKind> { + Ok(if self.eat(&token::Eq) { + ast::MetaItemKind::NameValue(self.parse_unsuffixed_lit()?) + } else if self.eat(&token::OpenDelim(token::Paren)) { + ast::MetaItemKind::List(self.parse_meta_seq()?) + } else { + ast::MetaItemKind::Word + }) + } + + /// Matches `meta_item_inner : (meta_item | UNSUFFIXED_LIT) ;`. + fn parse_meta_item_inner(&mut self) -> PResult<'a, ast::NestedMetaItem> { + match self.parse_unsuffixed_lit() { + Ok(lit) => { + return Ok(ast::NestedMetaItem::Literal(lit)) + } + Err(ref mut err) => err.cancel(), + } + + match self.parse_meta_item() { + Ok(mi) => { + return Ok(ast::NestedMetaItem::MetaItem(mi)) + } + Err(ref mut err) => err.cancel(), + } + + let found = self.this_token_to_string(); + let msg = format!("expected unsuffixed literal or identifier, found `{}`", found); + Err(self.diagnostic().struct_span_err(self.token.span, &msg)) + } + + /// Matches `meta_seq = ( COMMASEP(meta_item_inner) )`. + fn parse_meta_seq(&mut self) -> PResult<'a, Vec<ast::NestedMetaItem>> { + self.parse_seq_to_end(&token::CloseDelim(token::Paren), + SeqSep::trailing_allowed(token::Comma), + |p: &mut Parser<'a>| p.parse_meta_item_inner()) + } +} diff --git a/src/librustc_parse/parser/diagnostics.rs b/src/librustc_parse/parser/diagnostics.rs new file mode 100644 index 00000000000..38eae008537 --- /dev/null +++ b/src/librustc_parse/parser/diagnostics.rs @@ -0,0 +1,1549 @@ +use super::{BlockMode, PathStyle, SemiColonMode, TokenType, TokenExpectType, SeqSep, Parser}; + +use syntax::ast::{ + self, Param, BinOpKind, BindingMode, BlockCheckMode, Expr, ExprKind, Ident, Item, ItemKind, + Mutability, Pat, PatKind, PathSegment, QSelf, Ty, TyKind, +}; +use syntax::token::{self, TokenKind, token_can_begin_expr}; +use syntax::print::pprust; +use syntax::ptr::P; +use syntax::symbol::{kw, sym}; +use syntax::ThinVec; +use syntax::util::parser::AssocOp; +use syntax::struct_span_err; + +use errors::{PResult, Applicability, DiagnosticBuilder, DiagnosticId, pluralize}; +use rustc_data_structures::fx::FxHashSet; +use syntax_pos::{Span, DUMMY_SP, MultiSpan, SpanSnippetError}; +use log::{debug, trace}; +use std::mem; + +const TURBOFISH: &'static str = "use `::<...>` instead of `<...>` to specify type arguments"; + +/// Creates a placeholder argument. +pub(super) fn dummy_arg(ident: Ident) -> Param { + let pat = P(Pat { + id: ast::DUMMY_NODE_ID, + kind: PatKind::Ident(BindingMode::ByValue(Mutability::Immutable), ident, None), + span: ident.span, + }); + let ty = Ty { + kind: TyKind::Err, + span: ident.span, + id: ast::DUMMY_NODE_ID + }; + Param { + attrs: ThinVec::default(), + id: ast::DUMMY_NODE_ID, + pat, + span: ident.span, + ty: P(ty), + is_placeholder: false, + } +} + +pub enum Error { + FileNotFoundForModule { + mod_name: String, + default_path: String, + secondary_path: String, + dir_path: String, + }, + DuplicatePaths { + mod_name: String, + default_path: String, + secondary_path: String, + }, + UselessDocComment, + InclusiveRangeWithNoEnd, +} + +impl Error { + fn span_err<S: Into<MultiSpan>>( + self, + sp: S, + handler: &errors::Handler, + ) -> DiagnosticBuilder<'_> { + match self { + Error::FileNotFoundForModule { + ref mod_name, + ref default_path, + ref secondary_path, + ref dir_path, + } => { + let mut err = struct_span_err!( + handler, + sp, + E0583, + "file not found for module `{}`", + mod_name, + ); + err.help(&format!( + "name the file either {} or {} inside the directory \"{}\"", + default_path, + secondary_path, + dir_path, + )); + err + } + Error::DuplicatePaths { ref mod_name, ref default_path, ref secondary_path } => { + let mut err = struct_span_err!( + handler, + sp, + E0584, + "file for module `{}` found at both {} and {}", + mod_name, + default_path, + secondary_path, + ); + err.help("delete or rename one of them to remove the ambiguity"); + err + } + Error::UselessDocComment => { + let mut err = struct_span_err!( + handler, + sp, + E0585, + "found a documentation comment that doesn't document anything", + ); + err.help("doc comments must come before what they document, maybe a comment was \ + intended with `//`?"); + err + } + Error::InclusiveRangeWithNoEnd => { + let mut err = struct_span_err!( + handler, + sp, + E0586, + "inclusive range with no end", + ); + err.help("inclusive ranges must be bounded at the end (`..=b` or `a..=b`)"); + err + } + } + } +} + +pub(super) trait RecoverQPath: Sized + 'static { + const PATH_STYLE: PathStyle = PathStyle::Expr; + fn to_ty(&self) -> Option<P<Ty>>; + fn recovered(qself: Option<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<QSelf>, path: ast::Path) -> Self { + Self { + span: path.span, + kind: TyKind::Path(qself, path), + id: ast::DUMMY_NODE_ID, + } + } +} + +impl RecoverQPath for Pat { + fn to_ty(&self) -> Option<P<Ty>> { + self.to_ty() + } + fn recovered(qself: Option<QSelf>, path: ast::Path) -> Self { + Self { + span: path.span, + kind: PatKind::Path(qself, path), + id: ast::DUMMY_NODE_ID, + } + } +} + +impl RecoverQPath for Expr { + fn to_ty(&self) -> Option<P<Ty>> { + self.to_ty() + } + fn recovered(qself: Option<QSelf>, path: ast::Path) -> Self { + Self { + span: path.span, + kind: ExprKind::Path(qself, path), + attrs: ThinVec::new(), + id: ast::DUMMY_NODE_ID, + } + } +} + +/// Control whether the closing delimiter should be consumed when calling `Parser::consume_block`. +crate enum ConsumeClosingDelim { + Yes, + No, +} + +impl<'a> Parser<'a> { + pub fn fatal(&self, m: &str) -> DiagnosticBuilder<'a> { + self.span_fatal(self.token.span, m) + } + + crate fn span_fatal<S: Into<MultiSpan>>(&self, sp: S, m: &str) -> DiagnosticBuilder<'a> { + self.sess.span_diagnostic.struct_span_fatal(sp, m) + } + + pub(super) fn span_fatal_err<S: Into<MultiSpan>>( + &self, + sp: S, + err: Error, + ) -> DiagnosticBuilder<'a> { + err.span_err(sp, self.diagnostic()) + } + + pub(super) fn bug(&self, m: &str) -> ! { + self.sess.span_diagnostic.span_bug(self.token.span, m) + } + + pub(super) fn span_err<S: Into<MultiSpan>>(&self, sp: S, m: &str) { + self.sess.span_diagnostic.span_err(sp, m) + } + + pub fn struct_span_err<S: Into<MultiSpan>>(&self, sp: S, m: &str) -> DiagnosticBuilder<'a> { + self.sess.span_diagnostic.struct_span_err(sp, m) + } + + pub fn span_bug<S: Into<MultiSpan>>(&self, sp: S, m: &str) -> ! { + self.sess.span_diagnostic.span_bug(sp, m) + } + + pub(super) fn diagnostic(&self) -> &'a errors::Handler { + &self.sess.span_diagnostic + } + + pub(super) fn span_to_snippet(&self, span: Span) -> Result<String, SpanSnippetError> { + self.sess.source_map().span_to_snippet(span) + } + + pub(super) fn expected_ident_found(&self) -> DiagnosticBuilder<'a> { + let mut err = self.struct_span_err( + self.token.span, + &format!("expected identifier, found {}", self.this_token_descr()), + ); + if let token::Ident(name, false) = self.token.kind { + if Ident::new(name, self.token.span).is_raw_guess() { + err.span_suggestion( + self.token.span, + "you can escape reserved keywords to use them as identifiers", + format!("r#{}", name), + Applicability::MaybeIncorrect, + ); + } + } + if let Some(token_descr) = self.token_descr() { + err.span_label(self.token.span, format!("expected identifier, found {}", token_descr)); + } else { + err.span_label(self.token.span, "expected identifier"); + if self.token == token::Comma && self.look_ahead(1, |t| t.is_ident()) { + err.span_suggestion( + self.token.span, + "remove this comma", + String::new(), + Applicability::MachineApplicable, + ); + } + } + err + } + + pub(super) fn expected_one_of_not_found( + &mut self, + edible: &[TokenKind], + inedible: &[TokenKind], + ) -> PResult<'a, bool /* recovered */> { + 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(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 + }) + } + + let mut expected = edible.iter() + .map(|x| TokenType::Token(x.clone())) + .chain(inedible.iter().map(|x| TokenType::Token(x.clone()))) + .chain(self.expected_tokens.iter().cloned()) + .collect::<Vec<_>>(); + expected.sort_by_cached_key(|x| x.to_string()); + expected.dedup(); + let expect = tokens_to_string(&expected[..]); + let actual = self.this_token_descr(); + let (msg_exp, (label_sp, label_exp)) = if expected.len() > 1 { + let short_expect = if expected.len() > 6 { + format!("{} possible tokens", expected.len()) + } else { + expect.clone() + }; + (format!("expected one of {}, found {}", expect, actual), + (self.sess.source_map().next_point(self.prev_span), + format!("expected one of {}", short_expect))) + } else if expected.is_empty() { + (format!("unexpected token: {}", actual), + (self.prev_span, "unexpected token after this".to_string())) + } else { + (format!("expected {}, found {}", expect, actual), + (self.sess.source_map().next_point(self.prev_span), + format!("expected {}", expect))) + }; + self.last_unexpected_token_span = Some(self.token.span); + let mut err = self.fatal(&msg_exp); + if self.token.is_ident_named(sym::and) { + err.span_suggestion_short( + self.token.span, + "use `&&` instead of `and` for the boolean operator", + "&&".to_string(), + Applicability::MaybeIncorrect, + ); + } + if self.token.is_ident_named(sym::or) { + err.span_suggestion_short( + self.token.span, + "use `||` instead of `or` for the boolean operator", + "||".to_string(), + Applicability::MaybeIncorrect, + ); + } + 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_span + } else { + label_sp + }; + match self.recover_closing_delimiter(&expected.iter().filter_map(|tt| match tt { + TokenType::Token(t) => Some(t.clone()), + _ => None, + }).collect::<Vec<_>>(), err) { + Err(e) => err = e, + Ok(recovered) => { + return Ok(recovered); + } + } + + let sm = self.sess.source_map(); + if self.prev_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"); + } + self.maybe_annotate_with_ascription(&mut err, false); + Err(err) + } + + pub fn maybe_annotate_with_ascription( + &mut self, + err: &mut DiagnosticBuilder<'_>, + maybe_expected_semicolon: bool, + ) { + if let Some((sp, likely_path)) = self.last_type_ascription.take() { + let sm = self.sess.source_map(); + let next_pos = sm.lookup_char_pos(self.token.span.lo()); + let op_pos = sm.lookup_char_pos(sp.hi()); + + let allow_unstable = self.sess.unstable_features.is_nightly_build(); + + if likely_path { + err.span_suggestion( + sp, + "maybe write a path separator here", + "::".to_string(), + if allow_unstable { + Applicability::MaybeIncorrect + } else { + Applicability::MachineApplicable + }, + ); + } else if op_pos.line != next_pos.line && maybe_expected_semicolon { + err.span_suggestion( + sp, + "try using a semicolon", + ";".to_string(), + Applicability::MaybeIncorrect, + ); + } else if allow_unstable { + err.span_label(sp, "tried to parse a type due to this type ascription"); + } else { + err.span_label(sp, "tried to parse a type due to this"); + } + if allow_unstable { + // Give extra information about type ascription only if it's a nightly compiler. + err.note("`#![feature(type_ascription)]` lets you annotate an expression with a \ + type: `<expr>: <type>`"); + err.note("for more information, see \ + https://github.com/rust-lang/rust/issues/23416"); + } + } + } + + /// 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(ref mut 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 _ = vec![1, 2, 3].into_iter().collect::<Vec<usize>>>>(); + /// ^^ help: remove extra angle brackets + /// ``` + pub(super) fn check_trailing_angle_brackets(&mut self, segment: &PathSegment, end: TokenKind) { + // 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() + .map(|args| args.is_angle_bracketed()) + .unwrap_or(false); + + debug!( + "check_trailing_angle_brackets: parsed_angle_bracket_args={:?}", + parsed_angle_bracket_args, + ); + if !parsed_angle_bracket_args { + return; + } + + // 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; + } + + // 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); + *t == end + }) { + // 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 total_num_of_gt = number_of_gt + number_of_shr * 2; + self.diagnostic() + .struct_span_err( + span, + &format!("unmatched angle bracket{}", pluralize!(total_num_of_gt)), + ) + .span_suggestion( + span, + &format!("remove extra angle bracket{}", pluralize!(total_num_of_gt)), + String::new(), + Applicability::MachineApplicable, + ) + .emit(); + } + } + + /// 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: + /// + /// outer_op + /// / \ + /// inner_op r2 + /// / \ + /// l1 r1 + pub(super) fn check_no_chained_comparison( + &mut self, + lhs: &Expr, + outer_op: &AssocOp, + ) -> PResult<'a, Option<P<Expr>>> { + debug_assert!( + outer_op.is_comparison(), + "check_no_chained_comparison: {:?} is not comparison", + outer_op, + ); + + let mk_err_expr = |this: &Self, span| { + Ok(Some(this.mk_expr(span, ExprKind::Err, ThinVec::new()))) + }; + + match lhs.kind { + ExprKind::Binary(op, _, _) if op.node.is_comparison() => { + // Respan to include both operators. + let op_span = op.span.to(self.prev_span); + let mut err = self.struct_span_err( + op_span, + "chained comparison operators require parentheses", + ); + + let suggest = |err: &mut DiagnosticBuilder<'_>| { + err.span_suggestion_verbose( + op_span.shrink_to_lo(), + TURBOFISH, + "::".to_string(), + Applicability::MaybeIncorrect, + ); + }; + + if op.node == BinOpKind::Lt && + *outer_op == AssocOp::Less || // Include `<` to provide this recommendation + *outer_op == AssocOp::Greater // even in a case like the following: + { // Foo<Bar<Baz<Qux, ()>>> + if *outer_op == AssocOp::Less { + let snapshot = self.clone(); + 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(token::Paren), + token::ModSep, + ].contains(&self.token.kind) { + // We don't have `foo< bar >(` or `foo< bar >::`, so we rewind the + // parser and bail out. + mem::replace(self, snapshot.clone()); + } + } + return if token::ModSep == self.token.kind { + // We have some certainty that this was a bad turbofish at this point. + // `foo< bar >::` + suggest(&mut err); + + let snapshot = self.clone(); + 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. + err.emit(); + // 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, lhs.span.to(self.prev_span)) + } + Err(mut expr_err) => { + expr_err.cancel(); + // Not entirely sure now, but we bubble the error up with the + // suggestion. + mem::replace(self, snapshot); + Err(err) + } + } + } else if token::OpenDelim(token::Paren) == self.token.kind { + // We have high certainty that this was a bad turbofish at this point. + // `foo< bar >(` + suggest(&mut err); + // Consume the fn call arguments. + match self.consume_fn_args() { + Err(()) => Err(err), + Ok(()) => { + err.emit(); + // 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, lhs.span.to(self.prev_span)) + } + } + } else { + // 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); + err.help("or use `(...)` if you meant to specify fn arguments"); + // These cases cause too many knock-down errors, bail out (#61329). + Err(err) + }; + } + err.emit(); + } + _ => {} + } + Ok(None) + } + + fn consume_fn_args(&mut self) -> Result<(), ()> { + let snapshot = self.clone(); + self.bump(); // `(` + + // Consume the fn call arguments. + let modifiers = [ + (token::OpenDelim(token::Paren), 1), + (token::CloseDelim(token::Paren), -1), + ]; + self.consume_tts(1, &modifiers[..]); + + if self.token.kind == token::Eof { + // Not entirely sure that what we consumed were fn arguments, rollback. + mem::replace(self, snapshot); + Err(()) + } else { + // 99% certain that the suggestion is correct, continue parsing. + Ok(()) + } + } + + pub(super) fn maybe_report_ambiguous_plus( + &mut self, + allow_plus: bool, + impl_dyn_multi: bool, + ty: &Ty, + ) { + if !allow_plus && impl_dyn_multi { + let sum_with_parens = format!("({})", pprust::ty_to_string(&ty)); + self.struct_span_err(ty.span, "ambiguous `+` in a type") + .span_suggestion( + ty.span, + "use parentheses to disambiguate", + sum_with_parens, + Applicability::MachineApplicable, + ) + .emit(); + } + } + + pub(super) fn maybe_recover_from_bad_type_plus( + &mut self, + allow_plus: bool, + ty: &Ty, + ) -> PResult<'a, ()> { + // Do not add `+` to expected tokens. + if !allow_plus || !self.token.is_like_plus() { + return Ok(()); + } + + self.bump(); // `+` + let bounds = self.parse_generic_bounds(None)?; + let sum_span = ty.span.to(self.prev_span); + + let mut err = struct_span_err!( + self.sess.span_diagnostic, + sum_span, + E0178, + "expected a path on the left-hand side of `+`, not `{}`", + pprust::ty_to_string(ty) + ); + + match ty.kind { + TyKind::Rptr(ref lifetime, ref mut_ty) => { + let sum_with_parens = pprust::to_string(|s| { + s.s.word("&"); + s.print_opt_lifetime(lifetime); + s.print_mutability(mut_ty.mutbl); + s.popen(); + s.print_type(&mut_ty.ty); + s.print_type_bounds(" +", &bounds); + s.pclose() + }); + err.span_suggestion( + sum_span, + "try adding parentheses", + sum_with_parens, + Applicability::MachineApplicable, + ); + } + TyKind::Ptr(..) | TyKind::BareFn(..) => { + err.span_label(sum_span, "perhaps you forgot parentheses?"); + } + _ => { + err.span_label(sum_span, "expected a path"); + } + } + err.emit(); + Ok(()) + } + + /// 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>, + allow_recovery: bool, + ) -> PResult<'a, P<T>> { + // Do not add `::` to expected tokens. + if allow_recovery && self.token == token::ModSep { + 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::ModSep)?; + + let mut path = ast::Path { + segments: Vec::new(), + span: DUMMY_SP, + }; + self.parse_path_segments(&mut path.segments, T::PATH_STYLE)?; + path.span = ty_span.to(self.prev_span); + + let ty_str = self + .span_to_snippet(ty_span) + .unwrap_or_else(|_| pprust::ty_to_string(&ty)); + self.diagnostic() + .struct_span_err(path.span, "missing angle brackets in associated item path") + .span_suggestion( + // This is a best-effort recovery. + path.span, + "try", + format!("<{}>::{}", ty_str, pprust::path_to_string(&path)), + Applicability::MaybeIncorrect, + ) + .emit(); + + let path_span = ty_span.shrink_to_hi(); // Use an empty path since `position == 0`. + Ok(P(T::recovered( + Some(QSelf { + ty, + path_span, + position: 0, + }), + path, + ))) + } + + pub(super) fn maybe_consume_incorrect_semicolon(&mut self, items: &[P<Item>]) -> bool { + if self.eat(&token::Semi) { + let mut err = self.struct_span_err(self.prev_span, "expected item, found `;`"); + err.span_suggestion_short( + self.prev_span, + "remove this semicolon", + String::new(), + Applicability::MachineApplicable, + ); + if !items.is_empty() { + let previous_item = &items[items.len() - 1]; + let previous_item_kind_name = match previous_item.kind { + // Say "braced struct" because tuple-structs and + // braceless-empty-struct declarations do take a semicolon. + ItemKind::Struct(..) => Some("braced struct"), + ItemKind::Enum(..) => Some("enum"), + ItemKind::Trait(..) => Some("trait"), + ItemKind::Union(..) => Some("union"), + _ => None, + }; + if let Some(name) = previous_item_kind_name { + err.help(&format!( + "{} declarations are not followed by a semicolon", + name + )); + } + } + err.emit(); + true + } else { + false + } + } + + /// Creates a `DiagnosticBuilder` 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, bool /* recovered */> { + let token_str = pprust::token_kind_to_string(t); + let this_token_str = self.this_token_descr(); + 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.sess.source_map().next_point(self.token.span); + (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_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_span, self.token.span), + _ => (self.sess.source_map().next_point(self.prev_span), 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.struct_span_err(sp, &msg); + let label_exp = format!("expected `{}`", token_str); + match self.recover_closing_delimiter(&[t.clone()], err) { + Err(e) => err = e, + Ok(recovered) => { + return Ok(recovered); + } + } + let sm = self.sess.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) { + return Ok(()); + } + let sm = self.sess.source_map(); + let msg = format!("expected `;`, found `{}`", self.this_token_descr()); + let appl = Applicability::MachineApplicable; + if self.token.span == DUMMY_SP || self.prev_span == DUMMY_SP { + // Likely inside a macro, can't provide meaninful suggestions. + return self.expect(&token::Semi).map(|_| ()); + } else if !sm.is_multiline(self.prev_span.until(self.token.span)) { + // The current token is in the same line as the prior token, not recoverable. + } else if self.look_ahead(1, |t| t == &token::CloseDelim(token::Brace) + || token_can_begin_expr(t) && 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; + self.bump(); + let sp = self.prev_span; + self.struct_span_err(sp, &msg) + .span_suggestion(sp, "change this to `;`", ";".to_string(), appl) + .emit(); + return Ok(()) + } else if self.look_ahead(0, |t| t == &token::CloseDelim(token::Brace) || ( + token_can_begin_expr(t) + && t != &token::Semi + && t != &token::Pound // Avoid triggering with too many trailing `#` in raw string. + )) { + // 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 sp = self.prev_span.shrink_to_hi(); + self.struct_span_err(sp, &msg) + .span_label(self.token.span, "unexpected token") + .span_suggestion_short(sp, "add `;` here", ";".to_string(), appl) + .emit(); + return Ok(()) + } + self.expect(&token::Semi).map(|_| ()) // Error unconditionally + } + + pub(super) fn parse_semi_or_incorrect_foreign_fn_body( + &mut self, + ident: &Ident, + extern_sp: Span, + ) -> PResult<'a, ()> { + if self.token != token::Semi { + // This might be an incorrect fn definition (#62109). + let parser_snapshot = self.clone(); + match self.parse_inner_attrs_and_block() { + Ok((_, body)) => { + self.struct_span_err(ident.span, "incorrect `fn` inside `extern` block") + .span_label(ident.span, "can't have a body") + .span_label(body.span, "this body is invalid here") + .span_label( + extern_sp, + "`extern` blocks define existing foreign functions and `fn`s \ + inside of them cannot have a body") + .help("you might have meant to write a function accessible through ffi, \ + which can be done by writing `extern fn` outside of the \ + `extern` block") + .note("for more information, visit \ + https://doc.rust-lang.org/std/keyword.extern.html") + .emit(); + } + Err(mut err) => { + err.cancel(); + mem::replace(self, parser_snapshot); + self.expect_semi()?; + } + } + } else { + self.bump(); + } + Ok(()) + } + + /// Consumes alternative await syntaxes like `await!(<expr>)`, `await <expr>`, + /// `await? <expr>`, `await(<expr>)`, and `await { <expr> }`. + pub(super) fn parse_incorrect_await_syntax( + &mut self, + lo: Span, + await_sp: Span, + ) -> PResult<'a, (Span, ExprKind)> { + if self.token == token::Not { + // Handle `await!(<expr>)`. + self.expect(&token::Not)?; + self.expect(&token::OpenDelim(token::Paren))?; + let expr = self.parse_expr()?; + self.expect(&token::CloseDelim(token::Paren))?; + let sp = self.error_on_incorrect_await(lo, self.prev_span, &expr, false); + return Ok((sp, ExprKind::Await(expr))) + } + + let is_question = self.eat(&token::Question); // Handle `await? <expr>`. + let expr = if self.token == token::OpenDelim(token::Brace) { + // Handle `await { <expr> }`. + // This needs to be handled separatedly from the next arm to avoid + // interpreting `await { <expr> }?` as `<expr>?.await`. + self.parse_block_expr( + None, + self.token.span, + BlockCheckMode::Default, + ThinVec::new(), + ) + } else { + self.parse_expr() + }.map_err(|mut err| { + err.span_label(await_sp, "while parsing this incorrect await expression"); + err + })?; + let sp = self.error_on_incorrect_await(lo, expr.span, &expr, is_question); + Ok((sp, ExprKind::Await(expr))) + } + + fn error_on_incorrect_await(&self, lo: Span, hi: Span, expr: &Expr, is_question: bool) -> Span { + let expr_str = self.span_to_snippet(expr.span) + .unwrap_or_else(|_| pprust::expr_to_string(&expr)); + let suggestion = format!("{}.await{}", expr_str, if is_question { "?" } else { "" }); + let sp = lo.to(hi); + let app = match expr.kind { + ExprKind::Try(_) => Applicability::MaybeIncorrect, // `await <expr>?` + _ => Applicability::MachineApplicable, + }; + self.struct_span_err(sp, "incorrect use of `await`") + .span_suggestion(sp, "`await` is a postfix operation", suggestion, app) + .emit(); + sp + } + + /// If encountering `future.await()`, consumes and emits an error. + pub(super) fn recover_from_await_method_call(&mut self) { + if self.token == token::OpenDelim(token::Paren) && + self.look_ahead(1, |t| t == &token::CloseDelim(token::Paren)) + { + // future.await() + let lo = self.token.span; + self.bump(); // ( + let sp = lo.to(self.token.span); + self.bump(); // ) + self.struct_span_err(sp, "incorrect use of `await`") + .span_suggestion( + sp, + "`await` is not a method call, remove the parentheses", + String::new(), + Applicability::MachineApplicable, + ).emit() + } + } + + /// Recovers a situation like `for ( $pat in $expr )` + /// and suggest writing `for $pat in $expr` instead. + /// + /// This should be called before parsing the `$block`. + pub(super) fn recover_parens_around_for_head( + &mut self, + pat: P<Pat>, + expr: &Expr, + begin_paren: Option<Span>, + ) -> P<Pat> { + match (&self.token.kind, begin_paren) { + (token::CloseDelim(token::Paren), Some(begin_par_sp)) => { + self.bump(); + + let pat_str = self + // Remove the `(` from the span of the pattern: + .span_to_snippet(pat.span.trim_start(begin_par_sp).unwrap()) + .unwrap_or_else(|_| pprust::pat_to_string(&pat)); + + self.struct_span_err(self.prev_span, "unexpected closing `)`") + .span_label(begin_par_sp, "opening `(`") + .span_suggestion( + begin_par_sp.to(self.prev_span), + "remove parenthesis in `for` loop", + format!("{} in {}", pat_str, pprust::expr_to_string(&expr)), + // 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. + Applicability::MachineApplicable, + ) + .emit(); + + // Unwrap `(pat)` into `pat` to avoid the `unused_parens` lint. + pat.and_then(|pat| match pat.kind { + PatKind::Paren(pat) => pat, + _ => P(pat), + }) + } + _ => pat, + } + } + + pub(super) fn could_ascription_be_path(&self, node: &ast::ExprKind) -> bool { + (self.token == token::Lt && // `foo:<bar`, likely a typoed turbofish. + self.look_ahead(1, |t| t.is_ident() && !t.is_reserved_ident()) + ) || + self.token.is_ident() && + match node { + // `foo::` → `foo:` or `foo.bar::` → `foo.bar:` + ast::ExprKind::Path(..) | ast::ExprKind::Field(..) => true, + _ => false, + } && + !self.token.is_reserved_ident() && // v `foo:bar(baz)` + self.look_ahead(1, |t| t == &token::OpenDelim(token::Paren)) || + self.look_ahead(1, |t| t == &token::Lt) && // `foo:bar<baz` + self.look_ahead(2, |t| t.is_ident()) || + self.look_ahead(1, |t| t == &token::Colon) && // `foo:bar:baz` + self.look_ahead(2, |t| t.is_ident()) || + self.look_ahead(1, |t| t == &token::ModSep) && + (self.look_ahead(2, |t| t.is_ident()) || // `foo:bar::baz` + self.look_ahead(2, |t| t == &token::Lt)) // `foo:bar::<baz>` + } + + pub(super) fn recover_seq_parse_error( + &mut self, + delim: token::DelimToken, + lo: Span, + result: PResult<'a, P<Expr>>, + ) -> P<Expr> { + match result { + Ok(x) => x, + Err(mut err) => { + 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_span), ExprKind::Err, ThinVec::new()) + } + } + } + + pub(super) fn recover_closing_delimiter( + &mut self, + tokens: &[TokenKind], + mut err: DiagnosticBuilder<'a>, + ) -> PResult<'a, bool> { + let mut pos = None; + // We want to use the last closing delim that would apply. + for (i, unmatched) in self.unclosed_delims.iter().enumerate().rev() { + if tokens.contains(&token::CloseDelim(unmatched.expected_delim)) + && Some(self.token.span) > unmatched.unclosed_span + { + pos = Some(i); + } + } + match pos { + Some(pos) => { + // Recover and assume that the detected unclosed delimiter was meant for + // this location. Emit the diagnostic and act as if the delimiter was + // present for the parser's sake. + + // Don't attempt to recover from this unclosed delimiter more than once. + let unmatched = self.unclosed_delims.remove(pos); + let delim = TokenType::Token(token::CloseDelim(unmatched.expected_delim)); + if unmatched.found_delim.is_none() { + // We encountered `Eof`, set this fact here to avoid complaining about missing + // `fn main()` when we found place to suggest the closing brace. + *self.sess.reached_eof.borrow_mut() = true; + } + + // We want to suggest the inclusion of the closing delimiter where it makes + // the most sense, which is immediately after the last token: + // + // {foo(bar {}} + // - ^ + // | | + // | help: `)` may belong here + // | + // unclosed delimiter + if let Some(sp) = unmatched.unclosed_span { + err.span_label(sp, "unclosed delimiter"); + } + err.span_suggestion_short( + self.sess.source_map().next_point(self.prev_span), + &format!("{} may belong here", delim.to_string()), + delim.to_string(), + Applicability::MaybeIncorrect, + ); + if unmatched.found_delim.is_none() { + // Encountered `Eof` when lexing blocks. Do not recover here to avoid knockdown + // errors which would be emitted elsewhere in the parser and let other error + // recovery consume the rest of the file. + Err(err) + } else { + err.emit(); + self.expected_tokens.clear(); // Reduce the number of errors. + Ok(true) + } + } + _ => Err(err), + } + } + + /// Recovers from `pub` keyword in places where it seems _reasonable_ but isn't valid. + pub(super) fn eat_bad_pub(&mut self) { + // When `unclosed_delims` is populated, it means that the code being parsed is already + // quite malformed, which might mean that, for example, a pub struct definition could be + // parsed as being a trait item, which is invalid and this error would trigger + // unconditionally, resulting in misleading diagnostics. Because of this, we only attempt + // this nice to have recovery for code that is otherwise well formed. + if self.token.is_keyword(kw::Pub) && self.unclosed_delims.is_empty() { + match self.parse_visibility(false) { + Ok(vis) => { + self.diagnostic() + .struct_span_err(vis.span, "unnecessary visibility qualifier") + .span_label(vis.span, "`pub` not permitted here") + .emit(); + } + Err(mut err) => err.emit(), + } + } + } + + /// 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(token::DelimToken::Brace) => { + brace_depth += 1; + self.bump(); + if break_on_block == BlockMode::Break && + brace_depth == 1 && + bracket_depth == 0 { + in_block = true; + } + } + token::OpenDelim(token::DelimToken::Bracket) => { + bracket_depth += 1; + self.bump(); + } + token::CloseDelim(token::DelimToken::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(token::DelimToken::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 => + { + debug!("recover_stmt_ return - Semi"); + 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.struct_span_err(self.prev_span, "expected iterable, found keyword `in`") + .span_suggestion_short( + in_span.until(self.prev_span), + "remove the duplicated `in`", + String::new(), + Applicability::MachineApplicable, + ) + .emit(); + } + } + + pub(super) fn expected_semi_or_open_brace<T>(&mut self) -> PResult<'a, T> { + let token_str = self.this_token_descr(); + let mut err = self.fatal(&format!("expected `;` or `{{`, found {}", token_str)); + err.span_label(self.token.span, "expected `;` or `{`"); + Err(err) + } + + pub(super) fn eat_incorrect_doc_comment_for_param_type(&mut self) { + if let token::DocComment(_) = self.token.kind { + self.struct_span_err( + self.token.span, + "documentation comments cannot be applied to a function parameter's type", + ) + .span_label(self.token.span, "doc comments are not allowed here") + .emit(); + self.bump(); + } else if self.token == token::Pound && self.look_ahead(1, |t| { + *t == token::OpenDelim(token::Bracket) + }) { + let lo = self.token.span; + // Skip every token until next possible arg. + while self.token != token::CloseDelim(token::Bracket) { + self.bump(); + } + let sp = lo.to(self.token.span); + self.bump(); + self.struct_span_err( + sp, + "attributes cannot be applied to a function parameter's type", + ) + .span_label(sp, "attributes are not allowed here") + .emit(); + } + } + + pub(super) fn parameter_without_type( + &mut self, + err: &mut DiagnosticBuilder<'_>, + pat: P<ast::Pat>, + require_name: bool, + is_self_allowed: bool, + is_trait_item: 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(token::Paren) + }) { // `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", + String::from("<identifier>: <type>"), + Applicability::HasPlaceholders, + ); + return Some(ident); + } else if let PatKind::Ident(_, ident, _) = pat.kind { + if require_name && ( + is_trait_item || + self.token == token::Comma || + self.token == token::Lt || + self.token == token::CloseDelim(token::Paren) + ) { // `fn foo(a, b) {}`, `fn foo(a<x>, b<y>) {}` or `fn foo(usize, usize) {}` + if is_self_allowed { + err.span_suggestion( + pat.span, + "if this is a `self` type, give it a parameter name", + format!("self: {}", ident), + 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( + pat.span, + "if this was a parameter name, give it a type", + format!