diff options
Diffstat (limited to 'compiler/rustc_parse/src')
| -rw-r--r-- | compiler/rustc_parse/src/errors.rs | 4 | ||||
| -rw-r--r-- | compiler/rustc_parse/src/parser/expr.rs | 220 | ||||
| -rw-r--r-- | compiler/rustc_parse/src/parser/item.rs | 53 | ||||
| -rw-r--r-- | compiler/rustc_parse/src/parser/mod.rs | 32 | ||||
| -rw-r--r-- | compiler/rustc_parse/src/parser/nonterminal.rs | 16 | ||||
| -rw-r--r-- | compiler/rustc_parse/src/parser/pat.rs | 2 |
6 files changed, 225 insertions, 102 deletions
diff --git a/compiler/rustc_parse/src/errors.rs b/compiler/rustc_parse/src/errors.rs index 55cb7687823..dfdef018bc3 100644 --- a/compiler/rustc_parse/src/errors.rs +++ b/compiler/rustc_parse/src/errors.rs @@ -1696,10 +1696,10 @@ pub(crate) struct SelfArgumentPointer { #[derive(Diagnostic)] #[diag(parse_unexpected_token_after_dot)] -pub(crate) struct UnexpectedTokenAfterDot<'a> { +pub(crate) struct UnexpectedTokenAfterDot { #[primary_span] pub span: Span, - pub actual: Cow<'a, str>, + pub actual: String, } #[derive(Diagnostic)] diff --git a/compiler/rustc_parse/src/parser/expr.rs b/compiler/rustc_parse/src/parser/expr.rs index 00e75cc2c2f..424fee968e1 100644 --- a/compiler/rustc_parse/src/parser/expr.rs +++ b/compiler/rustc_parse/src/parser/expr.rs @@ -4,10 +4,10 @@ use core::mem; use core::ops::{Bound, ControlFlow}; use ast::mut_visit::{self, MutVisitor}; -use ast::token::{IdentIsRaw, MetaVarKind}; +use ast::token::IdentIsRaw; use ast::{CoroutineKind, ForLoopKind, GenBlockKind, MatchKind, Pat, Path, PathSegment, Recovered}; use rustc_ast::ptr::P; -use rustc_ast::token::{self, Delimiter, Token, TokenKind}; +use rustc_ast::token::{self, Delimiter, InvisibleOrigin, MetaVarKind, Token, TokenKind}; use rustc_ast::tokenstream::TokenTree; use rustc_ast::util::case::Case; use rustc_ast::util::classify; @@ -19,7 +19,6 @@ use rustc_ast::{ MetaItemLit, Movability, Param, RangeLimits, StmtKind, Ty, TyKind, UnOp, UnsafeBinderCastKind, YieldKind, }; -use rustc_ast_pretty::pprust; use rustc_data_structures::stack::ensure_sufficient_stack; use rustc_errors::{Applicability, Diag, PResult, StashKey, Subdiagnostic}; use rustc_lexer::unescape::unescape_char; @@ -605,7 +604,7 @@ impl<'a> Parser<'a> { // 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(), + _ => t.is_metavar_expr(), }; self.token.is_ident_named(sym::not) && self.look_ahead(1, token_cannot_continue_expr) } @@ -641,6 +640,13 @@ impl<'a> Parser<'a> { TokenKind::NtIdent(..) | TokenKind::NtLifetime(..) | TokenKind::Interpolated(..) => { self.prev_token.span } + TokenKind::CloseDelim(Delimiter::Invisible(InvisibleOrigin::MetaVar(_))) => { + // `expr.span` is the interpolated span, because invisible open + // and close delims both get marked with the same span, one + // that covers the entire thing between them. (See + // `rustc_expand::mbe::transcribe::transcribe`.) + self.prev_token.span + } _ => expr.span, } } @@ -979,12 +985,30 @@ impl<'a> Parser<'a> { } fn error_unexpected_after_dot(&self) { - let actual = pprust::token_to_string(&self.token); + let actual = super::token_descr(&self.token); let span = self.token.span; let sm = self.psess.source_map(); let (span, actual) = match (&self.token.kind, self.subparser_name) { - (token::Eof, Some(_)) if let Ok(actual) = sm.span_to_snippet(sm.next_point(span)) => { - (span.shrink_to_hi(), actual.into()) + (token::Eof, Some(_)) if let Ok(snippet) = sm.span_to_snippet(sm.next_point(span)) => { + (span.shrink_to_hi(), format!