5 pub use module::{ModulePath, ModulePathSuccess};
9 pub use path::PathStyle;
13 use diagnostics::Error;
15 use crate::lexer::UnmatchedBrace;
16 use crate::{Directory, DirectoryOwnership};
19 use rustc_ast_pretty::pprust;
20 use rustc_errors::{struct_span_err, Applicability, DiagnosticBuilder, FatalError, PResult};
21 use rustc_session::parse::ParseSess;
22 use rustc_span::source_map::respan;
23 use rustc_span::symbol::{kw, sym, Symbol};
24 use rustc_span::{FileName, Span, DUMMY_SP};
25 use syntax::ast::DUMMY_NODE_ID;
26 use syntax::ast::{self, AttrStyle, AttrVec, Const, CrateSugar, Extern, Ident, Unsafe};
27 use syntax::ast::{Async, MacArgs, MacDelimiter, Mutability, StrLit, Visibility, VisibilityKind};
29 use syntax::token::{self, DelimToken, Token, TokenKind};
30 use syntax::tokenstream::{self, DelimSpan, TokenStream, TokenTree, TreeAndJoint};
31 use syntax::util::comments::{doc_comment_style, strip_doc_comment_decoration};
33 use std::path::PathBuf;
34 use std::{cmp, mem, slice};
37 struct Restrictions: u8 {
38 const STMT_EXPR = 1 << 0;
39 const NO_STRUCT_LITERAL = 1 << 1;
43 #[derive(Clone, Copy, PartialEq, Debug)]
50 #[derive(Clone, Copy, PartialEq, Debug)]
56 /// Like `maybe_whole_expr`, but for things other than expressions.
58 macro_rules! maybe_whole {
59 ($p:expr, $constructor:ident, |$x:ident| $e:expr) => {
60 if let token::Interpolated(nt) = &$p.token.kind {
61 if let token::$constructor(x) = &**nt {
70 /// If the next tokens are ill-formed `$ty::` recover them as `<$ty>::`.
72 macro_rules! maybe_recover_from_interpolated_ty_qpath {
73 ($self: expr, $allow_qpath_recovery: expr) => {
74 if $allow_qpath_recovery && $self.look_ahead(1, |t| t == &token::ModSep) {
75 if let token::Interpolated(nt) = &$self.token.kind {
76 if let token::NtTy(ty) = &**nt {
79 return $self.maybe_recover_from_bad_qpath_stage_2($self.prev_span, ty);
87 pub struct Parser<'a> {
88 pub sess: &'a ParseSess,
89 /// The current normalized token.
90 /// "Normalized" means that some interpolated tokens
91 /// (`$i: ident` and `$l: lifetime` meta-variables) are replaced
92 /// with non-interpolated identifier and lifetime tokens they refer to.
93 /// Use span from this token if you need an isolated span.
95 /// The current non-normalized token if it's different from `token`.
96 /// Use span from this token if you need to concatenate it with some neighbouring spans.
97 unnormalized_token: Token,
98 /// The previous normalized token.
99 /// Use span from this token if you need an isolated span.
101 /// The previous non-normalized token if it's different from `prev_token`.
102 /// Use span from this token if you need to concatenate it with some neighbouring spans.
103 unnormalized_prev_token: Token,
104 /// Equivalent to `unnormalized_prev_token.span`.
105 /// FIXME: Remove in favor of `(unnormalized_)prev_token.span`.
107 restrictions: Restrictions,
108 /// Used to determine the path to externally loaded source files.
109 pub(super) directory: Directory,
110 /// `true` to parse sub-modules in other files.
111 // Public for rustfmt usage.
112 pub recurse_into_file_modules: bool,
113 /// Name of the root module this parser originated from. If `None`, then the
114 /// name is not known. This does not change while the parser is descending
115 /// into modules, and sub-parsers have new values for this name.
116 pub root_module_name: Option<String>,
117 expected_tokens: Vec<TokenType>,
118 token_cursor: TokenCursor,
119 desugar_doc_comments: bool,
120 /// `true` we should configure out of line modules as we parse.
121 // Public for rustfmt usage.
123 /// This field is used to keep track of how many left angle brackets we have seen. This is
124 /// required in order to detect extra leading left angle brackets (`<` characters) and error
127 /// See the comments in the `parse_path_segment` function for more details.
128 unmatched_angle_bracket_count: u32,
129 max_angle_bracket_count: u32,
130 /// A list of all unclosed delimiters found by the lexer. If an entry is used for error recovery
131 /// it gets removed from here. Every entry left at the end gets emitted as an independent
133 pub(super) unclosed_delims: Vec<UnmatchedBrace>,
134 last_unexpected_token_span: Option<Span>,
135 pub last_type_ascription: Option<(Span, bool /* likely path typo */)>,
136 /// If present, this `Parser` is not parsing Rust code but rather a macro call.
137 subparser_name: Option<&'static str>,
140 impl<'a> Drop for Parser<'a> {
142 emit_unclosed_delims(&mut self.unclosed_delims, &self.sess);
148 frame: TokenCursorFrame,
149 stack: Vec<TokenCursorFrame>,
153 struct TokenCursorFrame {
154 delim: token::DelimToken,
157 tree_cursor: tokenstream::Cursor,
159 last_token: LastToken,
162 /// This is used in `TokenCursorFrame` above to track tokens that are consumed
163 /// by the parser, and then that's transitively used to record the tokens that
164 /// each parse AST item is created with.
