1 // Copyright 2012-2013 The Rust Project Developers. See the COPYRIGHT
2 // file at the top-level directory of this distribution and at
3 // http://rust-lang.org/COPYRIGHT.
5 // Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
6 // http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
7 // <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
8 // option. This file may not be copied, modified, or distributed
9 // except according to those terms.
11 pub use self::BinOpToken::*;
12 pub use self::Nonterminal::*;
13 pub use self::DelimToken::*;
15 pub use self::Token::*;
21 use serialize::{Decodable, Decoder, Encodable, Encoder};
23 use syntax::parse::parse_stream_from_source_str;
24 use syntax_pos::{self, Span, FileName};
25 use syntax_pos::symbol::{self, Symbol};
26 use tokenstream::{self, DelimSpan, TokenStream, TokenTree};
30 use rustc_data_structures::sync::{Lrc, Lock};
32 #[derive(Clone, PartialEq, RustcEncodable, RustcDecodable, Hash, Debug, Copy)]
47 #[derive(Clone, PartialEq, RustcEncodable, RustcDecodable, Hash, Debug, Copy)]
49 /// A round parenthesis: `(` or `)`
51 /// A square bracket: `[` or `]`
53 /// A curly brace: `{` or `}`
55 /// An empty delimiter
60 pub fn len(self) -> usize {
61 if self == NoDelim { 0 } else { 1 }
64 pub fn is_empty(self) -> bool {
69 #[derive(Clone, PartialEq, RustcEncodable, RustcDecodable, Hash, Debug, Copy)]
76 StrRaw(ast::Name, u16), /* raw str delimited by n hash symbols */
78 ByteStrRaw(ast::Name, u16), /* raw byte str delimited by n hash symbols */
82 crate fn literal_name(&self) -> &'static str {
84 Byte(_) => "byte literal",
85 Char(_) => "char literal",
86 Integer(_) => "integer literal",
87 Float(_) => "float literal",
88 Str_(_) | StrRaw(..) => "string literal",
89 ByteStr(_) | ByteStrRaw(..) => "byte string literal"
93 // See comments in `interpolated_to_tokenstream` for why we care about
94 // *probably* equal here rather than actual equality
95 fn probably_equal_for_proc_macro(&self, other: &Lit) -> bool {
96 mem::discriminant(self) == mem::discriminant(other)
100 pub(crate) fn ident_can_begin_expr(ident: ast::Ident, is_raw: bool) -> bool {
101 let ident_token: Token = Ident(ident, is_raw);
103 !ident_token.is_reserved_ident() ||
104 ident_token.is_path_segment_keyword() ||
106 keywords::Async.name(),
108 keywords::Box.name(),
109 keywords::Break.name(),
110 keywords::Continue.name(),
111 keywords::False.name(),
112 keywords::For.name(),
114 keywords::Loop.name(),
115 keywords::Match.name(),
116 keywords::Move.name(),
117 keywords::Return.name(),
118 keywords::True.name(),
119 keywords::Unsafe.name(),
120 keywords::While.name(),
121 keywords::Yield.name(),
122 keywords::Static.name(),
123 ].contains(&ident.name)
126 fn ident_can_begin_type(ident: ast::Ident, is_raw: bool) -> bool {
127 let ident_token: Token = Ident(ident, is_raw);
129 !ident_token.is_reserved_ident() ||
130 ident_token.is_path_segment_keyword() ||
132 keywords::Underscore.name(),
133 keywords::For.name(),
134 keywords::Impl.name(),
136 keywords::Unsafe.name(),
137 keywords::Extern.name(),
138 keywords::Typeof.name(),
139 keywords::Dyn.name(),
140 ].contains(&ident.name)
143 #[derive(Clone, RustcEncodable, RustcDecodable, PartialEq, Debug)]
145 /* Expression-operator symbols. */
160 /* Structural symbols */
166 DotEq, // HACK(durka42) never produced by the parser, only used for libproc_macro
177 /// Used by proc macros for representing lifetimes, not generated by lexer right now.