("`{}`", snippet)) + } + (token::CloseDelim(Delimiter::Invisible(InvisibleOrigin::MetaVar(_))), _) => { + // No need to report an error. This case will only occur when parsing a pasted + // metavariable, and we should have emitted an error when parsing the macro call in + // the first place. E.g. in this code: + // ``` + // macro_rules! m { ($e:expr) => { $e }; } + // + // fn main() { + // let f = 1; + // m!(f.); + // } + // ``` + // we'll get an error "unexpected token: `)` when parsing the `m!(f.)`, so we don't + // want to issue a second error when parsing the expansion `«f.»` (where `«`/`»` + // represent the invisible delimiters). + self.dcx().span_delayed_bug(span, "bad dot expr in metavariable"); + return; } _ => (span, actual), }; @@ -1364,17 +1388,31 @@ impl<'a> Parser<'a> { let span = self.token.span; if let token::Interpolated(nt) = &self.token.kind { match &**nt { - token::NtExpr(e) | token::NtLiteral(e) => { - let e = e.clone(); - self.bump(); - return Ok(e); - } token::NtBlock(block) => { let block = block.clone(); self.bump(); return Ok(self.mk_expr(self.prev_token.span, ExprKind::Block(block, None))); } }; + } else if let Some(expr) = self.eat_metavar_seq_with_matcher( + |mv_kind| matches!(mv_kind, MetaVarKind::Expr { .. }), + |this| { + let expr = this.parse_expr(); + // FIXME(nnethercote) Sometimes with expressions we get a trailing comma, possibly + // related to the FIXME in `collect_tokens_for_expr`. Examples are the multi-line + // `assert_eq!` calls involving arguments annotated with `#[rustfmt::skip]` in + // `compiler/rustc_index/src/bit_set/tests.rs`. + if this.token.kind == token::Comma { + this.bump(); + } + expr + }, + ) { + return Ok(expr); + } else if let Some(lit) = + self.eat_metavar_seq(MetaVarKind::Literal, |this| this.parse_literal_maybe_minus()) + { + return Ok(lit); } else if let Some(path) = self.eat_metavar_seq(MetaVarKind::Path, |this| { this.collect_tokens_no_attrs(|this| this.parse_path(PathStyle::Type)) }) { @@ -2062,87 +2100,107 @@ impl<'a> Parser<'a> { .or_else(|()| self.handle_missing_lit(Parser::mk_meta_item_lit_char)) } - fn recover_after_dot(&mut self) -> Option<Token> { - let mut recovered = None; + fn recover_after_dot(&mut self) { if self.token == token::Dot { // Attempt to recover `.4` as `0.4`. We don't currently have any syntax where // dot would follow an optional literal, so we do this unconditionally. - recovered = self.look_ahead(1, |next_token| { + let recovered = self.look_ahead(1, |next_token| { + // If it's an integer that looks like a float, then recover as such. + // + // We will never encounter the exponent part of a floating + // point literal here, since there's no use of the exponent + // syntax that also constitutes a valid integer, so we need + // not check for that. if let token::Literal(token::Lit { kind: token::Integer, symbol, suffix }) = next_token.kind + && suffix.is_none_or(|s| s == sym::f32 || s == sym::f64) + && symbol.as_str().chars().all(|c| c.is_numeric() || c == '_') + && self.token.span.hi() == next_token.span.lo() { - // If this integer looks like a float, then