166 /// Right now this has two states, either collecting tokens or not collecting
167 /// tokens. If we're collecting tokens we just save everything off into a local
168 /// `Vec`. This should eventually though likely save tokens from the original
169 /// token stream and just use slicing of token streams to avoid creation of a
170 /// whole new vector.
172 /// The second state is where we're passively not recording tokens, but the last
173 /// token is still tracked for when we want to start recording tokens. This
174 /// "last token" means that when we start recording tokens we'll want to ensure
175 /// that this, the first token, is included in the output.
177 /// You can find some more example usage of this in the `collect_tokens` method
181 Collecting(Vec<TreeAndJoint>),
182 Was(Option<TreeAndJoint>),
185 impl TokenCursorFrame {
186 fn new(span: DelimSpan, delim: DelimToken, tts: &TokenStream) -> Self {
190 open_delim: delim == token::NoDelim,
191 tree_cursor: tts.clone().into_trees(),
192 close_delim: delim == token::NoDelim,
193 last_token: LastToken::Was(None),
199 fn next(&mut self) -> Token {
201 let tree = if !self.frame.open_delim {
202 self.frame.open_delim = true;
203 TokenTree::open_tt(self.frame.span, self.frame.delim)
204 } else if let Some(tree) = self.frame.tree_cursor.next() {
206 } else if !self.frame.close_delim {
207 self.frame.close_delim = true;
208 TokenTree::close_tt(self.frame.span, self.frame.delim)
209 } else if let Some(frame) = self.stack.pop() {
213 return Token::new(token::Eof, DUMMY_SP);
216 match self.frame.last_token {
217 LastToken::Collecting(ref mut v) => v.push(tree.clone().into()),
218 LastToken::Was(ref mut t) => *t = Some(tree.clone().into()),
222 TokenTree::Token(token) => return token,
223 TokenTree::Delimited(sp, delim, tts) => {
224 let frame = TokenCursorFrame::new(sp, delim, &tts);
225 self.stack.push(mem::replace(&mut self.frame, frame));
231 fn next_desugared(&mut self) -> Token {
232 let (name, sp) = match self.next() {
233 Token { kind: token::DocComment(name), span } => (name, span),
237 let stripped = strip_doc_comment_decoration(&name.as_str());
239 // Searches for the occurrences of `"#*` and returns the minimum number of `#`s
240 // required to wrap the text.
241 let mut num_of_hashes = 0;
243 for ch in stripped.chars() {
246 '#' if count > 0 => count + 1,
249 num_of_hashes = cmp::max(num_of_hashes, count);
252 let delim_span = DelimSpan::from_single(sp);
253 let body = TokenTree::Delimited(
257 TokenTree::token(token::Ident(sym::doc, false), sp),
258 TokenTree::token(token::Eq, sp),
260 TokenKind::lit(token::StrRaw(num_of_hashes), Symbol::intern(&stripped), None),
266 .collect::<TokenStream>()
270 self.stack.push(mem::replace(
272 TokenCursorFrame::new(
275 &if doc_comment_style(&name.as_str()) == AttrStyle::Inner {
276 [TokenTree::token(token::Pound, sp), TokenTree::token(token::Not, sp), body]
279 .collect::<TokenStream>()
281 [TokenTree::token(token::Pound, sp), body]
284 .collect::<TokenStream>()
293 #[derive(Clone, PartialEq)]
306 fn to_string(&self) -> String {
308 TokenType::Token(ref t) => format!("`{}`", pprust::token_kind_to_string(t)),
309 TokenType::Keyword(kw) => format!("`{}`", kw),
310 TokenType::Operator => "an operator".to_string(),
311 TokenType::Lifetime => "lifetime".to_string(),
312 TokenType::Ident => "identifier".to_string(),
313 TokenType::Path => "path".to_string(),
314 TokenType::Type => "type".to_string(),
315 TokenType::Const => "const".to_string(),
320 #[derive(Copy, Clone, Debug)]
321 enum TokenExpectType {
326 /// A sequence separator.
328 /// The separator token.
329 sep: Option<TokenKind>,
330 /// `true` if a trailing separator is allowed.
331 trailing_sep_allowed: bool,
335 fn trailing_allowed(t: TokenKind) -> SeqSep {
336 SeqSep { sep: Some(t), trailing_sep_allowed: true }
339 fn none() -> SeqSep {
340 SeqSep { sep: None, trailing_sep_allowed: false }
344 pub enum FollowedByType {
349 fn token_descr_opt(token: &Token) -> Option<&'static str> {
350 Some(match token.kind {
351 _ if token.is_special_ident() => "reserved identifier",
352 _ if token.is_used_keyword() => "keyword",
353 _ if token.is_unused_keyword() => "reserved keyword",
354 token::DocComment(..) => "doc comment",
359 pub(super) fn token_descr(token: &Token) -> String {
360 let token_str = pprust::token_to_string(token);
361 match token_descr_opt(token) {
362 Some(prefix) => format!("{} `{}`", prefix, token_str),
363 _ => format!("`{}`", token_str),
367 impl<'a> Parser<'a> {
371 directory: Option<Directory>,
372 recurse_into_file_modules: bool,
373 desugar_doc_comments: bool,
374 subparser_name: Option<&'static str>,
376 let mut parser = Parser {
378 token: Token::dummy(),
379 unnormalized_token: Token::dummy(),
380 prev_token: Token::dummy(),
381 unnormalized_prev_token: Token::dummy(),
383 restrictions: Restrictions::empty(),
384 recurse_into_file_modules,
385 directory: Directory {
386 path: PathBuf::new(),
387 ownership: DirectoryOwnership::Owned { relative: None },
389 root_module_name: None,
390 expected_tokens: Vec::new(),
391 token_cursor: TokenCursor {
392 frame: TokenCursorFrame::new(DelimSpan::dummy(), token::NoDelim, &tokens.into()),
395 desugar_doc_comments,
397 unmatched_angle_bracket_count: 0,
398 max_angle_bracket_count: 0,
399 unclosed_delims: Vec::new(),
400 last_unexpected_token_span: None,
401 last_type_ascription: None,
405 // Make parser point to the first token.
