13 use crate::lexer::UnmatchedBrace;
14 pub use attr_wrapper::AttrWrapper;
15 pub use diagnostics::AttemptLocalParseRecovery;
16 use diagnostics::Error;
17 pub(crate) use item::FnParseMode;
18 pub use pat::{CommaRecoveryMode, RecoverColon, RecoverComma};
19 pub use path::PathStyle;
21 use rustc_ast::ptr::P;
22 use rustc_ast::token::{self, DelimToken, Nonterminal, Token, TokenKind};
23 use rustc_ast::tokenstream::AttributesData;
24 use rustc_ast::tokenstream::{self, DelimSpan, Spacing};
25 use rustc_ast::tokenstream::{TokenStream, TokenTree};
26 use rustc_ast::AttrId;
27 use rustc_ast::DUMMY_NODE_ID;
28 use rustc_ast::{self as ast, AnonConst, AstLike, AttrStyle, AttrVec, Const, CrateSugar, Extern};
29 use rustc_ast::{Async, Expr, ExprKind, MacArgs, MacDelimiter, Mutability, StrLit, Unsafe};
30 use rustc_ast::{Visibility, VisibilityKind};
31 use rustc_ast_pretty::pprust;
32 use rustc_data_structures::fx::FxHashMap;
33 use rustc_data_structures::sync::Lrc;
34 use rustc_errors::PResult;
36 struct_span_err, Applicability, DiagnosticBuilder, ErrorGuaranteed, FatalError, MultiSpan,
38 use rustc_session::parse::ParseSess;
39 use rustc_span::source_map::{Span, DUMMY_SP};
40 use rustc_span::symbol::{kw, sym, Ident, Symbol};
44 use std::{cmp, mem, slice};
47 struct Restrictions: u8 {
48 const STMT_EXPR = 1 << 0;
49 const NO_STRUCT_LITERAL = 1 << 1;
50 const CONST_EXPR = 1 << 2;
54 #[derive(Clone, Copy, PartialEq, Debug)]
61 #[derive(Clone, Copy, PartialEq, Debug)]
67 /// Whether or not we should force collection of tokens for an AST node,
68 /// regardless of whether or not it has attributes
69 #[derive(Clone, Copy, PartialEq)]
70 pub enum ForceCollect {
75 #[derive(Debug, Eq, PartialEq)]
76 pub enum TrailingToken {
79 /// If the trailing token is a comma, then capture it
80 /// Otherwise, ignore the trailing token
84 /// Like `maybe_whole_expr`, but for things other than expressions.
86 macro_rules! maybe_whole {
87 ($p:expr, $constructor:ident, |$x:ident| $e:expr) => {
88 if let token::Interpolated(nt) = &$p.token.kind {
89 if let token::$constructor(x) = &**nt {
98 /// If the next tokens are ill-formed `$ty::` recover them as `<$ty>::`.
100 macro_rules! maybe_recover_from_interpolated_ty_qpath {
101 ($self: expr, $allow_qpath_recovery: expr) => {
102 if $allow_qpath_recovery
103 && $self.look_ahead(1, |t| t == &token::ModSep)
104 && let token::Interpolated(nt) = &$self.token.kind
105 && let token::NtTy(ty) = &**nt
109 return $self.maybe_recover_from_bad_qpath_stage_2($self.prev_token.span, ty);
115 pub struct Parser<'a> {
116 pub sess: &'a ParseSess,
117 /// The current token.
119 /// The spacing for the current token
120 pub token_spacing: Spacing,
121 /// The previous token.
122 pub prev_token: Token,
123 pub capture_cfg: bool,
124 restrictions: Restrictions,
125 expected_tokens: Vec<TokenType>,
126 // Important: This must only be advanced from `next_tok`
127 // to ensure that `token_cursor.num_next_calls` is updated properly
128 token_cursor: TokenCursor,
129 desugar_doc_comments: bool,
130 /// This field is used to keep track of how many left angle brackets we have seen. This is
131 /// required in order to detect extra leading left angle brackets (`<` characters) and error
134 /// See the comments in the `parse_path_segment` function for more details.
135 unmatched_angle_bracket_count: u32,
136 max_angle_bracket_count: u32,
137 /// A list of all unclosed delimiters found by the lexer. If an entry is used for error recovery
138 /// it gets removed from here. Every entry left at the end gets emitted as an independent
140 pub(super) unclosed_delims: Vec<UnmatchedBrace>,
141 last_unexpected_token_span: Option<Span>,
142 /// Span pointing at the `:` for the last type ascription the parser has seen, and whether it
143 /// looked like it could have been a mistyped path or literal `Option:Some(42)`).
144 pub last_type_ascription: Option<(Span, bool /* likely path typo */)>,
145 /// If present, this `Parser` is not parsing Rust code but rather a macro call.
146 subparser_name: Option<&'static str>,
147 capture_state: CaptureState,
148 /// This allows us to recover when the user forget to add braces around
149 /// multiple statements in the closure body.
150 pub current_closure: Option<ClosureSpans>,
153 /// Stores span information about a closure.
155 pub struct ClosureSpans {
156 pub whole_closure: Span,
157 pub closing_pipe: Span,
161 /// Indicates a range of tokens that should be replaced by
162 /// the tokens in the provided vector. This is used in two
163 /// places during token collection:
165 /// 1. During the parsing of an AST node that may have a `#[derive]`
166 /// attribute, we parse a nested AST node that has `#[cfg]` or `#[cfg_attr]`
167 /// In this case, we use a `ReplaceRange` to replace the entire inner AST node
168 /// with `FlatToken::AttrTarget`, allowing us to perform eager cfg-expansion
169 /// on an `AttrAnnotatedTokenStream`
171 /// 2. When we parse an inner attribute while collecting tokens. We
172 /// remove inner attributes from the token stream entirely, and
173 /// instead track them through the `attrs` field on the AST node.