recover as such. - // - // We will never encounter the exponent part of a floating - // point literal here, since there's no use of the exponent - // syntax that also constitutes a valid integer, so we need - // not check for that. - if suffix.is_none_or(|s| s == sym::f32 || s == sym::f64) - && symbol.as_str().chars().all(|c| c.is_numeric() || c == '_') - && self.token.span.hi() == next_token.span.lo() - { - let s = String::from("0.") + symbol.as_str(); - let kind = TokenKind::lit(token::Float, Symbol::intern(&s), suffix); - return Some(Token::new(kind, self.token.span.to(next_token.span))); - } + let s = String::from("0.") + symbol.as_str(); + let kind = TokenKind::lit(token::Float, Symbol::intern(&s), suffix); + Some(Token::new(kind, self.token.span.to(next_token.span))) + } else { + None } - None }); - if let Some(token) = &recovered { - self.bump(); + if let Some(recovered) = recovered { self.dcx().emit_err(errors::FloatLiteralRequiresIntegerPart { - span: token.span, - suggestion: token.span.shrink_to_lo(), + span: recovered.span, + suggestion: recovered.span.shrink_to_lo(), }); + self.bump(); + self.token = recovered; } } + } - recovered + /// Keep this in sync with `Token::can_begin_literal_maybe_minus` and + /// `Lit::from_token` (excluding unary negation). + fn eat_token_lit(&mut self) -> Option<token::Lit> { + match self.token.uninterpolate().kind { + token::Ident(name, IdentIsRaw::No) if name.is_bool_lit() => { + self.bump(); + Some(token::Lit::new(token::Bool, name, None)) + } + token::Literal(token_lit) => { + self.bump(); + Some(token_lit) + } + token::OpenDelim(Delimiter::Invisible(InvisibleOrigin::MetaVar( + MetaVarKind::Literal, + ))) => { + let lit = self + .eat_metavar_seq(MetaVarKind::Literal, |this| this.parse_literal_maybe_minus()) + .expect("metavar seq literal"); + let ast::ExprKind::Lit(token_lit) = lit.kind else { + panic!("didn't reparse a literal"); + }; + Some(token_lit) + } + token::OpenDelim(Delimiter::Invisible(InvisibleOrigin::MetaVar( + mv_kind @ MetaVarKind::Expr { can_begin_literal_maybe_minus: true, .. }, + ))) => { + let expr = self + .eat_metavar_seq(mv_kind, |this| this.parse_expr()) + .expect("metavar seq expr"); + let ast::ExprKind::Lit(token_lit) = expr.kind else { + panic!("didn't reparse an expr"); + }; + Some(token_lit) + } + _ => None, + } } /// Matches `lit = true | false | token_lit`. /// Returns `None` if the next token is not a literal. - pub(super) fn parse_opt_token_lit(&mut self) -> Option<(token::Lit, Span)> { - let recovered = self.recover_after_dot(); - let token = recovered.as_ref().unwrap_or(&self.token); - let span = token.span; - - token::Lit::from_token(token).map(|token_lit| { - self.bump(); - (token_lit, span) - }) + fn parse_opt_token_lit(&mut self) -> Option<(token::Lit, Span)> { + self.recover_after_dot(); + let span = self.token.span; + self.eat_token_lit().map(|token_lit| (token_lit, span)) } /// Matches `lit = true | false | token_lit`. /// Returns `None` if the next token is not a literal. - pub(super) fn parse_opt_meta_item_lit(&mut self) -> Option<MetaItemLit> { - let recovered = self.recover_after_dot(); - let token = recovered.as_ref().unwrap_or(&self.token); - match token::Lit::from_token(token) { - Some(lit) => { - match