("{}: TypeName", ident), + Applicability::HasPlaceholders, + ); + } + err.span_suggestion( + pat.span, + "if this is a type, explicitly ignore the parameter name", + format!("_: {}", ident), + Applicability::MachineApplicable, + ); + err.note("anonymous parameters are removed in the 2018 edition (see RFC 1685)"); + + // 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(Some("argument name"))?; + self.expect(&token::Colon)?; + let ty = self.parse_ty()?; + + self.diagnostic() + .struct_span_err_with_code( + pat.span, + "patterns aren't allowed in methods without bodies", + DiagnosticId::Error("E0642".into()), + ) + .span_suggestion_short( + pat.span, + "give this argument a name or use an underscore to ignore it", + "_".to_owned(), + Applicability::MachineApplicable, + ) + .emit(); + + // 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 + }); + Ok((pat, ty)) + } + + pub(super) fn recover_bad_self_param( + &mut self, + mut param: ast::Param, + is_trait_item: bool, + ) -> PResult<'a, ast::Param> { + let sp = param.pat.span; + param.ty.kind = TyKind::Err; + let mut err = self.struct_span_err(sp, "unexpected `self` parameter in function"); + if is_trait_item { + err.span_label(sp, "must be the first associated function parameter"); + } else { + err.span_label(sp, "not valid as function parameter"); + err.note("`self` is only valid as the first parameter of an associated function"); + } + err.emit(); + Ok(param) + } + + pub(super) fn consume_block( + &mut self, + delim: token::DelimToken, + 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 || self.eat(&token::CloseDelim(token::NoDelim)) { + return; + } else { + self.bump(); + } + } + } + + pub(super) fn expected_expression_found(&self) -> DiagnosticBuilder<'a> { + let (span, msg) = match (&self.token.kind, self.subparser_name) { + (&token::Eof, Some(origin)) => { + let sp = self.sess.source_map().next_point(self.token.span); + (sp, format!("expected expression, found end of {}", origin)) + } + _ => (self.token.span, format!( + "expected expression, found {}", + self.this_token_descr(), + )), + }; + let mut err = self.struct_span_err(span, &msg); + let sp = self.sess.source_map().start_point(self.token.span); + if let Some(sp) = self.sess.ambiguous_block_expr_parse.borrow().get(&sp) { + self.sess.expr_parentheses_needed(&mut err, *sp, None); + } + err.span_label(span, "expected expression"); + 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 unecessary 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 + /// unecessary 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 Vec<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); + } + } + } +} diff --git a/src/librustc_parse/parser/expr.rs b/src/librustc_parse/parser/expr.rs new file mode 100644 index 00000000000..dadb91f8b3c --- /dev/null +++ b/src/librustc_parse/parser/expr.rs @@ -0,0 +1,1963 @@ +use super::{Parser, Restrictions, PrevTokenKind, TokenType, PathStyle, BlockMode}; +use super::{SemiColonMode, SeqSep, TokenExpectType}; +use super::pat::{GateOr, PARAM_EXPECTED}; +use super::diagnostics::Error; +use crate::maybe_recover_from_interpolated_ty_qpath; + +use syntax::ast::{ + self, DUMMY_NODE_ID, Attribute, AttrStyle, Ident, CaptureBy, BlockCheckMode, + Expr, ExprKind, RangeLimits, Label, Movability, IsAsync, Arm, Ty, TyKind, + FunctionRetTy, Param, FnDecl, BinOpKind, BinOp, UnOp, Mac, AnonConst, Field, Lit, +}; +use syntax::token::{self, Token, TokenKind}; +use syntax::print::pprust; +use syntax::ptr::P; +use syntax::source_map::{self, Span}; +use syntax::util::classify; +use syntax::util::literal::LitError; +use syntax::util::parser::{AssocOp, Fixity, prec_let_scrutinee_needs_par}; +use syntax_pos::symbol::{kw, sym}; +use syntax_pos::Symbol; +use errors::{PResult, Applicability}; +use std::mem; +use rustc_data_structures::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.token.span, ExprKind::Path(None, path), ThinVec::new() + )); + } + token::NtBlock(block) => { + let block = block.clone(); + $p.bump(); + return Ok($p.mk_expr( + $p.token.span, ExprKind::Block(block, None), ThinVec::new() + )); + } + // N.B., `NtIdent(ident)` is normalized to `Ident` in `fn bump`. + _ => {}, + }; + } + } +} + +#[derive(Debug)] +pub(super) enum LhsExpr { + NotYetParsed, + AttributesParsed(ThinVec<Attribute>), + AlreadyParsed(P<Expr>), +} + +impl From<Option<ThinVec<Attribute>>> for LhsExpr { + /// Converts `Some(attrs)` into `LhsExpr::AttributesParsed(attrs)` + /// and `None` into `LhsExpr::NotYetParsed`. + /// + /// This conversion does not allocate. + fn from(o: Option<ThinVec<Attribute>>) -> 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)`. + /// + /// This conversion does not allocate. + fn from(expr: P<Expr>) -> Self { + LhsExpr::AlreadyParsed(expr) + } +} + +impl<'a> Parser<'a> { + /// Parses an expression. + #[inline] + pub fn parse_expr(&mut self) -> PResult<'a, P<Expr>> { + self.parse_expr_res(Restrictions::empty(), None) + } + + fn parse_paren_expr_seq(&mut self) -> PResult<'a, Vec<P<Expr>>> { + self.parse_paren_comma_seq(|p| { + match p.parse_expr() { + Ok(expr) => Ok(expr), + Err(mut err) => match p.token.kind { + token::Ident(name, false) + if name == kw::Underscore && p.look_ahead(1, |t| { + t == &token::Comma + }) => { + // Special-case handling of `foo(_, _, _)` + err.emit(); + let sp = p.token.span; + p.bump(); + Ok(p.mk_expr(sp, ExprKind::Err, ThinVec::new())) + } + _ => Err(err), + }, + } + }).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<ThinVec<Attribute>> + ) -> PResult<'a, P<Expr>> { + self.with_res(r, |this| this.parse_assoc_expr(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_assoc_expr( + &mut self, + already_parsed_attrs: Option<ThinVec<Attribute>>, + ) -> PResult<'a, P<Expr>> { + self.parse_assoc_expr_with(0, already_parsed_attrs.into()) + } + + /// Parses an associative expression with operators of at least `min_prec` precedence. + pub(super) fn parse_assoc_expr_with( + &mut self, + min_prec: usize, + lhs: LhsExpr, + ) -> PResult<'a, P<Expr>> { + let mut lhs = if let LhsExpr::AlreadyParsed(expr) = lhs { + expr + } else { + let attrs = match lhs { + LhsExpr::AttributesParsed(attrs) => Some(attrs), + _ => None, + }; + if [token::DotDot, token::DotDotDot, token::DotDotEq].contains(&self.token.kind) { + return self.parse_prefix_range_expr(attrs); + } else { + self.parse_prefix_expr(attrs)? + } + }; + let last_type_ascription_set = self.last_type_ascription.is_some(); + + match (self.expr_is_complete(&lhs), AssocOp::from_token(&self.token)) { + (true, None) => { + self.last_type_ascription = None; + // Semi-statement forms are odd. See https://github.com/rust-lang/rust/issues/29071 + return Ok(lhs); + } + (false, _) => {} // 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::LAnd)) | // `{ 42 } &&x` (#61475) + (true, Some(AssocOp::Add)) // `{ 42 } + 42 + // If the next token is a keyword, then the tokens above *are* unambiguously incorrect: + // `if x { a } else { b } && if y { c } else { d }` + if !self.look_ahead(1, |t| t.is_reserved_ident()) => { + self.last_type_ascription = None; + // These cases are ambiguous and can't be identified in the parser alone + let sp = self.sess.source_map().start_point(self.token.span); + self.sess.ambiguous_block_expr_parse.borrow_mut().insert(sp, lhs.span); + return Ok(lhs); + } + (true, Some(ref op)) if !op.can_continue_expr_unambiguously() => { + self.last_type_ascription = None; + return Ok(lhs); + } + (true, Some(_)) => { + // 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` + let mut err = self.struct_span_err(self.token.span, &format!( + "expected expression, found `{}`", + pprust::token_to_string(&self.token), + )); + err.span_label(self.token.span, "expected expression"); + self.sess.expr_parentheses_needed( + &mut err, + lhs.span, + Some(pprust::expr_to_string(&lhs), + )); + err.emit(); + } + } + self.expected_tokens.push(TokenType::Operator); + while let Some(op) = AssocOp::from_token(&self.token) { + + // Adjust the span for interpolated LHS to point to the `$lhs` token and not to what + // it refers to. Interpolated identifiers are unwrapped early and never show up here + // as `PrevTokenKind::Interpolated` so if LHS is a single identifier we always process + // it as "interpolated", it doesn't change the answer for non-interpolated idents. + let lhs_span = match (self.prev_token_kind, &lhs.kind) { + (PrevTokenKind::Interpolated, _) => self.prev_span, + (PrevTokenKind::Ident, &ExprKind::Path(None, ref path)) + if path.segments.len() == 1 => self.prev_span, + _ => lhs.span, + }; + + let cur_op_span = self.token.span; + let restrictions = if op.is_assign_like() { + self.restrictions & Restrictions::NO_STRUCT_LITERAL + } else { + self.restrictions + }; + let prec = op.precedence(); + if prec < min_prec { + break; + } + // Check for deprecated `...` syntax + if self.token == token::DotDotDot && op == 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.is_comparison() { + if let Some(expr) = self.check_no_chained_comparison(&lhs, &op)? { + return Ok(expr); + } + } + // Special cases: + if op == AssocOp::As { + lhs = self.parse_assoc_op_cast(lhs, lhs_span, ExprKind::Cast)?; + continue + } else if op == AssocOp::Colon { + let maybe_path = self.could_ascription_be_path(&lhs.kind); + self.last_type_ascription = Some((self.prev_span, maybe_path)); + + lhs = self.parse_assoc_op_cast(lhs, lhs_span, ExprKind::Type)?; + self.sess.gated_spans.gate(sym::type_ascription, lhs.span); + 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. + // + // We have 2 alternatives here: `x..y`/`x..=y` and `x..`/`x..=` The other + // two variants are handled with `parse_prefix_range_expr` call above. + let rhs = if self.is_at_start_of_range_notation_rhs() { + Some(self.parse_assoc_expr_with(prec + 1, LhsExpr::NotYetParsed)?) + } else { + None + }; + let (lhs_span, rhs_span) = (lhs.span, if let Some(ref x) = rhs { + x.span + } else { + cur_op_span + }); + let limits = if op == AssocOp::DotDot { + RangeLimits::HalfOpen + } else { + RangeLimits::Closed + }; + + let r = self.mk_range(Some(lhs), rhs, limits)?; + lhs = self.mk_expr(lhs_span.to(rhs_span), r, ThinVec::new()); + 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_assoc_expr_with(prec + prec_adjustment, LhsExpr::NotYetParsed) + )?; + + // Make sure that the span of the parent node is larger than the span of lhs and rhs, + // including the attributes. + let lhs_span = lhs + .attrs + .iter() + .filter(|a| a.style == AttrStyle::Outer) + .next() + .map_or(lhs_span, |a| a.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, ThinVec::new()) + } + AssocOp::Assign => self.mk_expr(span, ExprKind::Assign(lhs, rhs), ThinVec::new()), + 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, ThinVec::new()) + } + AssocOp::As | AssocOp::Colon | AssocOp::DotDot | AssocOp::DotDotEq => { + self.bug("AssocOp should have been handled by special case") + } + }; + + if let Fixity::None = fixity { break } + } + if last_type_ascription_set { + self.last_type_ascription = None; + } + Ok(lhs) + } + + /// 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_requires_semi_to_be_stmt(e) + } + + 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(token::Brace) { + return !self.restrictions.contains(Restrictions::NO_STRUCT_LITERAL); + } + true + } else { + false + } + } + + /// Parses prefix-forms of range notation: `..expr`, `..`, `..=expr`. + fn parse_prefix_range_expr( + &mut self, + already_parsed_attrs: Option<ThinVec<Attribute>> + ) -> PResult<'a, P<Expr>> { + // Check for deprecated `...` syntax. + if self.token == token::DotDotDot { + self.err_dotdotdot_syntax(self.token.span); + } + + debug_assert!([token::DotDot, token::DotDotDot, token::DotDotEq].contains(&self.token.kind), + "parse_prefix_range_expr: token {:?} is not DotDot/DotDotEq", + self.token); + let tok = self.token.clone(); + let attrs = self.parse_or_use_outer_attributes(already_parsed_attrs)?; + let lo = self.token.span; + let mut hi = self.token.span; + self.bump(); + let opt_end = if self.is_at_start_of_range_notation_rhs() { + // RHS must be parsed with more associativity than the dots. + let next_prec = AssocOp::from_token(&tok).unwrap().precedence() + 1; + Some(self.parse_assoc_expr_with(next_prec, LhsExpr::NotYetParsed) + .map(|x| { + hi = x.span; + x + })?) + } else { + None + }; + let limits = if tok == token::DotDot { + RangeLimits::HalfOpen + } else { + RangeLimits::Closed + }; + + let r = self.mk_range(None, opt_end, limits)?; + Ok(self.mk_expr(lo.to(hi), r, attrs)) + } + + /// Parses a prefix-unary-operator expr. + fn parse_prefix_expr( + &mut self, + already_parsed_attrs: Option<ThinVec<Attribute>> + ) -> PResult<'a, P<Expr>> { + let attrs = self.parse_or_use_outer_attributes(already_parsed_attrs)?; + let lo = self.token.span; + // Note: when adding new unary operators, don't forget to adjust TokenKind::can_begin_expr() + let (hi, ex) = match self.token.kind { + token::Not => { + self.bump(); + let e = self.parse_prefix_expr(None); + let (span, e) = self.interpolated_or_expr_span(e)?; + (lo.to(span), self.mk_unary(UnOp::Not, e)) + } + // Suggest `!` for bitwise negation when encountering a `~` + token::Tilde => { + self.bump(); + let e = self.parse_prefix_expr(None); + let (span, e) = self.interpolated_or_expr_span(e)?; + let span_of_tilde = lo; + self.struct_span_err(span_of_tilde, "`~` cannot be used as a unary operator") + .span_suggestion_short( + span_of_tilde, + "use `!` to perform bitwise not", + "!".to_owned(), + Applicability::MachineApplicable + ) + .emit(); + (lo.to(span), self.mk_unary(UnOp::Not, e)) + } + token::BinOp(token::Minus) => { + self.bump(); + let e = self.parse_prefix_expr(None); + let (span, e) = self.interpolated_or_expr_span(e)?; + (lo.to(span), self.mk_unary(UnOp::Neg, e)) + } + token::BinOp(token::Star) => { + self.bump(); + let e = self.parse_prefix_expr(None); + let (span, e) = self.interpolated_or_expr_span(e)?; + (lo.to(span), self.mk_unary(UnOp::Deref, e)) + } + token::BinOp(token::And) | token::AndAnd => { + self.expect_and()?; + let m = self.parse_mutability(); + let e = self.parse_prefix_expr(None); + let (span, e) = self.interpolated_or_expr_span(e)?; + (lo.to(span), ExprKind::AddrOf(m, e)) + } + token::Ident(..) if self.token.is_keyword(kw::Box) => { + self.bump(); + let e = self.parse_prefix_expr(None); + let (span, e) = self.interpolated_or_expr_span(e)?; + let span = lo.to(span); + self.sess.gated_spans.gate(sym::box_syntax, span); + (span, ExprKind::Box(e)) + } + token::Ident(..) if self.token.is_ident_named(sym::not) => { + // `not` is just an ordinary identifier in Rust-the-language, + // but as `rustc`-the-compiler, we can issue clever diagnostics + // for confused users who really want to say `!` + let token_cannot_continue_expr = |t: &Token| match t.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(), + }; + let cannot_continue_expr = self.look_ahead(1, token_cannot_continue_expr); + if cannot_continue_expr { + self.bump(); + // Emit the error ... + self.struct_span_err( + self.token.span, + &format!("unexpected {} after identifier",self.this_token_descr()) + ) + .span_suggestion_short( + // Span the `not` plus trailing whitespace to avoid + // trailing whitespace after the `!` in our suggestion + self.sess.source_map() + .span_until_non_whitespace(lo.to(self.token.span)), + "use `!` to perform logical negation", + "!".to_owned(), + Applicability::MachineApplicable + ) + .emit(); + // —and recover! (just as if we were in the block + // for the `token::Not` arm) + let e = self.parse_prefix_expr(None); + let (span, e) = self.interpolated_or_expr_span(e)?; + (lo.to(span), self.mk_unary(UnOp::Not, e)) + } else { + return self.parse_dot_or_call_expr(Some(attrs)); + } + } + _ => { return self.parse_dot_or_call_expr(Some(attrs)); } + }; + return Ok(self.mk_expr(lo.to(hi), ex, attrs)); + } + + /// Returns the span of expr, if it was not interpolated or the span of the interpolated token. + fn interpolated_or_expr_span( + &self, + expr: PResult<'a, P<Expr>>, + ) -> PResult<'a, (Span, P<Expr>)> { + expr.map(|e| { + if self.prev_token_kind == PrevTokenKind::Interpolated { + (self.prev_span, e) + } else { + (e.span, e) + } + }) + } + + 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, rhs: P<Ty>| { + this.mk_expr(lhs_span.to(rhs.span), expr_kind(lhs, rhs), ThinVec::new()) + }; + + // 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(); + match self.parse_ty_no_plus() { + Ok(rhs) => { + Ok(mk_expr(self, rhs)) + } + Err(mut 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 = self.clone(); + mem::replace(self, parser_snapshot_before_type); + + match self.parse_path(PathStyle::Expr) { + Ok(path) => { + let (op_noun, op_verb) = match self.token.kind { + token::Lt => ("comparison", "comparing"), + token::BinOp(token::Shl) => ("shift", "shifting"), + _ => { + // 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. + mem::replace(self, parser_snapshot_after_type); + return Err(type_err); + } + }; + + // Successfully parsed the type path leaving a `<` yet to parse. + type_err.cancel(); + + // Report non-fatal diagnostics, keep `x as usize` as an expression + // in AST and continue parsing. + let msg = format!( + "`<` is interpreted as a start of generic arguments for `{}`, not a {}", + pprust::path_to_string(&path), + op_noun, + ); + let span_after_type = parser_snapshot_after_type.token.span; + let expr = mk_expr(self, P(Ty { + span: path.span, + kind: TyKind::Path(None, path), + id: DUMMY_NODE_ID, + })); + + let expr_str = self.span_to_snippet(expr.span) + .unwrap_or_else(|_| pprust::expr_to_string(&expr)); + + self.struct_span_err(self.token.span, &msg) + .span_label( + self.look_ahead(1, |t| t.span).to(span_after_type), + "interpreted as generic arguments" + ) + .span_label(self.token.span, format!("not interpreted as {}", op_noun)) + .span_suggestion( + expr.span, + &format!("try {} the cast value", op_verb), + format!("({})", expr_str), + Applicability::MachineApplicable, + ) + .emit(); + + Ok(expr) + } + Err(mut path_err) => { + // Couldn't parse as a path, return original error and parser state. + path_err.cancel(); + mem::replace(self, parser_snapshot_after_type); + Err(type_err) + } + } + } + } + } + + /// Parses `a.b` or `a(13)` or `a[4]` or just `a`. + fn parse_dot_or_call_expr( + &mut self, + already_parsed_attrs: Option<ThinVec<Attribute>>, + ) -> PResult<'a, P<Expr>> { + let attrs = self.parse_or_use_outer_attributes(already_parsed_attrs)?; + + let b = self.parse_bottom_expr(); + let (span, b) = self.interpolated_or_expr_span(b)?; + self.parse_dot_or_call_expr_with(b, span, attrs) + } + + pub(super) fn parse_dot_or_call_expr_with( + &mut self, + e0: P<Expr>, + lo: Span, + mut attrs: ThinVec<Attribute>, + ) -> 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 + self.parse_dot_or_call_expr_with_(e0, lo).map(|expr| + expr.map(|mut expr| { + attrs.extend::<Vec<_>>(expr.attrs.into()); + expr.attrs = attrs; + match expr.kind { + ExprKind::If(..) if !expr.attrs.is_empty() => { + // Just point to the first attribute in there... + let span = expr.attrs[0].span; + self.span_err(span, "attributes are not yet allowed on `if` expressions"); + } + _ => {} + } + expr + }) + ) + } + + fn parse_dot_or_call_expr_with_(&mut self, e0: P<Expr>, lo: Span) -> PResult<'a, P<Expr>> { + let mut e = e0; + let mut hi; + loop { + // expr? + while self.eat(&token::Question) { + let hi = self.prev_span; + e = self.mk_expr(lo.to(hi), ExprKind::Try(e), ThinVec::new()); + } + + // expr.f + if self.eat(&token::Dot) { + match self.token.kind { + token::Ident(..) => { + e = self.parse_dot_suffix(e, lo)?; + } + token::Literal(token::Lit { kind: token::Integer, symbol, suffix }) => { + let span = self.token.span; + self.bump(); + let field = ExprKind::Field(e, Ident::new(symbol, span)); + e = self.mk_expr(lo.to(span), field, ThinVec::new()); + + self.expect_no_suffix(span, "a tuple index", suffix); + } + token::Literal(token::Lit { kind: token::Float, symbol, .. }) => { + self.bump(); + let fstr = symbol.as_str(); + let msg = format!("unexpected token: `{}`", symbol); + let mut err = self.diagnostic().struct_span_err(self.prev_span, &msg); + err.span_label(self.prev_span, "unexpected token"); + if fstr.chars().all(|x| "0123456789.".contains(x)) { + let float = match fstr.parse::<f64>().ok() { + Some(f) => f, + None => continue, + }; + let sugg = pprust::to_string(|s| { + s.popen(); + s.print_expr(&e); + s.s.word( "."); + s.print_usize(float.trunc() as usize); + s.pclose(); + s.s.word("."); + s.s.word(fstr.splitn(2, ".").last().unwrap().to_string()) + }); + err.span_suggestion( + lo.to(self.prev_span), + "try parenthesizing the first index", + sugg, + Applicability::MachineApplicable + ); + } + return Err(err); + + } + _ => { + // FIXME Could factor this out into non_fatal_unexpected or something. + let actual = self.this_token_to_string(); + self.span_err(self.token.span, &format!("unexpected token: `{}`", actual)); + } + } + continue; + } + if self.expr_is_complete(&e) { break; } + match self.token.kind { + // expr(...) + token::OpenDelim(token::Paren) => { + let seq = self.parse_paren_expr_seq().map(|es| { + let nd = self.mk_call(e, es); + let hi = self.prev_span; + self.mk_expr(lo.to(hi), nd, ThinVec::new()) + }); + e = self.recover_seq_parse_error(token::Paren, lo, seq); + } + + // expr[...] + // Could be either an index expression or a slicing expression. + token::OpenDelim(token::Bracket) => { + self.bump(); + let ix = self.parse_expr()?; + hi = self.token.span; + self.expect(&token::CloseDelim(token::Bracket))?; + let index = self.mk_index(e, ix); + e = self.mk_expr(lo.to(hi), index, ThinVec::new()) + } + _ => return Ok(e) + } + } + return Ok(e); + } + + /// 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.span.rust_2018() && self.eat_keyword(kw::Await) { + return self.mk_await_expr(self_arg, lo); + } + + let segment = self.parse_path_segment(PathStyle::Expr)?; + self.check_trailing_angle_brackets(&segment, token::OpenDelim(token::Paren)); + + Ok(match self.token.kind { + token::OpenDelim(token::Paren) => { + // Method call `expr.f()` + let mut args = self.parse_paren_expr_seq()?; + args.insert(0, self_arg); + + let span = lo.to(self.prev_span); + self.mk_expr(span, ExprKind::MethodCall(segment, args), ThinVec::new()) + } + _ => { + // Field access `expr.f` + if let Some(args) = segment.args { + self.span_err(args.span(), + "field expressions may not have generic arguments"); + } + + let span = lo.to(self.prev_span); + self.mk_expr(span, ExprKind::Field(self_arg, segment.ident), ThinVec::new()) + } + }) + } + + /// 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_bottom_expr(&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. + // + // Therefore, prevent sub-parser from parsing + // attributes by giving them a empty "already-parsed" list. + let mut attrs = ThinVec::new(); + + let lo = self.token.span; + let mut hi = self.token.span; + + let ex: ExprKind; + + macro_rules! parse_lit { + () => { + match self.parse_lit() { + Ok(literal) => { + hi = self.prev_span; + ex = ExprKind::Lit(literal); + } + Err(mut err) => { + err.cancel(); + return Err(self.expected_expression_found()); + } + } + } + } + + // Note: when adding new syntax here, don't forget to adjust `TokenKind::can_begin_expr()`. + match self.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. + token::Literal(_) => { + parse_lit!() + } + token::OpenDelim(token::Paren) => { + self.bump(); + + attrs.extend(self.parse_inner_attributes()?); + + // `(e)` is parenthesized `e`. + // `(e,)` is a tuple with only one field, `e`. + let mut es = vec![]; + let mut trailing_comma = false; + let mut recovered = false; + while self.token != token::CloseDelim(token::Paren) { + es.push(match self.parse_expr() { + Ok(es) => es, + Err(mut err) => { + // Recover from parse error in tuple list. + match self.token.kind { + token::Ident(name, false) + if name == kw::Underscore && self.look_ahead(1, |t| { + t == &token::Comma + }) => { + // Special-case handling of `Foo<(_, _, _)>` + err.emit(); + let sp = self.token.span; + self.bump(); + self.mk_expr(sp, ExprKind::Err, ThinVec::new()) + } + _ => return Ok( + self.recover_seq_parse_error(token::Paren, lo, Err(err)), + ), + } + } + }); + recovered = self.expect_one_of( + &[], + &[token::Comma, token::CloseDelim(token::Paren)], + )?; + if self.eat(&token::Comma) { + trailing_comma = true; + } else { + trailing_comma = false; + break; + } + } + if !recovered { + self.bump(); + } + + hi = self.prev_span; + ex = if es.len() == 1 && !trailing_comma { + ExprKind::Paren(es.into_iter().nth(0).unwrap()) + } else { + ExprKind::Tup(es) + }; + } + token::OpenDelim(token::Brace) => { + return self.parse_block_expr(None, lo, BlockCheckMode::Default, attrs); + } + token::BinOp(token::Or) | token::OrOr => { + return self.parse_closure_expr(attrs); + } + token::OpenDelim(token::Bracket) => { + self.bump(); + + attrs.extend(self.parse_inner_attributes()?); + + if self.eat(&token::CloseDelim(token::Bracket)) { + // Empty vector + ex = ExprKind::Array(Vec::new()); + } else { + // Non-empty vector + let first_expr = self.parse_expr()?; + if self.eat(&token::Semi) { + // Repeating array syntax: `[ 0; 512 ]` + let count = AnonConst { + id: DUMMY_NODE_ID, + value: self.parse_expr()?, + }; + self.expect(&token::CloseDelim(token::Bracket))?; + ex = ExprKind::Repeat(first_expr, count); + } else if self.eat(&token::Comma) { + // Vector with two or more elements + let remaining_exprs = self.parse_seq_to_end( + &token::CloseDelim(token::Bracket), + SeqSep::trailing_allowed(token::Comma), + |p| Ok(p.parse_expr()?) + )?; + let mut exprs = vec![first_expr]; + exprs.extend(remaining_exprs); + ex = ExprKind::Array(exprs); + } else { + // Vector with one element + self.expect(&token::CloseDelim(token::Bracket))?; + ex = ExprKind::Array(vec![first_expr]); + } + } + hi = self.prev_span; + } + _ => { + if self.eat_lt() { + let (qself, path) = self.parse_qpath(PathStyle::Expr)?; + hi = path.span; + return Ok(self.mk_expr(lo.to(hi), ExprKind::Path(Some(qself), path), attrs)); + } + if self.token.is_path_start() { + let path = self.parse_path(PathStyle::Expr)?; + + // `!`, as an operator, is prefix, so we know this isn't that. + if self.eat(&token::Not) { + // MACRO INVOCATION expression + let (delim, tts) = self.expect_delimited_token_tree()?; + hi = self.prev_span; + ex = ExprKind::Mac(Mac { + path, + tts, + delim, + span: lo.to(hi), + prior_type_ascription: self.last_type_ascription, + }); + } else if self.check(&token::OpenDelim(token::Brace)) { + if let Some(expr) = self.maybe_parse_struct_expr(lo, &path, &attrs) { + return expr; + } else { + hi = path.span; + ex = ExprKind::Path(None, path); + } + } else { + hi = path.span; + ex = ExprKind::Path(None, path); + } + + let expr = self.mk_expr(lo.to(hi), ex, attrs); + return self.maybe_recover_from_bad_qpath(expr, true); + } + if self.check_keyword(kw::Move) || self.check_keyword(kw::Static) { + return self.parse_closure_expr(attrs); + } + if self.eat_keyword(kw::If) { + return self.parse_if_expr(attrs); + } + if self.eat_keyword(kw::For) { + let lo = self.prev_span; + return self.parse_for_expr(None, lo, attrs); + } + if self.eat_keyword(kw::While) { + let lo = self.prev_span; + return self.parse_while_expr(None, lo, attrs); + } + if let Some(label) = self.eat_label() { + let lo = label.ident.span; + self.expect(&token::Colon)?; + if self.eat_keyword(kw::While) { + return self.parse_while_expr(Some(label), lo, attrs) + } + if self.eat_keyword(kw::For) { + return self.parse_for_expr(Some(label), lo, attrs) + } + if self.eat_keyword(kw::Loop) { + return self.parse_loop_expr(Some(label), lo, attrs) + } + if self.token == token::OpenDelim(token::Brace) { + return self.parse_block_expr(Some(label), + lo, + BlockCheckMode::Default, + attrs); + } + let msg = "expected `while`, `for`, `loop` or `{` after a label"; + let mut err = self.fatal(msg); + err.span_label(self.token.span, msg); + return Err(err); + } + if self.eat_keyword(kw::Loop) { + let lo = self.prev_span; + return self.parse_loop_expr(None, lo, attrs); + } + if self.eat_keyword(kw::Continue) { + let label = self.eat_label(); + let ex = ExprKind::Continue(label); + let hi = self.prev_span; + return Ok(self.mk_expr(lo.to(hi), ex, attrs)); + } + if self.eat_keyword(kw::Match) { + let match_sp = self.prev_span; + return self.parse_match_expr(attrs).map_err(|mut err| { + err.span_label(match_sp, "while parsing this match expression"); + err + }); + } + if self.eat_keyword(kw::Unsafe) { + return self.parse_block_expr( + None, + lo, + BlockCheckMode::Unsafe(ast::UserProvided), + attrs); + } + if self.is_do_catch_block() { + let mut db = self.fatal("found removed `do catch` syntax"); + db.help("following RFC #2388, the new non-placeholder syntax is `try`"); + return Err(db); + } + if self.is_try_block() { + let lo = self.token.span; + assert!(self.eat_keyword(kw::Try)); + return self.parse_try_block(lo, attrs); + } + + // `Span::rust_2018()` is somewhat expensive; don't get it repeatedly. + let is_span_rust_2018 = self.token.span.rust_2018(); + if is_span_rust_2018 && self.check_keyword(kw::Async) { + return if self.is_async_block() { // Check for `async {` and `async move {`. + self.parse_async_block(attrs) + } else { + self.parse_closure_expr(attrs) + }; + } + if self.eat_keyword(kw::Return) { + if self.token.can_begin_expr() { + let e = self.parse_expr()?; + hi = e.span; + ex = ExprKind::Ret(Some(e)); + } else { + ex = ExprKind::Ret(None); + } + } else if self.eat_keyword(kw::Break) { + let label = self.eat_label(); + let e = if self.token.can_begin_expr() + && !(self.token == token::OpenDelim(token::Brace) + && self.restrictions.contains( + Restrictions::NO_STRUCT_LITERAL)) { + Some(self.parse_expr()?) + } else { + None + }; + ex = ExprKind::Break(label, e); + hi = self.prev_span; + } else if self.eat_keyword(kw::Yield) { + if self.token.can_begin_expr() { + let e = self.parse_expr()?; + hi = e.span; + ex = ExprKind::Yield(Some(e)); + } else { + ex = ExprKind::Yield(None); + } + + let span = lo.to(hi); + self.sess.gated_spans.gate(sym::generators, span); + } else if self.eat_keyword(kw::Let) { + return self.parse_let_expr(attrs); + } else if is_span_rust_2018 && self.eat_keyword(kw::Await) { + let (await_hi, e_kind) = self.parse_incorrect_await_syntax(lo, self.prev_span)?; + hi = await_hi; + ex = e_kind; + } else { + if !self.unclosed_delims.is_empty() && self.check(&token::Semi) { + // Don't complain about bare semicolons after unclosed braces + // recovery in order to keep the error count down. Fixing the + // delimiters will possibly also fix the bare semicolon found in + // expression context. For example, silence the following error: + // + // error: expected expression, found `;` + // --> file.rs:2:13 + // | + // 2 | foo(bar(; + // | ^ expected expression + self.bump(); + return Ok(self.mk_expr(self.token.span, ExprKind::Err, ThinVec::new())); + } + parse_lit!() + } + } + } + + let expr = self.mk_expr(lo.to(hi), ex, attrs); + self.maybe_recover_from_bad_qpath(expr, true) + } + + /// Matches `lit = true | false | token_lit`. + pub(super) fn parse_lit(&mut self) -> PResult<'a, Lit> { + let mut recovered = None; + if self.token == token::Dot { + // Attempt to recover `.4` as `0.4`. + recovered = self.look_ahead(1, |next_token| { + if let token::Literal(token::Lit { kind: token::Integer, symbol, suffix }) + = next_token.kind { + if 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.struct_span_err(token.span, "float literals must have an integer part") + .span_suggestion( + token.span, + "must have an integer part", + pprust::token_to_string(token), + Applicability::MachineApplicable, + ) + .emit(); + } + } + + let token = recovered.as_ref().unwrap_or(&self.token); + match Lit::from_token(token) { + Ok(lit) => { + self.bump(); + Ok(lit) + } + Err(LitError::NotLiteral) => { + let msg = format!("unexpected token: {}", self.this_token_descr()); + Err(self.span_fatal(token.span, &msg)) + } + Err(err) => { + let span = token.span; + let lit = match token.kind { + token::Literal(lit) => lit, + _ => unreachable!(), + }; + self.bump(); + self.error_literal_from_token(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, symbol, lit.suffix); + Lit::from_lit_token(lit, span).map_err(|_| unreachable!()) + } + } + } + + fn error_literal_from_token(&self, err: LitError, lit: token::Lit, span: Span) { + // Checks if `s` looks like i32 or u1234 etc. + fn looks_like_width_suffix(first_chars: &[char], s: &str) -> bool { + s.len() > 1 + && s.starts_with(first_chars) + && s[1..].chars().all(|c| c.is_ascii_digit()) + } + + let token::Lit { kind, suffix, .. } = lit; + match err { + // `NotLiteral` is not an error by itself, so we don't report + // it and give the parser opportunity to try something else. + LitError::NotLiteral => {} + // `LexerError` *is* an error, but it was already reported + // by lexer, so here we don't report it the second time. + LitError::LexerError => {} + LitError::InvalidSuffix => { + self.expect_no_suffix( + span, + &format!