408 if let Some(directory) = directory {
409 parser.directory = directory;
410 } else if !parser.token.span.is_dummy() {
411 if let Some(FileName::Real(path)) =
412 &sess.source_map().lookup_char_pos(parser.token.span.lo()).file.unmapped_path
414 if let Some(directory_path) = path.parent() {
415 parser.directory.path = directory_path.to_path_buf();
423 fn next_tok(&mut self, fallback_span: Span) -> Token {
424 let mut next = if self.desugar_doc_comments {
425 self.token_cursor.next_desugared()
427 self.token_cursor.next()
429 if next.span.is_dummy() {
430 // Tweak the location for better diagnostics, but keep syntactic context intact.
431 next.span = fallback_span.with_ctxt(next.span.ctxt());
436 crate fn unexpected<T>(&mut self) -> PResult<'a, T> {
437 match self.expect_one_of(&[], &[]) {
439 // We can get `Ok(true)` from `recover_closing_delimiter`
440 // which is called in `expected_one_of_not_found`.
441 Ok(_) => FatalError.raise(),
445 /// Expects and consumes the token `t`. Signals an error if the next token is not `t`.
446 pub fn expect(&mut self, t: &TokenKind) -> PResult<'a, bool /* recovered */> {
447 if self.expected_tokens.is_empty() {
448 if self.token == *t {
452 self.unexpected_try_recover(t)
455 self.expect_one_of(slice::from_ref(t), &[])
459 /// Expect next token to be edible or inedible token. If edible,
460 /// then consume it; if inedible, then return without consuming
461 /// anything. Signal a fatal error if next token is unexpected.
462 pub fn expect_one_of(
464 edible: &[TokenKind],
465 inedible: &[TokenKind],
466 ) -> PResult<'a, bool /* recovered */> {
467 if edible.contains(&self.token.kind) {
470 } else if inedible.contains(&self.token.kind) {
471 // leave it in the input
473 } else if self.last_unexpected_token_span == Some(self.token.span) {
476 self.expected_one_of_not_found(edible, inedible)
480 // Public for rustfmt usage.
481 pub fn parse_ident(&mut self) -> PResult<'a, ast::Ident> {
482 self.parse_ident_common(true)
485 fn parse_ident_common(&mut self, recover: bool) -> PResult<'a, ast::Ident> {
486 match self.token.kind {
487 token::Ident(name, _) => {
488 if self.token.is_reserved_ident() {
489 let mut err = self.expected_ident_found();
496 let span = self.token.span;
498 Ok(Ident::new(name, span))
500 _ => Err(match self.prev_token.kind {
501 TokenKind::DocComment(..) => {
502 self.span_fatal_err(self.prev_span, Error::UselessDocComment)
504 _ => self.expected_ident_found(),
509 /// Checks if the next token is `tok`, and returns `true` if so.
511 /// This method will automatically add `tok` to `expected_tokens` if `tok` is not
513 fn check(&mut self, tok: &TokenKind) -> bool {
514 let is_present = self.token == *tok;
516 self.expected_tokens.push(TokenType::Token(tok.clone()));
521 /// Consumes a token 'tok' if it exists. Returns whether the given token was present.
522 pub fn eat(&mut self, tok: &TokenKind) -> bool {
523 let is_present = self.check(tok);
530 /// If the next token is the given keyword, returns `true` without eating it.
531 /// An expectation is also added for diagnostics purposes.
532 fn check_keyword(&mut self, kw: Symbol) -> bool {
533 self.expected_tokens.push(TokenType::Keyword(kw));
534 self.token.is_keyword(kw)
537 /// If the next token is the given keyword, eats it and returns `true`.
538 /// Otherwise, returns `false`. An expectation is also added for diagnostics purposes.
539 // Public for rustfmt usage.
540 pub fn eat_keyword(&mut self, kw: Symbol) -> bool {
541 if self.check_keyword(kw) {
549 fn eat_keyword_noexpect(&mut self, kw: Symbol) -> bool {
550 if self.token.is_keyword(kw) {
558 /// If the given word is not a keyword, signals an error.
559 /// If the next token is not the given word, signals an error.
560 /// Otherwise, eats it.
561 fn expect_keyword(&mut self, kw: Symbol) -> PResult<'a, ()> {
562 if !self.eat_keyword(kw) { self.unexpected() } else { Ok(()) }
565 /// Is the given keyword `kw` followed by a non-reserved identifier?