174 /// This allows us to easily manipulate them (for example, removing
175 /// the first macro inner attribute to invoke a proc-macro).
176 /// When create a `TokenStream`, the inner attributes get inserted
177 /// into the proper place in the token stream.
178 pub type ReplaceRange = (Range<u32>, Vec<(FlatToken, Spacing)>);
180 /// Controls how we capture tokens. Capturing can be expensive,
181 /// so we try to avoid performing capturing in cases where
182 /// we will never need an `AttrAnnotatedTokenStream`
183 #[derive(Copy, Clone)]
185 /// We aren't performing any capturing - this is the default mode.
187 /// We are capturing tokens
192 struct CaptureState {
193 capturing: Capturing,
194 replace_ranges: Vec<ReplaceRange>,
195 inner_attr_ranges: FxHashMap<AttrId, ReplaceRange>,
198 impl<'a> Drop for Parser<'a> {
200 emit_unclosed_delims(&mut self.unclosed_delims, &self.sess);
206 frame: TokenCursorFrame,
207 stack: Vec<TokenCursorFrame>,
208 desugar_doc_comments: bool,
209 // Counts the number of calls to `{,inlined_}next` or
210 // `{,inlined_}next_desugared`, depending on whether
211 // `desugar_doc_comments` is set.
212 num_next_calls: usize,
213 // During parsing, we may sometimes need to 'unglue' a
214 // glued token into two component tokens
215 // (e.g. '>>' into '>' and '>), so that the parser
216 // can consume them one at a time. This process
217 // bypasses the normal capturing mechanism
218 // (e.g. `num_next_calls` will not be incremented),
219 // since the 'unglued' tokens due not exist in
220 // the original `TokenStream`.
222 // If we end up consuming both unglued tokens,
223 // then this is not an issue - we'll end up
224 // capturing the single 'glued' token.
226 // However, in certain circumstances, we may
227 // want to capture just the first 'unglued' token.
228 // For example, capturing the `Vec<u8>`
229 // in `Option<Vec<u8>>` requires us to unglue
230 // the trailing `>>` token. The `break_last_token`
231 // field is used to track this token - it gets
232 // appended to the captured stream when
233 // we evaluate a `LazyTokenStream`
234 break_last_token: bool,
238 struct TokenCursorFrame {
239 delim: token::DelimToken,
242 tree_cursor: tokenstream::Cursor,
246 impl TokenCursorFrame {
247 fn new(span: DelimSpan, delim: DelimToken, tts: TokenStream) -> Self {
252 tree_cursor: tts.into_trees(),
259 fn next(&mut self) -> (Token, Spacing) {
263 /// This always-inlined version should only be used on hot code paths.
265 fn inlined_next(&mut self) -> (Token, Spacing) {
267 let (tree, spacing) = if !self.frame.open_delim {
268 self.frame.open_delim = true;
269 TokenTree::open_tt(self.frame.span, self.frame.delim).into()
270 } else if let Some(tree) = self.frame.tree_cursor.next_with_spacing() {
272 } else if !self.frame.close_delim {
273 self.frame.close_delim = true;
274 TokenTree::close_tt(self.frame.span, self.frame.delim).into()
275 } else if let Some(frame) = self.stack.pop() {
279 (TokenTree::Token(Token::new(token::Eof, DUMMY_SP)), Spacing::Alone)
283 TokenTree::Token(token) => {
284 return (token, spacing);
286 TokenTree::Delimited(sp, delim, tts) => {
287 let frame = TokenCursorFrame::new(sp, delim, tts);
288 self.stack.push(mem::replace(&mut self.frame, frame));
294 fn next_desugared(&mut self) -> (Token, Spacing) {
295 self.inlined_next_desugared()
298 /// This always-inlined version should only be used on hot code paths.
300 fn inlined_next_desugared(&mut self) -> (Token, Spacing) {
301 let (data, attr_style, sp) = match self.inlined_next() {
302 (Token { kind: token::DocComment(_, attr_style, data), span }, _) => {
303 (data, attr_style, span)
308 // Searches for the occurrences of `"#*` and returns the minimum number of `#`s
309 // required to wrap the text.
310 let mut num_of_hashes = 0;
312 for ch in data.as_str().chars() {
315 '#' if count > 0 => count + 1,
318 num_of_hashes = cmp::max(num_of_hashes, count);
321 let delim_span = DelimSpan::from_single(sp);
322 let body = TokenTree::Delimited(
326 TokenTree::token(token::Ident(sym::doc, false), sp),
327 TokenTree::token(token::Eq, sp),
328 TokenTree::token(TokenKind::lit(token::StrRaw(num_of_hashes), data, None), sp),
332 .collect::<TokenStream>(),
335 self.stack.push(mem::replace(
337 TokenCursorFrame::new(
340 if attr_style == AttrStyle::Inner {
341 [TokenTree::token(token::Pound, sp), TokenTree::token(token::Not, sp), body]
344 .collect::<TokenStream>()
346 [TokenTree::token(token::Pound, sp), body]
349 .collect::<TokenStream>()
358 #[derive(Debug, Clone, PartialEq)]
371 fn to_string(&self) -> String {
373 TokenType::Token(ref t) => format!("`{}`", pprust::token_kind_to_string(t)),
374 TokenType::Keyword(kw) => format!("`{}`", kw),
375 TokenType::Operator => "an operator".to_string(),
376 TokenType::Lifetime => "lifetime".to_string(),
377 TokenType::Ident => "identifier".to_string(),
378 TokenType::Path => "path".to_string(),
379 TokenType::Type => "type".to_string(),
380 TokenType::Const => "a const expression".to_string(),
385 #[derive(Copy, Clone, Debug)]
386 enum TokenExpectType {
391 /// A sequence separator.
393 /// The separator token.
394 sep: Option<TokenKind>,
395 /// `true` if a trailing separator is allowed.