("{} {} literal", kind.article(), kind.descr()), + suffix, + ); + } + LitError::InvalidIntSuffix => { + let suf = suffix.expect("suffix error with no suffix").as_str(); + if looks_like_width_suffix(&['i', 'u'], &suf) { + // If it looks like a width, try to be helpful. + let msg = format!("invalid width `{}` for integer literal", &suf[1..]); + self.struct_span_err(span, &msg) + .help("valid widths are 8, 16, 32, 64 and 128") + .emit(); + } else { + let msg = format!("invalid suffix `{}` for integer literal", suf); + self.struct_span_err(span, &msg) + .span_label(span, format!("invalid suffix `{}`", suf)) + .help("the suffix must be one of the integral types (`u32`, `isize`, etc)") + .emit(); + } + } + LitError::InvalidFloatSuffix => { + let suf = suffix.expect("suffix error with no suffix").as_str(); + if looks_like_width_suffix(&['f'], &suf) { + // If it looks like a width, try to be helpful. + let msg = format!("invalid width `{}` for float literal", &suf[1..]); + self.struct_span_err(span, &msg) + .help("valid widths are 32 and 64") + .emit(); + } else { + let msg = format!("invalid suffix `{}` for float literal", suf); + self.struct_span_err(span, &msg) + .span_label(span, format!("invalid suffix `{}`", suf)) + .help("valid suffixes are `f32` and `f64`") + .emit(); + } + } + LitError::NonDecimalFloat(base) => { + let descr = match base { + 16 => "hexadecimal", + 8 => "octal", + 2 => "binary", + _ => unreachable!(), + }; + self.struct_span_err(span, &format!("{} float literal is not supported", descr)) + .span_label(span, "not supported") + .emit(); + } + LitError::IntTooLarge => { + self.struct_span_err(span, "integer literal is too large") + .emit(); + } + } + } + + pub(super) fn expect_no_suffix(&self, sp: Span, kind: &str, suffix: Option<Symbol>) { + if let Some(suf) = suffix { + let mut err = if kind == "a tuple index" + && [sym::i32, sym::u32, sym::isize, sym::usize].contains(&suf) + { + // #59553: warn instead of reject out of hand to allow the fix to percolate + // through the ecosystem when people fix their macros + let mut err = self.sess.span_diagnostic.struct_span_warn( + sp, + &format!("suffixes on {} are invalid", kind), + ); + err.note(&format!( + "`{}` is *temporarily* accepted on tuple index fields as it was \ + incorrectly accepted on stable for a few releases", + suf, + )); + err.help( + "on proc macros, you'll want to use `syn::Index::from` or \ + `proc_macro::Literal::*_unsuffixed` for code that will desugar \ + to tuple field access", + ); + err.note( + "for more context, see https://github.com/rust-lang/rust/issues/60210", + ); + err + } else { + self.struct_span_err(sp, &format!("suffixes on {} are invalid", kind)) + }; + err.span_label(sp, format!("invalid suffix `{}`", suf)); + err.emit(); + } + } + + /// Matches `'-' lit | lit` (cf. `ast_validation::AstValidator::check_expr_within_pat`). + pub fn parse_literal_maybe_minus(&mut self) -> PResult<'a, P<Expr>> { + maybe_whole_expr!(self); + + let minus_lo = self.token.span; + let minus_present = self.eat(&token::BinOp(token::Minus)); + let lo = self.token.span; + let literal = self.parse_lit()?; + let hi = self.prev_span; + let expr = self.mk_expr(lo.to(hi), ExprKind::Lit(literal), ThinVec::new()); + + if minus_present { + let minus_hi = self.prev_span; + let unary = self.mk_unary(UnOp::Neg, expr); + Ok(self.mk_expr(minus_lo.to(minus_hi), unary, ThinVec::new())) + } else { + Ok(expr) + } + } + + /// Parses a block or unsafe block. + pub(super) fn parse_block_expr( + &mut self, + opt_label: Option<Label>, + lo: Span, + blk_mode: BlockCheckMode, + outer_attrs: ThinVec<Attribute>, + ) -> PResult<'a, P<Expr>> { + if let Some(label) = opt_label { + self.sess.gated_spans.gate(sym::label_break_value, label.ident.span); + } + + self.expect(&token::OpenDelim(token::Brace))?; + + let mut attrs = outer_attrs; + attrs.extend(self.parse_inner_attributes()?); + + let blk = self.parse_block_tail(lo, blk_mode)?; + Ok(self.mk_expr(blk.span, ExprKind::Block(blk, opt_label), attrs)) + } + + /// Parses a closure expression (e.g., `move |args| expr`). + fn parse_closure_expr(&mut self, attrs: ThinVec<Attribute>) -> PResult<'a, P<Expr>> { + let lo = self.token.span; + + let movability = if self.eat_keyword(kw::Static) { + Movability::Static + } else { + Movability::Movable + }; + + let asyncness = if self.token.span.rust_2018() { + self.parse_asyncness() + } else { + IsAsync::NotAsync + }; + if asyncness.is_async() { + // Feature-gate `async ||` closures. + self.sess.gated_spans.gate(sym::async_closure, self.prev_span); + } + + let capture_clause = self.parse_capture_clause(); + let decl = self.parse_fn_block_decl()?; + let decl_hi = self.prev_span; + let body = match decl.output { + FunctionRetTy::Default(_) => { + let restrictions = self.restrictions - Restrictions::STMT_EXPR; + self.parse_expr_res(restrictions, None)? + }, + _ => { + // If an explicit return type is given, require a block to appear (RFC 968). + let body_lo = self.token.span; + self.parse_block_expr(None, body_lo, BlockCheckMode::Default, ThinVec::new())? + } + }; + + Ok(self.mk_expr( + lo.to(body.span), + ExprKind::Closure(capture_clause, asyncness, movability, decl, body, lo.to(decl_hi)), + attrs)) + } + + /// Parses an optional `move` prefix to a closure lke construct. + fn parse_capture_clause(&mut self) -> CaptureBy { + if self.eat_keyword(kw::Move) { + CaptureBy::Value + } else { + CaptureBy::Ref + } + } + + /// Parses the `|arg, arg|` header of a closure. + fn parse_fn_block_decl(&mut self) -> PResult<'a, P<FnDecl>> { + let inputs_captures = { + if self.eat(&token::OrOr) { + Vec::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 output = self.parse_ret_ty(true)?; + + Ok(P(FnDecl { + inputs: inputs_captures, + output, + })) + } + + /// 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()?; + let pat = self.parse_pat(PARAM_EXPECTED)?; + let t = if self.eat(&token::Colon) { + self.parse_ty()? + } else { + P(Ty { + id: DUMMY_NODE_ID, + kind: TyKind::Infer, + span: self.prev_span, + }) + }; + let span = lo.to(self.token.span); + Ok(Param { + attrs: attrs.into(), + ty: t, + pat, + span, + id: DUMMY_NODE_ID, + is_placeholder: false, + }) + } + + /// Parses an `if` expression (`if` token already eaten). + fn parse_if_expr(&mut self, attrs: ThinVec<Attribute>) -> PResult<'a, P<Expr>> { + let lo = self.prev_span; + let cond = self.parse_cond_expr()?; + + // Verify that the parsed `if` condition makes sense as a condition. If it is a block, then + // verify that the last statement is either an implicit return (no `;`) or an explicit + // return. This won't catch blocks with an explicit `return`, but that would be caught by + // the dead code lint. + if self.eat_keyword(kw::Else) || !cond.returns() { + let sp = self.sess.source_map().next_point(lo); + let mut err = self.diagnostic() + .struct_span_err(sp, "missing condition for `if` expression"); + err.span_label(sp, "expected if condition here"); + return Err(err) + } + let not_block = self.token != token::OpenDelim(token::Brace); + let thn = self.parse_block().map_err(|mut err| { + if not_block { + err.span_label(lo, "this `if` statement has a condition, but no block"); + } + err + })?; + let mut els: Option<P<Expr>> = None; + let mut hi = thn.span; + if self.eat_keyword(kw::Else) { + let elexpr = self.parse_else_expr()?; + hi = elexpr.span; + els = Some(elexpr); + } + Ok(self.mk_expr(lo.to(hi), ExprKind::If(cond, thn, els), attrs)) + } + + /// Parses the condition of a `if` or `while` expression. + fn parse_cond_expr(&mut self) -> PResult<'a, P<Expr>> { + let cond = self.parse_expr_res(Restrictions::NO_STRUCT_LITERAL, None)?; + + if let ExprKind::Let(..) = cond.kind { + // Remove the last feature gating of a `let` expression since it's stable. + self.sess.gated_spans.ungate_last(sym::let_chains, cond.span); + } + + Ok(cond) + } + + /// Parses a `let $pat = $expr` pseudo-expression. + /// The `let` token has already been eaten. + fn parse_let_expr(&mut self, attrs: ThinVec<Attribute>) -> PResult<'a, P<Expr>> { + let lo = self.prev_span; + let pat = self.parse_top_pat(GateOr::No)?; + self.expect(&token::Eq)?; + let expr = self.with_res( + Restrictions::NO_STRUCT_LITERAL, + |this| this.parse_assoc_expr_with(1 + prec_let_scrutinee_needs_par(), None.into()) + )?; + let span = lo.to(expr.span); + self.sess.gated_spans.gate(sym::let_chains, span); + Ok(self.mk_expr(span, ExprKind::Let(pat, expr), attrs)) + } + + /// Parses an `else { ... }` expression (`else` token already eaten). + fn parse_else_expr(&mut self) -> PResult<'a, P<Expr>> { + if self.eat_keyword(kw::If) { + return self.parse_if_expr(ThinVec::new()); + } else { + let blk = self.parse_block()?; + return Ok(self.mk_expr(blk.span, ExprKind::Block(blk, None), ThinVec::new())); + } + } + + /// Parses a `for ... in` expression (`for` token already eaten). + fn parse_for_expr( + &mut self, + opt_label: Option<Label>, + span_lo: Span, + mut attrs: ThinVec<Attribute> + ) -> PResult<'a, P<Expr>> { + // Parse: `for <src_pat> in <src_expr> <src_loop_block>` + + // 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 begin_paren = match self.token.kind { + token::OpenDelim(token::Paren) => Some(self.token.span), + _ => None, + }; + + let pat = self.parse_top_pat(GateOr::Yes)?; + if !self.eat_keyword(kw::In) { + let in_span = self.prev_span.between(self.token.span); + self.struct_span_err(in_span, "missing `in` in `for` loop") + .span_suggestion_short( + in_span, + "try adding `in` here", " in ".into(), + // has been misleading, at least in the past (closed Issue #48492) + Applicability::MaybeIncorrect + ) + .emit(); + } + let in_span = self.prev_span; + self.check_for_for_in_in_typo(in_span); + let expr = self.parse_expr_res(Restrictions::NO_STRUCT_LITERAL, None)?; + + let pat = self.recover_parens_around_for_head(pat, &expr, begin_paren); + + let (iattrs, loop_block) = self.parse_inner_attrs_and_block()?; + attrs.extend(iattrs); + + let hi = self.prev_span; + Ok(self.mk_expr(span_lo.to(hi), ExprKind::ForLoop(pat, expr, loop_block, opt_label), attrs)) + } + + /// Parses a `while` or `while let` expression (`while` token already eaten). + fn parse_while_expr( + &mut self, + opt_label: Option<Label>, + span_lo: Span, + mut attrs: ThinVec<Attribute> + ) -> PResult<'a, P<Expr>> { + let cond = self.parse_cond_expr()?; + let (iattrs, body) = self.parse_inner_attrs_and_block()?; + attrs.extend(iattrs); + let span = span_lo.to(body.span); + Ok(self.mk_expr(span, ExprKind::While(cond, body, opt_label), attrs)) + } + + /// Parses `loop { ... }` (`loop` token already eaten). + fn parse_loop_expr( + &mut self, + opt_label: Option<Label>, + span_lo: Span, + mut attrs: ThinVec<Attribute> + ) -> PResult<'a, P<Expr>> { + let (iattrs, body) = self.parse_inner_attrs_and_block()?; + attrs.extend(iattrs); + let span = span_lo.to(body.span); + Ok(self.mk_expr(span, ExprKind::Loop(body, opt_label), attrs)) + } + + fn eat_label(&mut self) -> Option<Label> { + if let Some(ident) = self.token.lifetime() { + let span = self.token.span; + self.bump(); + Some(Label { ident: Ident::new(ident.name, span) }) + } else { + None + } + } + + /// Parses a `match ... { ... }` expression (`match` token already eaten). + fn parse_match_expr(&mut self, mut attrs: ThinVec<Attribute>) -> PResult<'a, P<Expr>> { + let match_span = self.prev_span; + let lo = self.prev_span; + let discriminant = self.parse_expr_res(Restrictions::NO_STRUCT_LITERAL, None)?; + if let Err(mut e) = self.expect(&token::OpenDelim(token::Brace)) { + if self.token == token::Semi { + e.span_suggestion_short( + match_span, + "try removing this `match`", + String::new(), + Applicability::MaybeIncorrect // speculative + ); + } + return Err(e) + } + attrs.extend(self.parse_inner_attributes()?); + + let mut arms: Vec<Arm> = Vec::new(); + while self.token != token::CloseDelim(token::Brace) { + match self.parse_arm() { + Ok(arm) => arms.push(arm), + Err(mut e) => { + // Recover by skipping to the end of the block. + e.emit(); + self.recover_stmt(); + let span = lo.to(self.token.span); + if self.token == token::CloseDelim(token::Brace) { + self.bump(); + } + return Ok(self.mk_expr(span, ExprKind::Match(discriminant, arms), attrs)); + } + } + } + let hi = self.token.span; + self.bump(); + return Ok(self.mk_expr(lo.to(hi), ExprKind::Match(discriminant, arms), attrs)); + } + + pub(super) fn parse_arm(&mut self) -> PResult<'a, Arm> { + let attrs = self.parse_outer_attributes()?; + let lo = self.token.span; + let pat = self.parse_top_pat(GateOr::No)?; + let guard = if self.eat_keyword(kw::If) { + Some(self.parse_expr()?) + } else { + None + }; + let arrow_span = self.token.span; + self.expect(&token::FatArrow)?; + let arm_start_span = self.token.span; + + let expr = self.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_requires_semi_to_be_stmt(&expr) + && self.token != token::CloseDelim(token::Brace); + + let hi = self.token.span; + + if require_comma { + let cm = self.sess.source_map(); + self.expect_one_of(&[token::Comma], &[token::CloseDelim(token::Brace)]) + .map_err(|mut err| { + match (cm.span_to_lines(expr.span), cm.span_to_lines(arm_start_span)) { + (Ok(ref expr_lines), Ok(ref arm_start_lines)) + if arm_start_lines.lines[0].end_col == expr_lines.lines[0].end_col + && expr_lines.lines.len() == 2 + && self.token == token::FatArrow => { + // 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( + cm.next_point(arm_start_span), + "missing a comma here to end this `match` arm", + ",".to_owned(), + Applicability::MachineApplicable + ); + } + _ => { + err.span_label(arrow_span, + "while parsing the `match` arm starting here"); + } + } + err + })?; + } else { + self.eat(&token::Comma); + } + + Ok(ast::Arm { + attrs, + pat, + guard, + body: expr, + span: lo.to(hi), + id: DUMMY_NODE_ID, + is_placeholder: false, + }) + } + + /// Parses a `try {...}` expression (`try` token already eaten). + fn parse_try_block( + &mut self, + span_lo: Span, + mut attrs: ThinVec<Attribute> + ) -> PResult<'a, P<Expr>> { + let (iattrs, body) = self.parse_inner_attrs_and_block()?; + attrs.extend(iattrs); + if self.eat_keyword(kw::Catch) { + let mut error = self.struct_span_err(self.prev_span, + "keyword `catch` cannot follow a `try` block"); + error.help("try using `match` on the result of the `try` block instead"); + error.emit(); + Err(error) + } else { + let span = span_lo.to(body.span); + self.sess.gated_spans.gate(sym::try_blocks, span); + Ok(self.mk_expr(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(token::Brace)) && + !self.restrictions.contains(Restrictions::NO_STRUCT_LITERAL) + } + + fn is_try_block(&self) -> bool { + self.token.is_keyword(kw::Try) && + self.look_ahead(1, |t| *t == token::OpenDelim(token::Brace)) && + self.token.span.rust_2018() && + // Prevent `while try {} {}`, `if try {} {} else {}`, etc. + !self.restrictions.contains(Restrictions::NO_STRUCT_LITERAL) + } + + /// Parses an `async move? {...}` expression. + fn parse_async_block(&mut self, mut attrs: ThinVec<Attribute>) -> PResult<'a, P<Expr>> { + let span_lo = self.token.span; + self.expect_keyword(kw::Async)?; + let capture_clause = self.parse_capture_clause(); + let (iattrs, body) = self.parse_inner_attrs_and_block()?; + attrs.extend(iattrs); + Ok(self.mk_expr( + span_lo.to(body.span), + ExprKind::Async(capture_clause, DUMMY_NODE_ID, body), attrs)) + } + + fn is_async_block(&self) -> bool { + self.token.is_keyword(kw::Async) && + ( + ( // `async move {` + self.is_keyword_ahead(1, &[kw::Move]) && + self.look_ahead(2, |t| *t == token::OpenDelim(token::Brace)) + ) || ( // `async {` + self.look_ahead(1, |t| *t == token::OpenDelim(token::Brace)) + ) + ) + } + + fn maybe_parse_struct_expr( + &mut self, + lo: Span, + path: &ast::Path, + attrs: &ThinVec<Attribute>, + ) -> Option<PResult<'a, P<Expr>>> { + let struct_allowed = !self.restrictions.contains(Restrictions::NO_STRUCT_LITERAL); + let certainly_not_a_block = || 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()) + ) + ); + + if struct_allowed || certainly_not_a_block() { + // This is a struct literal, but we don't can't accept them here. + let expr = self.parse_struct_expr(lo, path.clone(), attrs.clone()); + if let (Ok(expr), false) = (&expr, struct_allowed) { + self.struct_span_err( + expr.span, + "struct literals are not allowed here", + ) + .multipart_suggestion( + "surround the struct literal with parentheses", + vec![ + (lo.shrink_to_lo(), "(".to_string()), + (expr.span.shrink_to_hi(), ")".to_string()), + ], + Applicability::MachineApplicable, + ) + .emit(); + } + return Some(expr); + } + None + } + + pub(super) fn parse_struct_expr( + &mut self, + lo: Span, + pth: ast::Path, + mut attrs: ThinVec<Attribute> + ) -> PResult<'a, P<Expr>> { + let struct_sp = lo.to(self.prev_span); + self.bump(); + let mut fields = Vec::new(); + let mut base = None; + + attrs.extend(self.parse_inner_attributes()?); + + while self.token != token::CloseDelim(token::Brace) { + if self.eat(&token::DotDot) { + let exp_span = self.prev_span; + match self.parse_expr() { + Ok(e) => { + base = Some(e); + } + Err(mut e) => { + e.emit(); + self.recover_stmt(); + } + } + if self.token == token::Comma { + self.struct_span_err( + exp_span.to(self.prev_span), + "cannot use a comma after the base struct", + ) + .span_suggestion_short( + self.token.span, + "remove this comma", + String::new(), + Applicability::MachineApplicable + ) + .note("the base struct must always be the last field") + .emit(); + self.recover_stmt(); + } + break; + } + + let mut recovery_field = None; + if let token::Ident(name, _) = self.token.kind { + if !self.token.is_reserved_ident() && self.look_ahead(1, |t| *t == token::Colon) { + // Use in case of error after field-looking code: `S { foo: () with a }`. + recovery_field = Some(ast::Field { + ident: Ident::new(name, self.token.span), + span: self.token.span, + expr: self.mk_expr(self.token.span, ExprKind::Err, ThinVec::new()), + is_shorthand: false, + attrs: ThinVec::new(), + id: DUMMY_NODE_ID, + is_placeholder: false, + }); + } + } + let mut parsed_field = None; + match self.parse_field() { + Ok(f) => parsed_field = Some(f), + Err(mut e) => { + e.span_label(struct_sp, "while parsing this struct"); + e.emit(); + + // 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; + } + } + } + } + + match self.expect_one_of(&[token::Comma], + &[token::CloseDelim(token::Brace)]) { + Ok(_) => if let Some(f) = parsed_field.or(recovery_field) { + // Only include the field if there's no parse error for the field name. + fields.push(f); + } + Err(mut e) => { + if let Some(f) = recovery_field { + fields.push(f); + } + e.span_label(struct_sp, "while parsing this struct"); + e.emit(); + self.recover_stmt_(SemiColonMode::Comma, BlockMode::Ignore); + self.eat(&token::Comma); + } + } + } + + let span = lo.to(self.token.span); + self.expect(&token::CloseDelim(token::Brace))?; + return Ok(self.mk_expr(span, ExprKind::Struct(pth, fields, base), attrs)); + } + + /// Parses `ident (COLON expr)?`. + fn parse_field(&mut self) -> PResult<'a, Field> { + let attrs = self.parse_outer_attributes()?; + let lo = self.token.span; + + // Check if a colon exists one ahead. This means we're parsing a fieldname. + let (fieldname, expr, is_shorthand) = if self.look_ahead(1, |t| { + t == &token::Colon || t == &token::Eq + }) { + let fieldname = self.parse_field_name()?; + + // Check for an equals token. This means the source incorrectly attempts to + // initialize a field with an eq rather than a colon. + if self.token == token::Eq { + self.diagnostic() + .struct_span_err(self.token.span, "expected `:`, found `=`") + .span_suggestion( + fieldname.span.shrink_to_hi().to(self.token.span), + "replace equals symbol with a colon", + ":".to_string(), + Applicability::MachineApplicable, + ) + .emit(); + } + self.bump(); // `:` + (fieldname, self.parse_expr()?, false) + } else { + let fieldname = self.parse_ident_common(false)?; + + // Mimic `x: x` for the `x` field shorthand. + let path = ast::Path::from_ident(fieldname); + let expr = self.mk_expr(fieldname.span, ExprKind::Path(None, path), ThinVec::new()); + (fieldname, expr, true) + }; + Ok(ast::Field { + ident: fieldname, + span: lo.to(expr.span), + expr, + is_shorthand, + attrs: attrs.into(), + id: DUMMY_NODE_ID, + is_placeholder: false, + }) + } + + fn err_dotdotdot_syntax(&self, span: Span) { + self.struct_span_err(span, "unexpected token: `...`") + .span_suggestion( + span, + "use `..` for an exclusive range", "..".to_owned(), + Applicability::MaybeIncorrect + ) + .span_suggestion( + span, + "or `..=` for an inclusive range", "..=".to_owned(), + Applicability::MaybeIncorrect + ) + .emit(); + } + + fn err_larrow_operator(&self, span: Span) { + self.struct_span_err( + span, + "unexpected token: `<-`" + ).span_suggestion( + span, + "if you meant to write a comparison against a negative value, add a \ + space in between `<` and `-`", + "< -".to_string(), + Applicability::MaybeIncorrect + ).emit(); + } + + fn mk_assign_op(&self, binop: BinOp, lhs: P<Expr>, rhs: P<Expr>) -> ExprKind { + ExprKind::AssignOp(binop, lhs, rhs) + } + + fn mk_range( + &self, + start: Option<P<Expr>>, + end: Option<P<Expr>>, + limits: RangeLimits + ) -> PResult<'a, ExprKind> { + if end.is_none() && limits == RangeLimits::Closed { + Err(self.span_fatal_err(self.token.span, Error::InclusiveRangeWithNoEnd)) + } else { + Ok(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>) -> ExprKind { + ExprKind::Index(expr, idx) + } + + fn mk_call(&self, f: P<Expr>, args: Vec<P<Expr>>) -> ExprKind { + ExprKind::Call(f, args) + } + + fn mk_await_expr(&mut self, self_arg: P<Expr>, lo: Span) -> PResult<'a, P<Expr>> { + let span = lo.to(self.prev_span); + let await_expr = self.mk_expr(span, ExprKind::Await(self_arg), ThinVec::new()); + self.recover_from_await_method_call(); + Ok(await_expr) + } + + crate fn mk_expr(&self, span: Span, kind: ExprKind, attrs: ThinVec<Attribute>) -> P<Expr> { + P(Expr { kind, span, attrs, id: DUMMY_NODE_ID }) + } + + pub(super) fn mk_expr_err(&self, span: Span) -> P<Expr> { + self.mk_expr(span, ExprKind::Err, ThinVec::new()) + } +} diff --git a/src/librustc_parse/parser/generics.rs b/src/librustc_parse/parser/generics.rs new file mode 100644 index 00000000000..ba5eafc0ed7 --- /dev/null +++ b/src/librustc_parse/parser/generics.rs @@ -0,0 +1,308 @@ +use super::Parser; + +use syntax::ast::{self, WhereClause, GenericParam, GenericParamKind, GenericBounds, Attribute}; +use syntax::token; +use syntax::source_map::DUMMY_SP; +use syntax_pos::symbol::{kw, sym}; + +use errors::PResult; + +impl<'a> Parser<'a> { + /// Parses bounds of a lifetime parameter `BOUND + BOUND + BOUND`, possibly with trailing `+`. + /// + /// ``` + /// 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: Vec<Attribute>) + -> PResult<'a, GenericParam> { + let ident = self.parse_ident()?; + + // Parse optional colon and param bounds. + let bounds = if self.eat(&token::Colon) { + self.parse_generic_bounds(Some(self.prev_span))? + } 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.into(), + bounds, + kind: GenericParamKind::Type { + default, + }, + is_placeholder: false + }) + } + + fn parse_const_param(&mut self, preceding_attrs: Vec<Attribute>) -> PResult<'a, GenericParam> { + let lo = self.token.span; + + self.expect_keyword(kw::Const)?; + let ident = self.parse_ident()?; + self.expect(&token::Colon)?; + let ty = self.parse_ty()?; + + self.sess.gated_spans.gate(sym::const_generics, lo.to(self.prev_span)); + + Ok(GenericParam { + ident, + id: ast::DUMMY_NODE_ID, + attrs: preceding_attrs.into(), + bounds: Vec::new(), + kind: GenericParamKind::Const { + ty, + }, + is_placeholder: false + }) + } + + /// 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, Vec<ast::GenericParam>> { + let mut params = Vec::new(); + loop { + let attrs = self.parse_outer_attributes()?; + if self.check_lifetime() { + let lifetime = self.expect_lifetime(); + // Parse lifetime parameter. + let bounds = if self.eat(&token::Colon) { + self.parse_lt_param_bounds() + } else { + Vec::new() + }; + params.push(ast::GenericParam { + ident: lifetime.ident, + id: lifetime.id, + attrs: attrs.into(), + bounds, + kind: ast::GenericParamKind::Lifetime, + is_placeholder: false + }); + } else if self.check_keyword(kw::Const) { + // Parse const parameter. + params.push(self.parse_const_param(attrs)?); + } else if self.check_ident() { + // Parse type parameter. + params.push(self.parse_ty_param(attrs)?); + } else if self.token.can_begin_type() { + // Trying to write an associated type bound? (#26271) + let snapshot = self.clone(); + match self.parse_ty_where_predicate() { + Ok(where_predicate) => { + self.struct_span_err( + where_predicate.span(), + "bounds on associated types do not belong here", + ) + .span_label(where_predicate.span(), "belongs in `where` clause") + .emit(); + } + Err(mut err) => { + err.cancel(); + std::mem::replace(self, snapshot); + break + } + } + } else { + // Check for trailing attributes and stop parsing. + if !attrs.is_empty() { + if !params.is_empty() { + self.struct_span_err( + attrs[0].span, + "trailing attribute after generic parameter", + ) + .span_label(attrs[0].span, "attributes must go before parameters") + .emit(); + } else { + self.struct_span_err( + attrs[0].span, + &format!("attribute without generic parameters"), + ) + .span_label( + attrs[0].span, + "attributes are only permitted when preceding parameters", + ) + .emit(); + } + } + break + } + + if !self.eat(&token::Comma) { + 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()?; + (params, span_lo.to(self.prev_span)) + } else { + (vec![], self.prev_span.between(self.token.span)) + }; + Ok(ast::Generics { + params, + where_clause: WhereClause { + predicates: Vec::new(), + span: DUMMY_SP, + }, + span, + }) + } + + /// Parses an optional where-clause and places it in `generics`. + /// + /// ```ignore (only-for-syntax-highlight) + /// where T : Trait<U, V> + 'b, 'a : 'b + /// ``` + pub(super) fn parse_where_clause(&mut self) -> PResult<'a, WhereClause> { + let mut where_clause = WhereClause { + predicates: Vec::new(), + span: self.prev_span.to(self.prev_span), + }; + + if !self.eat_keyword(kw::Where) { + return Ok(where_clause); + } + let lo = self.prev_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() { + let generics = self.parse_generics()?; + self.struct_span_err( + generics.span, + "generic parameters on `where` clauses are reserved for future use", + ) + .span_label(generics.span, "currently unsupported") + .emit(); + } + + loop { + let 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: lo.to(self.prev_span), + lifetime, + bounds, + } + )); + } else if self.check_type() { + where_clause.predicates.push(self.parse_ty_where_predicate()?); + } else { + break + } + + if !self.eat(&token::Comma) { + break + } + } + + where_clause.span = lo.to(self.prev_span); + Ok(where_clause) + } + + 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()?; + if self.eat(&token::Colon) { + let bounds = self.parse_generic_bounds(Some(self.prev_span))?; + Ok(ast::WherePredicate::BoundPredicate( + ast::WhereBoundPredicate { + span: lo.to(self.prev_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_span), + lhs_ty: ty, + rhs_ty, + id: ast::DUMMY_NODE_ID, + } + )) + } else { + self.unexpected() + } + } + + pub(super) fn choose_generics_over_qpath(&self) -> 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 + // 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.token == token::Lt && + (self.look_ahead(1, |t| t == &token::Pound || t == &token::Gt) || + self.look_ahead(1, |t| t.is_lifetime() || t.is_ident()) && + self.look_ahead(2, |t| t == &token::Gt || t == &token::Comma || + t == &token::Colon || t == &token::Eq) || + self.is_keyword_ahead(1, &[kw::Const])) + } +} diff --git a/src/librustc_parse/parser/item.rs b/src/librustc_parse/parser/item.rs new file mode 100644 index 00000000000..3e21436d313 --- /dev/null +++ b/src/librustc_parse/parser/item.rs @@ -0,0 +1,2238 @@ +use super::{Parser, PathStyle}; +use super::diagnostics::{Error, dummy_arg, ConsumeClosingDelim}; + +use crate::maybe_whole; + +use syntax::ast::{self, Abi, DUMMY_NODE_ID, Ident, Attribute, AttrKind, AttrStyle, AnonConst, Item}; +use syntax::ast::{ItemKind, ImplItem, ImplItemKind, TraitItem, TraitItemKind, UseTree, UseTreeKind}; +use syntax::ast::{PathSegment, IsAuto, Constness, IsAsync, Unsafety, Defaultness}; +use syntax::ast::{Visibility, VisibilityKind, Mutability, FnHeader, ForeignItem, ForeignItemKind}; +use syntax::ast::{Ty, TyKind, Generics, GenericBounds, TraitRef, EnumDef, VariantData, StructField}; +use syntax::ast::{Mac, MacDelimiter, Block, BindingMode, FnDecl, FnSig, SelfKind, Param}; +use syntax::ptr::P; +use syntax::ThinVec; +use syntax::token; +use syntax::tokenstream::{TokenTree, TokenStream}; +use syntax::source_map::{self, respan, Span}; +use syntax_pos::BytePos; +use syntax_pos::symbol::{kw, sym}; + +use log::debug; +use std::mem; +use errors::{PResult, Applicability, DiagnosticBuilder, DiagnosticId, StashKey}; + +/// Whether the type alias or associated type is a concrete type or an opaque type. +#[derive(Debug)] +pub(super) enum AliasKind { + /// Just a new name for the same type. + Weak(P<Ty>), + /// Only trait impls of the type will be usable, not the actual type itself. + OpaqueTy(GenericBounds), +} + +pub(super) type ItemInfo = (Ident, ItemKind, Option<Vec<Attribute>>); + +impl<'a> Parser<'a> { + pub fn parse_item(&mut self) -> PResult<'a, Option<P<Item>>> { + let attrs = self.parse_outer_attributes()?; + self.parse_item_(attrs, true, false) + } + + pub(super) fn parse_item_( + &mut self, + attrs: Vec<Attribute>, + macros_allowed: bool, + attributes_allowed: bool, + ) -> PResult<'a, Option<P<Item>>> { + let mut unclosed_delims = vec![]; + let (ret, tokens) = self.collect_tokens(|this| { + let item = this.parse_item_implementation(attrs, macros_allowed, attributes_allowed); + unclosed_delims.append(&mut this.unclosed_delims); + item + })?; + self.unclosed_delims.append(&mut unclosed_delims); + + // Once we've parsed an item and recorded the tokens we got while + // parsing we may want to store `tokens` into the item we're about to + // return. Note, though, that we specifically didn't capture tokens + // related to outer attributes. The `tokens` field here may later be + // used with procedural macros to convert this item back into a token + // stream, but during expansion we may be removing attributes as we go + // along. + // + // If we've got inner attributes then the `tokens` we've got above holds + // these inner attributes. If an inner attribute is expanded we won't + // actually remove it from the token stream, so we'll just keep yielding + // it (bad!). To work around this case for now we just avoid recording + // `tokens` if we detect any inner attributes. This should help keep + // expansion correct, but we should fix this bug one day! + Ok(ret.map(|item| { + item.map(|mut i| { + if !i.attrs.iter().any(|attr| attr.style == AttrStyle::Inner) { + i.tokens = Some(tokens); + } + i + }) + })) + } + + /// Parses one of the items allowed by the flags. + fn parse_item_implementation( + &mut self, + attrs: Vec<Attribute>, + macros_allowed: bool, + attributes_allowed: bool, + ) -> PResult<'a, Option<P<Item>>> { + maybe_whole!