566 fn is_kw_followed_by_ident(&self, kw: Symbol) -> bool {
567 self.token.is_keyword(kw) && self.look_ahead(1, |t| t.is_ident() && !t.is_reserved_ident())
570 fn check_or_expected(&mut self, ok: bool, typ: TokenType) -> bool {
574 self.expected_tokens.push(typ);
579 fn check_ident(&mut self) -> bool {
580 self.check_or_expected(self.token.is_ident(), TokenType::Ident)
583 fn check_path(&mut self) -> bool {
584 self.check_or_expected(self.token.is_path_start(), TokenType::Path)
587 fn check_type(&mut self) -> bool {
588 self.check_or_expected(self.token.can_begin_type(), TokenType::Type)
591 fn check_const_arg(&mut self) -> bool {
592 self.check_or_expected(self.token.can_begin_const_arg(), TokenType::Const)
595 /// Checks to see if the next token is either `+` or `+=`.
596 /// Otherwise returns `false`.
597 fn check_plus(&mut self) -> bool {
598 self.check_or_expected(
599 self.token.is_like_plus(),
600 TokenType::Token(token::BinOp(token::Plus)),
604 /// Expects and consumes a `+`. if `+=` is seen, replaces it with a `=`
605 /// and continues. If a `+` is not seen, returns `false`.
607 /// This is used when token-splitting `+=` into `+`.
608 /// See issue #47856 for an example of when this may occur.
609 fn eat_plus(&mut self) -> bool {
610 self.expected_tokens.push(TokenType::Token(token::BinOp(token::Plus)));
611 match self.token.kind {
612 token::BinOp(token::Plus) => {
616 token::BinOpEq(token::Plus) => {
617 let start_point = self.sess.source_map().start_point(self.token.span);
618 self.bump_with(token::Eq, self.token.span.with_lo(start_point.hi()));
625 /// Expects and consumes an `&`. If `&&` is seen, replaces it with a single
626 /// `&` and continues. If an `&` is not seen, signals an error.
627 fn expect_and(&mut self) -> PResult<'a, ()> {
628 self.expected_tokens.push(TokenType::Token(token::BinOp(token::And)));
629 match self.token.kind {
630 token::BinOp(token::And) => {
635 let start_point = self.sess.source_map().start_point(self.token.span);
637 .bump_with(token::BinOp(token::And), self.token.span.with_lo(start_point.hi())))
639 _ => self.unexpected(),
643 /// Expects and consumes an `|`. If `||` is seen, replaces it with a single
644 /// `|` and continues. If an `|` is not seen, signals an error.
645 fn expect_or(&mut self) -> PResult<'a, ()> {
646 self.expected_tokens.push(TokenType::Token(token::BinOp(token::Or)));
647 match self.token.kind {
648 token::BinOp(token::Or) => {
653 let start_point = self.sess.source_map().start_point(self.token.span);
655 .bump_with(token::BinOp(token::Or), self.token.span.with_lo(start_point.hi())))
657 _ => self.unexpected(),
661 /// Attempts to consume a `<`. If `<<` is seen, replaces it with a single
662 /// `<` and continue. If `<-` is seen, replaces it with a single `<`
663 /// and continue. If a `<` is not seen, returns false.
665 /// This is meant to be used when parsing generics on a path to get the
667 fn eat_lt(&mut self) -> bool {
668 self.expected_tokens.push(TokenType::Token(token::Lt));
669 let ate = match self.token.kind {
674 token::BinOp(token::Shl) => {
675 let start_point = self.sess.source_map().start_point(self.token.span);
676 self.bump_with(token::Lt, self.token.span.with_lo(start_point.hi()));
680 let start_point = self.sess.source_map().start_point(self.token.span);
682 token::BinOp(token::Minus),
683 self.token.span.with_lo(start_point.hi()),
691 // See doc comment for `unmatched_angle_bracket_count`.
692 self.unmatched_angle_bracket_count += 1;
693 self.max_angle_bracket_count += 1;
694 debug!("eat_lt: (increment) count={:?}", self.unmatched_angle_bracket_count);
700 fn expect_lt(&mut self) -> PResult<'a, ()> {
701 if !self.eat_lt() { self.unexpected() } else { Ok(()) }
704 /// Expects and consumes a single `>` token. if a `>>` is seen, replaces it
705 /// with a single `>` and continues. If a `>` is not seen, signals an error.
706 fn expect_gt(&mut self) -> PResult<'a, ()> {
707 self.expected_tokens.push(TokenType::Token(token::Gt));
708 let ate = match self.token.kind {
713 token::BinOp(token::Shr) => {
714 let start_point = self.sess.source_map().start_point(self.token.span);
715 Some(self.bump_with(token::Gt, self.token.span.with_lo(start_point.hi())))
717 token::BinOpEq(token::Shr) => {
718 let start_point = self.sess.source_map().start_point(self.token.span);
719 Some(self.bump_with(token::Ge, self.token.span.with_lo(start_point.hi())))
722 let start_point = self.sess.source_map().start_point(self.token.span);
723 Some(self.bump_with(token::Eq, self.token.span.with_lo(start_point.hi())))
730 // See doc comment for `unmatched_angle_bracket_count`.
