13 use crate::lexer::UnmatchedBrace;
14 pub use attr_wrapper::AttrWrapper;
15 pub use diagnostics::AttemptLocalParseRecovery;
16 pub(crate) use item::FnParseMode;
17 pub use pat::{CommaRecoveryMode, RecoverColon, RecoverComma};
18 pub use path::PathStyle;
20 use rustc_ast::ptr::P;
21 use rustc_ast::token::{self, Delimiter, Nonterminal, Token, TokenKind};
22 use rustc_ast::tokenstream::AttributesData;
23 use rustc_ast::tokenstream::{self, DelimSpan, Spacing};
24 use rustc_ast::tokenstream::{TokenStream, TokenTree};
25 use rustc_ast::util::case::Case;
26 use rustc_ast::AttrId;
27 use rustc_ast::DUMMY_NODE_ID;
28 use rustc_ast::{self as ast, AnonConst, AttrStyle, Const, DelimArgs, Extern};
29 use rustc_ast::{Async, AttrArgs, AttrArgsEq, Expr, ExprKind, MacDelimiter, Mutability, StrLit};
30 use rustc_ast::{HasAttrs, HasTokens, Unsafe, Visibility, VisibilityKind};
31 use rustc_ast_pretty::pprust;
32 use rustc_data_structures::fx::FxHashMap;
33 use rustc_errors::PResult;
35 Applicability, DiagnosticBuilder, ErrorGuaranteed, FatalError, IntoDiagnostic, MultiSpan,
37 use rustc_session::parse::ParseSess;
38 use rustc_span::source_map::{Span, DUMMY_SP};
39 use rustc_span::symbol::{kw, sym, Ident, Symbol};
42 use std::{cmp, mem, slice};
45 DocCommentDoesNotDocumentAnything, IncorrectVisibilityRestriction, MismatchedClosingDelimiter,
50 struct Restrictions: u8 {
51 const STMT_EXPR = 1 << 0;
52 const NO_STRUCT_LITERAL = 1 << 1;
53 const CONST_EXPR = 1 << 2;
54 const ALLOW_LET = 1 << 3;
58 #[derive(Clone, Copy, PartialEq, Debug)]
65 #[derive(Clone, Copy, PartialEq, Debug)]
71 /// Whether or not we should force collection of tokens for an AST node,
72 /// regardless of whether or not it has attributes
73 #[derive(Clone, Copy, PartialEq)]
74 pub enum ForceCollect {
79 #[derive(Debug, Eq, PartialEq)]
80 pub enum TrailingToken {
84 /// If the trailing token is a comma, then capture it
85 /// Otherwise, ignore the trailing token
89 /// Like `maybe_whole_expr`, but for things other than expressions.
91 macro_rules! maybe_whole {
92 ($p:expr, $constructor:ident, |$x:ident| $e:expr) => {
93 if let token::Interpolated(nt) = &$p.token.kind {
94 if let token::$constructor(x) = &**nt {
103 /// If the next tokens are ill-formed `$ty::` recover them as `<$ty>::`.
105 macro_rules! maybe_recover_from_interpolated_ty_qpath {
106 ($self: expr, $allow_qpath_recovery: expr) => {
107 if $allow_qpath_recovery
108 && $self.may_recover()
109 && $self.look_ahead(1, |t| t == &token::ModSep)
110 && let token::Interpolated(nt) = &$self.token.kind
111 && let token::NtTy(ty) = &**nt
115 return $self.maybe_recover_from_bad_qpath_stage_2($self.prev_token.span, ty);
120 #[derive(Clone, Copy)]
127 pub struct Parser<'a> {
128 pub sess: &'a ParseSess,
129 /// The current token.
131 /// The spacing for the current token
132 pub token_spacing: Spacing,
133 /// The previous token.
134 pub prev_token: Token,
135 pub capture_cfg: bool,
136 restrictions: Restrictions,
137 expected_tokens: Vec<TokenType>,
138 // Important: This must only be advanced from `bump` to ensure that
139 // `token_cursor.num_next_calls` is updated properly.
140 token_cursor: TokenCursor,
141 desugar_doc_comments: bool,
142 /// This field is used to keep track of how many left angle brackets we have seen. This is
143 /// required in order to detect extra leading left angle brackets (`<` characters) and error
146 /// See the comments in the `parse_path_segment` function for more details.
147 unmatched_angle_bracket_count: u32,
148 max_angle_bracket_count: u32,
149 /// A list of all unclosed delimiters found by the lexer. If an entry is used for error recovery
150 /// it gets removed from here. Every entry left at the end gets emitted as an independent
152 pub(super) unclosed_delims: Vec<UnmatchedBrace>,
153 last_unexpected_token_span: Option<Span>,
154 /// Span pointing at the `:` for the last type ascription the parser has seen, and whether it
155 /// looked like it could have been a mistyped path or literal `Option:Some(42)`).
156 pub last_type_ascription: Option<(Span, bool /* likely path typo */)>,
157 /// If present, this `Parser` is not parsing Rust code but rather a macro call.
158 subparser_name: Option<&'static str>,
159 capture_state: CaptureState,
160 /// This allows us to recover when the user forget to add braces around
161 /// multiple statements in the closure body.
162 pub current_closure: Option<ClosureSpans>,
163 /// Whether the parser is allowed to do recovery.
164 /// This is disabled when parsing macro arguments, see #103534
165 pub recovery: Recovery,
168 // This type is used a lot, e.g. it's cloned when matching many declarative macro rules with nonterminals. Make sure
169 // it doesn't unintentionally get bigger.
170 #[cfg(all(target_arch = "x86_64", target_pointer_width = "64"))]
171 rustc_data_structures::static_assert_size!(Parser<'_>, 336);
173 /// Stores span information about a closure.
175 pub struct ClosureSpans {
176 pub whole_closure: Span,
177 pub closing_pipe: Span,
181 /// Indicates a range of tokens that should be replaced by
182 /// the tokens in the provided vector. This is used in two
183 /// places during token collection:
185 /// 1. During the parsing of an AST node that may have a `#[derive]`
186 /// attribute, we parse a nested AST node that has `#[cfg]` or `#[cfg_attr]`
187 /// In this case, we use a `ReplaceRange` to replace the entire inner AST node
188 /// with `FlatToken::AttrTarget`, allowing us to perform eager cfg-expansion
189 /// on an `AttrTokenStream`.