396 trailing_sep_allowed: bool,
400 fn trailing_allowed(t: TokenKind) -> SeqSep {
401 SeqSep { sep: Some(t), trailing_sep_allowed: true }
404 fn none() -> SeqSep {
405 SeqSep { sep: None, trailing_sep_allowed: false }
409 pub enum FollowedByType {
414 fn token_descr_opt(token: &Token) -> Option<&'static str> {
415 Some(match token.kind {
416 _ if token.is_special_ident() => "reserved identifier",
417 _ if token.is_used_keyword() => "keyword",
418 _ if token.is_unused_keyword() => "reserved keyword",
419 token::DocComment(..) => "doc comment",
424 pub(super) fn token_descr(token: &Token) -> String {
425 let token_str = pprust::token_to_string(token);
426 match token_descr_opt(token) {
427 Some(prefix) => format!("{} `{}`", prefix, token_str),
428 _ => format!("`{}`", token_str),
432 impl<'a> Parser<'a> {
436 desugar_doc_comments: bool,
437 subparser_name: Option<&'static str>,
439 let mut start_frame = TokenCursorFrame::new(DelimSpan::dummy(), token::NoDelim, tokens);
440 start_frame.open_delim = true;
441 start_frame.close_delim = true;
443 let mut parser = Parser {
445 token: Token::dummy(),
446 token_spacing: Spacing::Alone,
447 prev_token: Token::dummy(),
449 restrictions: Restrictions::empty(),
450 expected_tokens: Vec::new(),
451 token_cursor: TokenCursor {
455 desugar_doc_comments,
456 break_last_token: false,
458 desugar_doc_comments,
459 unmatched_angle_bracket_count: 0,
460 max_angle_bracket_count: 0,
461 unclosed_delims: Vec::new(),
462 last_unexpected_token_span: None,
463 last_type_ascription: None,
465 capture_state: CaptureState {
466 capturing: Capturing::No,
467 replace_ranges: Vec::new(),
468 inner_attr_ranges: Default::default(),
470 current_closure: None,
473 // Make parser point to the first token.
480 fn next_tok(&mut self, fallback_span: Span) -> (Token, Spacing) {
482 let (mut next, spacing) = if self.desugar_doc_comments {
483 self.token_cursor.inlined_next_desugared()
485 self.token_cursor.inlined_next()
487 self.token_cursor.num_next_calls += 1;
488 // We've retrieved an token from the underlying
489 // cursor, so we no longer need to worry about
490 // an unglued token. See `break_and_eat` for more details
491 self.token_cursor.break_last_token = false;
492 if next.span.is_dummy() {
493 // Tweak the location for better diagnostics, but keep syntactic context intact.
494 next.span = fallback_span.with_ctxt(next.span.ctxt());
498 token::OpenDelim(token::NoDelim) | token::CloseDelim(token::NoDelim)
502 return (next, spacing);
506 pub fn unexpected<T>(&mut self) -> PResult<'a, T> {
507 match self.expect_one_of(&[], &[]) {
509 // We can get `Ok(true)` from `recover_closing_delimiter`
510 // which is called in `expected_one_of_not_found`.
511 Ok(_) => FatalError.raise(),
515 /// Expects and consumes the token `t`. Signals an error if the next token is not `t`.
516 pub fn expect(&mut self, t: &TokenKind) -> PResult<'a, bool /* recovered */> {
517 if self.expected_tokens.is_empty() {
518 if self.token == *t {
522 self.unexpected_try_recover(t)
525 self.expect_one_of(slice::from_ref(t), &[])
529 /// Expect next token to be edible or inedible token. If edible,
530 /// then consume it; if inedible, then return without consuming
531 /// anything. Signal a fatal error if next token is unexpected.
532 pub fn expect_one_of(
534 edible: &[TokenKind],
535 inedible: &[TokenKind],
536 ) -> PResult<'a, bool /* recovered */> {
537 if edible.contains(&self.token.kind) {
540 } else if inedible.contains(&self.token.kind) {
541 // leave it in the input
543 } else if self.last_unexpected_token_span == Some(self.token.span) {
546 self.expected_one_of_not_found(edible, inedible)
550 // Public for rustfmt usage.
551 pub fn parse_ident(&mut self) -> PResult<'a, Ident> {
552 self.parse_ident_common(true)
555 fn ident_or_err(&mut self) -> PResult<'a, (Ident, /* is_raw */ bool)> {
556 self.token.ident().ok_or_else(|| match self.prev_token.kind {
557 TokenKind::DocComment(..) => {
558 self.span_err(self.prev_token.span, Error::UselessDocComment)
560 _ => self.expected_ident_found(),
564 fn parse_ident_common(&mut self, recover: bool) -> PResult<'a, Ident> {
565 let (ident, is_raw) = self.ident_or_err()?;
566 if !is_raw && ident.is_reserved() {
567 let mut err = self.expected_ident_found();
578 /// Checks if the next token is `tok`, and returns `true` if so.
580 /// This method will automatically add `tok` to `expected_tokens` if `tok` is not
582 fn check(&mut self, tok: &TokenKind) -> bool {
583 let is_present = self.token == *tok;
585 self.expected_tokens.push(TokenType::Token(tok.clone()));
590 /// Consumes a token 'tok' if it exists. Returns whether the given token was present.
591 pub fn eat(&mut self, tok: &TokenKind) -> bool {
592 let is_present = self.check(tok);
599 /// If the next token is the given keyword, returns `true` without eating it.
600 /// An expectation is also added for diagnostics purposes.
601 fn check_keyword(&mut self, kw: Symbol) -> bool {
602 self.expected_tokens.push(TokenType::Keyword(kw));
603 self.token.is_keyword(kw)
606 /// If the next token is the given keyword, eats it and returns `true`.
607 /// Otherwise, returns `false`. An expectation is also added for diagnostics purposes.
608 // Public for rustfmt usage.