731 if self.unmatched_angle_bracket_count > 0 {
732 self.unmatched_angle_bracket_count -= 1;
733 debug!("expect_gt: (decrement) count={:?}", self.unmatched_angle_bracket_count);
738 None => self.unexpected(),
742 fn expect_any_with_type(&mut self, kets: &[&TokenKind], expect: TokenExpectType) -> bool {
743 kets.iter().any(|k| match expect {
744 TokenExpectType::Expect => self.check(k),
745 TokenExpectType::NoExpect => self.token == **k,
749 fn parse_seq_to_before_tokens<T>(
753 expect: TokenExpectType,
754 mut f: impl FnMut(&mut Parser<'a>) -> PResult<'a, T>,
755 ) -> PResult<'a, (Vec<T>, bool /* trailing */, bool /* recovered */)> {
756 let mut first = true;
757 let mut recovered = false;
758 let mut trailing = false;
760 while !self.expect_any_with_type(kets, expect) {
761 if let token::CloseDelim(..) | token::Eof = self.token.kind {
764 if let Some(ref t) = sep.sep {
768 match self.expect(t) {
774 Err(mut expect_err) => {
775 let sp = self.prev_span.shrink_to_hi();
776 let token_str = pprust::token_kind_to_string(t);
778 // Attempt to keep parsing if it was a similar separator.
779 if let Some(ref tokens) = t.similar_tokens() {
780 if tokens.contains(&self.token.kind) {
785 // Attempt to keep parsing if it was an omitted separator.
788 // Parsed successfully, therefore most probably the code only
789 // misses a separator.
791 .span_suggestion_short(
793 &format!("missing `{}`", token_str),
795 Applicability::MaybeIncorrect,
803 // Parsing failed, therefore it must be something more serious
804 // than just a missing separator.
815 if sep.trailing_sep_allowed && self.expect_any_with_type(kets, expect) {
824 Ok((v, trailing, recovered))
827 /// Parses a sequence, not including the closing delimiter. The function
828 /// `f` must consume tokens until reaching the next separator or
830 fn parse_seq_to_before_end<T>(
834 f: impl FnMut(&mut Parser<'a>) -> PResult<'a, T>,
835 ) -> PResult<'a, (Vec<T>, bool, bool)> {
836 self.parse_seq_to_before_tokens(&[ket], sep, TokenExpectType::Expect, f)
839 /// Parses a sequence, including the closing delimiter. The function
840 /// `f` must consume tokens until reaching the next separator or
842 fn parse_seq_to_end<T>(
846 f: impl FnMut(&mut Parser<'a>) -> PResult<'a, T>,
847 ) -> PResult<'a, (Vec<T>, bool /* trailing */)> {
848 let (val, trailing, recovered) = self.parse_seq_to_before_end(ket, sep, f)?;
855 /// Parses a sequence, including the closing delimiter. The function
856 /// `f` must consume tokens until reaching the next separator or
858 fn parse_unspanned_seq<T>(
863 f: impl FnMut(&mut Parser<'a>) -> PResult<'a, T>,
864 ) -> PResult<'a, (Vec<T>, bool)> {
866 self.parse_seq_to_end(ket, sep, f)
869 fn parse_delim_comma_seq<T>(
872 f: impl FnMut(&mut Parser<'a>) -> PResult<'a, T>,
873 ) -> PResult<'a, (Vec<T>, bool)> {
874 self.parse_unspanned_seq(
875 &token::OpenDelim(delim),
876 &token::CloseDelim(delim),
877 SeqSep::trailing_allowed(token::Comma),
882 fn parse_paren_comma_seq<T>(
884 f: impl FnMut(&mut Parser<'a>) -> PResult<'a, T>,
885 ) -> PResult<'a, (Vec<T>, bool)> {
886 self.parse_delim_comma_seq(token::Paren, f)
889 // Interpolated identifier (`$i: ident`) and lifetime (`$l: lifetime`)
890 // tokens are replaced with usual identifier and lifetime tokens,
891 // so the former are never encountered during normal parsing.
892 crate fn set_token(&mut self, token: Token) {
893 self.unnormalized_token = token;
894 self.token = match &self.unnormalized_token.kind {
895 token::Interpolated(nt) => match **nt {
896 token::NtIdent(ident, is_raw) => {
897 Token::new(token::Ident(ident.name, is_raw), ident.span)
899 token::NtLifetime(ident) => Token::new(token::Lifetime(ident.name), ident.span),
900 _ => self.unnormalized_token.clone(),
902 _ => self.unnormalized_token.clone(),
906 /// Advance the parser by one token.
907 pub fn bump(&mut self) {
908 if self.prev_token.kind == TokenKind::Eof {
909 // Bumping after EOF is a bad sign, usually an infinite loop.
910 let msg = "attempted to bump the parser past EOF (may be stuck in a loop)";
911 self.span_bug(self.token.span, msg);
914 // Update the current and previous tokens.
915 self.prev_token = self.token.take();
916 self.unnormalized_prev_token = self.unnormalized_token.take();
917 let next_token = self.next_tok(self.unnormalized_prev_token.span);
918 self.set_token(next_token);
920 // Update fields derived from the previous token.
921 self.prev_span = self.unnormalized_prev_token.span;
923 self.expected_tokens.clear();
926 /// Advances the parser using provided token as a next one. Use this when
927 /// consuming a part of a token. For example a single `<` from `<<`.
928 /// FIXME: this function sets the previous token data to some semi-nonsensical values
929 /// which kind of work because they are currently used in very limited ways in practice.
930 /// Correct token kinds and spans need to be calculated instead.
931 fn bump_with(&mut self, next: TokenKind, span: Span) {
932 // Update the current and previous tokens.
933 self.prev_token = self.token.take();
934 self.unnormalized_prev_token = self.unnormalized_token.take();
935 self.set_token(Token::new(next, span));
937 // Update fields derived from the previous token.