MetaItemLit::from_token_lit(lit, token.span) { - Ok(lit) => { - self.bump(); - Some(lit) - } - Err(err) => { - let span = token.uninterpolated_span(); - self.bump(); - let guar = report_lit_error(self.psess, err, lit, span); - // Pack possible quotes and prefixes from the original literal into - // the error literal's symbol so they can be pretty-printed faithfully. - let suffixless_lit = token::Lit::new(lit.kind, lit.symbol, None); - let symbol = Symbol::intern(&suffixless_lit.to_string()); - let lit = token::Lit::new(token::Err(guar), symbol, lit.suffix); - Some( - MetaItemLit::from_token_lit(lit, span) - .unwrap_or_else(|_| unreachable!()), - ) - } + fn parse_opt_meta_item_lit(&mut self) -> Option<MetaItemLit> { + self.recover_after_dot(); + let span = self.token.span; + let uninterpolated_span = self.uninterpolated_token_span(); + self.eat_token_lit().map(|token_lit| { + match MetaItemLit::from_token_lit(token_lit, span) { + Ok(lit) => lit, + Err(err) => { + let guar = report_lit_error(&self.psess, err, token_lit, uninterpolated_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(token_lit.kind, token_lit.symbol, None); + let symbol = Symbol::intern(&suffixless_lit.to_string()); + let token_lit = token::Lit::new(token::Err(guar), symbol, token_lit.suffix); + MetaItemLit::from_token_lit(token_lit, uninterpolated_span).unwrap() } } - None => None, - } + }) } pub(super) fn expect_no_tuple_index_suffix(&self, span: Span, suffix: Symbol) { @@ -2166,9 +2224,10 @@ impl<'a> Parser<'a> { /// Matches `'-' lit | lit` (cf. `ast_validation::AstValidator::check_expr_within_pat`). /// Keep this in sync with `Token::can_begin_literal_maybe_minus`. pub fn parse_literal_maybe_minus(&mut self) -> PResult<'a, P<Expr>> { - if let token::Interpolated(nt) = &self.token.kind { - match &**nt { - // FIXME(nnethercote) The `NtExpr` case should only match if + if let Some(expr) = self.eat_metavar_seq_with_matcher( + |mv_kind| matches!(mv_kind, MetaVarKind::Expr { .. }), + |this| { + // FIXME(nnethercote) The `expr` case should only match if // `e` is an `ExprKind::Lit` or an `ExprKind::Unary` containing // an `UnOp::Neg` and an `ExprKind::Lit`, like how // `can_begin_literal_maybe_minus` works. But this method has @@ -2178,13 +2237,14 @@ impl<'a> Parser<'a> { // `ExprKind::Path` must be accepted when parsing range // patterns. That requires some care. So for now, we continue // being less strict here than we should be. - token::NtExpr(e) | token::NtLiteral(e) => { - let e = e.clone(); - self.bump(); - return Ok(e); - } - _ => {} - }; + this.parse_expr() + }, + ) { + return Ok(expr); + } else if let Some(lit) = + self.eat_metavar_seq(MetaVarKind::Literal, |this| this.parse_literal_maybe_minus()) + { + return Ok(lit); } let lo = self.token.span; diff --git a/compiler/rustc_parse/src/parser/item.rs b/compiler/rustc_parse/src/parser/item.rs index e93fb2473fb..20833a0f70f 100644 --- a/compiler/rustc_parse/src/parser/item.rs +++ b/compiler/rustc_parse/src/parser/item.rs @@ -1290,12 +1290,24 @@ impl<'a> Parser<'a> { } fn is_unsafe_foreign_mod(&self) -> bool { - self.token.is_keyword(kw::Unsafe) - && self.is_keyword_ahead(1, &[kw::Extern]) - && self.look_ahead( - 2 + self.look_ahead(2, |t| t.can_begin_string_literal() as usize), - |t| *t == token::OpenDelim(Delimiter::Brace), - ) + // Look