191 /// 2. When we parse an inner attribute while collecting tokens. We
192 /// remove inner attributes from the token stream entirely, and
193 /// instead track them through the `attrs` field on the AST node.
194 /// This allows us to easily manipulate them (for example, removing
195 /// the first macro inner attribute to invoke a proc-macro).
196 /// When create a `TokenStream`, the inner attributes get inserted
197 /// into the proper place in the token stream.
198 pub type ReplaceRange = (Range<u32>, Vec<(FlatToken, Spacing)>);
200 /// Controls how we capture tokens. Capturing can be expensive,
201 /// so we try to avoid performing capturing in cases where
202 /// we will never need an `AttrTokenStream`.
203 #[derive(Copy, Clone)]
205 /// We aren't performing any capturing - this is the default mode.
207 /// We are capturing tokens
212 struct CaptureState {
213 capturing: Capturing,
214 replace_ranges: Vec<ReplaceRange>,
215 inner_attr_ranges: FxHashMap<AttrId, ReplaceRange>,
218 impl<'a> Drop for Parser<'a> {
220 emit_unclosed_delims(&mut self.unclosed_delims, &self.sess);
226 // The current (innermost) frame. `frame` and `stack` could be combined,
227 // but it's faster to keep them separate and access `frame` directly
228 // rather than via something like `stack.last().unwrap()` or
229 // `stack[stack.len() - 1]`.
230 frame: TokenCursorFrame,
231 // Additional frames that enclose `frame`.
232 stack: Vec<TokenCursorFrame>,
233 desugar_doc_comments: bool,
234 // Counts the number of calls to `{,inlined_}next`.
235 num_next_calls: usize,
236 // During parsing, we may sometimes need to 'unglue' a
237 // glued token into two component tokens
238 // (e.g. '>>' into '>' and '>), so that the parser
239 // can consume them one at a time. This process
240 // bypasses the normal capturing mechanism
241 // (e.g. `num_next_calls` will not be incremented),
242 // since the 'unglued' tokens due not exist in
243 // the original `TokenStream`.
245 // If we end up consuming both unglued tokens,
246 // then this is not an issue - we'll end up
247 // capturing the single 'glued' token.
249 // However, in certain circumstances, we may
250 // want to capture just the first 'unglued' token.
251 // For example, capturing the `Vec<u8>`
252 // in `Option<Vec<u8>>` requires us to unglue
253 // the trailing `>>` token. The `break_last_token`
254 // field is used to track this token - it gets
255 // appended to the captured stream when
256 // we evaluate a `LazyAttrTokenStream`.
257 break_last_token: bool,
261 struct TokenCursorFrame {
262 // This is `None` only for the outermost frame.
263 delim_sp: Option<(Delimiter, DelimSpan)>,
264 tree_cursor: tokenstream::Cursor,
267 impl TokenCursorFrame {
268 fn new(delim_sp: Option<(Delimiter, DelimSpan)>, tts: TokenStream) -> Self {
269 TokenCursorFrame { delim_sp, tree_cursor: tts.into_trees() }
274 fn next(&mut self, desugar_doc_comments: bool) -> (Token, Spacing) {
275 self.inlined_next(desugar_doc_comments)
278 /// This always-inlined version should only be used on hot code paths.
280 fn inlined_next(&mut self, desugar_doc_comments: bool) -> (Token, Spacing) {
282 // FIXME: we currently don't return `Delimiter` open/close delims. To fix #67062 we will
283 // need to, whereupon the `delim != Delimiter::Invisible` conditions below can be
285 if let Some(tree) = self.frame.tree_cursor.next_ref() {
287 &TokenTree::Token(ref token, spacing) => match (desugar_doc_comments, token) {
288 (true, &Token { kind: token::DocComment(_, attr_style, data), span }) => {
289 let desugared = self.desugar(attr_style, data, span);
290 self.frame.tree_cursor.replace_prev_and_rewind(desugared);
291 // Continue to get the first token of the desugared doc comment.
294 debug_assert!(!matches!(
296 token::OpenDelim(_) | token::CloseDelim(_)
298 return (token.clone(), spacing);
301 &TokenTree::Delimited(sp, delim, ref tts) => {
302 // Set `open_delim` to true here because we deal with it immediately.
303 let frame = TokenCursorFrame::new(Some((delim, sp)), tts.clone());
304 self.stack.push(mem::replace(&mut self.frame, frame));
305 if delim != Delimiter::Invisible {
306 return (Token::new(token::OpenDelim(delim), sp.open), Spacing::Alone);
308 // No open delimiter to return; continue on to the next iteration.
311 } else if let Some(frame) = self.stack.pop() {
312 // We have exhausted this frame. Move back to its parent frame.
313 let (delim, span) = self.frame.delim_sp.unwrap();
315 if delim != Delimiter::Invisible {
316 return (Token::new(token::CloseDelim(delim), span.close), Spacing::Alone);
318 // No close delimiter to return; continue on to the next iteration.
320 // We have exhausted the outermost frame.
321 return (Token::new(token::Eof, DUMMY_SP), Spacing::Alone);
326 // Desugar a doc comment into something like `#[doc = r"foo"]`.
327 fn desugar(&mut self, attr_style: AttrStyle, data: Symbol, span: Span) -> Vec<TokenTree> {
328 // Searches for the occurrences of `"#*` and returns the minimum number of `#`s
329 // required to wrap the text. E.g.
330 // - `abc d` is wrapped as `r"abc d"` (num_of_hashes = 0)
331 // - `abc "d"` is wrapped as `r#"abc "d""#` (num_of_hashes = 1)
332 // - `abc "##d##"` is wrapped as `r###"abc ##"d"##"###` (num_of_hashes = 3)
333 let mut num_of_hashes = 0;
335 for ch in data.as_str().chars() {
338 '#' if count > 0 => count + 1,
341 num_of_hashes = cmp::max(num_of_hashes, count);
344 // `/// foo` becomes `doc = r"foo".