179 /// An opening delimiter, eg. `{`
180 OpenDelim(DelimToken),
181 /// A closing delimiter, eg. `}`
182 CloseDelim(DelimToken),
185 Literal(Lit, Option<ast::Name>),
187 /* Name components */
188 Ident(ast::Ident, /* is_raw */ bool),
189 Lifetime(ast::Ident),
191 // The `LazyTokenStream` is a pure function of the `Nonterminal`,
192 // and so the `LazyTokenStream` can be ignored by Eq, Hash, etc.
193 Interpolated(Lrc<(Nonterminal, LazyTokenStream)>),
194 // Can be expanded into several tokens.
196 DocComment(ast::Name),
198 // Junk. These carry no data because we don't really care about the data
199 // they *would* carry, and don't really want to allocate a new ident for
200 // them. Instead, users could extract that from the associated span.
212 pub fn interpolated(nt: Nonterminal) -> Token {
213 Token::Interpolated(Lrc::new((nt, LazyTokenStream::new())))
216 /// Recovers a `Token` from an `ast::Ident`. This creates a raw identifier if necessary.
217 pub fn from_ast_ident(ident: ast::Ident) -> Token {
218 Ident(ident, ident.is_raw_guess())
221 crate fn is_like_plus(&self) -> bool {
223 BinOp(Plus) | BinOpEq(Plus) => true,
228 /// Returns `true` if the token can appear at the start of an expression.
229 crate fn can_begin_expr(&self) -> bool {
231 Ident(ident, is_raw) =>
232 ident_can_begin_expr(ident, is_raw), // value name or keyword
233 OpenDelim(..) | // tuple, array or block
234 Literal(..) | // literal
235 Not | // operator not
236 BinOp(Minus) | // unary minus
237 BinOp(Star) | // dereference
238 BinOp(Or) | OrOr | // closure
239 BinOp(And) | // reference
240 AndAnd | // double reference
241 // DotDotDot is no longer supported, but we need some way to display the error
242 DotDot | DotDotDot | DotDotEq | // range notation
243 Lt | BinOp(Shl) | // associated path
244 ModSep | // global path
245 Lifetime(..) | // labeled loop
246 Pound => true, // expression attributes
247 Interpolated(ref nt) => match nt.0 {
253 NtLifetime(..) => true,
260 /// Returns `true` if the token can appear at the start of a type.
261 crate fn can_begin_type(&self) -> bool {
263 Ident(ident, is_raw) =>
264 ident_can_begin_type(ident, is_raw), // type name or keyword
265 OpenDelim(Paren) | // tuple
266 OpenDelim(Bracket) | // array
268 BinOp(Star) | // raw pointer
269 BinOp(And) | // reference
270 AndAnd | // double reference
271 Question | // maybe bound in trait object
272 Lifetime(..) | // lifetime bound in trait object
273 Lt | BinOp(Shl) | // associated path
274 ModSep => true, // global path
275 Interpolated(ref nt) => match nt.0 {
276 NtIdent(..) | NtTy(..) | NtPath(..) | NtLifetime(..) => true,
283 /// Returns `true` if the token can appear at the start of a generic bound.
284 crate fn can_begin_bound(&self) -> bool {
285 self.is_path_start() || self.is_lifetime() || self.is_keyword(keywords::For) ||
286 self == &Question || self == &OpenDelim(Paren)
289 /// Returns `true` if the token is any literal
290 crate fn is_lit(&self) -> bool {
297 /// Returns `true` if the token is any literal, a minus (which can follow a literal,
298 /// for example a '-42', or one of the boolean idents).
299 crate fn can_begin_literal_or_bool(&self) -> bool {
302 BinOp(Minus) => true,
303 Ident(ident, false) if ident.name == keywords::True.name() => true,
304 Ident(ident, false) if ident.name == keywords::False.name() => true,
305 Interpolated(ref nt) => match nt.0 {
306 NtLiteral(..) => true,
313 /// Returns an identifier if this token is an identifier.