for `unsafe`. + if !self.token.is_keyword(kw::Unsafe) { + return false; + } + // Look for `extern`. + if !self.is_keyword_ahead(1, &[kw::Extern]) { + return false; + } + + // Look for the optional ABI string literal. + let n = if self.look_ahead(2, |t| t.can_begin_string_literal()) { 3 } else { 2 }; + + // Look for the `{`. Use `tree_look_ahead` because the ABI (if present) + // might be a metavariable i.e. an invisible-delimited sequence, and + // `tree_look_ahead` will consider that a single element when looking + // ahead. + self.tree_look_ahead(n, |t| matches!(t, TokenTree::Delimited(_, _, Delimiter::Brace, _))) + == Some(true) } fn is_static_global(&mut self) -> bool { @@ -2604,13 +2616,36 @@ impl<'a> Parser<'a> { }) // `extern ABI fn` || self.check_keyword_case(exp!(Extern), case) + // Use `tree_look_ahead` because `ABI` might be a metavariable, + // i.e. an invisible-delimited sequence, and `tree_look_ahead` + // will consider that a single element when looking ahead. && self.look_ahead(1, |t| t.can_begin_string_literal()) - && (self.look_ahead(2, |t| t.is_keyword_case(kw::Fn, case)) || + && (self.tree_look_ahead(2, |tt| { + match tt { + TokenTree::Token(t, _) => t.is_keyword_case(kw::Fn, case), + TokenTree::Delimited(..) => false, + } + }) == Some(true) || // This branch is only for better diagnostics; `pub`, `unsafe`, etc. are not // allowed here. (self.may_recover() - && self.look_ahead(2, |t| ALL_QUALS.iter().any(|exp| t.is_keyword(exp.kw))) - && self.look_ahead(3, |t| t.is_keyword_case(kw::Fn, case)))) + && self.tree_look_ahead(2, |tt| { + match tt { + TokenTree::Token(t, _) => + ALL_QUALS.iter().any(|exp| { + t.is_keyword(exp.kw) + }), + TokenTree::Delimited(..) => false, + } + }) == Some(true) + && self.tree_look_ahead(3, |tt| { + match tt { + TokenTree::Token(t, _) => t.is_keyword_case(kw::Fn, case), + TokenTree::Delimited(..) => false, + } + }) == Some(true) + ) + ) } /// Parses all the "front matter" (or "qualifiers") for a `fn` declaration, diff --git a/compiler/rustc_parse/src/parser/mod.rs b/compiler/rustc_parse/src/parser/mod.rs index d865fd42764..a08fbde52a7 100644 --- a/compiler/rustc_parse/src/parser/mod.rs +++ b/compiler/rustc_parse/src/parser/mod.rs @@ -24,8 +24,8 @@ pub use pat::{CommaRecoveryMode, RecoverColon, RecoverComma}; use path::PathStyle; use rustc_ast::ptr::P; use rustc_ast::token::{ - self, Delimiter, IdentIsRaw, InvisibleOrigin, MetaVarKind, Nonterminal, NtPatKind, Token, - TokenKind, + self, Delimiter, IdentIsRaw, InvisibleOrigin, MetaVarKind, Nonterminal, NtExprKind, NtPatKind, + Token, TokenKind, }; use rustc_ast::tokenstream::{AttrsTarget, Spacing, TokenStream, TokenTree}; use rustc_ast::util::case::Case; @@ -101,6 +101,7 @@ pub enum ForceCollect { #[macro_export] macro_rules! maybe_whole { ($p:expr, $constructor:ident, |$x:ident| $e:expr) => { + #[allow(irrefutable_let_patterns)] // FIXME: temporary if let token::Interpolated(nt) = &$p.token.kind && let token::$constructor(x) = &**nt { @@ -299,6 +300,10 @@ impl TokenTreeCursor { self.stream.get(self.index) } + fn look_ahead(&self, n: usize) -> Option<&TokenTree> { + self.stream.get(self.index + n) + } + #[inline] fn bump(&mut self) { self.index += 1; @@ -1290,6 +1295,17 @@ impl<'a> Parser<'a> { looker(&token) } + /// Like `lookahead`, but skips over token trees rather than tokens. Useful + /// when looking past possible metavariable pasting sites. + pub fn tree_look_ahead<R>( + &self, + dist: usize, + looker: impl FnOnce(&TokenTree) -> R, + ) -> Option<R> { + assert_ne!