345 let delim_span = DelimSpan::from_single(span);
346 let body = TokenTree::Delimited(
350 TokenTree::token_alone(token::Ident(sym::doc, false), span),
351 TokenTree::token_alone(token::Eq, span),
352 TokenTree::token_alone(
353 TokenKind::lit(token::StrRaw(num_of_hashes), data, None),
358 .collect::<TokenStream>(),
361 if attr_style == AttrStyle::Inner {
363 TokenTree::token_alone(token::Pound, span),
364 TokenTree::token_alone(token::Not, span),
368 vec![TokenTree::token_alone(token::Pound, span), body]
373 #[derive(Debug, Clone, PartialEq)]
386 fn to_string(&self) -> String {
388 TokenType::Token(t) => format!("`{}`", pprust::token_kind_to_string(t)),
389 TokenType::Keyword(kw) => format!("`{}`", kw),
390 TokenType::Operator => "an operator".to_string(),
391 TokenType::Lifetime => "lifetime".to_string(),
392 TokenType::Ident => "identifier".to_string(),
393 TokenType::Path => "path".to_string(),
394 TokenType::Type => "type".to_string(),
395 TokenType::Const => "a const expression".to_string(),
400 #[derive(Copy, Clone, Debug)]
401 enum TokenExpectType {
406 /// A sequence separator.
408 /// The separator token.
409 sep: Option<TokenKind>,
410 /// `true` if a trailing separator is allowed.
411 trailing_sep_allowed: bool,
415 fn trailing_allowed(t: TokenKind) -> SeqSep {
416 SeqSep { sep: Some(t), trailing_sep_allowed: true }
419 fn none() -> SeqSep {
420 SeqSep { sep: None, trailing_sep_allowed: false }
424 pub enum FollowedByType {
429 #[derive(Clone, Copy, PartialEq, Eq)]
430 pub enum TokenDescription {
437 impl TokenDescription {
438 pub fn from_token(token: &Token) -> Option<Self> {
440 _ if token.is_special_ident() => Some(TokenDescription::ReservedIdentifier),
441 _ if token.is_used_keyword() => Some(TokenDescription::Keyword),
442 _ if token.is_unused_keyword() => Some(TokenDescription::ReservedKeyword),
443 token::DocComment(..) => Some(TokenDescription::DocComment),
449 pub(super) fn token_descr(token: &Token) -> String {
450 let name = pprust::token_to_string(token).to_string();
452 let kind = TokenDescription::from_token(token).map(|kind| match kind {
453 TokenDescription::ReservedIdentifier => "reserved identifier",
454 TokenDescription::Keyword => "keyword",
455 TokenDescription::ReservedKeyword => "reserved keyword",
456 TokenDescription::DocComment => "doc comment",
459 if let Some(kind) = kind { format!("{} `{}`", kind, name) } else { format!("`{}`", name) }
462 impl<'a> Parser<'a> {
466 desugar_doc_comments: bool,
467 subparser_name: Option<&'static str>,
469 let mut parser = Parser {
471 token: Token::dummy(),
472 token_spacing: Spacing::Alone,
473 prev_token: Token::dummy(),
475 restrictions: Restrictions::empty(),
476 expected_tokens: Vec::new(),
477 token_cursor: TokenCursor {
478 frame: TokenCursorFrame::new(None, tokens),
481 desugar_doc_comments,
482 break_last_token: false,
484 desugar_doc_comments,
485 unmatched_angle_bracket_count: 0,
486 max_angle_bracket_count: 0,
487 unclosed_delims: Vec::new(),
488 last_unexpected_token_span: None,
489 last_type_ascription: None,
491 capture_state: CaptureState {
492 capturing: Capturing::No,
493 replace_ranges: Vec::new(),
494 inner_attr_ranges: Default::default(),
496 current_closure: None,
497 recovery: Recovery::Allowed,
500 // Make parser point to the first token.
506 pub fn recovery(mut self, recovery: Recovery) -> Self {
507 self.recovery = recovery;
511 /// Whether the parser is allowed to recover from broken code.
513 /// If this returns false, recovering broken code into valid code (especially if this recovery does lookahead)
514 /// is not allowed. All recovery done by the parser must be gated behind this check.
516 /// Technically, this only needs to restrict eager recovery by doing lookahead at more tokens.
517 /// But making the distinction is very subtle, and simply forbidding all recovery is a lot simpler to uphold.
518 fn may_recover(&self) -> bool {
519 matches!(self.recovery, Recovery::Allowed)
522 pub fn unexpected<T>(&mut self) -> PResult<'a, T> {
523 match self.expect_one_of(&[], &[]) {
525 // We can get `Ok(true)` from `recover_closing_delimiter`
526 // which is called in `expected_one_of_not_found`.
527 Ok(_) => FatalError.raise(),
531 /// Expects and consumes the token `t`. Signals an error if the next token is not `t`.
532 pub fn expect(&mut self, t: &TokenKind) -> PResult<'a, bool /* recovered */> {
533 if self.expected_tokens.is_empty() {
534 if self.token == *t {
538 self.unexpected_try_recover(t)
541 self.expect_one_of(slice::from_ref(t), &[])
545 /// Expect next token to be edible or inedible token. If edible,
546 /// then consume it; if inedible, then return without consuming
547 /// anything. Signal a fatal error if next token is unexpected.
548 pub fn expect_one_of(
550 edible: &[TokenKind],
551 inedible: &[TokenKind],
552 ) -> PResult<'a, bool /* recovered */> {
553 if edible.contains(&self.token.kind) {
556 } else if inedible.contains(&self.token.kind) {
557 // leave it in the input
559 } else if self.last_unexpected_token_span == Some(self.token.span) {
562 self.expected_one_of_not_found(edible, inedible)
566 // Public for rustfmt usage.
567 pub fn parse_ident(&mut self) -> PResult<'a, Ident> {
568 self.parse_ident_common(true)
571 fn ident_or_err(&mut self) -> PResult<'a, (Ident, /* is_raw */ bool)> {
572 self.token.ident().ok_or_else(|| match self.prev_token.kind {
573 TokenKind::DocComment(..) => DocCommentDoesNotDocumentAnything {
574 span: self.prev_token.span,
577 .into_diagnostic(&self.sess.span_diagnostic),
578 _ => self.expected_ident_found(),
582 fn parse_ident_common(&mut self, recover: bool) -> PResult<'a, Ident> {
583 let (ident, is_raw) = self.ident_or_err()?;
584 if !is_raw && ident.is_reserved() {
585 let mut err = self.expected_ident_found();
596 /// Checks if the next token is `tok`, and returns `true` if so.