314 pub fn ident(&self) -> Option<(ast::Ident, /* is_raw */ bool)> {
316 Ident(ident, is_raw) => Some((ident, is_raw)),
317 Interpolated(ref nt) => match nt.0 {
318 NtIdent(ident, is_raw) => Some((ident, is_raw)),
324 /// Returns a lifetime identifier if this token is a lifetime.
325 pub fn lifetime(&self) -> Option<ast::Ident> {
327 Lifetime(ident) => Some(ident),
328 Interpolated(ref nt) => match nt.0 {
329 NtLifetime(ident) => Some(ident),
335 /// Returns `true` if the token is an identifier.
336 pub fn is_ident(&self) -> bool {
337 self.ident().is_some()
339 /// Returns `true` if the token is a lifetime.
340 crate fn is_lifetime(&self) -> bool {
341 self.lifetime().is_some()
344 /// Returns `true` if the token is a identifier whose name is the given
346 crate fn is_ident_named(&self, name: &str) -> bool {
348 Some((ident, _)) => ident.as_str() == name,
353 /// Returns `true` if the token is an interpolated path.
354 fn is_path(&self) -> bool {
355 if let Interpolated(ref nt) = *self {
356 if let NtPath(..) = nt.0 {
363 /// Returns `true` if the token is either the `mut` or `const` keyword.
364 crate fn is_mutability(&self) -> bool {
365 self.is_keyword(keywords::Mut) ||
366 self.is_keyword(keywords::Const)
369 crate fn is_qpath_start(&self) -> bool {
370 self == &Lt || self == &BinOp(Shl)
373 crate fn is_path_start(&self) -> bool {
374 self == &ModSep || self.is_qpath_start() || self.is_path() ||
375 self.is_path_segment_keyword() || self.is_ident() && !self.is_reserved_ident()
378 /// Returns `true` if the token is a given keyword, `kw`.
379 pub fn is_keyword(&self, kw: keywords::Keyword) -> bool {
380 self.ident().map(|(ident, is_raw)| ident.name == kw.name() && !is_raw).unwrap_or(false)
383 pub fn is_path_segment_keyword(&self) -> bool {
385 Some((id, false)) => id.is_path_segment_keyword(),
390 // Returns true for reserved identifiers used internally for elided lifetimes,
391 // unnamed method parameters, crate root module, error recovery etc.
392 pub fn is_special_ident(&self) -> bool {
394 Some((id, false)) => id.is_special(),
399 /// Returns `true` if the token is a keyword used in the language.
400 crate fn is_used_keyword(&self) -> bool {
402 Some((id, false)) => id.is_used_keyword(),
407 /// Returns `true` if the token is a keyword reserved for possible future use.
408 crate fn is_unused_keyword(&self) -> bool {
410 Some((id, false)) => id.is_unused_keyword(),
415 /// Returns `true` if the token is either a special identifier or a keyword.
416 pub fn is_reserved_ident(&self) -> bool {
418 Some((id, false)) => id.is_reserved(),
423 crate fn glue(self, joint: Token) -> Option<Token> {
434 BinOp(Minus) => LArrow,
447 BinOp(op) => match joint {
449 BinOp(And) if op == And => AndAnd,
450 BinOp(Or) if op == Or => OrOr,
451 Gt if op == Minus => RArrow,
460 DotDot => match joint {
465 Colon => match joint {
469 SingleQuote => match joint {
470 Ident(ident, false) => {
471 let name = Symbol::intern(&format!("'{}", ident));
472 Lifetime(symbol::Ident {
480 Le | EqEq | Ne | Ge | AndAnd | OrOr | Tilde | BinOpEq(..) | At | DotDotDot | DotEq |
481 DotDotEq | Comma | Semi | ModSep | RArrow | LArrow | FatArrow | Pound | Dollar |
482 Question | OpenDelim(..) | CloseDelim(..) => return None,
484 Literal(..) | Ident(..) | Lifetime(..) | Interpolated(..) | DocComment(..) |
485 Whitespace | Comment | Shebang(..) | Eof => return None,
489 /// Returns tokens that are likely to be typed accidentally instead of the current token.