(self, NtItem, |item| { + let mut item = item.into_inner(); + let mut attrs = attrs; + mem::swap(&mut item.attrs, &mut attrs); + item.attrs.extend(attrs); + Some(P(item)) + }); + + let lo = self.token.span; + + let vis = self.parse_visibility(false)?; + + if self.eat_keyword(kw::Use) { + // USE ITEM + let item_ = ItemKind::Use(P(self.parse_use_tree()?)); + self.expect_semi()?; + + let span = lo.to(self.prev_span); + let item = self.mk_item(span, Ident::invalid(), item_, vis, attrs); + return Ok(Some(item)); + } + + if self.eat_keyword(kw::Extern) { + let extern_sp = self.prev_span; + if self.eat_keyword(kw::Crate) { + return Ok(Some(self.parse_item_extern_crate(lo, vis, attrs)?)); + } + + let abi = self.parse_opt_abi()?; + + if self.eat_keyword(kw::Fn) { + // EXTERN FUNCTION ITEM + let fn_span = self.prev_span; + let header = FnHeader { + unsafety: Unsafety::Normal, + asyncness: respan(fn_span, IsAsync::NotAsync), + constness: respan(fn_span, Constness::NotConst), + abi, + }; + return self.parse_item_fn(lo, vis, attrs, header); + } else if self.check(&token::OpenDelim(token::Brace)) { + return Ok(Some( + self.parse_item_foreign_mod(lo, abi, vis, attrs, extern_sp)?, + )); + } + + self.unexpected()?; + } + + if self.is_static_global() { + self.bump(); + // STATIC ITEM + let m = self.parse_mutability(); + let info = self.parse_item_const(Some(m))?; + return self.mk_item_with_info(attrs, lo, vis, info); + } + + if self.eat_keyword(kw::Const) { + let const_span = self.prev_span; + if [kw::Fn, kw::Unsafe, kw::Extern].iter().any(|k| self.check_keyword(*k)) { + // CONST FUNCTION ITEM + let unsafety = self.parse_unsafety(); + + if self.check_keyword(kw::Extern) { + self.sess.gated_spans.gate(sym::const_extern_fn, lo.to(self.token.span)); + } + let abi = self.parse_extern_abi()?; + self.bump(); // `fn` + + let header = FnHeader { + unsafety, + asyncness: respan(const_span, IsAsync::NotAsync), + constness: respan(const_span, Constness::Const), + abi, + }; + return self.parse_item_fn(lo, vis, attrs, header); + } + + // CONST ITEM + if self.eat_keyword(kw::Mut) { + let prev_span = self.prev_span; + self.struct_span_err(prev_span, "const globals cannot be mutable") + .span_label(prev_span, "cannot be mutable") + .span_suggestion( + const_span, + "you might want to declare a static instead", + "static".to_owned(), + Applicability::MaybeIncorrect, + ) + .emit(); + } + + let info = self.parse_item_const(None)?; + return self.mk_item_with_info(attrs, lo, vis, info); + } + + // Parses `async unsafe? fn`. + if self.check_keyword(kw::Async) { + let async_span = self.token.span; + if self.is_keyword_ahead(1, &[kw::Fn]) + || self.is_keyword_ahead(2, &[kw::Fn]) + { + // ASYNC FUNCTION ITEM + self.bump(); // `async` + let unsafety = self.parse_unsafety(); // `unsafe`? + self.expect_keyword(kw::Fn)?; // `fn` + let fn_span = self.prev_span; + let asyncness = respan(async_span, IsAsync::Async { + closure_id: DUMMY_NODE_ID, + return_impl_trait_id: DUMMY_NODE_ID, + }); + self.ban_async_in_2015(async_span); + let header = FnHeader { + unsafety, + asyncness, + constness: respan(fn_span, Constness::NotConst), + abi: Abi::new(sym::Rust, fn_span), + }; + return self.parse_item_fn(lo, vis, attrs, header); + } + } + + if self.check_keyword(kw::Unsafe) && + self.is_keyword_ahead(1, &[kw::Trait, kw::Auto]) + { + // UNSAFE TRAIT ITEM + self.bump(); // `unsafe` + let info = self.parse_item_trait(lo, Unsafety::Unsafe)?; + return self.mk_item_with_info(attrs, lo, vis, info); + } + + if self.check_keyword(kw::Impl) || + self.check_keyword(kw::Unsafe) && + self.is_keyword_ahead(1, &[kw::Impl]) || + self.check_keyword(kw::Default) && + self.is_keyword_ahead(1, &[kw::Impl, kw::Unsafe]) + { + // IMPL ITEM + let defaultness = self.parse_defaultness(); + let unsafety = self.parse_unsafety(); + self.expect_keyword(kw::Impl)?; + let info = self.parse_item_impl(unsafety, defaultness)?; + return self.mk_item_with_info(attrs, lo, vis, info); + } + + if self.check_keyword(kw::Fn) { + // FUNCTION ITEM + self.bump(); + let fn_span = self.prev_span; + let header = FnHeader { + unsafety: Unsafety::Normal, + asyncness: respan(fn_span, IsAsync::NotAsync), + constness: respan(fn_span, Constness::NotConst), + abi: Abi::new(sym::Rust, fn_span), + }; + return self.parse_item_fn(lo, vis, attrs, header); + } + + if self.check_keyword(kw::Unsafe) + && self.look_ahead(1, |t| *t != token::OpenDelim(token::Brace)) + { + // UNSAFE FUNCTION ITEM + self.bump(); // `unsafe` + // `{` is also expected after `unsafe`; in case of error, include it in the diagnostic. + self.check(&token::OpenDelim(token::Brace)); + let abi = self.parse_extern_abi()?; + self.expect_keyword(kw::Fn)?; + let fn_span = self.prev_span; + let header = FnHeader { + unsafety: Unsafety::Unsafe, + asyncness: respan(fn_span, IsAsync::NotAsync), + constness: respan(fn_span, Constness::NotConst), + abi, + }; + return self.parse_item_fn(lo, vis, attrs, header); + } + + if self.eat_keyword(kw::Mod) { + // MODULE ITEM + let info = self.parse_item_mod(&attrs[..])?; + return self.mk_item_with_info(attrs, lo, vis, info); + } + + if let Some(type_) = self.eat_type() { + let (ident, alias, generics) = type_?; + // TYPE ITEM + let item_ = match alias { + AliasKind::Weak(ty) => ItemKind::TyAlias(ty, generics), + AliasKind::OpaqueTy(bounds) => ItemKind::OpaqueTy(bounds, generics), + }; + let span = lo.to(self.prev_span); + return Ok(Some(self.mk_item(span, ident, item_, vis, attrs))); + } + + if self.eat_keyword(kw::Enum) { + // ENUM ITEM + let info = self.parse_item_enum()?; + return self.mk_item_with_info(attrs, lo, vis, info); + } + + if self.check_keyword(kw::Trait) + || (self.check_keyword(kw::Auto) + && self.is_keyword_ahead(1, &[kw::Trait])) + { + // TRAIT ITEM + let info = self.parse_item_trait(lo, Unsafety::Normal)?; + return self.mk_item_with_info(attrs, lo, vis, info); + } + + if self.eat_keyword(kw::Struct) { + // STRUCT ITEM + let info = self.parse_item_struct()?; + return self.mk_item_with_info(attrs, lo, vis, info); + } + + if self.is_union_item() { + // UNION ITEM + self.bump(); + let info = self.parse_item_union()?; + return self.mk_item_with_info(attrs, lo, vis, info); + } + + if let Some(macro_def) = self.eat_macro_def(&attrs, &vis, lo)? { + return Ok(Some(macro_def)); + } + + // Verify whether we have encountered a struct or method definition where the user forgot to + // add the `struct` or `fn` keyword after writing `pub`: `pub S {}` + if vis.node.is_pub() && + self.check_ident() && + self.look_ahead(1, |t| *t != token::Not) + { + // Space between `pub` keyword and the identifier + // + // pub S {} + // ^^^ `sp` points here + let sp = self.prev_span.between(self.token.span); + let full_sp = self.prev_span.to(self.token.span); + let ident_sp = self.token.span; + if self.look_ahead(1, |t| *t == token::OpenDelim(token::Brace)) { + // possible public struct definition where `struct` was forgotten + let ident = self.parse_ident().unwrap(); + let msg = format!("add `struct` here to parse `{}` as a public struct", + ident); + let mut err = self.diagnostic() + .struct_span_err(sp, "missing `struct` for struct definition"); + err.span_suggestion_short( + sp, &msg, " struct ".into(), Applicability::MaybeIncorrect // speculative + ); + return Err(err); + } else if self.look_ahead(1, |t| *t == token::OpenDelim(token::Paren)) { + let ident = self.parse_ident().unwrap(); + self.bump(); // `(` + let kw_name = self.recover_first_param(); + self.consume_block(token::Paren, ConsumeClosingDelim::Yes); + let (kw, kw_name, ambiguous) = if self.check(&token::RArrow) { + self.eat_to_tokens(&[&token::OpenDelim(token::Brace)]); + self.bump(); // `{` + ("fn", kw_name, false) + } else if self.check(&token::OpenDelim(token::Brace)) { + self.bump(); // `{` + ("fn", kw_name, false) + } else if self.check(&token::Colon) { + let kw = "struct"; + (kw, kw, false) + } else { + ("fn` or `struct", "function or struct", true) + }; + + let msg = format!("missing `{}` for {} definition", kw, kw_name); + let mut err = self.diagnostic().struct_span_err(sp, &msg); + if !ambiguous { + self.consume_block(token::Brace, ConsumeClosingDelim::Yes); + let suggestion = format!("add `{}` here to parse `{}` as a public {}", + kw, + ident, + kw_name); + err.span_suggestion_short( + sp, &suggestion, format!(" {} ", kw), Applicability::MachineApplicable + ); + } else { + if let Ok(snippet) = self.span_to_snippet(ident_sp) { + err.span_suggestion( + full_sp, + "if you meant to call a macro, try", + format!("{}!", snippet), + // this is the `ambiguous` conditional branch + Applicability::MaybeIncorrect + ); + } else { + err.help("if you meant to call a macro, remove the `pub` \ + and add a trailing `!` after the identifier"); + } + } + return Err(err); + } else if self.look_ahead(1, |t| *t == token::Lt) { + let ident = self.parse_ident().unwrap(); + self.eat_to_tokens(&[&token::Gt]); + self.bump(); // `>` + let (kw, kw_name, ambiguous) = if self.eat(&token::OpenDelim(token::Paren)) { + ("fn", self.recover_first_param(), false) + } else if self.check(&token::OpenDelim(token::Brace)) { + ("struct", "struct", false) + } else { + ("fn` or `struct", "function or struct", true) + }; + let msg = format!("missing `{}` for {} definition", kw, kw_name); + let mut err = self.diagnostic().struct_span_err(sp, &msg); + if !ambiguous { + err.span_suggestion_short( + sp, + &format!("add `{}` here to parse `{}` as a public {}", kw, ident, kw_name), + format!(" {} ", kw), + Applicability::MachineApplicable, + ); + } + return Err(err); + } + } + self.parse_macro_use_or_failure(attrs, macros_allowed, attributes_allowed, lo, vis) + } + + pub(super) fn mk_item_with_info( + &self, + attrs: Vec<Attribute>, + lo: Span, + vis: Visibility, + info: ItemInfo, + ) -> PResult<'a, Option<P<Item>>> { + let (ident, item, extra_attrs) = info; + let span = lo.to(self.prev_span); + let attrs = Self::maybe_append(attrs, extra_attrs); + Ok(Some(self.mk_item(span, ident, item, vis, attrs))) + } + + fn maybe_append<T>(mut lhs: Vec<T>, mut rhs: Option<Vec<T>>) -> Vec<T> { + if let Some(ref mut rhs) = rhs { + lhs.append(rhs); + } + lhs + } + + /// This is the fall-through for parsing items. + fn parse_macro_use_or_failure( + &mut self, + attrs: Vec<Attribute> , + macros_allowed: bool, + attributes_allowed: bool, + lo: Span, + visibility: Visibility + ) -> PResult<'a, Option<P<Item>>> { + if macros_allowed && self.token.is_path_start() && + !(self.is_async_fn() && self.token.span.rust_2015()) { + // MACRO INVOCATION ITEM + + let prev_span = self.prev_span; + self.complain_if_pub_macro(&visibility.node, prev_span); + + let mac_lo = self.token.span; + + // Item macro + let path = self.parse_path(PathStyle::Mod)?; + self.expect(&token::Not)?; + let (delim, tts) = self.expect_delimited_token_tree()?; + if delim != MacDelimiter::Brace && !self.eat(&token::Semi) { + self.report_invalid_macro_expansion_item(); + } + + let hi = self.prev_span; + let mac = Mac { + path, + tts, + delim, + span: mac_lo.to(hi), + prior_type_ascription: self.last_type_ascription, + }; + let item = + self.mk_item(lo.to(hi), Ident::invalid(), ItemKind::Mac(mac), visibility, attrs); + return Ok(Some(item)); + } + + // FAILURE TO PARSE ITEM + match visibility.node { + VisibilityKind::Inherited => {} + _ => { + return Err(self.span_fatal(self.prev_span, "unmatched visibility `pub`")); + } + } + + if !attributes_allowed && !attrs.is_empty() { + self.expected_item_err(&attrs)?; + } + Ok(None) + } + + /// Emits an expected-item-after-attributes error. + fn expected_item_err(&mut self, attrs: &[Attribute]) -> PResult<'a, ()> { + let message = match attrs.last() { + Some(&Attribute { kind: AttrKind::DocComment(_), .. }) => + "expected item after doc comment", + _ => + "expected item after attributes", + }; + + let mut err = self.diagnostic().struct_span_err(self.prev_span, message); + if attrs.last().unwrap().is_doc_comment() { + err.span_label(self.prev_span, "this doc comment doesn't document anything"); + } + Err(err) + } + + pub(super) fn is_async_fn(&self) -> bool { + self.token.is_keyword(kw::Async) && + self.is_keyword_ahead(1, &[kw::Fn]) + } + + /// Parses a macro invocation inside a `trait`, `impl` or `extern` block. + fn parse_assoc_macro_invoc(&mut self, item_kind: &str, vis: Option<&Visibility>, + at_end: &mut bool) -> PResult<'a, Option<Mac>> + { + if self.token.is_path_start() && + !(self.is_async_fn() && self.token.span.rust_2015()) { + let prev_span = self.prev_span; + let lo = self.token.span; + let path = self.parse_path(PathStyle::Mod)?; + + if path.segments.len() == 1 { + if !self.eat(&token::Not) { + return Err(self.missing_assoc_item_kind_err(item_kind, prev_span)); + } + } else { + self.expect(&token::Not)?; + } + + if let Some(vis) = vis { + self.complain_if_pub_macro(&vis.node, prev_span); + } + + *at_end = true; + + // eat a matched-delimiter token tree: + let (delim, tts) = self.expect_delimited_token_tree()?; + if delim != MacDelimiter::Brace { + self.expect_semi()?; + } + + Ok(Some(Mac { + path, + tts, + delim, + span: lo.to(self.prev_span), + prior_type_ascription: self.last_type_ascription, + })) + } else { + Ok(None) + } + } + + fn missing_assoc_item_kind_err(&self, item_type: &str, prev_span: Span) + -> DiagnosticBuilder<'a> + { + let expected_kinds = if item_type == "extern" { + "missing `fn`, `type`, or `static`" + } else { + "missing `fn`, `type`, or `const`" + }; + + // Given this code `path(`, it seems like this is not + // setting the visibility of a macro invocation, but rather + // a mistyped method declaration. + // Create a diagnostic pointing out that `fn` is missing. + // + // x | pub path(&self) { + // | ^ missing `fn`, `type`, or `const` + // pub path( + // ^^ `sp` below will point to this + let sp = prev_span.between(self.prev_span); + let mut err = self.diagnostic().struct_span_err( + sp, + &format!("{} for {}-item declaration", + expected_kinds, item_type)); + err.span_label(sp, expected_kinds); + err + } + + /// Parses an implementation item, `impl` keyword is already parsed. + /// + /// impl<'a, T> TYPE { /* impl items */ } + /// impl<'a, T> TRAIT for TYPE { /* impl items */ } + /// impl<'a, T> !TRAIT for TYPE { /* impl items */ } + /// + /// We actually parse slightly more relaxed grammar for better error reporting and recovery. + /// `impl` GENERICS `!`? TYPE `for`? (TYPE | `..`) (`where` PREDICATES)? `{` BODY `}` + /// `impl` GENERICS `!`? TYPE (`where` PREDICATES)? `{` BODY `}` + fn parse_item_impl(&mut self, unsafety: Unsafety, defaultness: Defaultness) + -> PResult<'a, ItemInfo> { + // First, parse generic parameters if necessary. + let mut generics = if self.choose_generics_over_qpath() { + self.parse_generics()? + } else { + Generics::default() + }; + + // Disambiguate `impl !Trait for Type { ... }` and `impl ! { ... }` for the never type. + let polarity = if self.check(&token::Not) && self.look_ahead(1, |t| t.can_begin_type()) { + self.bump(); // `!` + ast::ImplPolarity::Negative + } else { + ast::ImplPolarity::Positive + }; + + // Parse both types and traits as a type, then reinterpret if necessary. + let err_path = |span| ast::Path::from_ident(Ident::new(kw::Invalid, span)); + let ty_first = if self.token.is_keyword(kw::For) && + self.look_ahead(1, |t| t != &token::Lt) { + let span = self.prev_span.between(self.token.span); + self.struct_span_err(span, "missing trait in a trait impl").emit(); + P(Ty { kind: TyKind::Path(None, err_path(span)), span, id: DUMMY_NODE_ID }) + } else { + self.parse_ty()? + }; + + // If `for` is missing we try to recover. + let has_for = self.eat_keyword(kw::For); + let missing_for_span = self.prev_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 + Some(self.mk_ty(self.prev_span, TyKind::Err)) + } 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, attrs) = self.parse_impl_body()?; + + let item_kind = match ty_second { + Some(ty_second) => { + // impl Trait for Type + if !has_for { + self.struct_span_err(missing_for_span, "missing `for` in a trait impl") + .span_suggestion_short( + missing_for_span, + "add `for` here", + " for ".to_string(), + Applicability::MachineApplicable, + ).emit(); + } + + 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, + _ => { + self.span_err(ty_first.span, "expected a trait, found type"); + err_path(ty_first.span) + } + }; + let trait_ref = TraitRef { path, ref_id: ty_first.id }; + + ItemKind::Impl(unsafety, polarity, defaultness, + generics, Some(trait_ref), ty_second, impl_items) + } + None => { + // impl Type + ItemKind::Impl(unsafety, polarity, defaultness, + generics, None, ty_first, impl_items) + } + }; + + Ok((Ident::invalid(), item_kind, Some(attrs))) + } + + fn parse_impl_body(&mut self) -> PResult<'a, (Vec<ImplItem>, Vec<Attribute>)> { + self.expect(&token::OpenDelim(token::Brace))?; + let attrs = self.parse_inner_attributes()?; + + let mut impl_items = Vec::new(); + while !self.eat(&token::CloseDelim(token::Brace)) { + let mut at_end = false; + match self.parse_impl_item(&mut at_end) { + Ok(impl_item) => impl_items.push(impl_item), + Err(mut err) => { + err.emit(); + if !at_end { + self.consume_block(token::Brace, ConsumeClosingDelim::Yes); + break; + } + } + } + } + Ok((impl_items, attrs)) + } + + /// Parses an impl item. + pub fn parse_impl_item(&mut self, at_end: &mut bool) -> PResult<'a, ImplItem> { + maybe_whole!(self, NtImplItem, |x| x); + let attrs = self.parse_outer_attributes()?; + let mut unclosed_delims = vec![]; + let (mut item, tokens) = self.collect_tokens(|this| { + let item = this.parse_impl_item_(at_end, attrs); + unclosed_delims.append(&mut this.unclosed_delims); + item + })?; + self.unclosed_delims.append(&mut unclosed_delims); + + // See `parse_item` for why this clause is here. + if !item.attrs.iter().any(|attr| attr.style == AttrStyle::Inner) { + item.tokens = Some(tokens); + } + Ok(item) + } + + fn parse_impl_item_( + &mut self, + at_end: &mut bool, + mut attrs: Vec<Attribute>, + ) -> PResult<'a, ImplItem> { + let lo = self.token.span; + let vis = self.parse_visibility(false)?; + let defaultness = self.parse_defaultness(); + let (name, kind, generics) = if let Some(type_) = self.eat_type() { + let (name, alias, generics) = type_?; + let kind = match alias { + AliasKind::Weak(typ) => ast::ImplItemKind::TyAlias(typ), + AliasKind::OpaqueTy(bounds) => ast::ImplItemKind::OpaqueTy(bounds), + }; + (name, kind, generics) + } else if self.is_const_item() { + self.parse_impl_const()? + } else if let Some(mac) = self.parse_assoc_macro_invoc("impl", Some(&vis), at_end)? { + // FIXME: code copied from `parse_macro_use_or_failure` -- use abstraction! + (Ident::invalid(), ast::ImplItemKind::Macro(mac), Generics::default()) + } else { + let (name, inner_attrs, generics, kind) = self.parse_impl_method(at_end)?; + attrs.extend(inner_attrs); + (name, kind, generics) + }; + + Ok(ImplItem { + id: DUMMY_NODE_ID, + span: lo.to(self.prev_span), + ident: name, + vis, + defaultness, + attrs, + generics, + kind, + tokens: None, + }) + } + + /// Parses defaultness (i.e., `default` or nothing). + fn parse_defaultness(&mut self) -> Defaultness { + // `pub` is included for better error messages + if self.check_keyword(kw::Default) && + self.is_keyword_ahead(1, &[ + kw::Impl, + kw::Const, + kw::Async, + kw::Fn, + kw::Unsafe, + kw::Extern, + kw::Type, + kw::Pub, + ]) + { + self.bump(); // `default` + Defaultness::Default + } else { + Defaultness::Final + } + } + + /// Returns `true` if we are looking at `const ID` + /// (returns `false` for things like `const fn`, etc.). + fn is_const_item(&self) -> bool { + self.token.is_keyword(kw::Const) && + !self.is_keyword_ahead(1, &[kw::Fn, kw::Unsafe]) + } + + /// This parses the grammar: + /// ImplItemConst = "const" Ident ":" Ty "=" Expr ";" + fn parse_impl_const(&mut self) -> PResult<'a, (Ident, ImplItemKind, Generics)> { + self.expect_keyword(kw::Const)?; + let name = self.parse_ident()?; + self.expect(&token::Colon)?; + let typ = self.parse_ty()?; + self.expect(&token::Eq)?; + let expr = self.parse_expr()?; + self.expect_semi()?; + Ok((name, ImplItemKind::Const(typ, expr), Generics::default())) + } + + /// Parses `auto? trait Foo { ... }` or `trait Foo = Bar;`. + fn parse_item_trait(&mut self, lo: Span, unsafety: Unsafety) -> PResult<'a, ItemInfo> { + // Parse optional `auto` prefix. + let is_auto = if self.eat_keyword(kw::Auto) { + IsAuto::Yes + } else { + IsAuto::No + }; + + self.expect_keyword(kw::Trait)?; + let ident = self.parse_ident()?; + let mut tps = self.parse_generics()?; + + // Parse optional colon and supertrait bounds. + let had_colon = self.eat(&token::Colon); + let span_at_colon = self.prev_span; + let bounds = if had_colon { + self.parse_generic_bounds(Some(self.prev_span))? + } else { + Vec::new() + }; + + let span_before_eq = self.prev_span; + if self.eat(&token::Eq) { + // It's a trait alias. + if had_colon { + let span = span_at_colon.to(span_before_eq); + self.struct_span_err(span, "bounds are not allowed on trait aliases") + .emit(); + } + + let bounds = self.parse_generic_bounds(None)?; + tps.where_clause = self.parse_where_clause()?; + self.expect_semi()?; + + let whole_span = lo.to(self.prev_span); + if is_auto == IsAuto::Yes { + let msg = "trait aliases cannot be `auto`"; + self.struct_span_err(whole_span, msg) + .span_label(whole_span, msg) + .emit(); + } + if unsafety != Unsafety::Normal { + let msg = "trait aliases cannot be `unsafe`"; + self.struct_span_err(whole_span, msg) + .span_label(whole_span, msg) + .emit(); + } + + self.sess.gated_spans.gate(sym::trait_alias, whole_span); + + Ok((ident, ItemKind::TraitAlias(tps, bounds), None)) + } else { + // It's a normal trait. + tps.where_clause = self.parse_where_clause()?; + self.expect(&token::OpenDelim(token::Brace))?; + let mut trait_items = vec![]; + while !self.eat(&token::CloseDelim(token::Brace)) { + if let token::DocComment(_) = self.token.kind { + if self.look_ahead(1, + |tok| tok == &token::CloseDelim(token::Brace)) { + self.diagnostic().struct_span_err_with_code( + self.token.span, + "found a documentation comment that doesn't document anything", + DiagnosticId::Error("E0584".into()), + ) + .help( + "doc comments must come before what they document, maybe a \ + comment was intended with `//`?", + ) + .emit(); + self.bump(); + continue; + } + } + let mut at_end = false; + match self.parse_trait_item(&mut at_end) { + Ok(item) => trait_items.push(item), + Err(mut e) => { + e.emit(); + if !at_end { + self.consume_block(token::Brace, ConsumeClosingDelim::Yes); + break; + } + } + } + } + Ok((ident, ItemKind::Trait(is_auto, unsafety, tps, bounds, trait_items), None)) + } + } + + /// Parses the items in a trait declaration. + pub fn parse_trait_item(&mut self, at_end: &mut bool) -> PResult<'a, TraitItem> { + maybe_whole!(self, NtTraitItem, |x| x); + let attrs = self.parse_outer_attributes()?; + let mut unclosed_delims = vec![]; + let (mut item, tokens) = self.collect_tokens(|this| { + let item = this.parse_trait_item_(at_end, attrs); + unclosed_delims.append(&mut this.unclosed_delims); + item + })?; + self.unclosed_delims.append(&mut unclosed_delims); + // See `parse_item` for why this clause is here. + if !item.attrs.iter().any(|attr| attr.style == AttrStyle::Inner) { + item.tokens = Some(tokens); + } + Ok(item) + } + + fn parse_trait_item_( + &mut self, + at_end: &mut bool, + mut attrs: Vec<Attribute>, + ) -> PResult<'a, TraitItem> { + let lo = self.token.span; + self.eat_bad_pub(); + let (name, kind, generics) = if self.eat_keyword(kw::Type) { + self.parse_trait_item_assoc_ty()? + } else if self.is_const_item() { + self.parse_trait_item_const()? + } else if let Some(mac) = self.parse_assoc_macro_invoc("trait", None, &mut false)? { + // trait item macro. + (Ident::invalid(), TraitItemKind::Macro(mac), Generics::default()) + } else { + self.parse_trait_item_method(at_end, &mut attrs)? + }; + + Ok(TraitItem { + id: DUMMY_NODE_ID, + ident: name, + attrs, + generics, + kind, + span: lo.to(self.prev_span), + tokens: None, + }) + } + + fn parse_trait_item_const(&mut self) -> PResult<'a, (Ident, TraitItemKind, Generics)> { + self.expect_keyword(kw::Const)?; + let ident = self.parse_ident()?; + self.expect(&token::Colon)?; + let ty = self.parse_ty()?; + let default = if self.eat(&token::Eq) { + Some(self.parse_expr()?) + } else { + None + }; + self.expect_semi()?; + Ok((ident, TraitItemKind::Const(ty, default), Generics::default())) + } + + /// Parses the following grammar: + /// + /// TraitItemAssocTy = Ident ["<"...">"] [":" [GenericBounds]] ["where" ...] ["=" Ty] + fn parse_trait_item_assoc_ty(&mut self) -> PResult<'a, (Ident, TraitItemKind, Generics)> { + 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(None)? + } else { + Vec::new() + }; + generics.where_clause = self.parse_where_clause()?; + + let default = if self.eat(&token::Eq) { + Some(self.parse_ty()?) + } else { + None + }; + self.expect_semi()?; + + Ok((ident, TraitItemKind::Type(bounds, default), generics)) + } + + /// Parses a `UseTree`. + /// + /// ``` + /// 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: Vec::new(), span: lo.shrink_to_lo() }; + let kind = if self.check(&token::OpenDelim(token::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::ModSep) { + 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::ModSep) { + self.parse_use_tree_glob_or_nested()? + } else { + UseTreeKind::Simple(self.parse_rename()?, DUMMY_NODE_ID, DUMMY_NODE_ID) + } + }; + + Ok(UseTree { prefix, kind, span: lo.to(self.prev_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 { + UseTreeKind::Nested(self.parse_use_tree_list()?) + }) + } + + /// Parses a `UseTreeKind::Nested(list)`. + /// + /// ``` + /// USE_TREE_LIST = Ø | (USE_TREE `,`)* USE_TREE [`,`] + /// ``` + fn parse_use_tree_list(&mut self) -> PResult<'a, Vec<(UseTree, ast::NodeId)>> { + self.parse_delim_comma_seq(token::Brace, |p| 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, ast::Ident> { + match self.token.kind { + token::Ident(name, false) if name == kw::Underscore => { + let span = self.token.span; + self.bump(); + Ok(Ident::new(name, span)) + } + _ => self.parse_ident(), + } + } + + /// Parses `extern crate` links. + /// + /// # Examples + /// + /// ``` + /// extern crate foo; + /// extern crate bar as foo; + /// ``` + fn parse_item_extern_crate( + &mut self, + lo: Span, + visibility: Visibility, + attrs: Vec<Attribute> + ) -> PResult<'a, P<Item>> { + // 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()?; + + let span = lo.to(self.prev_span); + Ok(self.mk_item(span, item_name, ItemKind::ExternCrate(orig_name), visibility, attrs)) + } + + fn parse_crate_name_with_dashes(&mut self) -> PResult<'a, ast::Ident> { + let error_msg = "crate name using dashes are not valid in `extern crate` statements"; + let suggestion_msg = "if the original crate name uses dashes you need to use underscores \ + in the code"; + let mut ident = if self.token.is_keyword(kw::SelfLower) { + self.parse_path_segment_ident() + } else { + self.parse_ident() + }?; + let mut idents = vec![]; + let mut replacement = vec![]; + let mut fixed_crate_name = false; + // Accept `extern crate name-like-this` for better diagnostics. + let dash = token::BinOp(token::BinOpToken::Minus); + if self.token == dash { // Do not include `-` as part of the expected tokens list. + while self.eat(&dash) { + fixed_crate_name = true; + replacement.push((self.prev_span, "_".to_string())); + idents.push(self.parse_ident()?); + } + } + if fixed_crate_name { + let fixed_name_sp = ident.span.to(idents.last().unwrap().span); + let mut fixed_name = format!("{}", ident.name); + for part in idents { + fixed_name.push_str(&format!("_{}", part.name)); + } + ident = Ident::from_str_and_span(&fixed_name, fixed_name_sp); + + self.struct_span_err(fixed_name_sp, error_msg) + .span_label(fixed_name_sp, "dash-separated idents are not valid") + .multipart_suggestion(suggestion_msg, replacement, Applicability::MachineApplicable) + .emit(); + } + Ok(ident) + } + + /// 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, + lo: Span, + abi: Abi, + visibility: Visibility, + mut attrs: Vec<Attribute>, + extern_sp: Span, + ) -> PResult<'a, P<Item>> { + self.expect(&token::OpenDelim(token::Brace))?; + + attrs.extend(self.parse_inner_attributes()?); + + let mut foreign_items = vec![]; + while !self.eat(&token::CloseDelim(token::Brace)) { + foreign_items.push(self.parse_foreign_item(extern_sp)?); + } + + let prev_span = self.prev_span; + let m = ast::ForeignMod { + abi, + items: foreign_items + }; + let invalid = Ident::invalid(); + Ok(self.mk_item(lo.to(prev_span), invalid, ItemKind::ForeignMod(m), visibility, attrs)) + } + + /// Parses a foreign item. + pub fn parse_foreign_item(&mut self, extern_sp: Span) -> PResult<'a, ForeignItem> { + maybe_whole!(self, NtForeignItem, |ni| ni); + + let attrs = self.parse_outer_attributes()?; + let lo = self.token.span; + let visibility = self.parse_visibility(false)?; + + // FOREIGN STATIC ITEM + // Treat `const` as `static` for error recovery, but don't add it to expected tokens. + if self.check_keyword(kw::Static) || self.token.is_keyword(kw::Const) { + if self.token.is_keyword(kw::Const) { + let mut err = self + .struct_span_err(self.token.span, "extern items cannot be `const`"); + + + // The user wrote 'const fn' + if self.is_keyword_ahead(1, &[kw::Fn, kw::Unsafe]) { + err.emit(); + // Consume `const` + self.bump(); + // Consume `unsafe` if present, since `extern` blocks + // don't allow it. This will leave behind a plain 'fn' + self.eat_keyword(kw::Unsafe); + // Treat 'const fn` as a plain `fn` for error recovery purposes. + // We've already emitted an error, so compilation is guaranteed + // to fail + return Ok(self.parse_item_foreign_fn(visibility, lo, attrs, extern_sp)?); + } + err.span_suggestion( + self.token.span, + "try using a static value", + "static".to_owned(), + Applicability::MachineApplicable + ); + err.emit(); + } + self.bump(); // `static` or `const` + return Ok(self.parse_item_foreign_static(visibility, lo, attrs)?); + } + // FOREIGN FUNCTION ITEM + if self.check_keyword(kw::Fn) { + return Ok(self.parse_item_foreign_fn(visibility, lo, attrs, extern_sp)?); + } + // FOREIGN TYPE ITEM + if self.check_keyword(kw::Type) { + return Ok(self.parse_item_foreign_type(visibility, lo, attrs)?); + } + + match self.parse_assoc_macro_invoc("extern", Some(&visibility), &mut false)? { + Some(mac) => { + Ok( + ForeignItem { + ident: Ident::invalid(), + span: lo.to(self.prev_span), + id: DUMMY_NODE_ID, + attrs, + vis: visibility, + kind: ForeignItemKind::Macro(mac), + } + ) + } + None => { + if !attrs.is_empty() { + self.expected_item_err(&attrs)?; + } + + self.unexpected() + } + } + } + + /// Parses a static item from a foreign module. + /// Assumes that the `static` keyword is already parsed. + fn parse_item_foreign_static(&mut self, vis: ast::Visibility, lo: Span, attrs: Vec<Attribute>) + -> PResult<'a, ForeignItem> { + let mutbl = self.parse_mutability(); + let ident = self.parse_ident()?; + self.expect(&token::Colon)?; + let ty = self.parse_ty()?; + let hi = self.token.span; + self.expect_semi()?; + Ok(ForeignItem { + ident, + attrs, + kind: ForeignItemKind::Static(ty, mutbl), + id: DUMMY_NODE_ID, + span: lo.to(hi), + vis, + }) + } + + /// Parses a type from a foreign module. + fn parse_item_foreign_type(&mut self, vis: ast::Visibility, lo: Span, attrs: Vec<Attribute>) + -> PResult<'a, ForeignItem> { + self.expect_keyword(kw::Type)?; + + let ident = self.parse_ident()?; + let hi = self.token.span; + self.expect_semi()?; + Ok(ast::ForeignItem { + ident, + attrs, + kind: ForeignItemKind::Ty, + id: DUMMY_NODE_ID, + span: lo.to(hi), + vis + }) + } + + 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; + } + match token.kind { + token::BinOp(token::Or) | token::OrOr => true, + _ => false, + } + }) + } else { + false + } + } + + /// Parse `["const" | ("static" "mut"?)] $ident ":" $ty = $expr` with + /// `["const" | ("static" "mut"?)]` already parsed and stored in `m`. + /// + /// When `m` is `"const"`, `$ident` may also be `"_"`. + fn parse_item_const(&mut self, m: Option<Mutability>) -> PResult<'a, ItemInfo> { + let id = if m.is_none() { self.parse_ident_or_underscore() } else { self.parse_ident() }?; + + // Parse the type of a `const` or `static mut?` item. + // That is, the `":" $ty` fragment. + let ty = if self.token == token::Eq { + self.recover_missing_const_type(id, m) + } else { + // Not `=` so expect `":"" $ty` as usual. + self.expect(&token::Colon)?; + self.parse_ty()? + }; + + self.expect(&token::Eq)?; + let e = self.parse_expr()?; + self.expect_semi()?; + let item = match m { + Some(m) => ItemKind::Static(ty, m, e), + None => ItemKind::Const(ty, e), + }; + Ok((id, item, None)) + } + + /// We were supposed to parse `:` but instead, we're already at `=`. + /// This means that the type is missing. + fn recover_missing_const_type(&mut self, id: Ident, 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::Mutable) => "static mut", + Some(Mutability::Immutable) => "static", + None => "const", + }; + let mut err = self.struct_span_err(id.span, &format!("missing type for `{}` item", kind)); + err.span_suggestion( + id.span, + "provide a type for the item", + format!("{}: <type>", id), + Applicability::HasPlaceholders, + ); + err.stash(id.