598 /// This method will automatically add `tok` to `expected_tokens` if `tok` is not
600 fn check(&mut self, tok: &TokenKind) -> bool {
601 let is_present = self.token == *tok;
603 self.expected_tokens.push(TokenType::Token(tok.clone()));
608 fn check_noexpect(&self, tok: &TokenKind) -> bool {
612 /// Consumes a token 'tok' if it exists. Returns whether the given token was present.
614 /// the main purpose of this function is to reduce the cluttering of the suggestions list
615 /// which using the normal eat method could introduce in some cases.
616 pub fn eat_noexpect(&mut self, tok: &TokenKind) -> bool {
617 let is_present = self.check_noexpect(tok);
624 /// Consumes a token 'tok' if it exists. Returns whether the given token was present.
625 pub fn eat(&mut self, tok: &TokenKind) -> bool {
626 let is_present = self.check(tok);
633 /// If the next token is the given keyword, returns `true` without eating it.
634 /// An expectation is also added for diagnostics purposes.
635 fn check_keyword(&mut self, kw: Symbol) -> bool {
636 self.expected_tokens.push(TokenType::Keyword(kw));
637 self.token.is_keyword(kw)
640 fn check_keyword_case(&mut self, kw: Symbol, case: Case) -> bool {
641 if self.check_keyword(kw) {
645 if case == Case::Insensitive
646 && let Some((ident, /* is_raw */ false)) = self.token.ident()
647 && ident.as_str().to_lowercase() == kw.as_str().to_lowercase() {
654 /// If the next token is the given keyword, eats it and returns `true`.
655 /// Otherwise, returns `false`. An expectation is also added for diagnostics purposes.
656 // Public for rustfmt usage.
657 pub fn eat_keyword(&mut self, kw: Symbol) -> bool {
658 if self.check_keyword(kw) {
666 /// Eats a keyword, optionally ignoring the case.
667 /// If the case differs (and is ignored) an error is issued.
668 /// This is useful for recovery.
669 fn eat_keyword_case(&mut self, kw: Symbol, case: Case) -> bool {
670 if self.eat_keyword(kw) {
674 if case == Case::Insensitive
675 && let Some((ident, /* is_raw */ false)) = self.token.ident()
676 && ident.as_str().to_lowercase() == kw.as_str().to_lowercase() {
678 .struct_span_err(ident.span, format!("keyword `{kw}` is written in a wrong case"))
681 "write it in the correct case",
683 Applicability::MachineApplicable
693 fn eat_keyword_noexpect(&mut self, kw: Symbol) -> bool {
694 if self.token.is_keyword(kw) {
702 /// If the given word is not a keyword, signals an error.
703 /// If the next token is not the given word, signals an error.
704 /// Otherwise, eats it.
705 fn expect_keyword(&mut self, kw: Symbol) -> PResult<'a, ()> {
706 if !self.eat_keyword(kw) { self.unexpected() } else { Ok(()) }
709 /// Is the given keyword `kw` followed by a non-reserved identifier?
710 fn is_kw_followed_by_ident(&self, kw: Symbol) -> bool {
711 self.token.is_keyword(kw) && self.look_ahead(1, |t| t.is_ident() && !t.is_reserved_ident())
714 fn check_or_expected(&mut self, ok: bool, typ: TokenType) -> bool {
718 self.expected_tokens.push(typ);
723 fn check_ident(&mut self) -> bool {
724 self.check_or_expected(self.token.is_ident(), TokenType::Ident)
727 fn check_path(&mut self) -> bool {
728 self.check_or_expected(self.token.is_path_start(), TokenType::Path)
731 fn check_type(&mut self) -> bool {
732 self.check_or_expected(self.token.can_begin_type(), TokenType::Type)
735 fn check_const_arg(&mut self) -> bool {
736 self.check_or_expected(self.token.can_begin_const_arg(), TokenType::Const)
739 fn check_const_closure(&self) -> bool {
740 self.is_keyword_ahead(0, &[kw::Const])
741 && self.look_ahead(1, |t| match &t.kind {
742 token::Ident(kw::Move | kw::Static | kw::Async, _)
744 | token::BinOp(token::Or) => true,
749 fn check_inline_const(&self, dist: usize) -> bool {
750 self.is_keyword_ahead(dist, &[kw::Const])
751 && self.look_ahead(dist + 1, |t| match &t.kind {
752 token::Interpolated(nt) => matches!(**nt, token::NtBlock(..)),
753 token::OpenDelim(Delimiter::Brace) => true,
758 /// Checks to see if the next token is either `+` or `+=`.
759 /// Otherwise returns `false`.
760 fn check_plus(&mut self) -> bool {
761 self.check_or_expected(
762 self.token.is_like_plus(),
763 TokenType::Token(token::BinOp(token::Plus)),
767 /// Eats the expected token if it's present possibly breaking
768 /// compound tokens like multi-character operators in process.
769 /// Returns `true` if the token was eaten.
770 fn break_and_eat(&mut self, expected: TokenKind) -> bool {
771 if self.token.kind == expected {
775 match self.token.kind.break_two_token_op() {
776 Some((first, second)) if first == expected => {
777 let first_span = self.sess.source_map().start_point(self.token.span);
778 let second_span = self.token.span.with_lo(first_span.hi());
779 self.token = Token::new(first, first_span);
780 // Keep track of this token - if we end token capturing now,
781 // we'll want to append this token to the captured stream.
783 // If we consume any additional tokens, then this token
784 // is not needed (we'll capture the entire 'glued' token),
785 // and `bump` will set this field to `None`
786 self.token_cursor.break_last_token = true;
787 // Use the spacing of the glued token as the spacing
788 // of the unglued second token.
789 self.bump_with((Token::new(second, second_span), self.token_spacing));
793 self.expected_tokens.push(TokenType::Token(expected));
799 /// Eats `+` possibly breaking tokens like `+=` in process.
800 fn eat_plus(&mut self) -> bool {
801 self.break_and_eat(token::BinOp(token::Plus))
804 /// Eats `&` possibly breaking tokens like `&&` in process.
805 /// Signals an error if `&` is not eaten.
806 fn expect_and(&mut self) -> PResult<'a, ()> {
807 if self.break_and_eat(token::BinOp(token::And)) { Ok(()) } else { self.unexpected() }
810 /// Eats `|` possibly breaking tokens like `||` in process.
811 /// Signals an error if `|` was not eaten.