609 pub fn eat_keyword(&mut self, kw: Symbol) -> bool {
610 if self.check_keyword(kw) {
618 fn eat_keyword_noexpect(&mut self, kw: Symbol) -> bool {
619 if self.token.is_keyword(kw) {
627 /// If the given word is not a keyword, signals an error.
628 /// If the next token is not the given word, signals an error.
629 /// Otherwise, eats it.
630 fn expect_keyword(&mut self, kw: Symbol) -> PResult<'a, ()> {
631 if !self.eat_keyword(kw) { self.unexpected() } else { Ok(()) }
634 /// Is the given keyword `kw` followed by a non-reserved identifier?
635 fn is_kw_followed_by_ident(&self, kw: Symbol) -> bool {
636 self.token.is_keyword(kw) && self.look_ahead(1, |t| t.is_ident() && !t.is_reserved_ident())
639 fn check_or_expected(&mut self, ok: bool, typ: TokenType) -> bool {
643 self.expected_tokens.push(typ);
648 fn check_ident(&mut self) -> bool {
649 self.check_or_expected(self.token.is_ident(), TokenType::Ident)
652 fn check_path(&mut self) -> bool {
653 self.check_or_expected(self.token.is_path_start(), TokenType::Path)
656 fn check_type(&mut self) -> bool {
657 self.check_or_expected(self.token.can_begin_type(), TokenType::Type)
660 fn check_const_arg(&mut self) -> bool {
661 self.check_or_expected(self.token.can_begin_const_arg(), TokenType::Const)
664 fn check_inline_const(&self, dist: usize) -> bool {
665 self.is_keyword_ahead(dist, &[kw::Const])
666 && self.look_ahead(dist + 1, |t| match t.kind {
667 token::Interpolated(ref nt) => matches!(**nt, token::NtBlock(..)),
668 token::OpenDelim(DelimToken::Brace) => true,
673 /// Checks to see if the next token is either `+` or `+=`.
674 /// Otherwise returns `false`.
675 fn check_plus(&mut self) -> bool {
676 self.check_or_expected(
677 self.token.is_like_plus(),
678 TokenType::Token(token::BinOp(token::Plus)),
682 /// Eats the expected token if it's present possibly breaking
683 /// compound tokens like multi-character operators in process.
684 /// Returns `true` if the token was eaten.
685 fn break_and_eat(&mut self, expected: TokenKind) -> bool {
686 if self.token.kind == expected {
690 match self.token.kind.break_two_token_op() {
691 Some((first, second)) if first == expected => {
692 let first_span = self.sess.source_map().start_point(self.token.span);
693 let second_span = self.token.span.with_lo(first_span.hi());
694 self.token = Token::new(first, first_span);
695 // Keep track of this token - if we end token capturing now,
696 // we'll want to append this token to the captured stream.
698 // If we consume any additional tokens, then this token
699 // is not needed (we'll capture the entire 'glued' token),
700 // and `next_tok` will set this field to `None`
701 self.token_cursor.break_last_token = true;
702 // Use the spacing of the glued token as the spacing
703 // of the unglued second token.
704 self.bump_with((Token::new(second, second_span), self.token_spacing));
708 self.expected_tokens.push(TokenType::Token(expected));
714 /// Eats `+` possibly breaking tokens like `+=` in process.
715 fn eat_plus(&mut self) -> bool {
716 self.break_and_eat(token::BinOp(token::Plus))
719 /// Eats `&` possibly breaking tokens like `&&` in process.
720 /// Signals an error if `&` is not eaten.
721 fn expect_and(&mut self) -> PResult<'a, ()> {
722 if self.break_and_eat(token::BinOp(token::And)) { Ok(()) } else { self.unexpected() }
725 /// Eats `|` possibly breaking tokens like `||` in process.
726 /// Signals an error if `|` was not eaten.
727 fn expect_or(&mut self) -> PResult<'a, ()> {
728 if self.break_and_eat(token::BinOp(token::Or)) { Ok(()) } else { self.unexpected() }
731 /// Eats `<` possibly breaking tokens like `<<` in process.
732 fn eat_lt(&mut self) -> bool {
733 let ate = self.break_and_eat(token::Lt);
735 // See doc comment for `unmatched_angle_bracket_count`.
736 self.unmatched_angle_bracket_count += 1;
737 self.max_angle_bracket_count += 1;
738 debug!("eat_lt: (increment) count={:?}", self.unmatched_angle_bracket_count);
743 /// Eats `<` possibly breaking tokens like `<<` in process.
744 /// Signals an error if `<` was not eaten.
745 fn expect_lt(&mut self) -> PResult<'a, ()> {
746 if self.eat_lt() { Ok(()) } else { self.unexpected() }
749 /// Eats `>` possibly breaking tokens like `>>` in process.
750 /// Signals an error if `>` was not eaten.
751 fn expect_gt(&mut self) -> PResult<'a, ()> {
752 if self.break_and_eat(token::Gt) {
753 // See doc comment for `unmatched_angle_bracket_count`.
754 if self.unmatched_angle_bracket_count > 0 {
755 self.unmatched_angle_bracket_count -= 1;
756 debug!("expect_gt: (decrement) count={:?}", self.unmatched_angle_bracket_count);
764 fn expect_any_with_type(&mut self, kets: &[&TokenKind], expect: TokenExpectType) -> bool {
765 kets.iter().any(|k| match expect {
766 TokenExpectType::Expect => self.check(k),
767 TokenExpectType::NoExpect => self.token == **k,
771 fn parse_seq_to_before_tokens<T>(
775 expect: TokenExpectType,
776 mut f: impl FnMut(&mut Parser<'a>) -> PResult<'a, T>,
777 ) -> PResult<'a, (Vec<T>, bool /* trailing */, bool /* recovered */)> {
778 let mut first = true;
779 let mut recovered = false;
780 let mut trailing = false;
782 let unclosed_delims = !self.unclosed_delims.is_empty();
784 while !self.expect_any_with_type(kets, expect) {
785 if let token::CloseDelim(..) | token::Eof = self.token.kind {
788 if let Some(ref t) = sep.sep {
792 match self.expect(t) {
794 self.current_closure.take();
797 self.current_closure.take();
801 Err(mut expect_err) => {
802 let sp = self.prev_token.span.shrink_to_hi();
803 let token_str = pprust::token_kind_to_string(t);
805 match self.current_closure.take() {
806 Some(closure_spans) if self.token.kind == TokenKind::Semi => {
807 // Finding a semicolon instead of a comma
808 // after a closure body indicates that the
809 // closure body may be a block but the user
810 // forgot to put braces around its
813 self.recover_missing_braces_around_closure_body(
822 // Attempt to keep parsing if it was a similar separator.