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.span, + id: ast::DUMMY_NODE_ID, + }) + } + + /// Parses `type Foo = Bar;` or returns `None` + /// without modifying the parser state. + fn eat_type(&mut self) -> Option<PResult<'a, (Ident, AliasKind, Generics)>> { + // This parses the grammar: + // Ident ["<"...">"] ["where" ...] ("=" | ":") Ty ";" + if self.eat_keyword(kw::Type) { + Some(self.parse_type_alias()) + } else { + None + } + } + + /// Parses a type alias or opaque type. + fn parse_type_alias(&mut self) -> PResult<'a, (Ident, AliasKind, Generics)> { + let ident = self.parse_ident()?; + let mut tps = self.parse_generics()?; + tps.where_clause = self.parse_where_clause()?; + self.expect(&token::Eq)?; + let alias = if self.check_keyword(kw::Impl) { + self.bump(); + let bounds = self.parse_generic_bounds(Some(self.prev_span))?; + AliasKind::OpaqueTy(bounds) + } else { + let ty = self.parse_ty()?; + AliasKind::Weak(ty) + }; + self.expect_semi()?; + Ok((ident, alias, tps)) + } + + /// Parses an enum declaration. + fn parse_item_enum(&mut self) -> PResult<'a, ItemInfo> { + let id = self.parse_ident()?; + let mut generics = self.parse_generics()?; + generics.where_clause = self.parse_where_clause()?; + self.expect(&token::OpenDelim(token::Brace))?; + + let enum_definition = self.parse_enum_def(&generics).map_err(|e| { + self.recover_stmt(); + self.eat(&token::CloseDelim(token::Brace)); + e + })?; + Ok((id, ItemKind::Enum(enum_definition, generics), None)) + } + + /// Parses the part of an enum declaration following the `{`. + fn parse_enum_def(&mut self, _generics: &Generics) -> PResult<'a, EnumDef> { + let mut variants = Vec::new(); + while self.token != token::CloseDelim(token::Brace) { + let variant_attrs = self.parse_outer_attributes()?; + let vlo = self.token.span; + + self.eat_bad_pub(); + let ident = self.parse_ident()?; + + let struct_def = if self.check(&token::OpenDelim(token::Brace)) { + // Parse a struct variant. + let (fields, recovered) = self.parse_record_struct_body()?; + VariantData::Struct(fields, recovered) + } else if self.check(&token::OpenDelim(token::Paren)) { + VariantData::Tuple( + self.parse_tuple_struct_body()?, + DUMMY_NODE_ID, + ) + } else { + VariantData::Unit(DUMMY_NODE_ID) + }; + + let disr_expr = if self.eat(&token::Eq) { + Some(AnonConst { + id: DUMMY_NODE_ID, + value: self.parse_expr()?, + }) + } else { + None + }; + + let vr = ast::Variant { + ident, + id: DUMMY_NODE_ID, + attrs: variant_attrs, + data: struct_def, + disr_expr, + span: vlo.to(self.prev_span), + is_placeholder: false, + }; + variants.push(vr); + + if !self.eat(&token::Comma) { + if self.token.is_ident() && !self.token.is_reserved_ident() { + let sp = self.sess.source_map().next_point(self.prev_span); + self.struct_span_err(sp, "missing comma") + .span_suggestion_short( + sp, + "missing comma", + ",".to_owned(), + Applicability::MaybeIncorrect, + ) + .emit(); + } else { + break; + } + } + } + self.expect(&token::CloseDelim(token::Brace))?; + + Ok(ast::EnumDef { variants }) + } + + /// 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) { + generics.where_clause = self.parse_where_clause()?; + 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()?; + VariantData::Struct(fields, recovered) + } + // 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(token::Brace) { + let (fields, recovered) = self.parse_record_struct_body()?; + VariantData::Struct(fields, recovered) + // Tuple-style struct definition with optional where-clause. + } else if self.token == token::OpenDelim(token::Paren) { + 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 token_str = self.this_token_descr(); + let mut err = self.fatal(&format!( + "expected `where`, `{{`, `(`, or `;` after struct name, found {}", + token_str + )); + err.span_label(self.token.span, "expected `where`, `{`, `(`, or `;` after struct name"); + return Err(err); + }; + + Ok((class_name, ItemKind::Struct(vdata, generics), None)) + } + + /// 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()?; + VariantData::Struct(fields, recovered) + } else if self.token == token::OpenDelim(token::Brace) { + let (fields, recovered) = self.parse_record_struct_body()?; + VariantData::Struct(fields, recovered) + } else { + let token_str = self.this_token_descr(); + let mut err = self.fatal(&format!( + "expected `where` or `{{` after union name, found {}", token_str)); + err.span_label(self.token.span, "expected `where` or `{` after union name"); + return Err(err); + }; + + Ok((class_name, ItemKind::Union(vdata, generics), None)) + } + + pub(super) fn is_union_item(&self) -> bool { + self.token.is_keyword(kw::Union) && + self.look_ahead(1, |t| t.is_ident() && !t.is_reserved_ident()) + } + + fn parse_record_struct_body( + &mut self, + ) -> PResult<'a, (Vec<StructField>, /* recovered */ bool)> { + let mut fields = Vec::new(); + let mut recovered = false; + if self.eat(&token::OpenDelim(token::Brace)) { + while self.token != token::CloseDelim(token::Brace) { + let field = self.parse_struct_decl_field().map_err(|e| { + self.consume_block(token::Brace, ConsumeClosingDelim::No); + recovered = true; + e + }); + match field { + Ok(field) => fields.push(field), + Err(mut err) => { + err.emit(); + break; + } + } + } + self.eat(&token::CloseDelim(token::Brace)); + } else { + let token_str = self.this_token_descr(); + let mut err = self.fatal(&format!( + "expected `where`, or `{{` after struct name, found {}", token_str)); + err.span_label(self.token.span, "expected `where`, or `{` after struct name"); + return Err(err); + } + + Ok((fields, recovered)) + } + + fn parse_tuple_struct_body(&mut self) -> PResult<'a, Vec<StructField>> { + // 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()?; + let lo = p.token.span; + let vis = p.parse_visibility(true)?; + let ty = p.parse_ty()?; + Ok(StructField { + span: lo.to(ty.span), + vis, + ident: None, + id: DUMMY_NODE_ID, + ty, + attrs, + is_placeholder: false, + }) + }).map(|(r, _)| r) + } + + /// Parses an element of a struct declaration. + fn parse_struct_decl_field(&mut self) -> PResult<'a, StructField> { + let attrs = self.parse_outer_attributes()?; + let lo = self.token.span; + let vis = self.parse_visibility(false)?; + self.parse_single_struct_field(lo, vis, attrs) + } + + /// Parses a structure field declaration. + fn parse_single_struct_field(&mut self, + lo: Span, + vis: Visibility, + attrs: Vec<Attribute> ) + -> PResult<'a, StructField> { + let mut seen_comma: bool = false; + let a_var = self.parse_name_and_ty(lo, vis, attrs)?; + if self.token == token::Comma { + seen_comma = true; + } + match self.token.kind { + token::Comma => { + self.bump(); + } + token::CloseDelim(token::Brace) => {} + token::DocComment(_) => { + let previous_span = self.prev_span; + let mut err = self.span_fatal_err(self.token.span, Error::UselessDocComment); + 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 == false && comma_after_doc_seen == true { + seen_comma = true; + } + if comma_after_doc_seen || self.token == token::CloseDelim(token::Brace) { + err.emit(); + } else { + if seen_comma == false { + let sp = self.sess.source_map().next_point(previous_span); + err.span_suggestion( + sp, + "missing comma here", + ",".into(), + Applicability::MachineApplicable + ); + } + return Err(err); + } + } + _ => { + let sp = self.sess.source_map().next_point(self.prev_span); + let mut err = self.struct_span_err(sp, &format!("expected `,`, or `}}`, found {}", + self.this_token_descr())); + if self.token.is_ident() { + // This is likely another field; emit the diagnostic and keep going + err.span_suggestion( + sp, + "try adding a comma", + ",".into(), + Applicability::MachineApplicable, + ); + err.emit(); + } else { + return Err(err) + } + } + } + Ok(a_var) + } + + /// Parses a structure field. + fn parse_name_and_ty( + &mut self, + lo: Span, + vis: Visibility, + attrs: Vec<Attribute> + ) -> PResult<'a, StructField> { + let name = self.parse_ident()?; + self.expect(&token::Colon)?; + let ty = self.parse_ty()?; + Ok(StructField { + span: lo.to(self.prev_span), + ident: Some(name), + vis, + id: DUMMY_NODE_ID, + ty, + attrs, + is_placeholder: false, + }) + } + + pub(super) fn eat_macro_def( + &mut self, + attrs: &[Attribute], + vis: &Visibility, + lo: Span + ) -> PResult<'a, Option<P<Item>>> { + let token_lo = self.token.span; + let (ident, def) = if self.eat_keyword(kw::Macro) { + let ident = self.parse_ident()?; + let tokens = if self.check(&token::OpenDelim(token::Brace)) { + match self.parse_token_tree() { + TokenTree::Delimited(_, _, tts) => tts, + _ => unreachable!(), + } + } else if self.check(&token::OpenDelim(token::Paren)) { + let args = self.parse_token_tree(); + let body = if self.check(&token::OpenDelim(token::Brace)) { + self.parse_token_tree() + } else { + self.unexpected()?; + unreachable!() + }; + TokenStream::new(vec![ + args.into(), + TokenTree::token(token::FatArrow, token_lo.to(self.prev_span)).into(), + body.into(), + ]) + } else { + self.unexpected()?; + unreachable!() + }; + + (ident, ast::MacroDef { tokens: tokens.into(), legacy: false }) + } else if self.check_keyword(sym::macro_rules) && + self.look_ahead(1, |t| *t == token::Not) && + self.look_ahead(2, |t| t.is_ident()) { + let prev_span = self.prev_span; + self.complain_if_pub_macro(&vis.node, prev_span); + self.bump(); + self.bump(); + + let ident = self.parse_ident()?; + let (delim, tokens) = self.expect_delimited_token_tree()?; + if delim != MacDelimiter::Brace && !self.eat(&token::Semi) { + self.report_invalid_macro_expansion_item(); + } + + (ident, ast::MacroDef { tokens, legacy: true }) + } else { + return Ok(None); + }; + + let span = lo.to(self.prev_span); + + if !def.legacy { + self.sess.gated_spans.gate(sym::decl_macro, span); + } + + Ok(Some(self.mk_item(span, ident, ItemKind::MacroDef(def), vis.clone(), attrs.to_vec()))) + } + + fn complain_if_pub_macro(&self, vis: &VisibilityKind, sp: Span) { + match *vis { + VisibilityKind::Inherited => {} + _ => { + let mut err = if self.token.is_keyword(sym::macro_rules) { + let mut err = self.diagnostic() + .struct_span_err(sp, "can't qualify macro_rules invocation with `pub`"); + err.span_suggestion( + sp, + "try exporting the macro", + "#[macro_export]".to_owned(), + Applicability::MaybeIncorrect // speculative + ); + err + } else { + let mut err = self.diagnostic() + .struct_span_err(sp, "can't qualify macro invocation with `pub`"); + err.help("try adjusting the macro to put `pub` inside the invocation"); + err + }; + err.emit(); + } + } + } + + fn report_invalid_macro_expansion_item(&self) { + self.struct_span_err( + self.prev_span, + "macros that expand to items must be delimited with braces or followed by a semicolon", + ).multipart_suggestion( + "change the delimiters to curly braces", + vec![ + (self.prev_span.with_hi(self.prev_span.lo() + BytePos(1)), String::from(" {")), + (self.prev_span.with_lo(self.prev_span.hi() - BytePos(1)), '}'.to_string()), + ], + Applicability::MaybeIncorrect, + ).span_suggestion( + self.sess.source_map().next_point(self.prev_span), + "add a semicolon", + ';'.to_string(), + Applicability::MaybeIncorrect, + ).emit(); + } + + fn mk_item(&self, span: Span, ident: Ident, kind: ItemKind, vis: Visibility, + attrs: Vec<Attribute>) -> P<Item> { + P(Item { + ident, + attrs, + id: DUMMY_NODE_ID, + kind, + vis, + span, + tokens: None, + }) + } +} + +/// The parsing configuration used to parse a parameter list (see `parse_fn_params`). +pub(super) struct ParamCfg { + /// Is `self` is allowed as the first parameter? + pub is_self_allowed: bool, + /// Is `...` allowed as the tail of the parameter list? + pub allow_c_variadic: bool, + /// `is_name_required` decides if, per-parameter, + /// the parameter must have a pattern or just a type. + pub is_name_required: fn(&token::Token) -> bool, +} + +/// Parsing of functions and methods. +impl<'a> Parser<'a> { + /// Parses an item-position function declaration. + fn parse_item_fn( + &mut self, + lo: Span, + vis: Visibility, + attrs: Vec<Attribute>, + header: FnHeader, + ) -> PResult<'a, Option<P<Item>>> { + let (ident, decl, generics) = self.parse_fn_sig(ParamCfg { + is_self_allowed: false, + allow_c_variadic: header.abi.symbol == sym::C && header.unsafety == Unsafety::Unsafe, + is_name_required: |_| true, + })?; + let (inner_attrs, body) = self.parse_inner_attrs_and_block()?; + let kind = ItemKind::Fn(FnSig { decl, header }, generics, body); + self.mk_item_with_info(attrs, lo, vis, (ident, kind, Some(inner_attrs))) + } + + /// Parses a function declaration from a foreign module. + fn parse_item_foreign_fn( + &mut self, + vis: ast::Visibility, + lo: Span, + attrs: Vec<Attribute>, + extern_sp: Span, + ) -> PResult<'a, ForeignItem> { + self.expect_keyword(kw::Fn)?; + let (ident, decl, generics) = self.parse_fn_sig(ParamCfg { + is_self_allowed: false, + allow_c_variadic: true, + is_name_required: |_| true, + })?; + let span = lo.to(self.token.span); + self.parse_semi_or_incorrect_foreign_fn_body(&ident, extern_sp)?; + Ok(ast::ForeignItem { + ident, + attrs, + kind: ForeignItemKind::Fn(decl, generics), + id: DUMMY_NODE_ID, + span, + vis, + }) + } + + /// Parses a method or a macro invocation in a trait impl. + fn parse_impl_method( + &mut self, + at_end: &mut bool, + ) -> PResult<'a, (Ident, Vec<Attribute>, Generics, ImplItemKind)> { + let (ident, sig, generics) = self.parse_method_sig(|_| true)?; + *at_end = true; + let (inner_attrs, body) = self.parse_inner_attrs_and_block()?; + Ok((ident, inner_attrs, generics, ast::ImplItemKind::Method(sig, body))) + } + + fn parse_trait_item_method( + &mut self, + at_end: &mut bool, + attrs: &mut Vec<Attribute>, + ) -> PResult<'a, (Ident, TraitItemKind, Generics)> { + // This is somewhat dubious; We don't want to allow + // argument names to be left off if there is a definition... + // + // We don't allow argument names to be left off in edition 2018. + let (ident, sig, generics) = self.parse_method_sig(|t| t.span.rust_2018())?; + let body = self.parse_trait_method_body(at_end, attrs)?; + Ok((ident, TraitItemKind::Method(sig, body), generics)) + } + + /// Parse the "body" of a method in a trait item definition. + /// This can either be `;` when there's no body, + /// or e.g. a block when the method is a provided one. + fn parse_trait_method_body( + &mut self, + at_end: &mut bool, + attrs: &mut Vec<Attribute>, + ) -> PResult<'a, Option<P<Block>>> { + Ok(match self.token.kind { + token::Semi => { + debug!("parse_trait_method_body(): parsing required method"); + self.bump(); + *at_end = true; + None + } + token::OpenDelim(token::Brace) => { + debug!("parse_trait_method_body(): parsing provided method"); + *at_end = true; + let (inner_attrs, body) = self.parse_inner_attrs_and_block()?; + attrs.extend(inner_attrs.iter().cloned()); + Some(body) + } + token::Interpolated(ref nt) => { + match **nt { + token::NtBlock(..) => { + *at_end = true; + let (inner_attrs, body) = self.parse_inner_attrs_and_block()?; + attrs.extend(inner_attrs.iter().cloned()); + Some(body) + } + _ => return self.expected_semi_or_open_brace(), + } + } + _ => return self.expected_semi_or_open_brace(), + }) + } + + /// Parse the "signature", including the identifier, parameters, and generics + /// of a method. The body is not parsed as that differs between `trait`s and `impl`s. + fn parse_method_sig( + &mut self, + is_name_required: fn(&token::Token) -> bool, + ) -> PResult<'a, (Ident, FnSig, Generics)> { + let header = self.parse_fn_front_matter()?; + let (ident, decl, generics) = self.parse_fn_sig(ParamCfg { + is_self_allowed: true, + allow_c_variadic: false, + is_name_required, + })?; + Ok((ident, FnSig { header, decl }, generics)) + } + + /// Parses all the "front matter" for a `fn` declaration, up to + /// and including the `fn` keyword: + /// + /// - `const fn` + /// - `unsafe fn` + /// - `const unsafe fn` + /// - `extern fn` + /// - etc. + fn parse_fn_front_matter(&mut self) -> PResult<'a, FnHeader> { + let is_const_fn = self.eat_keyword(kw::Const); + let const_span = self.prev_span; + let asyncness = self.parse_asyncness(); + if let IsAsync::Async { .. } = asyncness { + self.ban_async_in_2015(self.prev_span); + } + let asyncness = respan(self.prev_span, asyncness); + let unsafety = self.parse_unsafety(); + let (constness, unsafety, abi) = if is_const_fn { + (respan(const_span, Constness::Const), unsafety, Abi::default()) + } else { + let abi = self.parse_extern_abi()?; + (respan(self.prev_span, Constness::NotConst), unsafety, abi) + }; + if !self.eat_keyword(kw::Fn) { + // 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. + if !self.expect_one_of(&[], &[])? { unreachable!() } + } + Ok(FnHeader { constness, unsafety, asyncness, abi }) + } + + /// Parse the "signature", including the identifier, parameters, and generics of a function. + fn parse_fn_sig(&mut self, cfg: ParamCfg) -> PResult<'a, (Ident, P<FnDecl>, Generics)> { + let ident = self.parse_ident()?; + let mut generics = self.parse_generics()?; + let decl = self.parse_fn_decl(cfg, true)?; + generics.where_clause = self.parse_where_clause()?; + Ok((ident, decl, generics)) + } + + /// Parses the parameter list and result type of a function declaration. + pub(super) fn parse_fn_decl( + &mut self, + cfg: ParamCfg, + ret_allow_plus: bool, + ) -> PResult<'a, P<FnDecl>> { + Ok(P(FnDecl { + inputs: self.parse_fn_params(cfg)?, + output: self.parse_ret_ty(ret_allow_plus)?, + })) + } + + /// Parses the parameter list of a function, including the `(` and `)` delimiters. + fn parse_fn_params(&mut self, mut cfg: ParamCfg) -> PResult<'a, Vec<Param>> { + let sp = self.token.span; + let is_trait_item = cfg.is_self_allowed; + let mut c_variadic = false; + // Parse the arguments, starting out with `self` being possibly allowed... + let (params, _) = self.parse_paren_comma_seq(|p| { + let param = p.parse_param_general(&cfg, is_trait_item); + // ...now that we've parsed the first argument, `self` is no longer allowed. + cfg.is_self_allowed = false; + + match param { + Ok(param) => Ok( + if let TyKind::CVarArgs = param.ty.kind { + c_variadic = true; + if p.token != token::CloseDelim(token::Paren) { + p.span_err( + p.token.span, + "`...` must be the last argument of a C-variadic function", + ); + // FIXME(eddyb) this should probably still push `CVarArgs`. + // Maybe AST validation/HIR lowering should emit the above error? + None + } else { + Some(param) + } + } else { + Some(param) + } + ), + Err(mut e) => { + e.emit(); + let lo = p.prev_span; + // Skip every token until next possible arg or end. + p.eat_to_tokens(&[&token::Comma, &token::CloseDelim(token::Paren)]); + // Create a placeholder argument for proper arg count (issue #34264). + let span = lo.to(p.prev_span); + Ok(Some(dummy_arg(Ident::new(kw::Invalid, span)))) + } + } + })?; + + let mut params: Vec<_> = params.into_iter().filter_map(|x| x).collect(); + + // Replace duplicated recovered params with `_` pattern to avoid unecessary errors. + self.deduplicate_recovered_params_names(&mut params); + + if c_variadic && params.len() <= 1 { + self.span_err( + sp, + "C-variadic function must be declared with at least one named argument", + ); + } + + Ok(params) + } + + /// Skips unexpected attributes and doc comments in this position and emits an appropriate + /// error. + /// This version of parse param doesn't necessarily require identifier names. + fn parse_param_general(&mut self, cfg: &ParamCfg, is_trait_item: bool) -> PResult<'a, Param> { + let lo = self.token.span; + let attrs = self.parse_outer_attributes()?; + + // Possibly parse `self`. Recover if we parsed it and it wasn't allowed here. + if let Some(mut param) = self.parse_self_param()? { + param.attrs = attrs.into(); + return if cfg.is_self_allowed { + Ok(param) + } else { + self.recover_bad_self_param(param, is_trait_item) + }; + } + + let is_name_required = match self.token.kind { + token::DotDotDot => false, + _ => (cfg.is_name_required)(&self.token), + }; + let (pat, ty) = if is_name_required || self.is_named_param() { + debug!("parse_param_general parse_pat (is_name_required:{})", is_name_required); + + let pat = self.parse_fn_param_pat()?; + if let Err(mut err) = self.expect(&token::Colon) { + return if let Some(ident) = self.parameter_without_type( + &mut err, + pat, + is_name_required, + cfg.is_self_allowed, + is_trait_item, + ) { + err.emit(); + Ok(dummy_arg(ident)) + } else { + Err(err) + }; + } + + self.eat_incorrect_doc_comment_for_param_type(); + (pat, self.parse_ty_common(true, true, cfg.allow_c_variadic)?) + } else { + debug!("parse_param_general ident_to_pat"); + let parser_snapshot_before_ty = self.clone(); + self.eat_incorrect_doc_comment_for_param_type(); + let mut ty = self.parse_ty_common(true, true, cfg.allow_c_variadic); + if ty.is_ok() && self.token != token::Comma && + self.token != token::CloseDelim(token::Paren) { + // 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 = self.unexpected(); + } + match ty { + Ok(ty) => { + let ident = Ident::new(kw::Invalid, self.prev_span); + let bm = BindingMode::ByValue(Mutability::Immutable); + let pat = self.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 self.token == token::DotDotDot => return Err(err), + // Recover from attempting to parse the argument as a type without pattern. + Err(mut err) => { + err.cancel(); + mem::replace(self, parser_snapshot_before_ty); + self.recover_arg_parse()? + } + } + }; + + let span = lo.to(self.token.span); + + Ok(Param { + attrs: attrs.into(), + id: ast::DUMMY_NODE_ID, + is_placeholder: false, + pat, + span, + ty, + }) + } + + /// Returns the parsed optional self parameter and whether a self shortcut was used. + /// + /// See `parse_self_param_with_attrs` to collect attributes. + 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.kind { + // Preserve hygienic context. + token::Ident(name, _) => { + let span = this.token.span; + this.bump(); + Ident::new(name, span) + } + _ => 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::ModSep) + }; + // 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_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| { + let msg = "cannot pass `self` by raw pointer"; + let span = this.token.span; + this.struct_span_err(span, msg) + .span_label(span, msg) + .emit(); + + Ok((SelfKind::Value(Mutability::Immutable), expect_self_ident(this), this.prev_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.kind { + token::BinOp(token::And) => { + let eself = if is_isolated_self(self, 1) { + // `&self` + self.bump(); + SelfKind::Region(None, Mutability::Immutable) + } else if is_isolated_mut_self(self, 1) { + // `&mut self` + self.bump(); + self.bump(); + SelfKind::Region(None, Mutability::Mutable) + } 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::Immutable) + } 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::Mutable) + } else { + // `¬_self` + return Ok(None); + }; + (eself, expect_self_ident(self), self.prev_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::Immutable)? + } + // `mut self` and `mut self: TYPE` + token::Ident(..) if is_isolated_mut_self(self, 0) => { + self.bump(); + parse_self_possibly_typed(self, Mutability::Mutable)? + } + _ => return Ok(None), + }; + + let eself = source_map::respan(eself_lo.to(eself_hi), eself); + Ok(Some(Param::from_self(ThinVec::default(), eself, eself_ident))) + } + + fn is_named_param(&self) -> bool { + let offset = match self.token.kind { + token::Interpolated(ref 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_first_param(&mut self) -> &'static str { + match self.parse_outer_attributes() + .and_then(|_| self.parse_self_param()) + .map_err(|mut e| e.cancel()) + { + Ok(Some(_)) => "method", + _ => "function", + } + } +} diff --git a/src/librustc_parse/parser/mod.rs b/src/librustc_parse/parser/mod.rs new file mode 100644 index 00000000000..a491d91e20f --- /dev/null +++ b/src/librustc_parse/parser/mod.rs @@ -0,0 +1,1393 @@ +pub mod attr; +mod expr; +mod pat; +mod item; +mod module; +mod ty; +mod path; +pub use path::PathStyle; +mod stmt; +mod generics; +mod diagnostics; +use diagnostics::Error; + +use crate::{Directory, DirectoryOwnership}; +use crate::lexer::UnmatchedBrace; + +use syntax::ast::{ + self, Abi, DUMMY_NODE_ID, AttrStyle, Attribute, CrateSugar, Ident, + IsAsync, MacDelimiter, Mutability, StrStyle, Visibility, VisibilityKind, Unsafety, +}; + +use syntax::print::pprust; +use syntax::ptr::P; +use syntax::token::{self, Token, TokenKind, DelimToken}; +use syntax::tokenstream::{self, DelimSpan, TokenTree, TokenStream, TreeAndJoint}; +use syntax::sess::ParseSess; +use syntax::source_map::respan; +use syntax::struct_span_err; +use syntax::util::comments::{doc_comment_style, strip_doc_comment_decoration}; +use syntax_pos::symbol::{kw, sym, Symbol}; +use syntax_pos::{Span, BytePos, DUMMY_SP, FileName}; +use rustc_data_structures::thin_vec::ThinVec; +use errors::{PResult, Applicability, DiagnosticBuilder, DiagnosticId, FatalError}; +use log::debug; + +use std::borrow::Cow; +use std::{cmp, mem, slice}; +use std::path::PathBuf; + +bitflags::bitflags! { + struct Restrictions: u8 { + const STMT_EXPR = 1 << 0; + const NO_STRUCT_LITERAL = 1 << 1; + } +} + +#[derive(Clone, Copy, PartialEq, Debug)] +enum SemiColonMode { + Break, + Ignore, + Comma, +} + +#[derive(Clone, Copy, PartialEq, Debug)] +enum BlockMode { + Break, + Ignore, +} + +/// 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 { + if let token::$constructor(x) = &**nt { + let $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 && $self.look_ahead(1, |t| t == &token::ModSep) { + if let token::Interpolated(nt) = &$self.token.kind { + if let token::NtTy(ty) = &**nt { + let ty = ty.clone(); + $self.bump(); + return $self.maybe_recover_from_bad_qpath_stage_2($self.prev_span, ty); + } + } + } + } +} + +#[derive(Debug, Clone, Copy, PartialEq)] +enum PrevTokenKind { + DocComment, + Comma, + Plus, + Interpolated, + Eof, + Ident, + BitOr, + Other, +} + +// NOTE: `Ident`s are handled by `common.rs`. + +#[derive(Clone)] +pub struct Parser<'a> { + pub sess: &'a ParseSess, + /// The current normalized token. + /// "Normalized" means that some interpolated tokens + /// (`$i: ident` and `$l: lifetime` meta-variables) are replaced + /// with non-interpolated identifier and lifetime tokens they refer to. + /// Perhaps the normalized / non-normalized setup can be simplified somehow. + pub token: Token, + /// The span of the current non-normalized token. + meta_var_span: Option<Span>, + /// The span of the previous non-normalized token. + pub prev_span: Span, + /// The kind of the previous normalized token (in simplified form). + prev_token_kind: PrevTokenKind, + restrictions: Restrictions, + /// Used to determine the path to externally loaded source files. + pub(super) directory: Directory<'a>, + /// `true` to parse sub-modules in other files. + pub(super) recurse_into_file_modules: bool, + /// Name of the root module this parser originated from. If `None`, then the + /// name is not known. This does not change while the parser is descending + /// into modules, and sub-parsers have new values for this name. + pub root_module_name: Option<String>, + expected_tokens: Vec<TokenType>, + token_cursor: TokenCursor, + desugar_doc_comments: bool, + /// `true` we should configure out of line modules as we parse. + cfg_mods: 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: u32, + max_angle_bracket_count: u32, + /// A list of all unclosed delimiters found by the lexer. If an entry is used for error recovery + /// it gets removed from here. Every entry left at the end gets emitted as an independent + /// error. + pub(super) unclosed_delims: Vec<UnmatchedBrace>, + last_unexpected_token_span: Option<Span>, + pub last_type_ascription: Option<(Span, bool /* likely path typo */)>, + /// If present, this `Parser` is not parsing Rust code but rather a macro call. + subparser_name: Option<&'static str>, +} + +impl<'a> Drop for Parser<'a> { + fn drop(&mut self) { + emit_unclosed_delims(&mut self.unclosed_delims, &self.sess); + } +} + +#[derive(Clone)] +struct TokenCursor { + frame: TokenCursorFrame, + stack: Vec<TokenCursorFrame>, +} + +#[derive(Clone)] +struct TokenCursorFrame { + delim: token::DelimToken, + span: DelimSpan, + open_delim: bool, + tree_cursor: tokenstream::Cursor, + close_delim: bool, + last_token: LastToken, +} + +/// This is used in `TokenCursorFrame` above to track tokens that are consumed +/// by the parser, and then that's transitively used to record the tokens that +/// each parse AST item is created with. +/// +/// Right now this has two states, either collecting tokens or not collecting +/// tokens. If we're collecting tokens we just save everything off into a local +/// `Vec`. This should eventually though likely save tokens from the original +/// token stream and just use slicing of token streams to avoid creation of a +/// whole new vector. +/// +/// The second state is where we're passively not recording tokens, but the last +/// token is still tracked for when we want to start recording tokens. This +/// "last token" means that when we start recording tokens we'll want to ensure +/// that this, the first token, is included in the output. +/// +/// You can find some more example usage of this in the `collect_tokens` method +/// on the parser. +#[derive(Clone)] +enum LastToken { + Collecting(Vec<TreeAndJoint>), + Was(Option<TreeAndJoint>), +} + +impl TokenCursorFrame { + fn new(span: DelimSpan, delim: DelimToken, tts: &TokenStream) -> Self { + TokenCursorFrame { + delim, + span, + open_delim: delim == token::NoDelim, + tree_cursor: tts.clone().into_trees(), + close_delim: delim == token::NoDelim, + last_token: LastToken::Was(None), + } + } +} + +impl TokenCursor { + fn next(&mut self) -> Token { + loop { + let tree = if !self.frame.open_delim { + self.frame.open_delim = true; + TokenTree::open_tt(self.frame.span, self.frame.delim) + } else if let Some(tree) = self.frame.tree_cursor.next() { + tree + } else if !self.frame.close_delim { + self.frame.close_delim = true; + TokenTree::close_tt(self.frame.span, self.frame.delim) + } else if let Some(frame) = self.stack.pop() { + self.frame = frame; + continue + } else { + return Token::new(token::Eof, DUMMY_SP); + }; + + match self.frame.last_token { + LastToken::Collecting(ref mut v) => v.push(tree.clone().into()), + LastToken::Was(ref mut t) => *t = Some(tree.clone().into()), + } + + match tree { + TokenTree::Token(token) => return token, + TokenTree::Delimited(sp, delim, tts) => { + let frame = TokenCursorFrame::new(sp, delim, &tts); + self.stack.push(mem::replace(&mut self.frame, frame)); + } + } + } + } + + fn next_desugared(&mut self) -> Token { + let (name, sp) = match self.next() { + Token { kind: token::DocComment(name), span } => (name, span), + tok => return tok, + }; + + let stripped = strip_doc_comment_decoration(&name.as_str()); + + // Searches for the occurrences of `"#*` and returns the minimum number of `#`s + // required to wrap the text. + let mut num_of_hashes = 0; + let mut count = 0; + for ch in stripped.chars() { + count = match ch { + '"' => 1, + '#' if count > 0 => count + 1, + _ => 0, + }; + num_of_hashes = cmp::max(num_of_hashes, count); + } + + let delim_span = DelimSpan::from_single(sp); + let body = TokenTree::Delimited( + delim_span, + token::Bracket, + [ + TokenTree::token(token::Ident(sym::doc, false), sp), + TokenTree::token(token::Eq, sp), + TokenTree::token(TokenKind::lit( + token::StrRaw(num_of_hashes), Symbol::intern(&stripped), None + ), sp), + ] + .iter().cloned().collect::<TokenStream>().into(), + ); + + self.stack.push(mem::replace(&mut self.frame, TokenCursorFrame::new( + delim_span, + token::NoDelim, + &if doc_comment_style(&name.as_str()) == AttrStyle::Inner { + [TokenTree::token(token::Pound, sp), TokenTree::token(token::Not, sp), body] + .iter().cloned().collect::<TokenStream>() + } else { + [TokenTree::token(token::Pound, sp), body] + .iter().cloned().collect::<TokenStream>() + }, + ))); + + self.next() + } +} + +#[derive(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(ref 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 => "const".to_string(), + } + } +} + +#[derive(Copy, Clone, Debug)] +enum TokenExpectType { + Expect, + NoExpect, +} + +/// A sequence separator. +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, + } + } +} + +impl<'a> Parser<'a> { + pub fn new( + sess: &'a ParseSess, + tokens: TokenStream, + directory: Option<Directory<'a>>, + recurse_into_file_modules: bool, + desugar_doc_comments: bool, + subparser_name: Option<&'static str>, + ) -> Self { + let mut parser = Parser { + sess, + token: Token::dummy(), + prev_span: DUMMY_SP, + meta_var_span: None, + prev_token_kind: PrevTokenKind::Other, + restrictions: Restrictions::empty(), + recurse_into_file_modules, + directory: Directory { + path: Cow::from(PathBuf::new()), + ownership: DirectoryOwnership::Owned { relative: None } + }, + root_module_name: None, + expected_tokens: Vec::new(), + token_cursor: TokenCursor { + frame: TokenCursorFrame::new( + DelimSpan::dummy(), + token::NoDelim, + &tokens.into(), + ), + stack: Vec::new(), + }, + desugar_doc_comments, + cfg_mods: true, + unmatched_angle_bracket_count: 0, + max_angle_bracket_count: 0, + unclosed_delims: Vec::new(), + last_unexpected_token_span: None, + last_type_ascription: None, + subparser_name, + }; + + parser.token = parser.next_tok(); + + if let Some(directory) = directory { + parser.directory = directory; + } else if !parser.token.span.is_dummy() { + if let Some(FileName::Real(path)) = + &sess.source_map().lookup_char_pos(parser.token.span.lo()).file.unmapped_path { + if let Some(directory_path) = path.parent() { + parser.directory.path = Cow::from(directory_path.to_path_buf()); + } + } + } + + parser.process_potential_macro_variable(); + parser + } + + fn next_tok(&mut self) -> Token { + let mut next = if self.desugar_doc_comments { + self.token_cursor.next_desugared() + } else { + self.token_cursor.next() + }; + if next.span.is_dummy() { + // Tweak the location for better diagnostics, but keep syntactic context intact. + next.span = self.prev_span.with_ctxt(next.span.ctxt()); + } + next + } + + /// Converts the current token to a string using `self`'s reader. + pub fn this_token_to_string(&self) -> String { + pprust::token_to_string(&self.token) + } + + fn token_descr(&self) -> Option<&'static str> { + Some(match &self.token.kind { + _ if self.token.is_special_ident() => "reserved identifier", + _ if self.token.is_used_keyword() => "keyword", + _ if self.token.is_unused_keyword() => "reserved keyword", + token::DocComment(..) => "doc comment", + _ => return None, + }) + } + + pub(super) fn this_token_descr(&self) -> String { + if let Some(prefix) = self.token_descr() { + format!("{} `{}`", prefix, self.this_token_to_string()) + } else { + format!("`{}`", self.this_token_to_string()) + } + } + + crate fn unexpected<T>(&mut self) -> PResult<'a, T> { + match self.expect_one_of(&[], &[]) { + Err(e) => Err(e), + Ok(_) => unreachable!