490 /// Enables better error recovery when the wrong token is found.
491 crate fn similar_tokens(&self) -> Option<Vec<Token>> {
493 Comma => Some(vec![Dot, Lt]),
494 Semi => Some(vec![Colon]),
499 pub fn interpolated_to_tokenstream(&self, sess: &ParseSess, span: Span)
502 let nt = match *self {
503 Token::Interpolated(ref nt) => nt,
504 _ => panic!("only works on interpolated tokens"),
507 // An `Interpolated` token means that we have a `Nonterminal`
508 // which is often a parsed AST item. At this point we now need
509 // to convert the parsed AST to an actual token stream, e.g.
510 // un-parse it basically.
512 // Unfortunately there's not really a great way to do that in a
513 // guaranteed lossless fashion right now. The fallback here is
514 // to just stringify the AST node and reparse it, but this loses
515 // all span information.
517 // As a result, some AST nodes are annotated with the token
518 // stream they came from. Here we attempt to extract these
519 // lossless token streams before we fall back to the
521 let mut tokens = None;
524 Nonterminal::NtItem(ref item) => {
525 tokens = prepend_attrs(sess, &item.attrs, item.tokens.as_ref(), span);
527 Nonterminal::NtTraitItem(ref item) => {
528 tokens = prepend_attrs(sess, &item.attrs, item.tokens.as_ref(), span);
530 Nonterminal::NtImplItem(ref item) => {
531 tokens = prepend_attrs(sess, &item.attrs, item.tokens.as_ref(), span);
533 Nonterminal::NtIdent(ident, is_raw) => {
534 let token = Token::Ident(ident, is_raw);
535 tokens = Some(TokenTree::Token(ident.span, token).into());
537 Nonterminal::NtLifetime(ident) => {
538 let token = Token::Lifetime(ident);
539 tokens = Some(TokenTree::Token(ident.span, token).into());
541 Nonterminal::NtTT(ref tt) => {
542 tokens = Some(tt.clone().into());
547 let tokens_for_real = nt.1.force(|| {
548 // FIXME(#43081): Avoid this pretty-print + reparse hack
549 let source = pprust::token_to_string(self);
550 parse_stream_from_source_str(FileName::MacroExpansion, source, sess, Some(span))
553 // During early phases of the compiler the AST could get modified
554 // directly (e.g. attributes added or removed) and the internal cache
555 // of tokens my not be invalidated or updated. Consequently if the
556 // "lossless" token stream disagrees with our actual stringification
557 // (which has historically been much more battle-tested) then we go
558 // with the lossy stream anyway (losing span information).
560 // Note that the comparison isn't `==` here to avoid comparing spans,
561 // but it *also* is a "probable" equality which is a pretty weird
562 // definition. We mostly want to catch actual changes to the AST
563 // like a `#[cfg]` being processed or some weird `macro_rules!`
566 // What we *don't* want to catch is the fact that a user-defined
567 // literal like `0xf` is stringified as `15`, causing the cached token
568 // stream to not be literal `==` token-wise (ignoring spans) to the
569 // token stream we got from stringification.
571 // Instead the "probably equal" check here is "does each token
572 // recursively have the same discriminant?" We basically don't look at
573 // the token values here and assume that such fine grained token stream
574 // modifications, including adding/removing typically non-semantic
575 // tokens such as extra braces and commas, don't happen.