823 if let Some(ref tokens) = t.similar_tokens() {
824 if tokens.contains(&self.token.kind) && !unclosed_delims {
831 // If this was a missing `@` in a binding pattern
832 // bail with a suggestion
833 // https://github.com/rust-lang/rust/issues/72373
834 if self.prev_token.is_ident() && self.token.kind == token::DotDot {
836 "if you meant to bind the contents of \
837 the rest of the array pattern into `{}`, use `@`",
838 pprust::token_to_string(&self.prev_token)
841 .span_suggestion_verbose(
842 self.prev_token.span.shrink_to_hi().until(self.token.span),
845 Applicability::MaybeIncorrect,
851 // Attempt to keep parsing if it was an omitted separator.
854 // Parsed successfully, therefore most probably the code only
855 // misses a separator.
857 .span_suggestion_short(
859 &format!("missing `{}`", token_str),
861 Applicability::MaybeIncorrect,
869 // Parsing failed, therefore it must be something more serious
870 // than just a missing separator.
881 if sep.trailing_sep_allowed && self.expect_any_with_type(kets, expect) {
890 Ok((v, trailing, recovered))
893 fn recover_missing_braces_around_closure_body(
895 closure_spans: ClosureSpans,
896 mut expect_err: DiagnosticBuilder<'_, ErrorGuaranteed>,
897 ) -> PResult<'a, ()> {
898 let initial_semicolon = self.token.span;
900 while self.eat(&TokenKind::Semi) {
901 let _ = self.parse_stmt(ForceCollect::Yes)?;
904 expect_err.set_primary_message(
905 "closure bodies that contain statements must be surrounded by braces",
908 let preceding_pipe_span = closure_spans.closing_pipe;
909 let following_token_span = self.token.span;
911 let mut first_note = MultiSpan::from(vec![initial_semicolon]);
912 first_note.push_span_label(
914 "this `;` turns the preceding closure into a statement".to_string(),
916 first_note.push_span_label(
918 "this expression is a statement because of the trailing semicolon".to_string(),
920 expect_err.span_note(first_note, "statement found outside of a block");
922 let mut second_note = MultiSpan::from(vec![closure_spans.whole_closure]);
923 second_note.push_span_label(
924 closure_spans.whole_closure,
925 "this is the parsed closure...".to_string(),
927 second_note.push_span_label(
928 following_token_span,
929 "...but likely you meant the closure to end here".to_string(),
931 expect_err.span_note(second_note, "the closure body may be incorrectly delimited");
933 expect_err.set_span(vec![preceding_pipe_span, following_token_span]);
935 let opening_suggestion_str = " {".to_string();
936 let closing_suggestion_str = "}".to_string();
938 expect_err.multipart_suggestion(
941 (preceding_pipe_span.shrink_to_hi(), opening_suggestion_str),
942 (following_token_span.shrink_to_lo(), closing_suggestion_str),
944 Applicability::MaybeIncorrect,
952 /// Parses a sequence, not including the closing delimiter. The function
953 /// `f` must consume tokens until reaching the next separator or
955 fn parse_seq_to_before_end<T>(
959 f: impl FnMut(&mut Parser<'a>) -> PResult<'a, T>,
960 ) -> PResult<'a, (Vec<T>, bool, bool)> {
961 self.parse_seq_to_before_tokens(&[ket], sep, TokenExpectType::Expect, f)
964 /// Parses a sequence, including the closing delimiter. The function
965 /// `f` must consume tokens until reaching the next separator or
967 fn parse_seq_to_end<T>(
971 f: impl FnMut(&mut Parser<'a>) -> PResult<'a, T>,
972 ) -> PResult<'a, (Vec<T>, bool /* trailing */)> {
973 let (val, trailing, recovered) = self.parse_seq_to_before_end(ket, sep, f)?;
980 /// Parses a sequence, including the closing delimiter. The function
981 /// `f` must consume tokens until reaching the next separator or
983 fn parse_unspanned_seq<T>(
988 f: impl FnMut(&mut Parser<'a>) -> PResult<'a, T>,
989 ) -> PResult<'a, (Vec<T>, bool)> {
991 self.parse_seq_to_end(ket, sep, f)
994 fn parse_delim_comma_seq<T>(
997 f: impl FnMut(&mut Parser<'a>) -> PResult<'a, T>,
998 ) -> PResult<'a, (Vec<T>, bool)> {
999 self.parse_unspanned_seq(
1000 &token::OpenDelim(delim),
1001 &token::CloseDelim(delim),
1002 SeqSep::trailing_allowed(token::Comma),
1007 fn parse_paren_comma_seq<T>(
1009 f: impl FnMut(&mut Parser<'a>) -> PResult<'a, T>,
1010 ) -> PResult<'a, (Vec<T>, bool)> {
1011 self.parse_delim_comma_seq(token::Paren, f)
1014 /// Advance the parser by one token using provided token as the next one.
1015 fn bump_with(&mut self, next: (Token, Spacing)) {
1016 self.inlined_bump_with(next)
1019 /// This always-inlined version should only be used on hot code paths.
1021 fn inlined_bump_with(&mut self, (next_token, next_spacing): (Token, Spacing)) {
1022 // Bumping after EOF is a bad sign, usually an infinite loop.