812 fn expect_or(&mut self) -> PResult<'a, ()> {
813 if self.break_and_eat(token::BinOp(token::Or)) { Ok(()) } else { self.unexpected() }
816 /// Eats `<` possibly breaking tokens like `<<` in process.
817 fn eat_lt(&mut self) -> bool {
818 let ate = self.break_and_eat(token::Lt);
820 // See doc comment for `unmatched_angle_bracket_count`.
821 self.unmatched_angle_bracket_count += 1;
822 self.max_angle_bracket_count += 1;
823 debug!("eat_lt: (increment) count={:?}", self.unmatched_angle_bracket_count);
828 /// Eats `<` possibly breaking tokens like `<<` in process.
829 /// Signals an error if `<` was not eaten.
830 fn expect_lt(&mut self) -> PResult<'a, ()> {
831 if self.eat_lt() { Ok(()) } else { self.unexpected() }
834 /// Eats `>` possibly breaking tokens like `>>` in process.
835 /// Signals an error if `>` was not eaten.
836 fn expect_gt(&mut self) -> PResult<'a, ()> {
837 if self.break_and_eat(token::Gt) {
838 // See doc comment for `unmatched_angle_bracket_count`.
839 if self.unmatched_angle_bracket_count > 0 {
840 self.unmatched_angle_bracket_count -= 1;
841 debug!("expect_gt: (decrement) count={:?}", self.unmatched_angle_bracket_count);
849 fn expect_any_with_type(&mut self, kets: &[&TokenKind], expect: TokenExpectType) -> bool {
850 kets.iter().any(|k| match expect {
851 TokenExpectType::Expect => self.check(k),
852 TokenExpectType::NoExpect => self.token == **k,
856 fn parse_seq_to_before_tokens<T>(
860 expect: TokenExpectType,
861 mut f: impl FnMut(&mut Parser<'a>) -> PResult<'a, T>,
862 ) -> PResult<'a, (Vec<T>, bool /* trailing */, bool /* recovered */)> {
863 let mut first = true;
864 let mut recovered = false;
865 let mut trailing = false;
867 let unclosed_delims = !self.unclosed_delims.is_empty();
869 while !self.expect_any_with_type(kets, expect) {
870 if let token::CloseDelim(..) | token::Eof = self.token.kind {
873 if let Some(t) = &sep.sep {
877 match self.expect(t) {
879 self.current_closure.take();
882 self.current_closure.take();
886 Err(mut expect_err) => {
887 let sp = self.prev_token.span.shrink_to_hi();
888 let token_str = pprust::token_kind_to_string(t);
890 match self.current_closure.take() {
891 Some(closure_spans) if self.token.kind == TokenKind::Semi => {
892 // Finding a semicolon instead of a comma
893 // after a closure body indicates that the
894 // closure body may be a block but the user
895 // forgot to put braces around its
898 self.recover_missing_braces_around_closure_body(
907 // Attempt to keep parsing if it was a similar separator.
908 if let Some(tokens) = t.similar_tokens() {
909 if tokens.contains(&self.token.kind) && !unclosed_delims {
916 // If this was a missing `@` in a binding pattern
917 // bail with a suggestion
918 // https://github.com/rust-lang/rust/issues/72373
919 if self.prev_token.is_ident() && self.token.kind == token::DotDot {
921 "if you meant to bind the contents of \
922 the rest of the array pattern into `{}`, use `@`",
923 pprust::token_to_string(&self.prev_token)
926 .span_suggestion_verbose(
927 self.prev_token.span.shrink_to_hi().until(self.token.span),
930 Applicability::MaybeIncorrect,
936 // Attempt to keep parsing if it was an omitted separator.
939 // Parsed successfully, therefore most probably the code only
940 // misses a separator.
942 .span_suggestion_short(
944 &format!("missing `{}`", token_str),
946 Applicability::MaybeIncorrect,
954 // Parsing failed, therefore it must be something more serious
955 // than just a missing separator.
956 for xx in &e.children {
957 // propagate the help message from sub error 'e' to main error 'expect_err;
958 expect_err.children.push(xx.clone());
970 if sep.trailing_sep_allowed && self.expect_any_with_type(kets, expect) {
979 Ok((v, trailing, recovered))
982 fn recover_missing_braces_around_closure_body(
984 closure_spans: ClosureSpans,
985 mut expect_err: DiagnosticBuilder<'_, ErrorGuaranteed>,
986 ) -> PResult<'a, ()> {
987 let initial_semicolon = self.token.span;
989 while self.eat(&TokenKind::Semi) {
990 let _ = self.parse_stmt(ForceCollect::Yes)?;
993 expect_err.set_primary_message(
994 "closure bodies that contain statements must be surrounded by braces",
997 let preceding_pipe_span = closure_spans.closing_pipe;
998 let following_token_span = self.token.span;
1000 let mut first_note = MultiSpan::from(vec![initial_semicolon]);
1001 first_note.push_span_label(
1003 "this `;` turns the preceding closure into a statement",
1005 first_note.push_span_label(
1007 "this expression is a statement because of the trailing semicolon",
1009 expect_err.span_note(first_note, "statement found outside of a block");
1011 let mut second_note = MultiSpan::from(vec![closure_spans.whole_closure]);
1012 second_note.push_span_label(closure_spans.whole_closure, "this is the parsed closure...");
1013 second_note.push_span_label(
1014 following_token_span,
1015 "...but likely you meant the closure to end here",
1017 expect_err.span_note(second_note, "the closure body may be incorrectly delimited");
1019 expect_err.set_span(vec![preceding_pipe_span, following_token_span]);
1021 let opening_suggestion_str = " {".to_string();
1022 let closing_suggestion_str = "}".to_string();
1024 expect_err.multipart_suggestion(
1025 "try adding braces",
1027 (preceding_pipe_span.shrink_to_hi(), opening_suggestion_str),
1028 (following_token_span.shrink_to_lo(), closing_suggestion_str),
1030 Applicability::MaybeIncorrect,
1038 /// Parses a sequence, not including the closing delimiter. The function
1039 /// `f` must consume tokens until reaching the next separator or
1040 /// closing bracket.
1041 fn parse_seq_to_before_end<T>(
1045 f: impl FnMut(&mut Parser<'a>) -> PResult<'a, T>,
1046 ) -> PResult<'a, (Vec<T>, bool, bool)> {
1047 self.parse_seq_to_before_tokens(&[ket], sep, TokenExpectType::Expect, f)
1050 /// Parses a sequence, including the closing delimiter. The function
1051 /// `f` must consume tokens until reaching the next separator or
1052 /// closing bracket.