576 if let Some(tokens) = tokens {
577 if tokens.probably_equal_for_proc_macro(&tokens_for_real) {
580 info!("cached tokens found, but they're not \"probably equal\", \
581 going with stringified version");
583 return tokens_for_real
586 // See comments in `interpolated_to_tokenstream` for why we care about
587 // *probably* equal here rather than actual equality
588 crate fn probably_equal_for_proc_macro(&self, other: &Token) -> bool {
589 if mem::discriminant(self) != mem::discriminant(other) {
592 match (self, other) {
607 (&DotDotDot, &DotDotDot) |
608 (&DotDotEq, &DotDotEq) |
616 (&FatArrow, &FatArrow) |
619 (&Question, &Question) |
620 (&Whitespace, &Whitespace) |
621 (&Comment, &Comment) |
622 (&Eof, &Eof) => true,
624 (&BinOp(a), &BinOp(b)) |
625 (&BinOpEq(a), &BinOpEq(b)) => a == b,
627 (&OpenDelim(a), &OpenDelim(b)) |
628 (&CloseDelim(a), &CloseDelim(b)) => a == b,
630 (&DocComment(a), &DocComment(b)) |
631 (&Shebang(a), &Shebang(b)) => a == b,
633 (&Lifetime(a), &Lifetime(b)) => a.name == b.name,
634 (&Ident(a, b), &Ident(c, d)) => a.name == c.name && b == d,
636 (&Literal(ref a, b), &Literal(ref c, d)) => {
637 b == d && a.probably_equal_for_proc_macro(c)
640 (&Interpolated(_), &Interpolated(_)) => false,
642 _ => panic!("forgot to add a token?"),
647 #[derive(Clone, RustcEncodable, RustcDecodable)]
648 /// For interpolation during macro expansion.
649 pub enum Nonterminal {
650 NtItem(P<ast::Item>),
651 NtBlock(P<ast::Block>),
654 NtExpr(P<ast::Expr>),
656 NtIdent(ast::Ident, /* is_raw */ bool),
657 NtLifetime(ast::Ident),
658 NtLiteral(P<ast::Expr>),
659 /// Stuff inside brackets for attributes
660 NtMeta(ast::MetaItem),
662 NtVis(ast::Visibility),
664 // These are not exposed to macros, but are used by quasiquote.
666 NtImplItem(ast::ImplItem),
667 NtTraitItem(ast::TraitItem),
668 NtForeignItem(ast::ForeignItem),
669 NtGenerics(ast::Generics),
670 NtWhereClause(ast::WhereClause),
674 impl PartialEq for Nonterminal {
675 fn eq(&self, rhs: &Self) -> bool {
677 (NtIdent(ident_lhs, is_raw_lhs), NtIdent(ident_rhs, is_raw_rhs)) =>
678 ident_lhs == ident_rhs && is_raw_lhs == is_raw_rhs,
679 (NtLifetime(ident_lhs), NtLifetime(ident_rhs)) => ident_lhs == ident_rhs,
680 (NtTT(tt_lhs), NtTT(tt_rhs)) => tt_lhs == tt_rhs,
681 // FIXME: Assume that all "complex" nonterminal are not equal, we can't compare them
682 // correctly based on data from AST. This will prevent them from matching each other
683 // in macros. The comparison will become possible only when each nonterminal has an
684 // attached token stream from which it was parsed.