938 self.prev_span = self.unnormalized_prev_token.span.with_hi(span.lo());
940 self.expected_tokens.clear();
943 /// Look-ahead `dist` tokens of `self.token` and get access to that token there.
944 /// When `dist == 0` then the current token is looked at.
945 pub fn look_ahead<R>(&self, dist: usize, looker: impl FnOnce(&Token) -> R) -> R {
947 return looker(&self.token);
950 let frame = &self.token_cursor.frame;
951 looker(&match frame.tree_cursor.look_ahead(dist - 1) {
952 Some(tree) => match tree {
953 TokenTree::Token(token) => token,
954 TokenTree::Delimited(dspan, delim, _) => {
955 Token::new(token::OpenDelim(delim), dspan.open)
958 None => Token::new(token::CloseDelim(frame.delim), frame.span.close),
962 /// Returns whether any of the given keywords are `dist` tokens ahead of the current one.
963 fn is_keyword_ahead(&self, dist: usize, kws: &[Symbol]) -> bool {
964 self.look_ahead(dist, |t| kws.iter().any(|&kw| t.is_keyword(kw)))
967 /// Parses asyncness: `async` or nothing.
968 fn parse_asyncness(&mut self) -> Async {
969 if self.eat_keyword(kw::Async) {
970 let span = self.prev_span;
971 Async::Yes { span, closure_id: DUMMY_NODE_ID, return_impl_trait_id: DUMMY_NODE_ID }
977 /// Parses unsafety: `unsafe` or nothing.
978 fn parse_unsafety(&mut self) -> Unsafe {
979 if self.eat_keyword(kw::Unsafe) { Unsafe::Yes(self.prev_span) } else { Unsafe::No }
982 /// Parses constness: `const` or nothing.
983 fn parse_constness(&mut self) -> Const {
984 if self.eat_keyword(kw::Const) { Const::Yes(self.prev_span) } else { Const::No }
987 /// Parses mutability (`mut` or nothing).
988 fn parse_mutability(&mut self) -> Mutability {
989 if self.eat_keyword(kw::Mut) { Mutability::Mut } else { Mutability::Not }
992 /// Possibly parses mutability (`const` or `mut`).
993 fn parse_const_or_mut(&mut self) -> Option<Mutability> {
994 if self.eat_keyword(kw::Mut) {
995 Some(Mutability::Mut)
996 } else if self.eat_keyword(kw::Const) {
997 Some(Mutability::Not)
1003 fn parse_field_name(&mut self) -> PResult<'a, Ident> {
1004 if let token::Literal(token::Lit { kind: token::Integer, symbol, suffix }) = self.token.kind
1006 self.expect_no_suffix(self.token.span, "a tuple index", suffix);
1008 Ok(Ident::new(symbol, self.prev_span))
1010 self.parse_ident_common(false)
1014 fn parse_mac_args(&mut self) -> PResult<'a, P<MacArgs>> {
1015 self.parse_mac_args_common(true).map(P)
1018 fn parse_attr_args(&mut self) -> PResult<'a, MacArgs> {
1019 self.parse_mac_args_common(false)
1022 fn parse_mac_args_common(&mut self, delimited_only: bool) -> PResult<'a, MacArgs> {
1024 if self.check(&token::OpenDelim(DelimToken::Paren))
1025 || self.check(&token::OpenDelim(DelimToken::Bracket))
1026 || self.check(&token::OpenDelim(DelimToken::Brace))
1028 match self.parse_token_tree() {
1029 TokenTree::Delimited(dspan, delim, tokens) =>
1030 // We've confirmed above that there is a delimiter so unwrapping is OK.
1032 MacArgs::Delimited(dspan, MacDelimiter::from_token(delim).unwrap(), tokens)
1034 _ => unreachable!(),
1036 } else if !delimited_only {
1037 if self.eat(&token::Eq) {
1038 let eq_span = self.prev_span;
1039 let mut is_interpolated_expr = false;
1040 if let token::Interpolated(nt) = &self.token.kind {
1041 if let token::NtExpr(..) = **nt {
1042 is_interpolated_expr = true;
1045 let token_tree = if is_interpolated_expr {
1046 // We need to accept arbitrary interpolated expressions to continue
1047 // supporting things like `doc = $expr` that work on stable.
1048 // Non-literal interpolated expressions are rejected after expansion.
1049 self.parse_token_tree()
1051 self.parse_unsuffixed_lit()?.token_tree()
1054 MacArgs::Eq(eq_span, token_tree.into())
1059 return self.unexpected();
1064 fn parse_or_use_outer_attributes(
1066 already_parsed_attrs: Option<AttrVec>,
1067 ) -> PResult<'a, AttrVec> {
1068 if let Some(attrs) = already_parsed_attrs {
1071 self.parse_outer_attributes().map(|a| a.into())
1075 /// Parses a single token tree from the input.
1076 pub fn parse_token_tree(&mut self) -> TokenTree {
1077 match self.token.kind {
1078 token::OpenDelim(..) => {
1079 let frame = mem::replace(
1080 &mut self.token_cursor.frame,
1081 self.token_cursor.stack.pop().unwrap(),
1083 self.set_token(Token::new(TokenKind::CloseDelim(frame.delim), frame.span.close));
1085 TokenTree::Delimited(frame.span, frame.delim, frame.tree_cursor.stream.into())
1087 token::CloseDelim(_) | token::Eof => unreachable!(),
1090 TokenTree::Token(self.prev_token.clone())
1095 /// Parses a stream of tokens into a list of `TokenTree`s, up to EOF.