1053 fn parse_seq_to_end<T>(
1057 f: impl FnMut(&mut Parser<'a>) -> PResult<'a, T>,
1058 ) -> PResult<'a, (Vec<T>, bool /* trailing */)> {
1059 let (val, trailing, recovered) = self.parse_seq_to_before_end(ket, sep, f)?;
1066 /// Parses a sequence, including the closing delimiter. The function
1067 /// `f` must consume tokens until reaching the next separator or
1068 /// closing bracket.
1069 fn parse_unspanned_seq<T>(
1074 f: impl FnMut(&mut Parser<'a>) -> PResult<'a, T>,
1075 ) -> PResult<'a, (Vec<T>, bool)> {
1077 self.parse_seq_to_end(ket, sep, f)
1080 fn parse_delim_comma_seq<T>(
1083 f: impl FnMut(&mut Parser<'a>) -> PResult<'a, T>,
1084 ) -> PResult<'a, (Vec<T>, bool)> {
1085 self.parse_unspanned_seq(
1086 &token::OpenDelim(delim),
1087 &token::CloseDelim(delim),
1088 SeqSep::trailing_allowed(token::Comma),
1093 fn parse_paren_comma_seq<T>(
1095 f: impl FnMut(&mut Parser<'a>) -> PResult<'a, T>,
1096 ) -> PResult<'a, (Vec<T>, bool)> {
1097 self.parse_delim_comma_seq(Delimiter::Parenthesis, f)
1100 /// Advance the parser by one token using provided token as the next one.
1101 fn bump_with(&mut self, next: (Token, Spacing)) {
1102 self.inlined_bump_with(next)
1105 /// This always-inlined version should only be used on hot code paths.
1107 fn inlined_bump_with(&mut self, (next_token, next_spacing): (Token, Spacing)) {
1108 // Update the current and previous tokens.
1109 self.prev_token = mem::replace(&mut self.token, next_token);
1110 self.token_spacing = next_spacing;
1113 self.expected_tokens.clear();
1116 /// Advance the parser by one token.
1117 pub fn bump(&mut self) {
1118 // Note: destructuring here would give nicer code, but it was found in #96210 to be slower
1119 // than `.0`/`.1` access.
1120 let mut next = self.token_cursor.inlined_next(self.desugar_doc_comments);
1121 self.token_cursor.num_next_calls += 1;
1122 // We've retrieved an token from the underlying
1123 // cursor, so we no longer need to worry about
1124 // an unglued token. See `break_and_eat` for more details
1125 self.token_cursor.break_last_token = false;
1126 if next.0.span.is_dummy() {
1127 // Tweak the location for better diagnostics, but keep syntactic context intact.
1128 let fallback_span = self.token.span;
1129 next.0.span = fallback_span.with_ctxt(next.0.span.ctxt());
1131 debug_assert!(!matches!(
1133 token::OpenDelim(Delimiter::Invisible) | token::CloseDelim(Delimiter::Invisible)
1135 self.inlined_bump_with(next)
1138 /// Look-ahead `dist` tokens of `self.token` and get access to that token there.
1139 /// When `dist == 0` then the current token is looked at.
1140 pub fn look_ahead<R>(&self, dist: usize, looker: impl FnOnce(&Token) -> R) -> R {
1142 return looker(&self.token);
1145 let frame = &self.token_cursor.frame;
1146 if let Some((delim, span)) = frame.delim_sp && delim != Delimiter::Invisible {
1147 let all_normal = (0..dist).all(|i| {
1148 let token = frame.tree_cursor.look_ahead(i);
1149 !matches!(token, Some(TokenTree::Delimited(_, Delimiter::Invisible, _)))
1152 return match frame.tree_cursor.look_ahead(dist - 1) {
1153 Some(tree) => match tree {
1154 TokenTree::Token(token, _) => looker(token),
1155 TokenTree::Delimited(dspan, delim, _) => {
1156 looker(&Token::new(token::OpenDelim(*delim), dspan.open))
1159 None => looker(&Token::new(token::CloseDelim(delim), span.close)),
1164 let mut cursor = self.token_cursor.clone();
1166 let mut token = Token::dummy();
1168 token = cursor.next(/* desugar_doc_comments */ false).0;
1171 token::OpenDelim(Delimiter::Invisible) | token::CloseDelim(Delimiter::Invisible)
1177 return looker(&token);
1180 /// Returns whether any of the given keywords are `dist` tokens ahead of the current one.
1181 fn is_keyword_ahead(&self, dist: usize, kws: &[Symbol]) -> bool {
1182 self.look_ahead(dist, |t| kws.iter().any(|&kw| t.is_keyword(kw)))
1185 /// Parses asyncness: `async` or nothing.
1186 fn parse_asyncness(&mut self, case: Case) -> Async {
1187 if self.eat_keyword_case(kw::Async, case) {
1188 let span = self.prev_token.uninterpolated_span();
1189 Async::Yes { span, closure_id: DUMMY_NODE_ID, return_impl_trait_id: DUMMY_NODE_ID }
1195 /// Parses unsafety: `unsafe` or nothing.
1196 fn parse_unsafety(&mut self, case: Case) -> Unsafe {
1197 if self.eat_keyword_case(kw::Unsafe, case) {
1198 Unsafe::Yes(self.prev_token.uninterpolated_span())
1204 /// Parses constness: `const` or nothing.
1205 fn parse_constness(&mut self, case: Case) -> Const {
1206 // Avoid const blocks to be parsed as const items
1207 if self.look_ahead(1, |t| t != &token::OpenDelim(Delimiter::Brace))
1208 && self.eat_keyword_case(kw::Const, case)
1210 Const::Yes(self.prev_token.uninterpolated_span())
1216 /// Parses inline const expressions.
1217 fn parse_const_block(&mut self, span: Span, pat: bool) -> PResult<'a, P<Expr>> {
1219 self.sess.gated_spans.gate(sym::inline_const_pat, span);
1221 self.sess.gated_spans.gate(sym::inline_const, span);
1223 self.eat_keyword(kw::Const);
1224 let (attrs, blk) = self.parse_inner_attrs_and_block()?;
1225 let anon_const = AnonConst {
1227 value: self.mk_expr(blk.span, ExprKind::Block(blk, None)),
1229 let blk_span = anon_const.value.span;
1230 Ok(self.mk_expr_with_attrs(span.to(blk_span), ExprKind::ConstBlock(anon_const), attrs))
1233 /// Parses mutability (`mut` or nothing).