690 impl fmt::Debug for Nonterminal {
691 fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
693 NtItem(..) => f.pad("NtItem(..)"),
694 NtBlock(..) => f.pad("NtBlock(..)"),
695 NtStmt(..) => f.pad("NtStmt(..)"),
696 NtPat(..) => f.pad("NtPat(..)"),
697 NtExpr(..) => f.pad("NtExpr(..)"),
698 NtTy(..) => f.pad("NtTy(..)"),
699 NtIdent(..) => f.pad("NtIdent(..)"),
700 NtLiteral(..) => f.pad("NtLiteral(..)"),
701 NtMeta(..) => f.pad("NtMeta(..)"),
702 NtPath(..) => f.pad("NtPath(..)"),
703 NtTT(..) => f.pad("NtTT(..)"),
704 NtArm(..) => f.pad("NtArm(..)"),
705 NtImplItem(..) => f.pad("NtImplItem(..)"),
706 NtTraitItem(..) => f.pad("NtTraitItem(..)"),
707 NtForeignItem(..) => f.pad("NtForeignItem(..)"),
708 NtGenerics(..) => f.pad("NtGenerics(..)"),
709 NtWhereClause(..) => f.pad("NtWhereClause(..)"),
710 NtArg(..) => f.pad("NtArg(..)"),
711 NtVis(..) => f.pad("NtVis(..)"),
712 NtLifetime(..) => f.pad("NtLifetime(..)"),
717 crate fn is_op(tok: &Token) -> bool {
719 OpenDelim(..) | CloseDelim(..) | Literal(..) | DocComment(..) |
720 Ident(..) | Lifetime(..) | Interpolated(..) |
721 Whitespace | Comment | Shebang(..) | Eof => false,
727 pub struct LazyTokenStream(Lock<Option<TokenStream>>);
729 impl cmp::Eq for LazyTokenStream {}
730 impl PartialEq for LazyTokenStream {
731 fn eq(&self, _other: &LazyTokenStream) -> bool {
736 impl fmt::Debug for LazyTokenStream {
737 fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
738 fmt::Debug::fmt(&self.clone().0.into_inner(), f)
742 impl LazyTokenStream {
744 LazyTokenStream(Lock::new(None))
747 fn force<F: FnOnce() -> TokenStream>(&self, f: F) -> TokenStream {
748 let mut opt_stream = self.0.lock();
749 if opt_stream.is_none() {
750 *opt_stream = Some(f());
752 opt_stream.clone().unwrap()
756 impl Encodable for LazyTokenStream {
757 fn encode<S: Encoder>(&self, _: &mut S) -> Result<(), S::Error> {
762 impl Decodable for LazyTokenStream {
763 fn decode<D: Decoder>(_: &mut D) -> Result<LazyTokenStream, D::Error> {
764 Ok(LazyTokenStream::new())
768 impl ::std::hash::Hash for LazyTokenStream {
769 fn hash<H: ::std::hash::Hasher>(&self, _hasher: &mut H) {}
772 fn prepend_attrs(sess: &ParseSess,
773 attrs: &[ast::Attribute],
774 tokens: Option<&tokenstream::TokenStream>,
775 span: syntax_pos::Span)
776 -> Option<tokenstream::TokenStream>
778 let tokens = tokens?;
779 if attrs.len() == 0 {
780 return Some(tokens.clone())
782 let mut builder = tokenstream::TokenStreamBuilder::new();
784 assert_eq!(attr.style, ast::AttrStyle::Outer,
785 "inner attributes should prevent cached tokens from existing");
787 if attr.is_sugared_doc {
788 let stream = parse_stream_from_source_str(
789 FileName::MacroExpansion,
790 pprust::attr_to_string(attr),
794 builder.push(stream);
798 // synthesize # [ $path $tokens ] manually here
799 let mut brackets = tokenstream::TokenStreamBuilder::new();
801 // For simple paths, push the identifier directly
802 if attr.path.segments.len() == 1 && attr.path.segments[0].args.is_none() {
803 let ident = attr.path.segments[0].ident;
804 let token = Ident(ident, ident.as_str().starts_with("r#"));
805 brackets.push(tokenstream::TokenTree::Token(ident.span, token));
807 // ... and for more complicated paths, fall back to a reparse hack that
808 // should eventually be removed.
810 let stream = parse_stream_from_source_str(
811 FileName::MacroExpansion,
812 pprust::path_to_string(&attr.path),
816 brackets.push(stream);
819 brackets.push(attr.tokens.clone());
821 let tokens = tokenstream::Delimited {
822 delim: DelimToken::Bracket,
823 tts: brackets.build().into(),
825 // The span we list here for `#` and for `[ ... ]` are both wrong in
826 // that it encompasses more than each token, but it hopefully is "good
827 // enough" for now at least.
828 builder.push(tokenstream::TokenTree::Token(attr.span, Pound));
829 let delim_span = DelimSpan::from_single(attr.span);
830 builder.push(tokenstream::TokenTree::Delimited(delim_span, tokens));
832 builder.push(tokens.clone());
833 Some(builder.build())