1096 pub fn parse_all_token_trees(&mut self) -> PResult<'a, Vec<TokenTree>> {
1097 let mut tts = Vec::new();
1098 while self.token != token::Eof {
1099 tts.push(self.parse_token_tree());
1104 pub fn parse_tokens(&mut self) -> TokenStream {
1105 let mut result = Vec::new();
1107 match self.token.kind {
1108 token::Eof | token::CloseDelim(..) => break,
1109 _ => result.push(self.parse_token_tree().into()),
1112 TokenStream::new(result)
1115 /// Evaluates the closure with restrictions in place.
1117 /// Afters the closure is evaluated, restrictions are reset.
1118 fn with_res<T>(&mut self, res: Restrictions, f: impl FnOnce(&mut Self) -> T) -> T {
1119 let old = self.restrictions;
1120 self.restrictions = res;
1122 self.restrictions = old;
1126 fn is_crate_vis(&self) -> bool {
1127 self.token.is_keyword(kw::Crate) && self.look_ahead(1, |t| t != &token::ModSep)
1130 /// Parses `pub`, `pub(crate)` and `pub(in path)` plus shortcuts `crate` for `pub(crate)`,
1131 /// `pub(self)` for `pub(in self)` and `pub(super)` for `pub(in super)`.
1132 /// If the following element can't be a tuple (i.e., it's a function definition), then
1133 /// it's not a tuple struct field), and the contents within the parentheses isn't valid,
1134 /// so emit a proper diagnostic.
1135 pub fn parse_visibility(&mut self, fbt: FollowedByType) -> PResult<'a, Visibility> {
1136 maybe_whole!(self, NtVis, |x| x);
1138 self.expected_tokens.push(TokenType::Keyword(kw::Crate));
1139 if self.is_crate_vis() {
1140 self.bump(); // `crate`
1141 self.sess.gated_spans.gate(sym::crate_visibility_modifier, self.prev_span);
1142 return Ok(respan(self.prev_span, VisibilityKind::Crate(CrateSugar::JustCrate)));
1145 if !self.eat_keyword(kw::Pub) {
1146 // We need a span for our `Spanned<VisibilityKind>`, but there's inherently no
1147 // keyword to grab a span from for inherited visibility; an empty span at the
1148 // beginning of the current token would seem to be the "Schelling span".
1149 return Ok(respan(self.token.span.shrink_to_lo(), VisibilityKind::Inherited));
1151 let lo = self.prev_span;
1153 if self.check(&token::OpenDelim(token::Paren)) {
1154 // We don't `self.bump()` the `(` yet because this might be a struct definition where
1155 // `()` or a tuple might be allowed. For example, `struct Struct(pub (), pub (usize));`.
1156 // Because of this, we only `bump` the `(` if we're assured it is appropriate to do so
1157 // by the following tokens.
1158 if self.is_keyword_ahead(1, &[kw::Crate]) && self.look_ahead(2, |t| t != &token::ModSep)
1159 // account for `pub(crate::foo)`
1161 // Parse `pub(crate)`.
1163 self.bump(); // `crate`
1164 self.expect(&token::CloseDelim(token::Paren))?; // `)`
1165 let vis = VisibilityKind::Crate(CrateSugar::PubCrate);
1166 return Ok(respan(lo.to(self.prev_span), vis));
1167 } else if self.is_keyword_ahead(1, &[kw::In]) {
1168 // Parse `pub(in path)`.
1170 self.bump(); // `in`
1171 let path = self.parse_path(PathStyle::Mod)?; // `path`
1172 self.expect(&token::CloseDelim(token::Paren))?; // `)`
1173 let vis = VisibilityKind::Restricted { path: P(path), id: ast::DUMMY_NODE_ID };
1174 return Ok(respan(lo.to(self.prev_span), vis));
1175 } else if self.look_ahead(2, |t| t == &token::CloseDelim(token::Paren))
1176 && self.is_keyword_ahead(1, &[kw::Super, kw::SelfLower])
1178 // Parse `pub(self)` or `pub(super)`.
1180 let path = self.parse_path(PathStyle::Mod)?; // `super`/`self`
1181 self.expect(&token::CloseDelim(token::Paren))?; // `)`
1182 let vis = VisibilityKind::Restricted { path: P(path), id: ast::DUMMY_NODE_ID };
1183 return Ok(respan(lo.to(self.prev_span), vis));
1184 } else if let FollowedByType::No = fbt {
1185 // Provide this diagnostic if a type cannot follow;
1186 // in particular, if this is not a tuple struct.
1187 self.recover_incorrect_vis_restriction()?;
1188 // Emit diagnostic, but continue with public visibility.
1192 Ok(respan(lo, VisibilityKind::Public))
1195 /// Recovery for e.g. `pub(something) fn ...` or `struct X { pub(something) y: Z }`
1196 fn recover_incorrect_vis_restriction(&mut self) -> PResult<'a, ()> {
1198 let path = self.parse_path(PathStyle::Mod)?;
1199 self.expect(&token::CloseDelim(token::Paren))?; // `)`
1201 let msg = "incorrect visibility restriction";
1202 let suggestion = r##"some possible visibility restrictions are:
1203 `pub(crate)`: visible only on the current crate
1204 `pub(super)`: visible only in the current module's parent
1205 `pub(in path::to::module)`: visible only on the specified path"##;
1207 let path_str = pprust::path_to_string(&path);
1209 struct_span_err!(self.sess.span_diagnostic, path.span, E0704, "{}", msg)
1213 &format!("make this visible only to module `{}` with `in`", path_str),
1214 format!("in {}", path_str),
1215 Applicability::MachineApplicable,
1222 /// Parses `extern string_literal?`.