(dist, 0); + self.token_cursor.curr.look_ahead(dist - 1).map(looker) + } + /// Returns whether any of the given keywords are `dist` tokens ahead of the current one. pub(crate) fn is_keyword_ahead(&self, dist: usize, kws: &[Symbol]) -> bool { self.look_ahead(dist, |t| kws.iter().any(|&kw| t.is_keyword(kw))) @@ -1706,6 +1722,16 @@ impl<'a> Parser<'a> { pub fn approx_token_stream_pos(&self) -> u32 { self.num_bump_calls } + + pub fn uninterpolated_token_span(&self) -> Span { + match &self.token.kind { + token::Interpolated(nt) => nt.use_span(), + token::OpenDelim(Delimiter::Invisible(InvisibleOrigin::MetaVar(_))) => { + self.look_ahead(1, |t| t.span) + } + _ => self.token.span, + } + } } pub(crate) fn make_unclosed_delims_error( @@ -1758,6 +1784,8 @@ pub enum ParseNtResult { Item(P<ast::Item>), Stmt(P<ast::Stmt>), Pat(P<ast::Pat>, NtPatKind), + Expr(P<ast::Expr>, NtExprKind), + Literal(P<ast::Expr>), Ty(P<ast::Ty>), Meta(P<ast::AttrItem>), Path(P<ast::Path>), diff --git a/compiler/rustc_parse/src/parser/nonterminal.rs b/compiler/rustc_parse/src/parser/nonterminal.rs index 1123755ce00..b4e540d670d 100644 --- a/compiler/rustc_parse/src/parser/nonterminal.rs +++ b/compiler/rustc_parse/src/parser/nonterminal.rs @@ -48,10 +48,6 @@ impl<'a> Parser<'a> { /// Old variant of `may_be_ident`. Being phased out. fn nt_may_be_ident(nt: &Nonterminal) -> bool { match nt { - NtExpr(_) - | NtLiteral(_) // `true`, `false` - => true, - NtBlock(_) => false, } } @@ -95,7 +91,7 @@ impl<'a> Parser<'a> { token::OpenDelim(Delimiter::Brace) => true, token::NtLifetime(..) => true, token::Interpolated(nt) => match &**nt { - NtBlock(_) | NtExpr(_) | NtLiteral(_) => true, + NtBlock(_) => true, }, token::OpenDelim(Delimiter::Invisible(InvisibleOrigin::MetaVar(k))) => match k { MetaVarKind::Block @@ -179,10 +175,14 @@ impl<'a> Parser<'a> { pat_kind, )); } - NonterminalKind::Expr(_) => NtExpr(self.parse_expr_force_collect()?), + NonterminalKind::Expr(expr_kind) => { + return Ok(ParseNtResult::Expr(self.parse_expr_force_collect()?, expr_kind)); + } NonterminalKind::Literal => { - // The `:literal` matcher does not support attributes - NtLiteral(self.collect_tokens_no_attrs(|this| this.parse_literal_maybe_minus())?) + // The `:literal` matcher does not support attributes. + return Ok(ParseNtResult::Literal( + self.collect_tokens_no_attrs(|this| this.parse_literal_maybe_minus())?, + )); } NonterminalKind::Ty => { return Ok(ParseNtResult::Ty( diff --git a/compiler/rustc_parse/src/parser/pat.rs b/compiler/rustc_parse/src/parser/pat.rs index 174cc929fa7..9612f71b2af 100644 --- a/compiler/rustc_parse/src/parser/pat.rs +++ b/compiler/rustc_parse/src/parser/pat.rs @@ -1252,7 +1252,7 @@ impl<'a> Parser<'a> { || *t == token::Dot // e.g. `.5` for recovery; || matches!(t.kind, token::Literal(..) | token::Minus) || t.is_bool_lit() - || t.is_whole_expr() + || t.is_metavar_expr() || t.is_lifetime() // recover `'a` instead of `'a'` || (self.may_recover() // recover leading `(` && *t == token::OpenDelim(Delimiter::Parenthesis) |