1234 fn parse_mutability(&mut self) -> Mutability {
1235 if self.eat_keyword(kw::Mut) { Mutability::Mut } else { Mutability::Not }
1238 /// Possibly parses mutability (`const` or `mut`).
1239 fn parse_const_or_mut(&mut self) -> Option<Mutability> {
1240 if self.eat_keyword(kw::Mut) {
1241 Some(Mutability::Mut)
1242 } else if self.eat_keyword(kw::Const) {
1243 Some(Mutability::Not)
1249 fn parse_field_name(&mut self) -> PResult<'a, Ident> {
1250 if let token::Literal(token::Lit { kind: token::Integer, symbol, suffix }) = self.token.kind
1252 if let Some(suffix) = suffix {
1253 self.expect_no_tuple_index_suffix(self.token.span, suffix);
1256 Ok(Ident::new(symbol, self.prev_token.span))
1258 self.parse_ident_common(true)
1262 fn parse_delim_args(&mut self) -> PResult<'a, P<DelimArgs>> {
1263 if let Some(args) = self.parse_delim_args_inner() { Ok(P(args)) } else { self.unexpected() }
1266 fn parse_attr_args(&mut self) -> PResult<'a, AttrArgs> {
1267 Ok(if let Some(args) = self.parse_delim_args_inner() {
1268 AttrArgs::Delimited(args)
1270 if self.eat(&token::Eq) {
1271 let eq_span = self.prev_token.span;
1272 AttrArgs::Eq(eq_span, AttrArgsEq::Ast(self.parse_expr_force_collect()?))
1279 fn parse_delim_args_inner(&mut self) -> Option<DelimArgs> {
1280 if self.check(&token::OpenDelim(Delimiter::Parenthesis))
1281 || self.check(&token::OpenDelim(Delimiter::Bracket))
1282 || self.check(&token::OpenDelim(Delimiter::Brace))
1284 match self.parse_token_tree() {
1285 // We've confirmed above that there is a delimiter so unwrapping is OK.
1286 TokenTree::Delimited(dspan, delim, tokens) => Some(DelimArgs {
1288 delim: MacDelimiter::from_token(delim).unwrap(),
1291 _ => unreachable!(),
1298 fn parse_or_use_outer_attributes(
1300 already_parsed_attrs: Option<AttrWrapper>,
1301 ) -> PResult<'a, AttrWrapper> {
1302 if let Some(attrs) = already_parsed_attrs {
1305 self.parse_outer_attributes()
1309 /// Parses a single token tree from the input.
1310 pub(crate) fn parse_token_tree(&mut self) -> TokenTree {
1311 match self.token.kind {
1312 token::OpenDelim(..) => {
1313 // Grab the tokens from this frame.
1314 let frame = &self.token_cursor.frame;
1315 let stream = frame.tree_cursor.stream.clone();
1316 let (delim, span) = frame.delim_sp.unwrap();
1318 // Advance the token cursor through the entire delimited
1319 // sequence. After getting the `OpenDelim` we are *within* the
1320 // delimited sequence, i.e. at depth `d`. After getting the
1321 // matching `CloseDelim` we are *after* the delimited sequence,
1322 // i.e. at depth `d - 1`.
1323 let target_depth = self.token_cursor.stack.len() - 1;
1325 // Advance one token at a time, so `TokenCursor::next()`
1326 // can capture these tokens if necessary.
1328 if self.token_cursor.stack.len() == target_depth {
1329 debug_assert!(matches!(self.token.kind, token::CloseDelim(_)));
1334 // Consume close delimiter
1336 TokenTree::Delimited(span, delim, stream)
1338 token::CloseDelim(_) | token::Eof => unreachable!(),
1341 TokenTree::Token(self.prev_token.clone(), Spacing::Alone)
1346 /// Parses a stream of tokens into a list of `TokenTree`s, up to EOF.
1347 pub fn parse_all_token_trees(&mut self) -> PResult<'a, Vec<TokenTree>> {
1348 let mut tts = Vec::new();
1349 while self.token != token::Eof {
1350 tts.push(self.parse_token_tree());
1355 pub fn parse_tokens(&mut self) -> TokenStream {
1356 let mut result = Vec::new();
1358 match self.token.kind {
1359 token::Eof | token::CloseDelim(..) => break,
1360 _ => result.push(self.parse_token_tree()),
1363 TokenStream::new(result)
1366 /// Evaluates the closure with restrictions in place.
1368 /// Afters the closure is evaluated, restrictions are reset.
1369 fn with_res<T>(&mut self, res: Restrictions, f: impl FnOnce(&mut Self) -> T) -> T {
1370 let old = self.restrictions;
1371 self.restrictions = res;
1373 self.restrictions = old;
1377 /// Parses `pub` and `pub(in path)` plus shortcuts `pub(crate)` for `pub(in crate)`, `pub(self)`
1378 /// for `pub(in self)` and `pub(super)` for `pub(in super)`.
1379 /// If the following element can't be a tuple (i.e., it's a function definition), then
1380 /// it's not a tuple struct field), and the contents within the parentheses aren't valid,
1381 /// so emit a proper diagnostic.
1382 // Public for rustfmt usage.
1383 pub fn parse_visibility(&mut self, fbt: FollowedByType) -> PResult<'a, Visibility> {
1384 maybe_whole!(self, NtVis, |x| x.into_inner());
1386 if !self.eat_keyword(kw::Pub) {
1387 // We need a span for our `Spanned<VisibilityKind>`, but there's inherently no
1388 // keyword to grab a span from for inherited visibility; an empty span at the
1389 // beginning of the current token would seem to be the "Schelling span".
1390 return Ok(Visibility {
1391 span: self.token.span.shrink_to_lo(),
1392 kind: VisibilityKind::Inherited,
1396 let lo = self.prev_token.span;
1398 if self.check(&token::OpenDelim(Delimiter::Parenthesis)) {
1399 // We don't `self.bump()` the `(` yet because this might be a struct definition where
1400 // `()` or a tuple might be allowed. For example, `struct Struct(pub (), pub (usize));`.