1223 fn parse_extern(&mut self) -> PResult<'a, Extern> {
1224 Ok(if self.eat_keyword(kw::Extern) {
1225 Extern::from_abi(self.parse_abi())
1231 /// Parses a string literal as an ABI spec.
1232 fn parse_abi(&mut self) -> Option<StrLit> {
1233 match self.parse_str_lit() {
1234 Ok(str_lit) => Some(str_lit),
1235 Err(Some(lit)) => match lit.kind {
1236 ast::LitKind::Err(_) => None,
1238 self.struct_span_err(lit.span, "non-string ABI literal")
1241 "specify the ABI with a string literal",
1242 "\"C\"".to_string(),
1243 Applicability::MaybeIncorrect,
1253 fn collect_tokens<R>(
1255 f: impl FnOnce(&mut Self) -> PResult<'a, R>,
1256 ) -> PResult<'a, (R, TokenStream)> {
1257 // Record all tokens we parse when parsing this item.
1258 let mut tokens = Vec::new();
1259 let prev_collecting = match self.token_cursor.frame.last_token {
1260 LastToken::Collecting(ref mut list) => Some(mem::take(list)),
1261 LastToken::Was(ref mut last) => {
1262 tokens.extend(last.take());
1266 self.token_cursor.frame.last_token = LastToken::Collecting(tokens);
1267 let prev = self.token_cursor.stack.len();
1269 let last_token = if self.token_cursor.stack.len() == prev {
1270 &mut self.token_cursor.frame.last_token
1271 } else if self.token_cursor.stack.get(prev).is_none() {
1272 // This can happen due to a bad interaction of two unrelated recovery mechanisms with
1273 // mismatched delimiters *and* recovery lookahead on the likely typo `pub ident(`
1275 return Ok((ret?, TokenStream::default()));
1277 &mut self.token_cursor.stack[prev].last_token
1280 // Pull out the tokens that we've collected from the call to `f` above.
1281 let mut collected_tokens = match *last_token {
1282 LastToken::Collecting(ref mut v) => mem::take(v),
1283 LastToken::Was(ref was) => {
1284 let msg = format!("our vector went away? - found Was({:?})", was);
1285 debug!("collect_tokens: {}", msg);
1286 self.sess.span_diagnostic.delay_span_bug(self.token.span, &msg);
1287 // This can happen due to a bad interaction of two unrelated recovery mechanisms
1288 // with mismatched delimiters *and* recovery lookahead on the likely typo
1289 // `pub ident(` (#62895, different but similar to the case above).
1290 return Ok((ret?, TokenStream::default()));
1294 // If we're not at EOF our current token wasn't actually consumed by
1295 // `f`, but it'll still be in our list that we pulled out. In that case
1297 let extra_token = if self.token != token::Eof { collected_tokens.pop() } else { None };
1299 // If we were previously collecting tokens, then this was a recursive
1300 // call. In that case we need to record all the tokens we collected in
1301 // our parent list as well. To do that we push a clone of our stream
1302 // onto the previous list.
1303 match prev_collecting {
1305 list.extend(collected_tokens.iter().cloned());
1306 list.extend(extra_token);
1307 *last_token = LastToken::Collecting(list);
1310 *last_token = LastToken::Was(extra_token);
1314 Ok((ret?, TokenStream::new(collected_tokens)))
1318 fn is_import_coupler(&mut self) -> bool {
1319 self.check(&token::ModSep)
1320 && self.look_ahead(1, |t| {
1321 *t == token::OpenDelim(token::Brace) || *t == token::BinOp(token::Star)
1326 crate fn make_unclosed_delims_error(
1327 unmatched: UnmatchedBrace,
1329 ) -> Option<DiagnosticBuilder<'_>> {
1330 // `None` here means an `Eof` was found. We already emit those errors elsewhere, we add them to
1331 // `unmatched_braces` only for error recovery in the `Parser`.
1332 let found_delim = unmatched.found_delim?;
1333 let mut err = sess.span_diagnostic.struct_span_err(
1334 unmatched.found_span,
1336 "mismatched closing delimiter: `{}`",
1337 pprust::token_kind_to_string(&token::CloseDelim(found_delim)),
1340 err.span_label(unmatched.found_span, "mismatched closing delimiter");
1341 if let Some(sp) = unmatched.candidate_span {
1342 err.span_label(sp, "closing delimiter possibly meant for this");
1344 if let Some(sp) = unmatched.unclosed_span {
1345 err.span_label(sp, "unclosed delimiter");
1350 pub fn emit_unclosed_delims(unclosed_delims: &mut Vec<UnmatchedBrace>, sess: &ParseSess) {
1351 *sess.reached_eof.borrow_mut() |=
1352 unclosed_delims.iter().any(|unmatched_delim| unmatched_delim.found_delim.is_none());
1353 for unmatched in unclosed_delims.drain(..) {
1354 make_unclosed_delims_error(unmatched, sess).map(|mut e| {