1023 if self.prev_token.kind == TokenKind::Eof {
1024 let msg = "attempted to bump the parser past EOF (may be stuck in a loop)";
1025 self.span_bug(self.token.span, msg);
1028 // Update the current and previous tokens.
1029 self.prev_token = mem::replace(&mut self.token, next_token);
1030 self.token_spacing = next_spacing;
1033 self.expected_tokens.clear();
1036 /// Advance the parser by one token.
1037 pub fn bump(&mut self) {
1038 let next_token = self.next_tok(self.token.span);
1039 self.inlined_bump_with(next_token);
1042 /// Look-ahead `dist` tokens of `self.token` and get access to that token there.
1043 /// When `dist == 0` then the current token is looked at.
1044 pub fn look_ahead<R>(&self, dist: usize, looker: impl FnOnce(&Token) -> R) -> R {
1046 return looker(&self.token);
1049 let frame = &self.token_cursor.frame;
1050 if frame.delim != DelimToken::NoDelim {
1051 let all_normal = (0..dist).all(|i| {
1052 let token = frame.tree_cursor.look_ahead(i);
1053 !matches!(token, Some(TokenTree::Delimited(_, DelimToken::NoDelim, _)))
1056 return match frame.tree_cursor.look_ahead(dist - 1) {
1057 Some(tree) => match tree {
1058 TokenTree::Token(token) => looker(token),
1059 TokenTree::Delimited(dspan, delim, _) => {
1060 looker(&Token::new(token::OpenDelim(*delim), dspan.open))
1063 None => looker(&Token::new(token::CloseDelim(frame.delim), frame.span.close)),
1068 let mut cursor = self.token_cursor.clone();
1070 let mut token = Token::dummy();
1072 token = cursor.next().0;
1075 token::OpenDelim(token::NoDelim) | token::CloseDelim(token::NoDelim)
1081 return looker(&token);
1084 /// Returns whether any of the given keywords are `dist` tokens ahead of the current one.
1085 fn is_keyword_ahead(&self, dist: usize, kws: &[Symbol]) -> bool {
1086 self.look_ahead(dist, |t| kws.iter().any(|&kw| t.is_keyword(kw)))
1089 /// Parses asyncness: `async` or nothing.
1090 fn parse_asyncness(&mut self) -> Async {
1091 if self.eat_keyword(kw::Async) {
1092 let span = self.prev_token.uninterpolated_span();
1093 Async::Yes { span, closure_id: DUMMY_NODE_ID, return_impl_trait_id: DUMMY_NODE_ID }
1099 /// Parses unsafety: `unsafe` or nothing.
1100 fn parse_unsafety(&mut self) -> Unsafe {
1101 if self.eat_keyword(kw::Unsafe) {
1102 Unsafe::Yes(self.prev_token.uninterpolated_span())
1108 /// Parses constness: `const` or nothing.
1109 fn parse_constness(&mut self) -> Const {
1110 // Avoid const blocks to be parsed as const items
1111 if self.look_ahead(1, |t| t != &token::OpenDelim(DelimToken::Brace))
1112 && self.eat_keyword(kw::Const)
1114 Const::Yes(self.prev_token.uninterpolated_span())
1120 /// Parses inline const expressions.
1121 fn parse_const_block(&mut self, span: Span, pat: bool) -> PResult<'a, P<Expr>> {
1123 self.sess.gated_spans.gate(sym::inline_const_pat, span);
1125 self.sess.gated_spans.gate(sym::inline_const, span);
1127 self.eat_keyword(kw::Const);
1128 let blk = self.parse_block()?;
1129 let anon_const = AnonConst {
1131 value: self.mk_expr(blk.span, ExprKind::Block(blk, None), AttrVec::new()),
1133 let blk_span = anon_const.value.span;
1134 Ok(self.mk_expr(span.to(blk_span), ExprKind::ConstBlock(anon_const), AttrVec::new()))
1137 /// Parses mutability (`mut` or nothing).
1138 fn parse_mutability(&mut self) -> Mutability {
1139 if self.eat_keyword(kw::Mut) { Mutability::Mut } else { Mutability::Not }
1142 /// Possibly parses mutability (`const` or `mut`).
1143 fn parse_const_or_mut(&mut self) -> Option<Mutability> {
1144 if self.eat_keyword(kw::Mut) {
1145 Some(Mutability::Mut)
1146 } else if self.eat_keyword(kw::Const) {
1147 Some(Mutability::Not)
1153 fn parse_field_name(&mut self) -> PResult<'a, Ident> {
1154 if let token::Literal(token::Lit { kind: token::Integer, symbol, suffix }) = self.token.kind
1156 self.expect_no_suffix(self.token.span, "a tuple index", suffix);
1158 Ok(Ident::new(symbol, self.prev_token.span))
1160 self.parse_ident_common(true)
1164 fn parse_mac_args(&mut self) -> PResult<'a, P<MacArgs>> {
1165 self.parse_mac_args_common(true).map(P)
1168 fn parse_attr_args(&mut self) -> PResult<'a, MacArgs> {
1169 self.parse_mac_args_common(false)
1172 fn parse_mac_args_common(&mut self, delimited_only: bool) -> PResult<'a, MacArgs> {
1174 if self.check(&token::OpenDelim(DelimToken::Paren))
1175 || self.check(&token::OpenDelim(DelimToken::Bracket))
1176 || self.check(&token::OpenDelim(DelimToken::Brace))
1178 match self.parse_token_tree() {
1179 TokenTree::Delimited(dspan, delim, tokens) =>
1180 // We've confirmed above that there is a delimiter so unwrapping is OK.
1182 MacArgs::Delimited(dspan, MacDelimiter::from_token(delim).unwrap(), tokens)
1184 _ => unreachable!(),
1186 } else if !delimited_only {
1187 if self.eat(&token::Eq) {
1188 let eq_span = self.prev_token.span;
1190 // Collect tokens because they are used during lowering to HIR.