1401 // Because of this, we only `bump` the `(` if we're assured it is appropriate to do so
1402 // by the following tokens.
1403 if self.is_keyword_ahead(1, &[kw::In]) {
1404 // Parse `pub(in path)`.
1406 self.bump(); // `in`
1407 let path = self.parse_path(PathStyle::Mod)?; // `path`
1408 self.expect(&token::CloseDelim(Delimiter::Parenthesis))?; // `)`
1409 let vis = VisibilityKind::Restricted {
1411 id: ast::DUMMY_NODE_ID,
1414 return Ok(Visibility {
1415 span: lo.to(self.prev_token.span),
1419 } else if self.look_ahead(2, |t| t == &token::CloseDelim(Delimiter::Parenthesis))
1420 && self.is_keyword_ahead(1, &[kw::Crate, kw::Super, kw::SelfLower])
1422 // Parse `pub(crate)`, `pub(self)`, or `pub(super)`.
1424 let path = self.parse_path(PathStyle::Mod)?; // `crate`/`super`/`self`
1425 self.expect(&token::CloseDelim(Delimiter::Parenthesis))?; // `)`
1426 let vis = VisibilityKind::Restricted {
1428 id: ast::DUMMY_NODE_ID,
1431 return Ok(Visibility {
1432 span: lo.to(self.prev_token.span),
1436 } else if let FollowedByType::No = fbt {
1437 // Provide this diagnostic if a type cannot follow;
1438 // in particular, if this is not a tuple struct.
1439 self.recover_incorrect_vis_restriction()?;
1440 // Emit diagnostic, but continue with public visibility.
1444 Ok(Visibility { span: lo, kind: VisibilityKind::Public, tokens: None })
1447 /// Recovery for e.g. `pub(something) fn ...` or `struct X { pub(something) y: Z }`
1448 fn recover_incorrect_vis_restriction(&mut self) -> PResult<'a, ()> {
1450 let path = self.parse_path(PathStyle::Mod)?;
1451 self.expect(&token::CloseDelim(Delimiter::Parenthesis))?; // `)`
1453 let path_str = pprust::path_to_string(&path);
1454 self.sess.emit_err(IncorrectVisibilityRestriction { span: path.span, inner_str: path_str });
1459 /// Parses `extern string_literal?`.
1460 fn parse_extern(&mut self, case: Case) -> Extern {
1461 if self.eat_keyword_case(kw::Extern, case) {
1462 let mut extern_span = self.prev_token.span;
1463 let abi = self.parse_abi();
1464 if let Some(abi) = abi {
1465 extern_span = extern_span.to(abi.span);
1467 Extern::from_abi(abi, extern_span)
1473 /// Parses a string literal as an ABI spec.
1474 fn parse_abi(&mut self) -> Option<StrLit> {
1475 match self.parse_str_lit() {
1476 Ok(str_lit) => Some(str_lit),
1477 Err(Some(lit)) => match lit.kind {
1478 ast::LitKind::Err => None,
1480 self.sess.emit_err(NonStringAbiLiteral { span: lit.span });
1488 pub fn collect_tokens_no_attrs<R: HasAttrs + HasTokens>(
1490 f: impl FnOnce(&mut Self) -> PResult<'a, R>,
1491 ) -> PResult<'a, R> {
1492 // The only reason to call `collect_tokens_no_attrs` is if you want tokens, so use
1493 // `ForceCollect::Yes`
1494 self.collect_tokens_trailing_token(
1495 AttrWrapper::empty(),
1497 |this, _attrs| Ok((f(this)?, TrailingToken::None)),
1502 fn is_import_coupler(&mut self) -> bool {
1503 self.check(&token::ModSep)
1504 && self.look_ahead(1, |t| {
1505 *t == token::OpenDelim(Delimiter::Brace) || *t == token::BinOp(token::Star)
1509 pub fn clear_expected_tokens(&mut self) {
1510 self.expected_tokens.clear();
1513 pub fn approx_token_stream_pos(&self) -> usize {
1514 self.token_cursor.num_next_calls
1518 pub(crate) fn make_unclosed_delims_error(
1519 unmatched: UnmatchedBrace,
1521 ) -> Option<DiagnosticBuilder<'_, ErrorGuaranteed>> {
1522 // `None` here means an `Eof` was found. We already emit those errors elsewhere, we add them to
1523 // `unmatched_braces` only for error recovery in the `Parser`.
1524 let found_delim = unmatched.found_delim?;
1525 let mut spans = vec![unmatched.found_span];
1526 if let Some(sp) = unmatched.unclosed_span {
1529 let err = MismatchedClosingDelimiter {
1531 delimiter: pprust::token_kind_to_string(&token::CloseDelim(found_delim)).to_string(),
1532 unmatched: unmatched.found_span,
1533 opening_candidate: unmatched.candidate_span,
1534 unclosed: unmatched.unclosed_span,
1536 .into_diagnostic(&sess.span_diagnostic);
1540 pub fn emit_unclosed_delims(unclosed_delims: &mut Vec<UnmatchedBrace>, sess: &ParseSess) {
1541 *sess.reached_eof.borrow_mut() |=
1542 unclosed_delims.iter().any(|unmatched_delim| unmatched_delim.found_delim.is_none());
1543 for unmatched in unclosed_delims.drain(..) {
1544 if let Some(mut e) = make_unclosed_delims_error(unmatched, sess) {
1550 /// A helper struct used when building an `AttrTokenStream` from
1551 /// a `LazyAttrTokenStream`. Both delimiter and non-delimited tokens
1552 /// are stored as `FlatToken::Token`. A vector of `FlatToken`s
1553 /// is then 'parsed' to build up an `AttrTokenStream` with nested
1554 /// `AttrTokenTree::Delimited` tokens.
1555 #[derive(Debug, Clone)]
1556 pub enum FlatToken {
1557 /// A token - this holds both delimiter (e.g. '{' and '}')
1558 /// and non-delimiter tokens
1560 /// Holds the `AttributesData` for an AST node. The
1561 /// `AttributesData` is inserted directly into the
1562 /// constructed `AttrTokenStream` as
1563 /// an `AttrTokenTree::Attributes`.
1564 AttrTarget(AttributesData),
1565 /// A special 'empty' token that is ignored during the conversion
1566 /// to an `AttrTokenStream`. This is used to simplify the
1567 /// handling of replace ranges.