(), + } + } + + /// 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, bool /* recovered */> { + if self.expected_tokens.is_empty() { + if self.token == *t { + self.bump(); + Ok(false) + } 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, bool /* recovered */> { + if edible.contains(&self.token.kind) { + self.bump(); + Ok(false) + } else if inedible.contains(&self.token.kind) { + // leave it in the input + Ok(false) + } else if self.last_unexpected_token_span == Some(self.token.span) { + FatalError.raise(); + } else { + self.expected_one_of_not_found(edible, inedible) + } + } + + fn parse_ident(&mut self) -> PResult<'a, ast::Ident> { + self.parse_ident_common(true) + } + + fn parse_ident_common(&mut self, recover: bool) -> PResult<'a, ast::Ident> { + match self.token.kind { + token::Ident(name, _) => { + if self.token.is_reserved_ident() { + let mut err = self.expected_ident_found(); + if recover { + err.emit(); + } else { + return Err(err); + } + } + let span = self.token.span; + self.bump(); + Ok(Ident::new(name, span)) + } + _ => { + Err(if self.prev_token_kind == PrevTokenKind::DocComment { + self.span_fatal_err(self.prev_span, Error::UselessDocComment) + } else { + self.expected_ident_found() + }) + } + } + } + + /// 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. + 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 + } + + /// Consumes a token 'tok' if it exists. Returns whether the given token was present. + 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. + fn check_keyword(&mut self, kw: Symbol) -> bool { + self.expected_tokens.push(TokenType::Keyword(kw)); + self.token.is_keyword(kw) + } + + /// If the next token is the given keyword, eats it and returns `true`. + /// Otherwise, returns `false`. An expectation is also added for diagnostics purposes. + fn eat_keyword(&mut self, kw: Symbol) -> bool { + if self.check_keyword(kw) { + self.bump(); + true + } else { + false + } + } + + 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(()) + } + } + + 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) + } + + /// Checks to see if the next token is either `+` or `+=`. + /// Otherwise returns `false`. + fn check_plus(&mut self) -> bool { + self.check_or_expected( + self.token.is_like_plus(), + TokenType::Token(token::BinOp(token::Plus)), + ) + } + + /// Expects and consumes a `+`. if `+=` is seen, replaces it with a `=` + /// and continues. If a `+` is not seen, returns `false`. + /// + /// This is used when token-splitting `+=` into `+`. + /// See issue #47856 for an example of when this may occur. + fn eat_plus(&mut self) -> bool { + self.expected_tokens.push(TokenType::Token(token::BinOp(token::Plus))); + match self.token.kind { + token::BinOp(token::Plus) => { + self.bump(); + true + } + token::BinOpEq(token::Plus) => { + let span = self.token.span.with_lo(self.token.span.lo() + BytePos(1)); + self.bump_with(token::Eq, span); + true + } + _ => false, + } + } + + /// Expects and consumes an `&`. If `&&` is seen, replaces it with a single + /// `&` and continues. If an `&` is not seen, signals an error. + fn expect_and(&mut self) -> PResult<'a, ()> { + self.expected_tokens.push(TokenType::Token(token::BinOp(token::And))); + match self.token.kind { + token::BinOp(token::And) => { + self.bump(); + Ok(()) + } + token::AndAnd => { + let span = self.token.span.with_lo(self.token.span.lo() + BytePos(1)); + Ok(self.bump_with(token::BinOp(token::And), span)) + } + _ => self.unexpected() + } + } + + /// Expects and consumes an `|`. If `||` is seen, replaces it with a single + /// `|` and continues. If an `|` is not seen, signals an error. + fn expect_or(&mut self) -> PResult<'a, ()> { + self.expected_tokens.push(TokenType::Token(token::BinOp(token::Or))); + match self.token.kind { + token::BinOp(token::Or) => { + self.bump(); + Ok(()) + } + token::OrOr => { + let span = self.token.span.with_lo(self.token.span.lo() + BytePos(1)); + Ok(self.bump_with(token::BinOp(token::Or), span)) + } + _ => self.unexpected() + } + } + + /// Attempts to consume a `<`. If `<<` is seen, replaces it with a single + /// `<` and continue. If `<-` is seen, replaces it with a single `<` + /// and continue. If a `<` is not seen, returns false. + /// + /// This is meant to be used when parsing generics on a path to get the + /// starting token. + fn eat_lt(&mut self) -> bool { + self.expected_tokens.push(TokenType::Token(token::Lt)); + let ate = match self.token.kind { + token::Lt => { + self.bump(); + true + } + token::BinOp(token::Shl) => { + let span = self.token.span.with_lo(self.token.span.lo() + BytePos(1)); + self.bump_with(token::Lt, span); + true + } + token::LArrow => { + let span = self.token.span.with_lo(self.token.span.lo() + BytePos(1)); + self.bump_with(token::BinOp(token::Minus), span); + true + } + _ => false, + }; + + if ate { + // See doc comment for `unmatched_angle_bracket_count`. + self.unmatched_angle_bracket_count += 1; + self.max_angle_bracket_count += 1; + debug!("eat_lt: (increment) count={:?}", self.unmatched_angle_bracket_count); + } + + ate + } + + fn expect_lt(&mut self) -> PResult<'a, ()> { + if !self.eat_lt() { + self.unexpected() + } else { + Ok(()) + } + } + + /// Expects and consumes a single `>` token. if a `>>` is seen, replaces it + /// with a single `>` and continues. If a `>` is not seen, signals an error. + fn expect_gt(&mut self) -> PResult<'a, ()> { + self.expected_tokens.push(TokenType::Token(token::Gt)); + let ate = match self.token.kind { + token::Gt => { + self.bump(); + Some(()) + } + token::BinOp(token::Shr) => { + let span = self.token.span.with_lo(self.token.span.lo() + BytePos(1)); + Some(self.bump_with(token::Gt, span)) + } + token::BinOpEq(token::Shr) => { + let span = self.token.span.with_lo(self.token.span.lo() + BytePos(1)); + Some(self.bump_with(token::Ge, span)) + } + token::Ge => { + let span = self.token.span.with_lo(self.token.span.lo() + BytePos(1)); + Some(self.bump_with(token::Eq, span)) + } + _ => None, + }; + + match ate { + Some(_) => { + // 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(()) + }, + None => self.unexpected(), + } + } + + /// Parses a sequence, including 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, Vec<T>> { + let (val, _, recovered) = self.parse_seq_to_before_end(ket, sep, f)?; + if !recovered { + self.bump(); + } + Ok(val) + } + + /// Parses a sequence, not including the closing delimiter. 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, (Vec<T>, bool, bool)> { + self.parse_seq_to_before_tokens(&[ket], sep, TokenExpectType::Expect, f) + } + + 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.token == **k, + } + }) + } + + 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, (Vec<T>, bool /* trailing */, bool /* recovered */)> { + let mut first = true; + let mut recovered = false; + let mut trailing = false; + let mut v = vec![]; + while !self.expect_any_with_type(kets, expect) { + if let token::CloseDelim(..) | token::Eof = self.token.kind { + break + } + if let Some(ref t) = sep.sep { + if first { + first = false; + } else { + match self.expect(t) { + Ok(false) => {} + Ok(true) => { + recovered = true; + break; + } + Err(mut e) => { + // Attempt to keep parsing if it was a similar separator. + if let Some(ref tokens) = t.similar_tokens() { + if tokens.contains(&self.token.kind) { + self.bump(); + } + } + e.emit(); + // Attempt to keep parsing if it was an omitted separator. + match f(self) { + Ok(t) => { + v.push(t); + continue; + }, + Err(mut e) => { + e.cancel(); + break; + } + } + } + } + } + } + if sep.trailing_sep_allowed && self.expect_any_with_type(kets, expect) { + trailing = true; + break; + } + + let t = f(self)?; + v.push(t); + } + + Ok((v, trailing, recovered)) + } + + /// Parses a sequence, including the closing delimiter. 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, (Vec<T>, bool)> { + self.expect(bra)?; + let (result, trailing, recovered) = self.parse_seq_to_before_end(ket, sep, f)?; + if !recovered { + self.eat(ket); + } + Ok((result, trailing)) + } + + fn parse_delim_comma_seq<T>( + &mut self, + delim: DelimToken, + f: impl FnMut(&mut Parser<'a>) -> PResult<'a, T>, + ) -> PResult<'a, (Vec<T>, bool)> { + self.parse_unspanned_seq( + &token::OpenDelim(delim), + &token::CloseDelim(delim), + SeqSep::trailing_allowed(token::Comma), + f, + ) + } + + fn parse_paren_comma_seq<T>( + &mut self, + f: impl FnMut(&mut Parser<'a>) -> PResult<'a, T>, + ) -> PResult<'a, (Vec<T>, bool)> { + self.parse_delim_comma_seq(token::Paren, f) + } + + /// Advance the parser by one token. + pub fn bump(&mut self) { + if self.prev_token_kind == PrevTokenKind::Eof { + // Bumping after EOF is a bad sign, usually an infinite loop. + self.bug("attempted to bump the parser past EOF (may be stuck in a loop)"); + } + + self.prev_span = self.meta_var_span.take().unwrap_or(self.token.span); + + // Record last token kind for possible error recovery. + self.prev_token_kind = match self.token.kind { + token::DocComment(..) => PrevTokenKind::DocComment, + token::Comma => PrevTokenKind::Comma, + token::BinOp(token::Plus) => PrevTokenKind::Plus, + token::BinOp(token::Or) => PrevTokenKind::BitOr, + token::Interpolated(..) => PrevTokenKind::Interpolated, + token::Eof => PrevTokenKind::Eof, + token::Ident(..) => PrevTokenKind::Ident, + _ => PrevTokenKind::Other, + }; + + self.token = self.next_tok(); + self.expected_tokens.clear(); + // Check after each token. + self.process_potential_macro_variable(); + } + + /// Advances the parser using provided token as a next one. Use this when + /// consuming a part of a token. For example a single `<` from `<<`. + fn bump_with(&mut self, next: TokenKind, span: Span) { + self.prev_span = self.token.span.with_hi(span.lo()); + // It would be incorrect to record the kind of the current token, but + // fortunately for tokens currently using `bump_with`, the + // `prev_token_kind` will be of no use anyway. + self.prev_token_kind = PrevTokenKind::Other; + self.token = Token::new(next, span); + self.expected_tokens.clear(); + } + + /// Look-ahead `dist` tokens of `self.token` and get access to that token there. + /// When `dist == 0` then the current token is looked at. + pub fn look_ahead<R>(&self, dist: usize, looker: impl FnOnce(&Token) -> R) -> R { + if dist == 0 { + return looker(&self.token); + } + + let frame = &self.token_cursor.frame; + looker(&match frame.tree_cursor.look_ahead(dist - 1) { + Some(tree) => match tree { + TokenTree::Token(token) => token, + TokenTree::Delimited(dspan, delim, _) => + Token::new(token::OpenDelim(delim), dspan.open), + } + None => Token::new(token::CloseDelim(frame.delim), frame.span.close) + }) + } + + /// Returns whether any of the given keywords are `dist` tokens ahead of the current one. + 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_asyncness(&mut self) -> IsAsync { + if self.eat_keyword(kw::Async) { + IsAsync::Async { + closure_id: DUMMY_NODE_ID, + return_impl_trait_id: DUMMY_NODE_ID, + } + } else { + IsAsync::NotAsync + } + } + + /// Parses unsafety: `unsafe` or nothing. + fn parse_unsafety(&mut self) -> Unsafety { + if self.eat_keyword(kw::Unsafe) { + Unsafety::Unsafe + } else { + Unsafety::Normal + } + } + + /// Parses mutability (`mut` or nothing). + fn parse_mutability(&mut self) -> Mutability { + if self.eat_keyword(kw::Mut) { + Mutability::Mutable + } else { + Mutability::Immutable + } + } + + /// Possibly parses mutability (`const` or `mut`). + fn parse_const_or_mut(&mut self) -> Option<Mutability> { + if self.eat_keyword(kw::Mut) { + Some(Mutability::Mutable) + } else if self.eat_keyword(kw::Const) { + Some(Mutability::Immutable) + } 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 { + self.expect_no_suffix(self.token.span, "a tuple index", suffix); + self.bump(); + Ok(Ident::new(symbol, self.prev_span)) + } else { + self.parse_ident_common(false) + } + } + + fn expect_delimited_token_tree(&mut self) -> PResult<'a, (MacDelimiter, TokenStream)> { + let delim = match self.token.kind { + token::OpenDelim(delim) => delim, + _ => { + let msg = "expected open delimiter"; + let mut err = self.fatal(msg); + err.span_label(self.token.span, msg); + return Err(err) + } + }; + let tts = match self.parse_token_tree() { + TokenTree::Delimited(_, _, tts) => tts, + _ => unreachable!(), + }; + let delim = match delim { + token::Paren => MacDelimiter::Parenthesis, + token::Bracket => MacDelimiter::Bracket, + token::Brace => MacDelimiter::Brace, + token::NoDelim => self.bug("unexpected no delimiter"), + }; + Ok((delim, tts.into())) + } + + fn parse_or_use_outer_attributes( + &mut self, + already_parsed_attrs: Option<ThinVec<Attribute>>, + ) -> PResult<'a, ThinVec<Attribute>> { + if let Some(attrs) = already_parsed_attrs { + Ok(attrs) + } else { + self.parse_outer_attributes().map(|a| a.into()) + } + } + + pub fn process_potential_macro_variable(&mut self) { + self.token = match self.token.kind { + token::Dollar if self.token.span.from_expansion() && + self.look_ahead(1, |t| t.is_ident()) => { + self.bump(); + let name = match self.token.kind { + token::Ident(name, _) => name, + _ => unreachable!() + }; + let span = self.prev_span.to(self.token.span); + self.diagnostic() + .struct_span_fatal(span, &format!("unknown macro variable `{}`", name)) + .span_label(span, "unknown macro variable") + .emit(); + self.bump(); + return + } + token::Interpolated(ref nt) => { + self.meta_var_span = Some(self.token.span); + // Interpolated identifier and lifetime tokens are replaced with usual identifier + // and lifetime tokens, so the former are never encountered during normal parsing. + match **nt { + token::NtIdent(ident, is_raw) => + Token::new(token::Ident(ident.name, is_raw), ident.span), + token::NtLifetime(ident) => + Token::new(token::Lifetime(ident.name), ident.span), + _ => return, + } + } + _ => return, + }; + } + + /// Parses a single token tree from the input. + pub fn parse_token_tree(&mut self) -> TokenTree { + match self.token.kind { + token::OpenDelim(..) => { + let frame = mem::replace(&mut self.token_cursor.frame, + self.token_cursor.stack.pop().unwrap()); + self.token.span = frame.span.entire(); + self.bump(); + TokenTree::Delimited( + frame.span, + frame.delim, + frame.tree_cursor.stream.into(), + ) + }, + token::CloseDelim(_) | token::Eof => unreachable!(), + _ => { + let token = self.token.take(); + self.bump(); + TokenTree::Token(token) + } + } + } + + /// Parses a stream of tokens into a list of `TokenTree`s, up to EOF. + pub fn parse_all_token_trees(&mut self) -> PResult<'a, Vec<TokenTree>> { + let mut tts = Vec::new(); + while self.token != token::Eof { + tts.push(self.parse_token_tree()); + } + Ok(tts) + } + + 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().into()), + } + } + 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 + } + + fn is_crate_vis(&self) -> bool { + self.token.is_keyword(kw::Crate) && self.look_ahead(1, |t| t != &token::ModSep) + } + + /// Parses `pub`, `pub(crate)` and `pub(in path)` plus shortcuts `crate` for `pub(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 isn't valid, + /// so emit a proper diagnostic. + pub fn parse_visibility(&mut self, can_take_tuple: bool) -> PResult<'a, Visibility> { + maybe_whole!(self, NtVis, |x| x); + + self.expected_tokens.push(TokenType::Keyword(kw::Crate)); + if self.is_crate_vis() { + self.bump(); // `crate` + self.sess.gated_spans.gate(sym::crate_visibility_modifier, self.prev_span); + return Ok(respan(self.prev_span, VisibilityKind::Crate(CrateSugar::JustCrate))); + } + + 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(respan(self.token.span.shrink_to_lo(), VisibilityKind::Inherited)) + } + let lo = self.prev_span; + + if self.check(&token::OpenDelim(token::Paren)) { + // 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::Crate]) + && self.look_ahead(2, |t| t != &token::ModSep) // account for `pub(crate::foo)` + { + // Parse `pub(crate)`. + self.bump(); // `(` + self.bump(); // `crate` + self.expect(&token::CloseDelim(token::Paren))?; // `)` + let vis = VisibilityKind::Crate(CrateSugar::PubCrate); + return Ok(respan(lo.to(self.prev_span), vis)); + } else 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(token::Paren))?; // `)` + let vis = VisibilityKind::Restricted { + path: P(path), + id: ast::DUMMY_NODE_ID, + }; + return Ok(respan(lo.to(self.prev_span), vis)); + } else if self.look_ahead(2, |t| t == &token::CloseDelim(token::Paren)) + && self.is_keyword_ahead(1, &[kw::Super, kw::SelfLower]) + { + // Parse `pub(self)` or `pub(super)`. + self.bump(); // `(` + let path = self.parse_path(PathStyle::Mod)?; // `super`/`self` + self.expect(&token::CloseDelim(token::Paren))?; // `)` + let vis = VisibilityKind::Restricted { + path: P(path), + id: ast::DUMMY_NODE_ID, + }; + return Ok(respan(lo.to(self.prev_span), vis)); + } else if !can_take_tuple { // Provide this diagnostic if this is not a tuple struct. + self.recover_incorrect_vis_restriction()?; + // Emit diagnostic, but continue with public visibility. + } + } + + Ok(respan(lo, VisibilityKind::Public)) + } + + /// 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(token::Paren))?; // `)` + + let msg = "incorrect visibility restriction"; + let suggestion = r##"some possible visibility restrictions are: +`pub(crate)`: visible only on the current crate +`pub(super)`: visible only in the current module's parent +`pub(in path::to::module)`: visible only on the specified path"##; + + let path_str = pprust::path_to_string(&path); + + struct_span_err!(self.sess.span_diagnostic, path.span, E0704, "{}", msg) + .help(suggestion) + .span_suggestion( + path.span, + &format!("make this visible only to module `{}` with `in`", path_str), + format!("in {}", path_str), + Applicability::MachineApplicable, + ) + .emit(); + + Ok(()) + } + + /// Parses `extern string_literal?`. + /// If `extern` is not found, the Rust ABI is used. + /// If `extern` is found and a `string_literal` does not follow, the C ABI is used. + fn parse_extern_abi(&mut self) -> PResult<'a, Abi> { + Ok(if self.eat_keyword(kw::Extern) { + self.parse_opt_abi()? + } else { + Abi::default() + }) + } + + /// Parses a string literal as an ABI spec. + /// If one is not found, the "C" ABI is used. + fn parse_opt_abi(&mut self) -> PResult<'a, Abi> { + let span = if self.token.can_begin_literal_or_bool() { + let ast::Lit { span, kind, .. } = self.parse_lit()?; + match kind { + ast::LitKind::Str(symbol, _) => return Ok(Abi::new(symbol, span)), + ast::LitKind::Err(_) => {} + _ => { + self.struct_span_err(span, "non-string ABI literal") + .span_suggestion( + span, + "specify the ABI with a string literal", + "\"C\"".to_string(), + Applicability::MaybeIncorrect, + ) + .emit(); + } + } + span + } else { + self.prev_span + }; + Ok(Abi::new(sym::C, span)) + } + + /// We are parsing `async fn`. If we are on Rust 2015, emit an error. + fn ban_async_in_2015(&self, async_span: Span) { + if async_span.rust_2015() { + self.diagnostic() + .struct_span_err_with_code( + async_span, + "`async fn` is not permitted in the 2015 edition", + DiagnosticId::Error("E0670".into()) + ) + .emit(); + } + } + + fn collect_tokens<R>( + &mut self, + f: impl FnOnce(&mut Self) -> PResult<'a, R>, + ) -> PResult<'a, (R, TokenStream)> { + // Record all tokens we parse when parsing this item. + let mut tokens = Vec::new(); + let prev_collecting = match self.token_cursor.frame.last_token { + LastToken::Collecting(ref mut list) => { + Some(mem::take(list)) + } + LastToken::Was(ref mut last) => { + tokens.extend(last.take()); + None + } + }; + self.token_cursor.frame.last_token = LastToken::Collecting(tokens); + let prev = self.token_cursor.stack.len(); + let ret = f(self); + let last_token = if self.token_cursor.stack.len() == prev { + &mut self.token_cursor.frame.last_token + } else if self.token_cursor.stack.get(prev).is_none() { + // This can happen due to a bad interaction of two unrelated recovery mechanisms with + // mismatched delimiters *and* recovery lookahead on the likely typo `pub ident(` + // (#62881). + return Ok((ret?, TokenStream::default())); + } else { + &mut self.token_cursor.stack[prev].last_token + }; + + // Pull out the tokens that we've collected from the call to `f` above. + let mut collected_tokens = match *last_token { + LastToken::Collecting(ref mut v) => mem::take(v), + LastToken::Was(ref was) => { + let msg = format!("our vector went away? - found Was({:?})", was); + debug!("collect_tokens: {}", msg); + self.sess.span_diagnostic.delay_span_bug(self.token.span, &msg); + // This can happen due to a bad interaction of two unrelated recovery mechanisms + // with mismatched delimiters *and* recovery lookahead on the likely typo + // `pub ident(` (#62895, different but similar to the case above). + return Ok((ret?, TokenStream::default())); + } + }; + + // If we're not at EOF our current token wasn't actually consumed by + // `f`, but it'll still be in our list that we pulled out. In that case + // put it back. + let extra_token = if self.token != token::Eof { + collected_tokens.pop() + } else { + None + }; + + // If we were previously collecting tokens, then this was a recursive + // call. In that case we need to record all the tokens we collected in + // our parent list as well. To do that we push a clone of our stream + // onto the previous list. + match prev_collecting { + Some(mut list) => { + list.extend(collected_tokens.iter().cloned()); + list.extend(extra_token); + *last_token = LastToken::Collecting(list); + } + None => { + *last_token = LastToken::Was(extra_token); + } + } + + Ok((ret?, TokenStream::new(collected_tokens))) + } + + /// `::{` or `::*` + fn is_import_coupler(&mut self) -> bool { + self.check(&token::ModSep) && + self.look_ahead(1, |t| *t == token::OpenDelim(token::Brace) || + *t == token::BinOp(token::Star)) + } + + fn parse_optional_str(&mut self) -> Option<(Symbol, ast::StrStyle, Option<ast::Name>)> { + let ret = match self.token.kind { + token::Literal(token::Lit { kind: token::Str, symbol, suffix }) => + (symbol, ast::StrStyle::Cooked, suffix), + token::Literal(token::Lit { kind: token::StrRaw(n), symbol, suffix }) => + (symbol, ast::StrStyle::Raw(n), suffix), + _ => return None + }; + self.bump(); + Some(ret) + } + + pub fn parse_str(&mut self) -> PResult<'a, (Symbol, StrStyle)> { + match self.parse_optional_str() { + Some((s, style, suf)) => { + let sp = self.prev_span; + self.expect_no_suffix(sp, "a string literal", suf); + Ok((s, style)) + } + _ => { + let msg = "expected string literal"; + let mut err = self.fatal(msg); + err.span_label(self.token.span, msg); + Err(err) + } + } + } +} + +crate fn make_unclosed_delims_error( + unmatched: UnmatchedBrace, + sess: &ParseSess, +) -> Option<DiagnosticBuilder<'_>> { + // `None` here means an `Eof` was found. We already emit those errors elsewhere, we add them to + // `unmatched_braces` only for error recovery in the `Parser`. + let found_delim = unmatched.found_delim?; + let mut err = sess.span_diagnostic.struct_span_err(unmatched.found_span, &format!( + "incorrect close delimiter: `{}`", + pprust::token_kind_to_string(&token::CloseDelim(found_delim)), + )); + err.span_label(unmatched.found_span, "incorrect close delimiter"); + if let Some(sp) = unmatched.candidate_span { + err.span_label(sp, "close delimiter possibly meant for this"); + } + if let Some(sp) = unmatched.unclosed_span { + err.span_label(sp, "un-closed delimiter"); + } + Some(err) +} + +pub fn emit_unclosed_delims(unclosed_delims: &mut Vec<UnmatchedBrace>, sess: &ParseSess) { + *sess.reached_eof.borrow_mut() |= unclosed_delims.iter() + .any(|unmatched_delim| unmatched_delim.found_delim.is_none()); + for unmatched in unclosed_delims.drain(..) { + make_unclosed_delims_error(unmatched, sess).map(|mut e| e.emit()); + } +} diff --git a/src/librustc_parse/parser/module.rs b/src/librustc_parse/parser/module.rs new file mode 100644 index 00000000000..3110f15e80b --- /dev/null +++ b/src/librustc_parse/parser/module.rs @@ -0,0 +1,316 @@ +use super::Parser; +use super::item::ItemInfo; +use super::diagnostics::Error; + +use crate::{new_sub_parser_from_file, DirectoryOwnership}; + +use syntax::attr; +use syntax::ast::{self, Ident, Attribute, ItemKind, Mod, Crate}; +use syntax::token::{self, TokenKind}; +use syntax::source_map::{SourceMap, Span, DUMMY_SP, FileName}; + +use syntax_pos::symbol::sym; +use errors::PResult; + +use std::path::{self, Path, PathBuf}; + +/// Information about the path to a module. +pub(super) struct ModulePath { + name: String, + path_exists: bool, + pub result: Result<ModulePathSuccess, Error>, +} + +pub(super) struct ModulePathSuccess { + pub path: PathBuf, + pub directory_ownership: DirectoryOwnership, +} + +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, Crate> { + let lo = self.token.span; + let krate = Ok(ast::Crate { + attrs: self.parse_inner_attributes()?, + module: self.parse_mod_items(&token::Eof, lo)?, + span: lo.to(self.token.span), + }); + krate + } + + /// Parses a `mod <foo> { ... }` or `mod <foo>;` item. + pub(super) fn parse_item_mod(&mut self, outer_attrs: &[Attribute]) -> PResult<'a, ItemInfo> { + // HACK(Centril): See documentation on `ParseSess::process_cfg_mod`. + let (in_cfg, outer_attrs) = (self.sess.process_cfg_mod)( + self.sess, + self.cfg_mods, + outer_attrs, + ); + + let id_span = self.token.span; + let id = self.parse_ident()?; + if self.eat(&token::Semi) { + if in_cfg && self.recurse_into_file_modules { + // This mod is in an external file. Let's go get it! + let ModulePathSuccess { path, directory_ownership } = + self.submod_path(id, &outer_attrs, id_span)?; + let (module, attrs) = + self.eval_src_mod(path, directory_ownership, id.to_string(), id_span)?; + Ok((id, ItemKind::Mod(module), Some(attrs))) + } else { + let placeholder = ast::Mod { + inner: DUMMY_SP, + items: Vec::new(), + inline: false + }; + Ok((id, ItemKind::Mod(placeholder), None)) + } + } else { + let old_directory = self.directory.clone(); + self.push_directory(id, &outer_attrs); + + self.expect(&token::OpenDelim(token::Brace))?; + let mod_inner_lo = self.token.span; + let attrs = self.parse_inner_attributes()?; + let module = self.parse_mod_items(&token::CloseDelim(token::Brace), mod_inner_lo)?; + + self.directory = old_directory; + Ok((id, ItemKind::Mod(module), Some(attrs))) + } + } + + /// Given a termination token, parses all of the items in a module. + fn parse_mod_items(&mut self, term: &TokenKind, inner_lo: Span) -> PResult<'a, Mod> { + let mut items = vec![]; + while let Some(item) = self.parse_item()? { + items.push(item); + self.maybe_consume_incorrect_semicolon(&items); + } + + if !self.eat(term) { + let token_str = self.this_token_descr(); + if !self.maybe_consume_incorrect_semicolon(&items) { + let mut err = self.fatal(&format!("expected item, found {}", token_str)); + err.span_label(self.token.span, "expected item"); + return Err(err); + } + } + + let hi = if self.token.span.is_dummy() { + inner_lo + } else { + self.prev_span + }; + + Ok(Mod { + inner: inner_lo.to(hi), + items, + inline: true + }) + } + + fn submod_path( + &mut self, + id: ast::Ident, + outer_attrs: &[Attribute], + id_sp: Span + ) -> PResult<'a, ModulePathSuccess> { + if let Some(path) = Parser::submod_path_from_attr(outer_attrs, &self.directory.path) { + return Ok(ModulePathSuccess { + directory_ownership: match path.file_name().and_then(|s| s.to_str()) { + // All `#[path]` files are treated as though they are a `mod.rs` file. + // This means that `mod foo;` declarations inside `#[path]`-included + // files are siblings, + // + // Note that this will produce weirdness when a file named `foo.rs` is + // `#[path]` included and contains a `mod foo;` declaration. + // If you encounter this, it's your own darn fault :P + Some(_) => DirectoryOwnership::Owned { relative: None }, + _ => DirectoryOwnership::UnownedViaMod, + }, + path, + }); + } + + let relative = match self.directory.ownership { + DirectoryOwnership::Owned { relative } => relative, + DirectoryOwnership::UnownedViaBlock | + DirectoryOwnership::UnownedViaMod => None, + }; + let paths = Parser::default_submod_path( + id, relative, &self.directory.path, self.sess.source_map()); + + match self.directory.ownership { + DirectoryOwnership::Owned { .. } => { + paths.result.map_err(|err| self.span_fatal_err(id_sp, err)) + }, + DirectoryOwnership::UnownedViaBlock => { + let msg = + "Cannot declare a non-inline module inside a block \ + unless it has a path attribute"; + let mut err = self.diagnostic().struct_span_err(id_sp, msg); + if paths.path_exists { + let msg = format!("Maybe `use` the module `{}` instead of redeclaring it", + paths.name); + err.span_note(id_sp, &msg); + } + Err(err) + } + DirectoryOwnership::UnownedViaMod => { + let mut err = self.diagnostic().struct_span_err(id_sp, + "cannot declare a new module at this location"); + if !id_sp.is_dummy() { + let src_path = self.sess.source_map().span_to_filename(id_sp); + if let FileName::Real(src_path) = src_path { + if let Some(stem) = src_path.file_stem() { + let mut dest_path = src_path.clone(); + dest_path.set_file_name(stem); + dest_path.push("mod.rs"); + err.span_note(id_sp, + &format!("maybe move this module `{}` to its own \ + directory via `{}`", src_path.display(), + dest_path.display())); + } + } + } + if paths.path_exists { + err.span_note(id_sp, + &format!("... or maybe `use` the module `{}` instead \ + of possibly redeclaring it", + paths.name)); + } + Err(err) + } + } + } + + pub(super) fn submod_path_from_attr(attrs: &[Attribute], dir_path: &Path) -> Option<PathBuf> { + if let Some(s) = attr::first_attr_value_str_by_name(attrs, sym::path) { + let s = s.as_str(); + + // On windows, the base path might have the form + // `\\?\foo\bar` in which case it does not tolerate + // mixed `/` and `\` separators, so canonicalize + // `/` to `\`. + #[cfg(windows)] + let s = s.replace("/", "\\"); + Some(dir_path.join(&*s)) + } else { + None + } + } + + /// Returns a path to a module. + pub(super) fn default_submod_path( + id: ast::Ident, + relative: Option<ast::Ident>, + dir_path: &Path, + source_map: &SourceMap) -> ModulePath + { + // If we're in a foo.rs file instead of a mod.rs file, + // we need to look for submodules in + // `./foo/<id>.rs` and `./foo/<id>/mod.rs` rather than + // `./<id>.rs` and `./<id>/mod.rs`. + let relative_prefix_string; + let relative_prefix = if let Some(ident) = relative { + relative_prefix_string = format!("{}{}", ident, path::MAIN_SEPARATOR); + &relative_prefix_string + } else { + "" + }; + + let mod_name = id.to_string(); + let default_path_str = format!("{}{}.rs", relative_prefix, mod_name); + let secondary_path_str = format!