1191 let expr = self.parse_expr_force_collect()?;
1192 let span = expr.span;
1194 let token_kind = token::Interpolated(Lrc::new(token::NtExpr(expr)));
1195 MacArgs::Eq(eq_span, Token::new(token_kind, span))
1200 return self.unexpected();
1205 fn parse_or_use_outer_attributes(
1207 already_parsed_attrs: Option<AttrWrapper>,
1208 ) -> PResult<'a, AttrWrapper> {
1209 if let Some(attrs) = already_parsed_attrs {
1212 self.parse_outer_attributes()
1216 /// Parses a single token tree from the input.
1217 pub(crate) fn parse_token_tree(&mut self) -> TokenTree {
1218 match self.token.kind {
1219 token::OpenDelim(..) => {
1220 let depth = self.token_cursor.stack.len();
1222 // We keep advancing the token cursor until we hit
1223 // the matching `CloseDelim` token.
1224 while !(depth == self.token_cursor.stack.len()
1225 && matches!(self.token.kind, token::CloseDelim(_)))
1227 // Advance one token at a time, so `TokenCursor::next()`
1228 // can capture these tokens if necessary.
1231 // We are still inside the frame corresponding
1232 // to the delimited stream we captured, so grab
1233 // the tokens from this frame.
1234 let frame = &self.token_cursor.frame;
1235 let stream = frame.tree_cursor.stream.clone();
1236 let span = frame.span;
1237 let delim = frame.delim;
1238 // Consume close delimiter
1240 TokenTree::Delimited(span, delim, stream)
1242 token::CloseDelim(_) | token::Eof => unreachable!(),
1245 TokenTree::Token(self.prev_token.clone())
1250 /// Parses a stream of tokens into a list of `TokenTree`s, up to EOF.
1251 pub fn parse_all_token_trees(&mut self) -> PResult<'a, Vec<TokenTree>> {
1252 let mut tts = Vec::new();
1253 while self.token != token::Eof {
1254 tts.push(self.parse_token_tree());
1259 pub fn parse_tokens(&mut self) -> TokenStream {
1260 let mut result = Vec::new();
1262 match self.token.kind {
1263 token::Eof | token::CloseDelim(..) => break,
1264 _ => result.push(self.parse_token_tree().into()),
1267 TokenStream::new(result)
1270 /// Evaluates the closure with restrictions in place.
1272 /// Afters the closure is evaluated, restrictions are reset.
1273 fn with_res<T>(&mut self, res: Restrictions, f: impl FnOnce(&mut Self) -> T) -> T {
1274 let old = self.restrictions;
1275 self.restrictions = res;
1277 self.restrictions = old;
1281 fn is_crate_vis(&self) -> bool {
1282 self.token.is_keyword(kw::Crate) && self.look_ahead(1, |t| t != &token::ModSep)
1285 /// Parses `pub`, `pub(crate)` and `pub(in path)` plus shortcuts `crate` for `pub(crate)`,
1286 /// `pub(self)` for `pub(in self)` and `pub(super)` for `pub(in super)`.
1287 /// If the following element can't be a tuple (i.e., it's a function definition), then
1288 /// it's not a tuple struct field), and the contents within the parentheses aren't valid,
1289 /// so emit a proper diagnostic.
1290 // Public for rustfmt usage.
1291 pub fn parse_visibility(&mut self, fbt: FollowedByType) -> PResult<'a, Visibility> {
1292 maybe_whole!(self, NtVis, |x| x.into_inner());
1294 self.expected_tokens.push(TokenType::Keyword(kw::Crate));
1295 if self.is_crate_vis() {
1296 self.bump(); // `crate`
1297 self.sess.gated_spans.gate(sym::crate_visibility_modifier, self.prev_token.span);
1298 return Ok(Visibility {
1299 span: self.prev_token.span,
1300 kind: VisibilityKind::Crate(CrateSugar::JustCrate),
1305 if !self.eat_keyword(kw::Pub) {
1306 // We need a span for our `Spanned<VisibilityKind>`, but there's inherently no
1307 // keyword to grab a span from for inherited visibility; an empty span at the
1308 // beginning of the current token would seem to be the "Schelling span".
1309 return Ok(Visibility {
1310 span: self.token.span.shrink_to_lo(),
1311 kind: VisibilityKind::Inherited,
1315 let lo = self.prev_token.span;
1317 if self.check(&token::OpenDelim(token::Paren)) {
1318 // We don't `self.bump()` the `(` yet because this might be a struct definition where
1319 // `()` or a tuple might be allowed. For example, `struct Struct(pub (), pub (usize));`.
1320 // Because of this, we only `bump` the `(` if we're assured it is appropriate to do so
1321 // by the following tokens.
1322 if self.is_keyword_ahead(1, &[kw::Crate]) && self.look_ahead(2, |t| t != &token::ModSep)
1323 // account for `pub(crate::foo)`
1325 // Parse `pub(crate)`.
1327 self.bump(); // `crate`
1328 self.expect(&token::CloseDelim(token::Paren))?; // `)`
1329 let vis = VisibilityKind::Crate(CrateSugar::PubCrate);
1330 return Ok(Visibility {
1331 span: lo.to(self.prev_token.span),
1335 } else if self.is_keyword_ahead(1, &[kw::In]) {
1336 // Parse `pub(in path)`.
1338 self.bump(); // `in`
1339 let path = self.parse_path(PathStyle::Mod)?; // `path`
1340 self.expect(&token::CloseDelim(token::Paren))?; // `)`
1341 let vis = VisibilityKind::Restricted { path: P(path), id: ast::DUMMY_NODE_ID };
1342 return Ok(Visibility {
1343 span: lo.to(self.prev_token.span),
1347 } else if self.look_ahead(2, |t| t == &token::CloseDelim(token::Paren))
1348 && self.is_keyword_ahead(1, &[kw::Super, kw::SelfLower])
1350 // Parse `pub(self)` or `pub(super)`.