("{}{}{}mod.rs", + relative_prefix, mod_name, path::MAIN_SEPARATOR); + let default_path = dir_path.join(&default_path_str); + let secondary_path = dir_path.join(&secondary_path_str); + let default_exists = source_map.file_exists(&default_path); + let secondary_exists = source_map.file_exists(&secondary_path); + + let result = match (default_exists, secondary_exists) { + (true, false) => Ok(ModulePathSuccess { + path: default_path, + directory_ownership: DirectoryOwnership::Owned { + relative: Some(id), + }, + }), + (false, true) => Ok(ModulePathSuccess { + path: secondary_path, + directory_ownership: DirectoryOwnership::Owned { + relative: None, + }, + }), + (false, false) => Err(Error::FileNotFoundForModule { + mod_name: mod_name.clone(), + default_path: default_path_str, + secondary_path: secondary_path_str, + dir_path: dir_path.display().to_string(), + }), + (true, true) => Err(Error::DuplicatePaths { + mod_name: mod_name.clone(), + default_path: default_path_str, + secondary_path: secondary_path_str, + }), + }; + + ModulePath { + name: mod_name, + path_exists: default_exists || secondary_exists, + result, + } + } + + /// Reads a module from a source file. + fn eval_src_mod( + &mut self, + path: PathBuf, + directory_ownership: DirectoryOwnership, + name: String, + id_sp: Span, + ) -> PResult<'a, (Mod, Vec<Attribute>)> { + let mut included_mod_stack = self.sess.included_mod_stack.borrow_mut(); + if let Some(i) = included_mod_stack.iter().position(|p| *p == path) { + let mut err = String::from("circular modules: "); + let len = included_mod_stack.len(); + for p in &included_mod_stack[i.. len] { + err.push_str(&p.to_string_lossy()); + err.push_str(" -> "); + } + err.push_str(&path.to_string_lossy()); + return Err(self.span_fatal(id_sp, &err[..])); + } + included_mod_stack.push(path.clone()); + drop(included_mod_stack); + + let mut p0 = + new_sub_parser_from_file(self.sess, &path, directory_ownership, Some(name), id_sp); + p0.cfg_mods = self.cfg_mods; + let mod_inner_lo = p0.token.span; + let mod_attrs = p0.parse_inner_attributes()?; + let mut m0 = p0.parse_mod_items(&token::Eof, mod_inner_lo)?; + m0.inline = false; + self.sess.included_mod_stack.borrow_mut().pop(); + Ok((m0, mod_attrs)) + } + + fn push_directory(&mut self, id: Ident, attrs: &[Attribute]) { + if let Some(path) = attr::first_attr_value_str_by_name(attrs, sym::path) { + self.directory.path.to_mut().push(&*path.as_str()); + self.directory.ownership = DirectoryOwnership::Owned { relative: None }; + } else { + // We have to push on the current module name in the case of relative + // paths in order to ensure that any additional module paths from inline + // `mod x { ... }` come after the relative extension. + // + // For example, a `mod z { ... }` inside `x/y.rs` should set the current + // directory path to `/x/y/z`, not `/x/z` with a relative offset of `y`. + if let DirectoryOwnership::Owned { relative } = &mut self.directory.ownership { + if let Some(ident) = relative.take() { // remove the relative offset + self.directory.path.to_mut().push(&*ident.as_str()); + } + } + self.directory.path.to_mut().push(&*id.as_str()); + } + } +} diff --git a/src/librustc_parse/parser/pat.rs b/src/librustc_parse/parser/pat.rs new file mode 100644 index 00000000000..b068a4f16a5 --- /dev/null +++ b/src/librustc_parse/parser/pat.rs @@ -0,0 +1,1015 @@ +use super::{Parser, PathStyle}; +use crate::{maybe_recover_from_interpolated_ty_qpath, maybe_whole}; +use syntax::ast::{self, Attribute, Pat, PatKind, FieldPat, RangeEnd, RangeSyntax, Mac}; +use syntax::ast::{BindingMode, Ident, Mutability, Path, QSelf, Expr, ExprKind}; +use syntax::mut_visit::{noop_visit_pat, noop_visit_mac, MutVisitor}; +use syntax::ptr::P; +use syntax::print::pprust; +use syntax::ThinVec; +use syntax::token; +use syntax::source_map::{respan, Span, Spanned}; +use syntax_pos::symbol::{kw, sym}; +use errors::{PResult, Applicability, DiagnosticBuilder}; + +type Expected = Option<&'static str>; + +/// `Expected` for function and lambda parameter patterns. +pub(super) const PARAM_EXPECTED: Expected = Some("parameter name"); + +const WHILE_PARSING_OR_MSG: &str = "while parsing this or-pattern starting here"; + +/// Whether or not an or-pattern should be gated when occurring in the current context. +#[derive(PartialEq)] +pub(super) enum GateOr { Yes, No } + +/// Whether or not to recover a `,` when parsing or-patterns. +#[derive(PartialEq, Copy, Clone)] +enum RecoverComma { Yes, No } + +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(&mut self, expected: Expected) -> PResult<'a, P<Pat>> { + self.parse_pat_with_range_pat(true, expected) + } + + /// Entry point to the main pattern parser. + /// Corresponds to `top_pat` in RFC 2535 and allows or-pattern at the top level. + pub(super) fn parse_top_pat(&mut self, gate_or: GateOr) -> PResult<'a, P<Pat>> { + // Allow a '|' before the pats (RFCs 1925, 2530, and 2535). + let gated_leading_vert = self.eat_or_separator(None) && gate_or == GateOr::Yes; + let leading_vert_span = self.prev_span; + + // Parse the possibly-or-pattern. + let pat = self.parse_pat_with_or(None, gate_or, RecoverComma::Yes)?; + + // If we parsed a leading `|` which should be gated, + // and no other gated or-pattern has been parsed thus far, + // then we should really gate the leading `|`. + // This complicated procedure is done purely for diagnostics UX. + if gated_leading_vert && self.sess.gated_spans.is_ungated(sym::or_patterns) { + self.sess.gated_spans.gate(sym::or_patterns, leading_vert_span); + } + + Ok(pat) + } + + /// Parse the pattern for a function or function pointer parameter. + /// Special recovery is provided for or-patterns and leading `|`. + pub(super) fn parse_fn_param_pat(&mut self) -> PResult<'a, P<Pat>> { + self.recover_leading_vert(None, "not allowed in a parameter pattern"); + let pat = self.parse_pat_with_or(PARAM_EXPECTED, GateOr::No, RecoverComma::No)?; + + if let PatKind::Or(..) = &pat.kind { + self.ban_illegal_fn_param_or_pat(&pat); + } + + Ok(pat) + } + + /// Ban `A | B` immediately in a parameter pattern and suggest wrapping in parens. + fn ban_illegal_fn_param_or_pat(&self, pat: &Pat) { + let msg = "wrap the pattern in parenthesis"; + let fix = format!("({})", pprust::pat_to_string(pat)); + self.struct_span_err(pat.span, "an or-pattern parameter must be wrapped in parenthesis") + .span_suggestion(pat.span, msg, fix, Applicability::MachineApplicable) + .emit(); + } + + /// Parses a pattern, that may be a or-pattern (e.g. `Foo | Bar` in `Some(Foo | Bar)`). + /// Corresponds to `pat<allow_top_alt>` in RFC 2535. + fn parse_pat_with_or( + &mut self, + expected: Expected, + gate_or: GateOr, + rc: RecoverComma, + ) -> PResult<'a, P<Pat>> { + // Parse the first pattern (`p_0`). + let first_pat = self.parse_pat(expected)?; + self.maybe_recover_unexpected_comma(first_pat.span, rc)?; + + // 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 { + return Ok(first_pat) + } + + // Parse the patterns `p_1 | ... | p_n` where `n > 0`. + let lo = first_pat.span; + let mut pats = vec![first_pat]; + while self.eat_or_separator(Some(lo)) { + let pat = self.parse_pat(expected).map_err(|mut err| { + err.span_label(lo, WHILE_PARSING_OR_MSG); + err + })?; + self.maybe_recover_unexpected_comma(pat.span, rc)?; + pats.push(pat); + } + let or_pattern_span = lo.to(self.prev_span); + + // Feature gate the or-pattern if instructed: + if gate_or == GateOr::Yes { + self.sess.gated_spans.gate(sym::or_patterns, or_pattern_span); + } + + Ok(self.mk_pat(or_pattern_span, PatKind::Or(pats))) + } + + /// 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>) -> bool { + if self.recover_trailing_vert(lo) { + return false; + } + + match self.token.kind { + token::OrOr => { + // Found `||`; Recover and pretend we parsed `|`. + self.ban_unexpected_or_or(lo); + self.bump(); + true + } + _ => self.eat(&token::BinOp(token::Or)), + } + } + + /// 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| match &token.kind { + token::FatArrow // e.g. `a | => 0,`. + | token::Ident(kw::If, false) // 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(token::Bracket) // e.g. `let [a | ]`. + | token::CloseDelim(token::Paren) // e.g. `let (a | )`. + | token::CloseDelim(token::Brace) => true, // e.g. `let A { f: a | }`. + _ => false, + }); + match (is_end_ahead, &self.token.kind) { + (true, token::BinOp(token::Or)) | (true, token::OrOr) => { + self.ban_illegal_vert(lo, "trailing", "not allowed in an or-pattern"); + self.bump(); + true + } + _ => false, + } + } + + /// We have parsed `||` instead of `|`. Error and suggest `|` instead. + fn ban_unexpected_or_or(&mut self, lo: Option<Span>) { + let mut err = self.struct_span_err(self.token.span, "unexpected token `||` after pattern"); + err.span_suggestion( + self.token.span, + "use a single `|` to separate multiple alternative patterns", + "|".to_owned(), + Applicability::MachineApplicable + ); + if let Some(lo) = lo { + err.span_label(lo, WHILE_PARSING_OR_MSG); + } + err.emit(); + } + + /// Some special error handling for the "top-level" patterns in a match arm, + /// `for` loop, `let`, &c. (in contrast to subpatterns within such). + fn maybe_recover_unexpected_comma(&mut self, lo: Span, rc: RecoverComma) -> PResult<'a, ()> { + if rc == RecoverComma::No || 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(mut 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_span); + let mut err = self.struct_span_err(comma_span, "unexpected `,` in pattern"); + if let Ok(seq_snippet) = self.span_to_snippet(seq_span) { + err.span_suggestion( + seq_span, + "try adding parentheses to match on a tuple..", + format!("({})", seq_snippet), + Applicability::MachineApplicable + ) + .span_suggestion( + seq_span, + "..or a vertical bar to match on multiple alternatives", + format!("{}", seq_snippet.replace(",", " |")), + Applicability::MachineApplicable + ); + } + Err(err) + } + + /// Parse and throw away a parentesized comma separated + /// sequence of patterns until `)` is reached. + fn skip_pat_list(&mut self) -> PResult<'a, ()> { + while !self.check(&token::CloseDelim(token::Paren)) { + self.parse_pat(None)?; + if !self.eat(&token::Comma) { + return Ok(()) + } + } + Ok(()) + } + + /// Recursive possibly-or-pattern parser with recovery for an erroneous leading `|`. + /// See `parse_pat_with_or` for details on parsing or-patterns. + fn parse_pat_with_or_inner(&mut self) -> PResult<'a, P<Pat>> { + self.recover_leading_vert(None, "only allowed in a top-level pattern"); + self.parse_pat_with_or(None, GateOr::Yes, RecoverComma::No) + } + + /// Recover if `|` or `||` is here. + /// The user is thinking that a leading `|` is allowed in this position. + fn recover_leading_vert(&mut self, lo: Option<Span>, ctx: &str) { + if let token::BinOp(token::Or) | token::OrOr = self.token.kind { + self.ban_illegal_vert(lo, "leading", ctx); + self.bump(); + } + } + + /// A `|` or possibly `||` token shouldn't be here. Ban it. + fn ban_illegal_vert(&mut self, lo: Option<Span>, pos: &str, ctx: &str) { + let span = self.token.span; + let mut err = self.struct_span_err(span, &format!("a {} `|` is {}", pos, ctx)); + err.span_suggestion( + span, + &format!("remove the `{}`", pprust::token_to_string(&self.token)), + String::new(), + Applicability::MachineApplicable, + ); + if let Some(lo) = lo { + err.span_label(lo, WHILE_PARSING_OR_MSG); + } + if let token::OrOr = self.token.kind { + err.note("alternatives in or-patterns are separated with `|`, not `||`"); + } + err.emit(); + } + + /// 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: Expected, + ) -> PResult<'a, P<Pat>> { + maybe_recover_from_interpolated_ty_qpath!(self, true); + maybe_whole!(self, NtPat, |x| x); + + let lo = self.token.span; + let pat = match self.token.kind { + token::BinOp(token::And) | token::AndAnd => self.parse_pat_deref(expected)?, + token::OpenDelim(token::Paren) => self.parse_pat_tuple_or_parens()?, + token::OpenDelim(token::Bracket) => { + // Parse `[pat, pat,...]` as a slice pattern. + let (pats, _) = self.parse_delim_comma_seq( + token::Bracket, + |p| p.parse_pat_with_or_inner(), + )?; + PatKind::Slice(pats) + } + token::DotDot => { + self.bump(); + if self.is_pat_range_end_start() { + // Parse `..42` for recovery. + self.parse_pat_range_to(RangeEnd::Excluded, "..")? + } else { + // A rest pattern `..`. + PatKind::Rest + } + } + token::DotDotEq => { + // Parse `..=42` for recovery. + self.bump(); + self.parse_pat_range_to(RangeEnd::Included(RangeSyntax::DotDotEq), "..=")? + } + token::DotDotDot => { + // Parse `...42` for recovery. + self.bump(); + self.parse_pat_range_to(RangeEnd::Included(RangeSyntax::DotDotDot), "...")? + } + // At this point, token != `&`, `&&`, `(`, `[`, `..`, `..=`, or `...`. + _ => 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) { + // Parse ref ident @ pat / ref mut ident @ pat + let mutbl = self.parse_mutability(); + self.parse_pat_ident(BindingMode::ByRef(mutbl))? + } else if self.eat_keyword(kw::Box) { + // Parse `box pat` + let pat = self.parse_pat_with_range_pat(false, None)?; + self.sess.gated_spans.gate(sym::box_patterns, lo.to(self.prev_span)); + PatKind::Box(pat) + } else if self.can_be_ident_pat() { + // Parse `ident @ pat` + // This can give false positives and parse nullary enums, + // they are dealt with later in resolve. + self.parse_pat_ident(BindingMode::ByValue(Mutability::Immutable))? + } 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::Expr)?; + (Some(qself), path) + } else { + // Parse an unqualified path + (None, self.parse_path(PathStyle::Expr)?) + }; + match self.token.kind { + token::Not if qself.is_none() => self.parse_pat_mac_invoc(lo, path)?, + token::DotDotDot | token::DotDotEq | token::DotDot => { + self.parse_pat_range_starting_with_path(lo, qself, path)? + } + token::OpenDelim(token::Brace) => self.parse_pat_struct(qself, path)?, + token::OpenDelim(token::Paren) => self.parse_pat_tuple_struct(qself, path)?, + _ => PatKind::Path(qself, path), + } + } else { + // Try to parse everything else as literal with optional minus + match self.parse_literal_maybe_minus() { + Ok(begin) + if self.check(&token::DotDot) + || self.check(&token::DotDotEq) + || self.check(&token::DotDotDot) => + { + self.parse_pat_range_starting_with_lit(begin)? + } + Ok(begin) => PatKind::Lit(begin), + Err(err) => return self.fatal_unexpected_non_pat(err, expected), + } + } + }; + + let pat = self.mk_pat(lo.to(self.prev_span), pat); + let pat = self.maybe_recover_from_bad_qpath(pat, true)?; + let pat = self.recover_intersection_pat(pat)?; + + if !allow_range_pat { + self.ban_pat_range_if_ambiguous(&pat)? + } + + Ok(pat) + } + + /// 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(None)?; + let sp = lhs.span.to(rhs.span); + + if let PatKind::Ident(_, _, ref mut sub @ None) = rhs.kind { + // The user inverted the order, so help them fix that. + let mut applicability = Applicability::MachineApplicable; + lhs.walk(&mut |p| match p.kind { + // `check_match` is unhappy if the subpattern has a binding anywhere. + PatKind::Ident(..) => { + applicability = Applicability::MaybeIncorrect; + false // Short-circuit. + }, + _ => true, + }); + + 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.struct_span_err(sp, "pattern on wrong side of `@`") + .span_label(lhs_span, "pattern on the left, should be on the right") + .span_label(rhs.span, "binding on the right, should be on the left") + .span_suggestion(sp, "switch the order", pprust::pat_to_string(&rhs), applicability) + .emit(); + } else { + // The special case above doesn't apply so we may have e.g. `A(x) @ B(y)`. + rhs.kind = PatKind::Wild; + self.struct_span_err(sp, "left-hand side of `@` must be a binding") + .span_label(lhs.span, "interpreted as a pattern, not a binding") + .span_label(rhs.span, "also a pattern") + .note("bindings are `x`, `mut x`, `ref x`, and `ref mut x`") + .emit(); + } + + rhs.span = sp; + Ok(rhs) + } + + /// Ban a range pattern if it has an ambiguous interpretation. + fn ban_pat_range_if_ambiguous(&self, pat: &Pat) -> PResult<'a, ()> { + match pat.kind { + PatKind::Range( + .., Spanned { node: RangeEnd::Included(RangeSyntax::DotDotDot), .. } + ) => return Ok(()), + PatKind::Range(..) => {} + _ => return Ok(()), + } + + let mut err = self.struct_span_err( + pat.span, + "the range pattern here has ambiguous interpretation", + ); + err.span_suggestion( + pat.span, + "add parentheses to clarify the precedence", + format!("({})", pprust::pat_to_string(&pat)), + // "ambiguous interpretation" implies that we have to be guessing + Applicability::MaybeIncorrect + ); + Err(err) + } + + /// Parse `&pat` / `&mut pat`. + fn parse_pat_deref(&mut self, expected: Expected) -> PResult<'a, PatKind> { + self.expect_and()?; + let mutbl = self.parse_mutability(); + + if let token::Lifetime(name) = self.token.kind { + let mut err = self.fatal(&format!("unexpected lifetime `{}` in pattern", name)); + err.span_label(self.token.span, "unexpected lifetime"); + return Err(err); + } + + let subpat = self.parse_pat_with_range_pat(false, expected)?; + Ok(PatKind::Ref(subpat, mutbl)) + } + + /// Parse a tuple or parenthesis pattern. + fn parse_pat_tuple_or_parens(&mut self) -> PResult<'a, PatKind> { + let (fields, trailing_comma) = self.parse_paren_comma_seq(|p| p.parse_pat_with_or_inner())?; + + // Here, `(pat,)` is a tuple pattern. + // For backward compatibility, `(..)` is a tuple pattern as well. + Ok(if fields.len() == 1 && !(trailing_comma || fields[0].is_rest()) { + PatKind::Paren(fields.into_iter().nth(0).unwrap()) + } else { + 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_span; + + if self.eat_keyword(kw::Ref) { + return self.recover_mut_ref_ident(mut_span) + } + + self.recover_additional_muts(); + + // Make sure we don't allow e.g. `let mut $p;` where `$p:pat`. + if let token::Interpolated(ref nt) = self.token.kind { + if let token::NtPat(_) = **nt { + self.expected_ident_found().emit(); + } + } + + // Parse the pattern we hope to be an identifier. + let mut pat = self.parse_pat(Some("identifier"))?; + + // Add `mut` to any binding in the parsed pattern. + let changed_any_binding = Self::make_all_value_bindings_mutable(&mut pat); + + // Unwrap; If we don't have `mut $ident`, error. + let pat = pat.into_inner(); + match &pat.kind { + PatKind::Ident(..) => {} + _ => self.ban_mut_general_pat(mut_span, &pat, changed_any_binding), + } + + Ok(pat.kind) + } + + /// Recover on `mut ref? ident @ pat` and suggest + /// that the order of `mut` and `ref` is incorrect. + fn recover_mut_ref_ident(&mut self, lo: Span) -> PResult<'a, PatKind> { + let mutref_span = lo.to(self.prev_span); + self.struct_span_err(mutref_span, "the order of `mut` and `ref` is incorrect") + .span_suggestion( + mutref_span, + "try switching the order", + "ref mut".into(), + Applicability::MachineApplicable + ) + .emit(); + + self.parse_pat_ident(BindingMode::ByRef(Mutability::Mutable)) + } + + /// 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_mac(&mut self, mac: &mut Mac) { + noop_visit_mac(mac, self); + } + + fn visit_pat(&mut self, pat: &mut P<Pat>) { + if let PatKind::Ident(BindingMode::ByValue(ref mut m @ Mutability::Immutable), ..) + = pat.kind + { + *m = Mutability::Mutable; + self.0 = true; + } + 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) { + let span = lo.to(pat.span); + let fix = pprust::pat_to_string(&pat); + let (problem, suggestion) = if changed_any_binding { + ("`mut` must be attached to each individual binding", "add `mut` to each binding") + } else { + ("`mut` must be followed by a named binding", "remove the `mut` prefix") + }; + self.struct_span_err(span, problem) + .span_suggestion(span, suggestion, fix, Applicability::MachineApplicable) + .note("`mut` may be followed by `variable` and `variable @ pattern`") + .emit() + } + + /// 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; + } + + let span = lo.to(self.prev_span); + self.struct_span_err(span, "`mut` on a binding may not be repeated") + .span_suggestion( + span, + "remove the additional `mut`s", + String::new(), + Applicability::MachineApplicable, + ) + .emit(); + } + + /// Parse macro invocation + fn parse_pat_mac_invoc(&mut self, lo: Span, path: Path) -> PResult<'a, PatKind> { + self.bump(); + let (delim, tts) = self.expect_delimited_token_tree()?; + let mac = Mac { + path, + tts, + delim, + span: lo.to(self.prev_span), + prior_type_ascription: self.last_type_ascription, + }; + Ok(PatKind::Mac(mac)) + } + + fn excluded_range_end(&self, span: Span) -> RangeEnd { + self.sess.gated_spans.gate(sym::exclusive_range_pattern, span); + RangeEnd::Excluded + } + + /// Parse a range pattern `$path $form $end?` where `$form = ".." | "..." | "..=" ;`. + /// The `$path` has already been parsed and the next token is the `$form`. + fn parse_pat_range_starting_with_path( + &mut self, + lo: Span, + qself: Option<QSelf>, + path: Path + ) -> PResult<'a, PatKind> { + let (end_kind, form) = match self.token.kind { + token::DotDot => (self.excluded_range_end(self.token.span), ".."), + token::DotDotDot => (RangeEnd::Included(RangeSyntax::DotDotDot), "..."), + token::DotDotEq => (RangeEnd::Included(RangeSyntax::DotDotEq), "..="), + _ => panic!("can only parse `..`/`...`/`..=` for ranges (checked above)"), + }; + let op_span = self.token.span; + // Parse range + let span = lo.to(self.prev_span); + let begin = self.mk_expr(span, ExprKind::Path(qself, path), ThinVec::new()); + self.bump(); + let end = self.parse_pat_range_end_opt(&begin, form)?; + Ok(PatKind::Range(begin, end, respan(op_span, end_kind))) + } + + /// Parse a range pattern `$literal $form $end?` where `$form = ".." | "..." | "..=" ;`. + /// The `$path` has already been parsed and the next token is the `$form`. + fn parse_pat_range_starting_with_lit(&mut self, begin: P<Expr>) -> PResult<'a, PatKind> { + let op_span = self.token.span; + let (end_kind, form) = 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) { + (self.excluded_range_end(op_span), "..") + } else { + panic!("impossible case: we already matched on a range-operator token") + }; + let end = self.parse_pat_range_end_opt(&begin, form)?; + Ok(PatKind::Range(begin, end, respan(op_span, end_kind))) + } + + fn fatal_unexpected_non_pat( + &mut self, + mut err: DiagnosticBuilder<'a>, + expected: Expected, + ) -> PResult<'a, P<Pat>> { + err.cancel(); + + let expected = expected.unwrap_or("pattern"); + let msg = format!("expected {}, found {}", expected, self.this_token_descr()); + + let mut err = self.fatal(&msg); + err.span_label(self.token.span, format!("expected {}", expected)); + + let sp = self.sess.source_map().start_point(self.token.span); + if let Some(sp) = self.sess.ambiguous_block_expr_parse.borrow().get(&sp) { + self.sess.expr_parentheses_needed(&mut err, *sp, None); + } + + Err(err) + } + + /// Is the current token suitable as the start of a range patterns end? + fn is_pat_range_end_start(&self) -> bool { + self.token.is_path_start() // e.g. `MY_CONST`; + || self.token == token::Dot // e.g. `.5` for recovery; + || self.token.can_begin_literal_or_bool() // e.g. `42`. + || self.token.is_whole_expr() + } + + /// Parse a range-to pattern, e.g. `..X` and `..=X` for recovery. + fn parse_pat_range_to(&mut self, re: RangeEnd, form: &str) -> PResult<'a, PatKind> { + let lo = self.prev_span; + let end = self.parse_pat_range_end()?; + let range_span = lo.to(end.span); + let begin = self.mk_expr(range_span, ExprKind::Err, ThinVec::new()); + + self.diagnostic() + .struct_span_err(range_span, &format!("`{}X` range patterns are not supported", form)) + .span_suggestion( + range_span, + "try using the minimum value for the type", + format!("MIN{}{}", form, pprust::expr_to_string(&end)), + Applicability::HasPlaceholders, + ) + .emit(); + + Ok(PatKind::Range(begin, end, respan(lo, re))) + } + + /// Parse the end of a `X..Y`, `X..=Y`, or `X...Y` range pattern or recover + /// if that end is missing treating it as `X..`, `X..=`, or `X...` respectively. + fn parse_pat_range_end_opt(&mut self, begin: &Expr, form: &str) -> PResult<'a, P<Expr>> { + if self.is_pat_range_end_start() { + // Parsing e.g. `X..=Y`. + self.parse_pat_range_end() + } else { + // Parsing e.g. `X..`. + let range_span = begin.span.to(self.prev_span); + + self.diagnostic() + .struct_span_err( + range_span, + &format!("`X{}` range patterns are not supported", form), + ) + .span_suggestion( + range_span, + "try using the maximum value for the type", + format!("{}{}MAX", pprust::expr_to_string(&begin), form), + Applicability::HasPlaceholders, + ) + .emit(); + + Ok(self.mk_expr(range_span, ExprKind::Err, ThinVec::new())) + } + } + + fn parse_pat_range_end(&mut self) -> PResult<'a, P<Expr>> { + if self.token.is_path_start() { + let lo = self.token.span; + let (qself, path) = if self.eat_lt() { + // Parse a qualified path + let (qself, path) = self.parse_qpath(PathStyle::Expr)?; + (Some(qself), path) + } else { + // Parse an unqualified path + (None, self.parse_path(PathStyle::Expr)?) + }; + let hi = self.prev_span; + Ok(self.mk_expr(lo.to(hi), ExprKind::Path(qself, path), ThinVec::new())) + } else { + self.parse_literal_maybe_minus() + } + } + + /// 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) + && self.look_ahead(1, |t| match t.kind { // Try to do something more complex? + token::OpenDelim(token::Paren) // A tuple struct pattern. + | token::OpenDelim(token::Brace) // A struct pattern. + | token::DotDotDot | token::DotDotEq | token::DotDot // A range pattern. + | token::ModSep // A tuple / struct variant pattern. + | token::Not => false, // A macro expanding to a pattern. + _ => true, + }) + } + + /// 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_mode: BindingMode) -> PResult<'a, PatKind> { + let ident = self.parse_ident()?; + let sub = if self.eat(&token::At) { + Some(self.parse_pat(Some("binding pattern"))?) + } 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(token::Paren) { + return Err(self.span_fatal( + self.prev_span, + "expected identifier, found enum pattern", + )) + } + + Ok(PatKind::Ident(binding_mode, ident, sub)) + } + + /// Parse a struct ("record") pattern (e.g. `Foo { ... }` or `Foo::Bar { ... }`). + fn parse_pat_struct(&mut self, qself: Option<QSelf>, path: Path) -> PResult<'a, PatKind> { + if qself.is_some() { + let msg = "unexpected `{` after qualified path"; + let mut err = self.fatal(msg); + err.span_label(self.token.span, msg); + return Err(err); + } + + self.bump(); + let (fields, etc) = self.parse_pat_fields().unwrap_or_else(|mut e| { + e.emit(); + self.recover_stmt(); + (vec![], true) + }); + self.bump(); + Ok(PatKind::Struct(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<QSelf>, path: Path) -> PResult<'a, PatKind> { + if qself.is_some() { + let msg = "unexpected `(` after qualified path"; + let mut err = self.fatal(msg); + err.span_label(self.token.span, msg); + return Err(err); + } + let (fields, _) = self.parse_paren_comma_seq(|p| p.parse_pat_with_or_inner())?; + Ok(PatKind::TupleStruct(path, fields)) + } + + /// Parses the fields of a struct-like pattern. + fn parse_pat_fields(&mut self) -> PResult<'a, (Vec<FieldPat>, bool)> { + let mut fields = Vec::new(); + let mut etc = false; + let mut ate_comma = true; + let mut delayed_err: Option<DiagnosticBuilder<'a>> = None; + let mut etc_span = None; + + while self.token != token::CloseDelim(token::Brace) { + let attrs = match self.parse_outer_attributes() { + Ok(attrs) => attrs, + Err(err) => { + if let Some(mut 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 err = self.struct_span_err(self.prev_span, "expected `,`"); + if let Some(mut delayed) = delayed_err { + delayed.emit(); + } + return Err(err); + } + ate_comma = false; + + if self.check(&token::DotDot) || self.token == token::DotDotDot { + etc = true; + let mut etc_sp = self.token.span; + + self.recover_one_fewer_dotdot(); + self.bump(); // `..` || `...` + + if self.token == token::CloseDelim(token::Brace) { + etc_span = Some(etc_sp); + break; + } + let token_str = self.this_token_descr(); + let mut err = self.fatal(&format!("expected `}}`, found {}", token_str)); + + err.span_label(self.token.span, "expected `}`"); + let mut comma_sp = None; + if self.token == token::Comma { // Issue #49257 + let nw_span = self.sess.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; + } + + etc_span = Some(etc_sp.until(self.token.span)); + if self.token == token::CloseDelim(token::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", + String::new(), + 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(mut delayed_err) = delayed_err { + delayed_err.emit(); + return Err(err); + } else { + delayed_err = Some(err); + } + } else { + if let Some(mut err) = delayed_err { + err.emit(); + } + return Err(err); + } + } + + fields.push(match self.parse_pat_field(lo, attrs) { + Ok(field) => field, + Err(err) => { + if let Some(mut delayed_err) = delayed_err { + delayed_err.emit(); + } + return Err(err); + } + }); + ate_comma = self.eat(&token::Comma); + } + + if let Some(mut err) = delayed_err { + if let Some(etc_span) = etc_span { + err.multipart_suggestion( + "move the `..` to the end of the field list", + vec![ + (etc_span, String::new()), + (self.token.span, format!("{}.. }}", if ate_comma { "" } else { ", " })), + ], + Applicability::MachineApplicable, + ); + } + err.emit(); + } + return Ok((fields, etc)); + } + + /// Recover on `...` as if it were `..` to avoid further errors. + /// See issue #46718. + fn recover_one_fewer_dotdot(&self) { + if self.token != token::DotDotDot { + return; + } + + self.struct_span_err(self.token.span, "expected field pattern, found `...`") + .span_suggestion( + self.token.span, + "to omit remaining fields, use one fewer `.`", + "..".to_owned(), + Applicability::MachineApplicable + ) + .emit(); + } + + fn parse_pat_field(&mut self, lo: Span, attrs: Vec<Attribute>) -> PResult<'a, FieldPat> { + // 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_with_or_inner()?; + 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 is_ref = self.eat_keyword(kw::Ref); + let is_mut = self.eat_keyword(kw::Mut); + let fieldname = self.parse_ident()?; + hi = self.prev_span; + + let bind_type = match (is_ref, is_mut) { + (true, true) => BindingMode::ByRef(Mutability::Mutable), + (true, false) => BindingMode::ByRef(Mutability::Immutable), + (false, true) => BindingMode::ByValue(Mutability::Mutable), + (false, false) => BindingMode::ByValue(Mutability::Immutable), + }; + + let fieldpat = self.mk_pat_ident(boxed_span.to(hi), bind_type, fieldname); + let subpat = if is_box { + self.mk_pat(lo.to(hi), PatKind::Box(fieldpat)) + } else { + fieldpat + }; + (subpat, fieldname, true) + }; + + Ok(FieldPat { + ident: fieldname, + pat: subpat, + is_shorthand, + attrs: attrs.into(), + id: ast::DUMMY_NODE_ID, + span: lo.to(hi), + is_placeholder: false, + }) + } + + pub(super) fn mk_pat_ident(&self, span: Span, bm: BindingMode, ident: Ident) -> P<Pat> { + self.mk_pat(span, PatKind::Ident(bm, ident, None)) + } + + fn mk_pat(&self, span: Span, kind: PatKind) -> P<Pat> { + P(Pat { kind, span, id: ast::DUMMY_NODE_ID }) + } +} diff --git a/src/librustc_parse/parser/path.rs b/src/librustc_parse/parser/path.rs new file mode 100644 index 00000000000..68307440712 --- /dev/null +++ b/src/librustc_parse/parser/path.rs @@ -0,0 +1,497 @@ +use super::{Parser, TokenType}; +use crate::maybe_whole; +use syntax::ast::{self, QSelf, Path, PathSegment, Ident, ParenthesizedArgs, AngleBracketedArgs}; +use syntax::ast::{AnonConst, GenericArg, AssocTyConstraint, AssocTyConstraintKind, BlockCheckMode}; +use syntax::ThinVec; +use syntax::token::{self, Token}; +use syntax::source_map::{Span, BytePos}; +use syntax_pos::symbol::{kw, sym}; + +use std::mem; +use log::debug; +use errors::{PResult, Applicability, pluralize}; + +/// 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. + Expr, + /// 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<'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, (QSelf, Path)> { + let lo = self.prev_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_span); + } else { + path_span = self.token.span.to(self.token.span); + path = ast::Path { segments: Vec::new(), span: path_span }; + } + + // 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); + } + + self.expect(&token::ModSep)?; + + let qself = QSelf { ty, path_span, position: path.segments.len() }; + self.parse_path_segments(&mut path.segments, style)?; + + Ok((qself, Path { segments: path.segments, span: lo.to(self.prev_span) })) + } + + /// 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 fn parse_path(&mut self, style: PathStyle) -> PResult<'a, Path> { + maybe_whole!(self, NtPath, |path| { + if style == PathStyle::Mod && + path.segments.iter().any(|segment| segment.args.is_some()) { + self.diagnostic().span_err(path.span, "unexpected generic arguments in path"); + } + path + }); + + let lo = self.meta_var_span.unwrap_or(self.token.span); + let mut segments = Vec::new(); + let mod_sep_ctxt = self.token.span.ctxt(); + if self.eat(&token::ModSep) { + segments.push(PathSegment::path_root(lo.shrink_to_lo().with_ctxt(mod_sep_ctxt))); + } + self.parse_path_segments(&mut segments, style)?; + + Ok(Path { segments, span: lo.to(self.prev_span) }) + } + + /// Like `parse_path`, but also supports parsing `Word` meta items into paths for + /// backwards-compatibility. This is used when parsing derive macro paths in `#[derive]` + /// attributes. + fn parse_path_allowing_meta(&mut self, style: PathStyle) -> PResult<'a, Path> { + let meta_ident = match self.token.kind { + token::Interpolated(ref nt) => match **nt { + token::NtMeta(ref item) => match item.tokens.is_empty() { + true => Some(item.path.clone()), + false => None, + }, + _ => None, + }, + _ => None, + }; + if let Some(path) = meta_ident { + self.bump(); + return Ok(path); + } + self.parse_path(style) + } + + /// Parse a list of paths inside `#[derive(path_0, ..., path_n)]`. + pub fn parse_derive_paths(&mut self) -> PResult<'a, Vec<Path>> { + self.expect(&token::OpenDelim(token::Paren))?; + let mut list = Vec::new(); + while !self.eat(&token::CloseDelim(token::Paren)) { + let path = self.parse_path_allowing_meta(PathStyle::Mod)?; + list.push(path); + if !self.eat(&token::Comma) { + self.expect(&token::CloseDelim(token::Paren))?; + break + } + } + Ok(list) + } + + pub(super) fn parse_path_segments( + &mut self, + segments: &mut Vec<PathSegment>, + style: PathStyle, + ) -> PResult<'a, ()> { + loop { + let segment = self.parse_path_segment(style)?; + if style == PathStyle::Expr { + // 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::ModSep); + } + segments.push(segment); + + if self.is_import_coupler() || !self.eat(&token::ModSep) { + return Ok(()); + } + } + } + + pub(super) fn parse_path_segment(&mut self, style: PathStyle) -> PResult<'a, PathSegment> { + let ident = self.parse_path_segment_ident()?; + + let is_args_start = |token: &Token| match token.kind { + token::Lt | token::BinOp(token::Shl) | token::OpenDelim(token::Paren) + | token::LArrow => true, + _ => false, + }; + let check_args_start = |this: &mut Self| { + this.expected_tokens.extend_from_slice( + &[TokenType::Token(token::Lt), TokenType::Token(token::OpenDelim(token::Paren))] + ); + is_args_start(&this.token) + }; + + Ok(if style == PathStyle::Type && check_args_start(self) || + style != PathStyle::Mod && self.check(&token::ModSep) + && 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; + self.max_angle_bracket_count = 0; + } + + // Generic arguments are found - `<`, `(`, `::<` or `::(`. + self.eat(&token::ModSep); + let lo = self.token.span; + let args = if self.eat_lt() { + // `<'a, T, A = U>` + let (args, constraints) = + self.parse_generic_args_with_leaning_angle_bracket_recovery(style, lo)?; + self.expect_gt()?; + let span = lo.to(self.prev_span); + AngleBracketedArgs { args, constraints, span }.into() + } else { + // `(T, U) -> R` + let (inputs, _) = self.parse_paren_comma_seq(|p| p.parse_ty())?; + let span = ident.span.to(self.prev_span); + let output = if self.eat(&token::RArrow) { + Some(self.parse_ty_common(false, false, false)?) + } else { + None + }; + ParenthesizedArgs { inputs, output, span }.into() + }; + + PathSegment { ident, 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.kind { + token::Ident(name, _) if name.is_path_segment_keyword() => { + let span = self.token.span; + self.bump(); + Ok(Ident::new(name, span)) + } + _ => self.parse_ident(), + } + } + + /// 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_generic_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_generic_args_with_leaning_angle_bracket_recovery( + &mut self, + style: PathStyle, + lo: Span, + ) -> PResult<'a, (Vec<GenericArg>, Vec<AssocTyConstraint>)> { + // 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 = if is_first_invocation { + Some(self.clone()) + } else { + None + }; + + debug!("parse_generic_args_with_leading_angle_bracket_recovery: (snapshotting)"); + match self.parse_generic_args() { + Ok(value) => Ok(value), + Err(ref mut e) if is_first_invocation && self.unmatched_angle_bracket_count > 0 => { + // 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(); + + // Swap `self` with our backup of the parser state before attempting to parse + // generic arguments. + let snapshot = mem::replace(self, snapshot.unwrap()); + + debug!( + "parse_generic_args_with_leading_angle_bracket_recovery: (snapshot failure) \ + snapshot.count={:?}", + snapshot.unmatched_angle_bracket_count, + ); + + // Eat the unmatched angle brackets. + for _ in 0..snapshot.unmatched_angle_bracket_count { + self.eat_lt(); + } + + // Make a span over ${unmatched angle bracket count} characters. + let span = lo.with_hi( + lo.lo() + BytePos(snapshot.unmatched_angle_bracket_count) + ); + self.diagnostic() + .struct_span_err( + span, + &format!( + "unmatched angle bracket{}", + pluralize!(snapshot.unmatched_angle_bracket_count) + ), + ) + .span_suggestion( + span, + &format!( + "remove extra angle bracket{}", + pluralize!(snapshot.unmatched_angle_bracket_count) + ), + String::new(), + Applicability::MachineApplicable, + ) + .emit(); + + // Try again without unmatched angle bracket characters. + self.parse_generic_args() + }, + Err(e) => Err(e), + } + } + + /// Parses (possibly empty) list of lifetime and type arguments and associated type bindings, + /// possibly including trailing comma. + fn parse_generic_args(&mut self) -> PResult<'a, (Vec<GenericArg>, Vec<AssocTyConstraint>)> { + let mut args = Vec::new(); + let mut constraints = Vec::new(); + let mut misplaced_assoc_ty_constraints: Vec<Span> = Vec::new(); + let mut assoc_ty_constraints: Vec<Span> = Vec::new(); + + let args_lo = self.token.span; + + loop { + if self.check_lifetime() && self.look_ahead(1, |t| !t.is_like_plus()) { + // Parse lifetime argument. + args.push(GenericArg::Lifetime(self.expect_lifetime())); + misplaced_assoc_ty_constraints.append(&mut assoc_ty_constraints); + } else if self.check_ident() + && self.look_ahead(1, |t| t == &token::Eq || t == &token::Colon) + { + // Parse associated type constraint. + let lo = self.token.span; + let ident = self.parse_ident()?; + let kind = if self.eat(&token::Eq) { + AssocTyConstraintKind::Equality { + ty: self.parse_ty()?, + } + } else if self.eat(&token::Colon) { + AssocTyConstraintKind::Bound { + bounds: self.parse_generic_bounds(Some(self.prev_span))?, + } + } else { + unreachable!(); + }; + + let span = lo.to(self.prev_span); + + // Gate associated type bounds, e.g., `Iterator<Item: Ord>`. + if let AssocTyConstraintKind::Bound { .. } = kind { + self.sess.gated_spans.gate(sym::associated_type_bounds, span); + } + + constraints.push(AssocTyConstraint { + id: ast::DUMMY_NODE_ID, + ident, + kind, + span, + }); + assoc_ty_constraints.push(span); + } else if self.check_const_arg() { + // Parse const argument. + let expr = if let token::OpenDelim(token::Brace) = self.token.kind { + self.parse_block_expr( + None, self.token.span, BlockCheckMode::Default, ThinVec::new() + )? + } else if self.token.is_ident() { + // FIXME(const_generics): to distinguish between idents for types and consts, + // we should introduce a GenericArg::Ident in the AST and distinguish when + // lowering to the HIR. For now, idents for const args are not permitted. + if self.token.is_bool_lit() { + self.parse_literal_maybe_minus()? + } else { + return Err( + self.fatal("identifiers may currently not be used for const generics") + ); + } + } else { + self.parse_literal_maybe_minus()? + }; + let value = AnonConst { + id: ast::DUMMY_NODE_ID, + value: expr, + }; + args.push(GenericArg::Const(value)); + misplaced_assoc_ty_constraints.append(&mut assoc_ty_constraints); + } else if self.check_type() { + // Parse type argument. + args.push(GenericArg::Type(self.parse_ty()?)); + misplaced_assoc_ty_constraints.append(&mut assoc_ty_constraints); + } else { + break + } + + if !self.eat(&token::Comma) { + break + } + } + + // FIXME: we would like to report this in ast_validation instead, but we currently do not + // preserve ordering of generic parameters with respect to associated type binding, so we + // lose that information after parsing. + if misplaced_assoc_ty_constraints.len() > 0 { + let mut err = self.struct_span_err( + args_lo.to(self.prev_span), + "associated type bindings must be declared after generic parameters", + ); + for span in misplaced_assoc_ty_constraints { + err.span_label( + span, + "this associated type binding should be moved after the generic parameters", + ); + } + err.emit(); + } + + Ok((args, constraints)) + } +} diff --git a/src/librustc_parse/parser/stmt.rs b/src/librustc_parse/parser/stmt.rs new file mode 100644 index 00000000000..a5f20691d07 --- /dev/null +++ b/src/librustc_parse/parser/stmt.rs @@ -0,0 +1,482 @@ +use super::{Parser, Restrictions, PrevTokenKind, SemiColonMode, BlockMode}; +use super::expr::LhsExpr; +use super::path::PathStyle; +use super::pat::GateOr; +use super::diagnostics::Error; +use crate::maybe_whole; +use crate::DirectoryOwnership; + +use syntax::ThinVec; +use syntax::ptr::P; +use syntax::ast; +use syntax::ast::{DUMMY_NODE_ID, Stmt, StmtKind, Local, Block, BlockCheckMode, Expr, ExprKind}; +use syntax::ast::{Attribute, AttrStyle, VisibilityKind, MacStmtStyle, Mac, MacDelimiter}; +use syntax::util::classify; +use syntax::token; +use syntax::source_map::{respan, Span}; +use syntax::symbol::{kw, sym}; + +use std::mem; +use errors::{PResult, Applicability}; + +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. + pub fn parse_stmt(&mut self) -> PResult<'a, Option<Stmt>> { + Ok(self.parse_stmt_(true)) + } + + fn parse_stmt_(&mut self, macro_legacy_warnings: bool) -> Option<Stmt> { + self.parse_stmt_without_recovery(macro_legacy_warnings).unwrap_or_else(|mut e| { + e.emit(); + self.recover_stmt_(SemiColonMode::Break, BlockMode::Ignore); + None + }) + } + + fn parse_stmt_without_recovery( + &mut self, + macro_legacy_warnings: bool, + ) -> PResult<'a, Option<Stmt>> { + maybe_whole!(self, NtStmt, |x| Some(x)); + + let attrs = self.parse_outer_attributes()?; + let lo = self.token.span; + + Ok(Some(if self.eat_keyword(kw::Let) { + Stmt { + id: DUMMY_NODE_ID, + kind: StmtKind::Local(self.parse_local(attrs.into())?), + span: lo.to(self.prev_span), + } + } else if let Some(macro_def) = self.eat_macro_def( + &attrs, + &respan(lo, VisibilityKind::Inherited), + lo, + )? { + Stmt { + id: DUMMY_NODE_ID, + kind: StmtKind::Item(macro_def), + span: lo.to(self.prev_span), + } + // Starts like a simple path, being careful to avoid contextual keywords + // such as a union items, item with `crate` visibility or auto trait items. + // Our goal here is to parse an arbitrary path `a::b::c` but not something that starts + // like a path (1 token), but it fact not a path. + // `union::b::c` - path, `union U { ... }` - not a path. + // `crate::b::c` - path, `crate struct S;` - not a path. + } else if self.token.is_path_start() && + !self.token.is_qpath_start() && + !self.is_union_item() && + !self.is_crate_vis() && + !self.is_auto_trait_item() && + !self.is_async_fn() { + let path = self.parse_path(PathStyle::Expr)?; + + if !self.eat(&token::Not) { + let expr = if self.check(&token::OpenDelim(token::Brace)) { + self.parse_struct_expr(lo, path, ThinVec::new())? + } else { + let hi = self.prev_span; + self.mk_expr(lo.to(hi), ExprKind::Path(None, path), ThinVec::new()) + }; + + let expr = self.with_res(Restrictions::STMT_EXPR, |this| { + let expr = this.parse_dot_or_call_expr_with(expr, lo, attrs.into())?; + this.parse_assoc_expr_with(0, LhsExpr::AlreadyParsed(expr)) + })?; + + return Ok(Some(Stmt { + id: DUMMY_NODE_ID, + kind: StmtKind::Expr(expr), + span: lo.to(self.prev_span), + })); + } + + let (delim, tts) = self.expect_delimited_token_tree()?; + let hi = self.prev_span; + + let style = if delim == MacDelimiter::Brace { + MacStmtStyle::Braces + } else { + MacStmtStyle::NoBraces + }; + + let mac = Mac { + path, + tts, + delim, + span: lo.to(hi), + prior_type_ascription: self.last_type_ascription, + }; + let kind = if delim == MacDelimiter::Brace || + self.token == token::Semi || self.token == token::Eof { + StmtKind::Mac(P((mac, style, attrs.into()))) + } + // We used to incorrectly stop parsing macro-expanded statements here. + // If the next token will be an error anyway but could have parsed with the + // earlier behavior, stop parsing here and emit a warning to avoid breakage. + else if macro_legacy_warnings && self.token.can_begin_expr() && + match self.token.kind { + // These can continue an expression, so we can't stop parsing and warn. + token::OpenDelim(token::Paren) | token::OpenDelim(token::Bracket) | + token::BinOp(token::Minus) | token::BinOp(token::Star) | + token::BinOp(token::And) | token::BinOp(token::Or) | + token::AndAnd | token::OrOr | + token::DotDot | token::DotDotDot | token::DotDotEq => false, + _ => true, + } + { + self.warn_missing_semicolon(); + StmtKind::Mac(P((mac, style, attrs.into()))) + } else { + let e = self.mk_expr(mac.span, ExprKind::Mac(mac), ThinVec::new()); + let e = self.maybe_recover_from_bad_qpath(e, true)?; + let e = self.parse_dot_or_call_expr_with(e, lo, attrs.into())?; + let e = self.parse_assoc_expr_with(0, LhsExpr::AlreadyParsed(e))?; + StmtKind::Expr(e) + }; + Stmt { + id: DUMMY_NODE_ID, + span: lo.to(hi), + kind, + } + } else { + // FIXME: Bad copy of attrs + let old_directory_ownership = + mem::replace(&mut self.directory.ownership, DirectoryOwnership::UnownedViaBlock); + let item = self.parse_item_(attrs.clone(), false, true)?; + self.directory.ownership = old_directory_ownership; + + match item { + Some(i) => Stmt { + id: DUMMY_NODE_ID, + span: lo.to(i.span), + kind: StmtKind::Item(i), + }, + None => { + let unused_attrs = |attrs: &[Attribute], s: &mut Self| { + if !attrs.is_empty() { + if s.prev_token_kind == PrevTokenKind::DocComment { + s.span_fatal_err(s.prev_span, Error::UselessDocComment).emit(); + } else if attrs.iter().any(|a| a.style == AttrStyle::Outer) { + s.span_err( + s.token.span, "expected statement after outer attribute" + ); + } + } + }; + + // Do not attempt to parse an expression if we're done here. + if self.token == token::Semi { + unused_attrs(&attrs, self); + self.bump(); + let mut last_semi = lo; + while self.token == token::Semi { + last_semi = self.token.span; + self.bump(); + } + // We are encoding a string of semicolons as an + // an empty tuple that spans the excess semicolons + // to preserve this info until the lint stage + return Ok(Some(Stmt { + id: DUMMY_NODE_ID, + span: lo.to(last_semi), + kind: StmtKind::Semi(self.mk_expr(lo.to(last_semi), + ExprKind::Tup(Vec::new()), + ThinVec::new() + )), + })); + } + + if self.token == token::CloseDelim(token::Brace) { + unused_attrs(&attrs, self); + return Ok(None); + } + + // Remainder are line-expr stmts. + let e = self.parse_expr_res( + Restrictions::STMT_EXPR, Some(attrs.into()))?; + Stmt { + id: DUMMY_NODE_ID, + span: lo.to(e.span), + kind: StmtKind::Expr(e), + } + } + } + })) + } + + /// Parses a local variable declaration. + fn parse_local(&mut self, attrs: ThinVec<Attribute>) -> PResult<'a, P<Local>> { + let lo = self.prev_span; + let pat = self.parse_top_pat(GateOr::Yes)?; + + let (err, ty) = if self.eat(&token::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_span; + match self.parse_ty() { + Ok(ty) => (None, Some(ty)), + Err(mut err) => { + // Rewind to before attempting to parse the type and continue parsing. + let parser_snapshot_after_type = self.clone(); + mem::replace(self, parser_snapshot_before_type); + + let snippet = self.span_to_snippet(pat.span).unwrap(); + err.span_label(pat.span, format!("while parsing the type for `{}`", snippet)); + (Some((parser_snapshot_after_type, colon_sp, err)), None) + } + } + } else { + (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", + " =".to_string(), + 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(mut 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>; + mem::replace(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 hi = if self.token == token::Semi { + self.token.span + } else { + self.prev_span + }; + Ok(P(ast::Local { + ty, + pat, + init, + id: DUMMY_NODE_ID, + span: lo.to(hi), + attrs, + })) + } + + /// Parses the RHS of a local variable declaration (e.g., '= 14;'). + fn parse_initializer(&mut self, skip_eq: bool) -> PResult<'a, Option<P<Expr>>> { + if self.eat(&token::Eq) { + Ok(Some(self.parse_expr()?)) + } else if skip_eq { + Ok(Some(self.parse_expr()?)) + } else { + Ok(None) + } + } + + fn is_auto_trait_item(&self) -> bool { + // auto trait + (self.token.is_keyword(kw::Auto) && + self.is_keyword_ahead(1, &[kw::Trait])) + || // unsafe auto trait + (self.token.is_keyword(kw::Unsafe) && + self.is_keyword_ahead(1, &[kw::Auto]) && + self.is_keyword_ahead(2, &[kw::Trait])) + } + + /// Parses a block. No inner attributes are allowed. + pub fn parse_block(&mut self) -> PResult<'a, P<Block>> { + maybe_whole!(self, NtBlock, |x| x); + + let lo = self.token.span; + + if !self.eat(&token::OpenDelim(token::Brace)) { + let sp = self.token.span; + let tok = self.this_token_descr(); + let mut e = self.span_fatal(sp, &format!("expected `{{`, found {}", tok)); + let do_not_suggest_help = + self.token.is_keyword(kw::In) || self.token == token::Colon; + + if self.token.is_ident_named(sym::and) { + e.span_suggestion_short( + self.token.span, + "use `&&` instead of `and` for the boolean operator", + "&&".to_string(), + Applicability::MaybeIncorrect, + ); + } + if self.token.is_ident_named(sym::or) { + e.span_suggestion_short( + self.token.span, + "use `||` instead of `or` for the boolean operator", + "||".to_string(), + Applicability::MaybeIncorrect, + ); + } + + // 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) { + Ok(Some(stmt)) => { + if self.look_ahead(1, |t| t == &token::OpenDelim(token::Brace)) + || do_not_suggest_help { + // If the next token is an open brace (e.g., `if a b {`), the place- + // inside-a-block suggestion would be more likely wrong than right. + e.span_label(sp, "expected `{`"); + return Err(e); + } + let mut stmt_span = stmt.span; + // Expand the span to include the semicolon, if it exists. + if self.eat(&token::Semi) { + stmt_span = stmt_span.with_hi(self.prev_span.hi()); + } + if let Ok(snippet) = self.span_to_snippet(stmt_span) { + e.span_suggestion( + stmt_span, + "try placing this code inside a block", + format!("{{ {} }}", snippet), + // Speculative; has been misleading in the past (#46836). + Applicability::MaybeIncorrect, + ); + } + } + Err(mut e) => { + self.recover_stmt_(SemiColonMode::Break, BlockMode::Ignore); + e.cancel(); + } + _ => () + } + e.span_label(sp, "expected `{`"); + return Err(e); + } + + self.parse_block_tail(lo, BlockCheckMode::Default) + } + + /// Parses a block. Inner attributes are allowed. + pub(super) fn parse_inner_attrs_and_block( + &mut self + ) -> PResult<'a, (Vec<Attribute>, P<Block>)> { + maybe_whole!(self, NtBlock, |x| (Vec::new(), x)); + + let lo = self.token.span; + self.expect(&token::OpenDelim(token::Brace))?; + Ok((self.parse_inner_attributes()?, + self.parse_block_tail(lo, BlockCheckMode::Default)?)) + } + + /// Parses the rest of a block expression or function body. + /// Precondition: already parsed the '{'. + pub(super) fn parse_block_tail( + &mut self, + lo: Span, + s: BlockCheckMode + ) -> PResult<'a, P<Block>> { + let mut stmts = vec![]; + while !self.eat(&token::CloseDelim(token::Brace)) { + if self.token == token::Eof { + break; + } + let stmt = match self.parse_full_stmt(false) { + Err(mut err) => { + self.maybe_annotate_with_ascription(&mut err, false); + err.emit(); + self.recover_stmt_(SemiColonMode::Ignore, BlockMode::Ignore); + Some(Stmt { + id: DUMMY_NODE_ID, + kind: StmtKind::Expr(self.mk_expr_err(self.token.span)), + span: self.token.span, + }) + } + Ok(stmt) => stmt, + }; + if let Some(stmt) = stmt { + stmts.push(stmt); + } else { + // Found only `;` or `}`. + continue; + }; + } + Ok(P(ast::Block { + stmts, + id: DUMMY_NODE_ID, + rules: s, + span: lo.to(self.prev_span), + })) + } + + /// Parses a statement, including the trailing semicolon. + pub fn parse_full_stmt(&mut self, macro_legacy_warnings: bool) -> PResult<'a, Option<Stmt>> { + // Skip looking for a trailing semicolon when we have an interpolated statement. + maybe_whole!(self, NtStmt, |x| Some(x)); + + let mut stmt = match self.parse_stmt_without_recovery(macro_legacy_warnings)? { + Some(stmt) => stmt, + None => return Ok(None), + }; + + let mut eat_semi = true; + match stmt.kind { + StmtKind::Expr(ref expr) if self.token != token::Eof => { + // expression without semicolon + if classify::expr_requires_semi_to_be_stmt(expr) { + // Just check for errors and recover; do not eat semicolon yet. + if let Err(mut e) = + self.expect_one_of(&[], &[token::Semi, token::CloseDelim(token::Brace)]) + { + e.emit(); + self.recover_stmt(); + // Don't complain about type errors in body tail after parse error (#57383). + let sp = expr.span.to(self.prev_span); + stmt.kind = StmtKind::Expr(self.mk_expr_err(sp)); + } + } + } + StmtKind::Local(..) => { + // We used to incorrectly allow a macro-expanded let statement to lack a semicolon. + if macro_legacy_warnings && self.token != token::Semi { + self.warn_missing_semicolon(); + } else { + self.expect_semi()?; + eat_semi = false; + } + } + _ => {} + } + + if eat_semi && self.eat(&token::Semi) { + stmt = stmt.add_trailing_semicolon(); + } + stmt.span = stmt.span.to(self.prev_span); + Ok(Some(stmt)) + } + + fn warn_missing_semicolon(&self) { + self.diagnostic().struct_span_warn(self.token.span, { + &format!("expected `;`, found {}", self.this_token_descr()) + }).note({ + "this was erroneously allowed and will become a hard error in a future release" + }).emit(); + } +} diff --git a/src/librustc_parse/parser/ty.rs b/src/librustc_parse/parser/ty.rs new file mode 100644 index 00000000000..fccabee2404 --- /dev/null +++ b/src/librustc_parse/parser/ty.rs @@ -0,0 +1,460 @@ +use super::{Parser, PathStyle, PrevTokenKind, TokenType}; +use super::item::ParamCfg; + +use crate::{maybe_whole, maybe_recover_from_interpolated_ty_qpath}; + +use syntax::ptr::P; +use syntax::ast::{self, Ty, TyKind, MutTy, BareFnTy, FunctionRetTy, GenericParam, Lifetime, Ident}; +use syntax::ast::{TraitBoundModifier, TraitObjectSyntax, GenericBound, GenericBounds, PolyTraitRef}; +use syntax::ast::{Mutability, AnonConst, Mac}; +use syntax::token::{self, Token}; +use syntax::source_map::Span; +use syntax::struct_span_fatal; +use syntax_pos::symbol::kw; + +use errors::{PResult, Applicability, pluralize}; + +/// 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::ModSep || t == &token::Lt || + t == &token::BinOp(token::Shl) +} + +impl<'a> Parser<'a> { + /// Parses a type. + pub fn parse_ty(&mut self) -> PResult<'a, P<Ty>> { + self.parse_ty_common(true, true, false) + } + + /// 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(false, true, false) + } + + /// Parses an optional return type `[ -> TY ]` in a function declaration. + pub(super) fn parse_ret_ty(&mut self, allow_plus: bool) -> PResult<'a, FunctionRetTy> { + if self.eat(&token::RArrow) { + Ok(FunctionRetTy::Ty(self.parse_ty_common(allow_plus, true, false)?)) + } else { + Ok(FunctionRetTy::Default(self.token.span.shrink_to_lo())) + } + } + + pub(super) fn parse_ty_common(&mut self, allow_plus: bool, allow_qpath_recovery: bool, + allow_c_variadic: bool) -> PResult<'a, P<Ty>> { + maybe_recover_from_interpolated_ty_qpath!(self, allow_qpath_recovery); + maybe_whole!(self, NtTy, |x| x); + + let lo = self.token.span; + let mut impl_dyn_multi = false; + let kind = if self.eat(&token::OpenDelim(token::Paren)) { + // `(TYPE)` is a parenthesized type. + // `(TYPE,)` is a tuple with a single field of type TYPE. + let mut ts = vec![]; + let mut last_comma = false; + while self.token != token::CloseDelim(token::Paren) { + ts.push(self.parse_ty()?); + if self.eat(&token::Comma) { + last_comma = true; + } else { + last_comma = false; + break; + } + } + let trailing_plus = self.prev_token_kind == PrevTokenKind::Plus; + self.expect(&token::CloseDelim(token::Paren))?; + + if ts.len() == 1 && !last_comma { + let ty = ts.into_iter().nth(0).unwrap().into_inner(); + let maybe_bounds = allow_plus && self.token.is_like_plus(); + match ty.kind { + // `(TY_BOUND_NOPAREN) + BOUND + ...`. + TyKind::Path(None, ref path) if maybe_bounds => { + self.parse_remaining_bounds(Vec::new(), path.clone(), lo, true)? + } + TyKind::TraitObject(ref bounds, TraitObjectSyntax::None) + if maybe_bounds && bounds.len() == 1 && !trailing_plus => { + let path = match bounds[0] { + GenericBound::Trait(ref pt, ..) => pt.trait_ref.path.clone(), + GenericBound::Outlives(..) => self.bug("unexpected lifetime bound"), + }; + self.parse_remaining_bounds(Vec::new(), path, lo, true)? + } + // `(TYPE)` + _ => TyKind::Paren(P(ty)) + } + } else { + TyKind::Tup(ts) + } + } else if self.eat(&token::Not) { + // Never type `!` + TyKind::Never + } else if self.eat(&token::BinOp(token::Star)) { + // Raw pointer + TyKind::Ptr(self.parse_ptr()?) + } else if self.eat(&token::OpenDelim(token::Bracket)) { + // Array or slice + let t = self.parse_ty()?; + // Parse optional `; EXPR` in `[TYPE; EXPR]` + let t = match self.maybe_parse_fixed_length_of_vec()? { + None => TyKind::Slice(t), + Some(length) => TyKind::Array(t, AnonConst { + id: ast::DUMMY_NODE_ID, + value: length, + }), + }; + self.expect(&token::CloseDelim(token::Bracket))?; + t + } 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) { + // `typeof(EXPR)` + // In order to not be ambiguous, the type must be surrounded by parens. + self.expect(&token::OpenDelim(token::Paren))?; + let e = AnonConst { + id: ast::DUMMY_NODE_ID, + value: self.parse_expr()?, + }; + self.expect(&token::CloseDelim(token::Paren))?; + TyKind::Typeof(e) + } else if self.eat_keyword(kw::Underscore) { + // A type to be inferred `_` + TyKind::Infer + } else if self.token_is_bare_fn_keyword() { + // Function pointer type + self.parse_ty_bare_fn(Vec::new())? + } else if self.check_keyword(kw::For) { + // 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 lo = self.token.span; + let lifetime_defs = self.parse_late_bound_lifetime_defs()?; + if self.token_is_bare_fn_keyword() { + self.parse_ty_bare_fn(lifetime_defs)? + } else { + let path = self.parse_path(PathStyle::Type)?; + let parse_plus = allow_plus && self.check_plus(); + self.parse_remaining_bounds(lifetime_defs, path, lo, parse_plus)? + } + } else if self.eat_keyword(kw::Impl) { + // Always parse bounds greedily for better error recovery. + let bounds = self.parse_generic_bounds(None)?; + impl_dyn_multi = bounds.len() > 1 || self.prev_token_kind == PrevTokenKind::Plus; + TyKind::ImplTrait(ast::DUMMY_NODE_ID, bounds) + } else if self.check_keyword(kw::Dyn) && + (self.token.span.rust_2018() || + self.look_ahead(1, |t| t.can_begin_bound() && + !can_continue_type_after_non_fn_ident(t))) { + self.bump(); // `dyn` + // Always parse bounds greedily for better error recovery. + let bounds = self.parse_generic_bounds(None)?; + impl_dyn_multi = bounds.len() > 1 || self.prev_token_kind == PrevTokenKind::Plus; + TyKind::TraitObject(bounds, TraitObjectSyntax::Dyn) + } else if self.check(&token::Question) || + self.check_lifetime() && self.look_ahead(1, |t| t.is_like_plus()) { + // Bound list (trait object type) + TyKind::TraitObject(self.parse_generic_bounds_common(allow_plus, None)?, + TraitObjectSyntax::None) + } else if self.eat_lt() { + // Qualified path + let (qself, path) = self.parse_qpath(PathStyle::Type)?; + TyKind::Path(Some(qself), path) + } else if self.token.is_path_start() { + // Simple path + let path = self.parse_path(PathStyle::Type)?; + if self.eat(&token::Not) { + // Macro invocation in type position + let (delim, tts) = self.expect_delimited_token_tree()?; + let mac = Mac { + path, + tts, + delim, + span: lo.to(self.prev_span), + prior_type_ascription: self.last_type_ascription, + }; + TyKind::Mac(mac) + } else { + // Just a type path or bound list (trait object type) starting with a trait. + // `Type` + // `Trait1 + Trait2 + 'a` + if allow_plus && self.check_plus() { + self.parse_remaining_bounds(Vec::new(), path, lo, true)? + } else { + TyKind::Path(None, path) + } + } + } else if self.check(&token::DotDotDot) { + if allow_c_variadic { + self.eat(&token::DotDotDot); + TyKind::CVarArgs + } else { + return Err(struct_span_fatal!( + self.sess.span_diagnostic, + self.token.span, + E0743, + "only foreign functions are allowed to be C-variadic", + )); + } + } else { + let msg = format!("expected type, found {}", self.this_token_descr()); + let mut err = self.fatal(&msg); + err.span_label(self.token.span, "expected type"); + self.maybe_annotate_with_ascription(&mut err, true); + return Err(err); + }; + + let span = lo.to(self.prev_span); + let ty = self.mk_ty(span, kind); + + // Try to recover from use of `+` with incorrect priority. + self.maybe_report_ambiguous_plus(allow_plus, impl_dyn_multi, &ty); + self.maybe_recover_from_bad_type_plus(allow_plus, &ty)?; + self.maybe_recover_from_bad_qpath(ty, allow_qpath_recovery) + } + + fn parse_remaining_bounds(&mut self, generic_params: Vec<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_span)); + let mut bounds = vec![GenericBound::Trait(poly_trait_ref, TraitBoundModifier::None)]; + if parse_plus { + self.eat_plus(); // `+`, or `+=` gets split and `+` is discarded + bounds.append(&mut self.parse_generic_bounds(Some(self.prev_span))?); + } + Ok(TyKind::TraitObject(bounds, TraitObjectSyntax::None)) + } + + fn parse_ptr(&mut self) -> PResult<'a, MutTy> { + let mutbl = self.parse_const_or_mut().unwrap_or_else(|| { + let span = self.prev_span; + let msg = "expected mut or const in raw pointer type"; + self.struct_span_err(span, msg) + .span_label(span, msg) + .help("use `*mut T` or `*const T` as appropriate") + .emit(); + Mutability::Immutable + }); + let t = self.parse_ty_no_plus()?; + Ok(MutTy { ty: t, mutbl }) + } + + fn maybe_parse_fixed_length_of_vec(&mut self) -> PResult<'a, Option<P<ast::Expr>>> { + if self.eat(&token::Semi) { + Ok(Some(self.parse_expr()?)) + } else { + Ok(None) + } + } + + fn parse_borrowed_pointee(&mut self) -> PResult<'a, TyKind> { + let opt_lifetime = if self.check_lifetime() { Some(self.expect_lifetime()) } else { None }; + let mutbl = self.parse_mutability(); + let ty = self.parse_ty_no_plus()?; + return Ok(TyKind::Rptr(opt_lifetime, MutTy { ty, mutbl })); + } + + /// Is the current token one of the keywords that signals a bare function type? + fn token_is_bare_fn_keyword(&mut self) -> bool { + self.check_keyword(kw::Fn) || + self.check_keyword(kw::Unsafe) || + self.check_keyword(kw::Extern) + } + + /// Parses a `TyKind::BareFn` type. + fn parse_ty_bare_fn(&mut self, generic_params: Vec<GenericParam>) -> PResult<'a, TyKind> { + /* + + [unsafe] [extern "ABI"] fn (S) -> T + ^~~~^ ^~~~^ ^~^ ^ + | | | | + | | | Return type + | | Argument types + | | + | ABI + Function Style + */ + + let unsafety = self.parse_unsafety(); + let abi = self.parse_extern_abi()?; + self.expect_keyword(kw::Fn)?; + let cfg = ParamCfg { + is_self_allowed: false, + allow_c_variadic: true, + is_name_required: |_| false, + }; + let decl = self.parse_fn_decl(cfg, false)?; + Ok(TyKind::BareFn(P(BareFnTy { + abi, + unsafety, + generic_params, + decl, + }))) + } + + pub(super) fn parse_generic_bounds(&mut self, + colon_span: Option<Span>) -> PResult<'a, GenericBounds> { + self.parse_generic_bounds_common(true, colon_span) + } + + /// Parses bounds of a type parameter `BOUND + BOUND + ...`, possibly with trailing `+`. + /// + /// ``` + /// BOUND = TY_BOUND | LT_BOUND + /// LT_BOUND = LIFETIME (e.g., `'a`) + /// TY_BOUND = TY_BOUND_NOPAREN | (TY_BOUND_NOPAREN) + /// TY_BOUND_NOPAREN = [?] [for<LT_PARAM_DEFS>] SIMPLE_PATH (e.g., `?for<'a: 'b> m::Trait<'a>`) + /// ``` + fn parse_generic_bounds_common(&mut self, + allow_plus: bool, + colon_span: Option<Span>) -> PResult<'a, GenericBounds> { + let mut bounds = Vec::new(); + let mut negative_bounds = Vec::new(); + let mut last_plus_span = None; + let mut was_negative = false; + loop { + // This needs to be synchronized with `TokenKind::can_begin_bound`. + let is_bound_start = self.check_path() || self.check_lifetime() || + self.check(&token::Not) || // used for error reporting only + self.check(&token::Question) || + self.check_keyword(kw::For) || + self.check(&token::OpenDelim(token::Paren)); + if is_bound_start { + let lo = self.token.span; + let has_parens = self.eat(&token::OpenDelim(token::Paren)); + let inner_lo = self.token.span; + let is_negative = self.eat(&token::Not); + let question = if self.eat(&token::Question) { Some(self.prev_span) } else { None }; + if self.token.is_lifetime() { + if let Some(question_span) = question { + self.span_err(question_span, + "`?` may only modify trait bounds, not lifetime bounds"); + } + bounds.push(GenericBound::Outlives(self.expect_lifetime())); + if has_parens { + let inner_span = inner_lo.to(self.prev_span); + self.expect(&token::CloseDelim(token::Paren))?; + let mut err = self.struct_span_err( + lo.to(self.prev_span), + "parenthesized lifetime bounds are not supported" + ); + if let Ok(snippet) = self.span_to_snippet(inner_span) { + err.span_suggestion_short( + lo.to(self.prev_span), + "remove the parentheses", + snippet.to_owned(), + Applicability::MachineApplicable + ); + } + err.emit(); + } + } else { + let lifetime_defs = self.parse_late_bound_lifetime_defs()?; + let path = self.parse_path(PathStyle::Type)?; + if has_parens { + self.expect(&token::CloseDelim(token::Paren))?; + } + let poly_span = lo.to(self.prev_span); + if is_negative { + was_negative = true; + if let Some(sp) = last_plus_span.or(colon_span) { + negative_bounds.push(sp.to(poly_span)); + } + } else { + let poly_trait = PolyTraitRef::new(lifetime_defs, path, poly_span); + let modifier = if question.is_some() { + TraitBoundModifier::Maybe + } else { + TraitBoundModifier::None + }; + bounds.push(GenericBound::Trait(poly_trait, modifier)); + } + } + } else { + break + } + + if !allow_plus || !self.eat_plus() { + break + } else { + last_plus_span = Some(self.prev_span); + } + } + + if !negative_bounds.is_empty() || was_negative { + let negative_bounds_len = negative_bounds.len(); + let last_span = negative_bounds.last().map(|sp| *sp); + let mut err = self.struct_span_err( + negative_bounds, + "negative trait bounds are not supported", + ); + if let Some(sp) = last_span { + err.span_label(sp, "negative trait bounds are not supported"); + } + if let Some(bound_list) = colon_span { + let bound_list = bound_list.to(self.prev_span); + let mut new_bound_list = String::new(); + if !bounds.is_empty() { + let mut snippets = bounds.iter().map(|bound| bound.span()) + .map(|span| self.span_to_snippet(span)); + while let Some(Ok(snippet)) = snippets.next() { + new_bound_list.push_str(" + "); + new_bound_list.push_str(&snippet); + } + new_bound_list = new_bound_list.replacen(" +", ":", 1); + } + err.span_suggestion_hidden( + bound_list, + &format!("remove the trait bound{}", pluralize!(negative_bounds_len)), + new_bound_list, + Applicability::MachineApplicable, + ); + } + err.emit(); + } + + return Ok(bounds); + } + + pub(super) fn parse_late_bound_lifetime_defs(&mut self) -> PResult<'a, Vec<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(Vec::new()) + } + } + + pub fn check_lifetime(&mut self) -> bool { + self.expected_tokens.push(TokenType::Lifetime); + self.token.is_lifetime() + } + + /// Parses a single lifetime `'a` or panics. + pub fn expect_lifetime(&mut self) -> Lifetime { + if let Some(ident) = self.token.lifetime() { + let span = self.token.span; + self.bump(); + Lifetime { ident: Ident::new(ident.name, span), id: ast::DUMMY_NODE_ID } + } else { + self.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 }) + } +} |