1352 let path = self.parse_path(PathStyle::Mod)?; // `super`/`self`
1353 self.expect(&token::CloseDelim(token::Paren))?; // `)`
1354 let vis = VisibilityKind::Restricted { path: P(path), id: ast::DUMMY_NODE_ID };
1355 return Ok(Visibility {
1356 span: lo.to(self.prev_token.span),
1360 } else if let FollowedByType::No = fbt {
1361 // Provide this diagnostic if a type cannot follow;
1362 // in particular, if this is not a tuple struct.
1363 self.recover_incorrect_vis_restriction()?;
1364 // Emit diagnostic, but continue with public visibility.
1368 Ok(Visibility { span: lo, kind: VisibilityKind::Public, tokens: None })
1371 /// Recovery for e.g. `pub(something) fn ...` or `struct X { pub(something) y: Z }`
1372 fn recover_incorrect_vis_restriction(&mut self) -> PResult<'a, ()> {
1374 let path = self.parse_path(PathStyle::Mod)?;
1375 self.expect(&token::CloseDelim(token::Paren))?; // `)`
1377 let msg = "incorrect visibility restriction";
1378 let suggestion = r##"some possible visibility restrictions are:
1379 `pub(crate)`: visible only on the current crate
1380 `pub(super)`: visible only in the current module's parent
1381 `pub(in path::to::module)`: visible only on the specified path"##;
1383 let path_str = pprust::path_to_string(&path);
1385 struct_span_err!(self.sess.span_diagnostic, path.span, E0704, "{}", msg)
1389 &format!("make this visible only to module `{}` with `in`", path_str),
1390 format!("in {}", path_str),
1391 Applicability::MachineApplicable,
1398 /// Parses `extern string_literal?`.
1399 fn parse_extern(&mut self) -> Extern {
1400 if self.eat_keyword(kw::Extern) { Extern::from_abi(self.parse_abi()) } else { Extern::None }
1403 /// Parses a string literal as an ABI spec.
1404 fn parse_abi(&mut self) -> Option<StrLit> {
1405 match self.parse_str_lit() {
1406 Ok(str_lit) => Some(str_lit),
1407 Err(Some(lit)) => match lit.kind {
1408 ast::LitKind::Err(_) => None,
1410 self.struct_span_err(lit.span, "non-string ABI literal")
1413 "specify the ABI with a string literal",
1414 "\"C\"".to_string(),
1415 Applicability::MaybeIncorrect,
1425 pub fn collect_tokens_no_attrs<R: AstLike>(
1427 f: impl FnOnce(&mut Self) -> PResult<'a, R>,
1428 ) -> PResult<'a, R> {
1429 // The only reason to call `collect_tokens_no_attrs` is if you want tokens, so use
1430 // `ForceCollect::Yes`
1431 self.collect_tokens_trailing_token(
1432 AttrWrapper::empty(),
1434 |this, _attrs| Ok((f(this)?, TrailingToken::None)),
1439 fn is_import_coupler(&mut self) -> bool {
1440 self.check(&token::ModSep)
1441 && self.look_ahead(1, |t| {
1442 *t == token::OpenDelim(token::Brace) || *t == token::BinOp(token::Star)
1446 pub fn clear_expected_tokens(&mut self) {
1447 self.expected_tokens.clear();
1451 crate fn make_unclosed_delims_error(
1452 unmatched: UnmatchedBrace,
1454 ) -> Option<DiagnosticBuilder<'_, ErrorGuaranteed>> {
1455 // `None` here means an `Eof` was found. We already emit those errors elsewhere, we add them to
1456 // `unmatched_braces` only for error recovery in the `Parser`.
1457 let found_delim = unmatched.found_delim?;
1458 let span: MultiSpan = if let Some(sp) = unmatched.unclosed_span {
1459 vec![unmatched.found_span, sp].into()
1461 unmatched.found_span.into()
1463 let mut err = sess.span_diagnostic.struct_span_err(
1466 "mismatched closing delimiter: `{}`",
1467 pprust::token_kind_to_string(&token::CloseDelim(found_delim)),
1470 err.span_label(unmatched.found_span, "mismatched closing delimiter");
1471 if let Some(sp) = unmatched.candidate_span {
1472 err.span_label(sp, "closing delimiter possibly meant for this");
1474 if let Some(sp) = unmatched.unclosed_span {
1475 err.span_label(sp, "unclosed delimiter");
1480 pub fn emit_unclosed_delims(unclosed_delims: &mut Vec<UnmatchedBrace>, sess: &ParseSess) {
1481 *sess.reached_eof.borrow_mut() |=
1482 unclosed_delims.iter().any(|unmatched_delim| unmatched_delim.found_delim.is_none());
1483 for unmatched in unclosed_delims.drain(..) {
1484 if let Some(mut e) = make_unclosed_delims_error(unmatched, sess) {
1490 /// A helper struct used when building an `AttrAnnotatedTokenStream` from
1491 /// a `LazyTokenStream`. Both delimiter and non-delimited tokens
1492 /// are stored as `FlatToken::Token`. A vector of `FlatToken`s
1493 /// is then 'parsed' to build up an `AttrAnnotatedTokenStream` with nested
1494 /// `AttrAnnotatedTokenTree::Delimited` tokens
1495 #[derive(Debug, Clone)]
1496 pub enum FlatToken {
1497 /// A token - this holds both delimiter (e.g. '{' and '}')
1498 /// and non-delimiter tokens
1500 /// Holds the `AttributesData` for an AST node. The
1501 /// `AttributesData` is inserted directly into the
1502 /// constructed `AttrAnnotatedTokenStream` as
1503 /// an `AttrAnnotatedTokenTree::Attributes`
1504 AttrTarget(AttributesData),
1505 /// A special 'empty' token that is ignored during the conversion
1506 /// to an `AttrAnnotatedTokenStream`. This is used to simplify the
1507 /// handling of replace ranges.