1 // Copyright 2012-2017 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 //! This is an NFA-based parser, which calls out to the main rust parser for named nonterminals
12 //! (which it commits to fully when it hits one in a grammar). There's a set of current NFA threads
13 //! and a set of next ones. Instead of NTs, we have a special case for Kleene star. The big-O, in
14 //! pathological cases, is worse than traditional use of NFA or Earley parsing, but it's an easier
15 //! fit for Macro-by-Example-style rules.
17 //! (In order to prevent the pathological case, we'd need to lazily construct the resulting
18 //! `NamedMatch`es at the very end. It'd be a pain, and require more memory to keep around old
19 //! items, but it would also save overhead)
21 //! We don't say this parser uses the Earley algorithm, because it's unnecessarily inaccurate.
22 //! The macro parser restricts itself to the features of finite state automata. Earley parsers
23 //! can be described as an extension of NFAs with completion rules, prediction rules, and recursion.
25 //! Quick intro to how the parser works:
27 //! A 'position' is a dot in the middle of a matcher, usually represented as a
28 //! dot. For example `· a $( a )* a b` is a position, as is `a $( · a )* a b`.
30 //! The parser walks through the input a character at a time, maintaining a list
31 //! of threads consistent with the current position in the input string: `cur_items`.
33 //! As it processes them, it fills up `eof_items` with threads that would be valid if
34 //! the macro invocation is now over, `bb_items` with threads that are waiting on
35 //! a Rust nonterminal like `$e:expr`, and `next_items` with threads that are waiting
36 //! on a particular token. Most of the logic concerns moving the · through the
37 //! repetitions indicated by Kleene stars. The rules for moving the · without
38 //! consuming any input are called epsilon transitions. It only advances or calls
39 //! out to the real Rust parser when no `cur_items` threads remain.
44 //! Start parsing a a a a b against [· a $( a )* a b].
46 //! Remaining input: a a a a b
47 //! next: [· a $( a )* a b]
49 //! - - - Advance over an a. - - -
51 //! Remaining input: a a a b
52 //! cur: [a · $( a )* a b]
53 //! Descend/Skip (first item).
54 //! next: [a $( · a )* a b] [a $( a )* · a b].
56 //! - - - Advance over an a. - - -
58 //! Remaining input: a a b
59 //! cur: [a $( a · )* a b] [a $( a )* a · b]
60 //! Follow epsilon transition: Finish/Repeat (first item)
61 //! next: [a $( a )* · a b] [a $( · a )* a b] [a $( a )* a · b]
63 //! - - - Advance over an a. - - - (this looks exactly like the last step)
65 //! Remaining input: a b
66 //! cur: [a $( a · )* a b] [a $( a )* a · b]
67 //! Follow epsilon transition: Finish/Repeat (first item)
68 //! next: [a $( a )* · a b] [a $( · a )* a b] [a $( a )* a · b]
70 //! - - - Advance over an a. - - - (this looks exactly like the last step)
72 //! Remaining input: b
73 //! cur: [a $( a · )* a b] [a $( a )* a · b]
74 //! Follow epsilon transition: Finish/Repeat (first item)
75 //! next: [a $( a )* · a b] [a $( · a )* a b] [a $( a )* a · b]
77 //! - - - Advance over a b. - - -
79 //! Remaining input: ''
80 //! eof: [a $( a )* a b ·]
83 pub use self::NamedMatch::*;
84 pub use self::ParseResult::*;
85 use self::TokenTreeOrTokenTreeSlice::*;
88 use syntax_pos::{self, Span};
89 use errors::FatalError;
90 use ext::tt::quoted::{self, TokenTree};
91 use parse::{Directory, ParseSess};
92 use parse::parser::{Parser, PathStyle};
93 use parse::token::{self, DocComment, Nonterminal, Token};
95 use smallvec::SmallVec;
97 use tokenstream::{DelimSpan, TokenStream};
99 use rustc_data_structures::fx::FxHashMap;
100 use std::collections::hash_map::Entry::{Occupied, Vacant};
102 use std::ops::{Deref, DerefMut};
105 // To avoid costly uniqueness checks, we require that `MatchSeq` always has a nonempty body.
107 /// Either a sequence of token trees or a single one. This is used as the representation of the
108 /// sequence of tokens that make up a matcher.
110 enum TokenTreeOrTokenTreeSlice<'tt> {
112 TtSeq(&'tt [TokenTree]),
115 impl<'tt> TokenTreeOrTokenTreeSlice<'tt> {
116 /// Returns the number of constituent top-level token trees of `self` (top-level in that it
117 /// will not recursively descend into subtrees).
118 fn len(&self) -> usize {
120 TtSeq(ref v) => v.len(),
121 Tt(ref tt) => tt.len(),
125 /// The `index`-th token tree of `self`.
126 fn get_tt(&self, index: usize) -> TokenTree {
128 TtSeq(ref v) => v[index].clone(),
129 Tt(ref tt) => tt.get_tt(index),
134 /// An unzipping of `TokenTree`s... see the `stack` field of `MatcherPos`.
136 /// This is used by `inner_parse_loop` to keep track of delimited submatchers that we have
139 struct MatcherTtFrame<'tt> {
140 /// The "parent" matcher that we are descending into.
141 elts: TokenTreeOrTokenTreeSlice<'tt>,
142 /// The position of the "dot" in `elts` at the time we descended.
146 type NamedMatchVec = SmallVec<[NamedMatch; 4]>;
148 /// Represents a single "position" (aka "matcher position", aka "item"), as
149 /// described in the module documentation.
153 /// - `'root` represents the lifetime of the stack slot that holds the root
154 /// `MatcherPos`. As described in `MatcherPosHandle`, the root `MatcherPos`
155 /// structure is stored on the stack, but subsequent instances are put into
157 /// - `'tt` represents the lifetime of the token trees that this matcher
158 /// position refers to.
160 /// It is important to distinguish these two lifetimes because we have a
161 /// `SmallVec<TokenTreeOrTokenTreeSlice<'tt>>` below, and the destructor of
162 /// that is considered to possibly access the data from its elements (it lacks
163 /// a `#[may_dangle]` attribute). As a result, the compiler needs to know that
164 /// all the elements in that `SmallVec` strictly outlive the root stack slot
165 /// lifetime. By separating `'tt` from `'root`, we can show that.
167 struct MatcherPos<'root, 'tt: 'root> {
168 /// The token or sequence of tokens that make up the matcher
169 top_elts: TokenTreeOrTokenTreeSlice<'tt>,
171 /// The position of the "dot" in this matcher
174 /// The first span of source source that the beginning of this matcher corresponds to. In other
175 /// words, the token in the source whose span is `sp_open` is matched against the first token of
179 /// For each named metavar in the matcher, we keep track of token trees matched against the
180 /// metavar by the black box parser. In particular, there may be more than one match per
181 /// metavar if we are in a repetition (each repetition matches each of the variables).
182 /// Moreover, matchers and repetitions can be nested; the `matches` field is shared (hence the
183 /// `Rc`) among all "nested" matchers. `match_lo`, `match_cur`, and `match_hi` keep track of
184 /// the current position of the `self` matcher position in the shared `matches` list.
186 /// Also, note that while we are descending into a sequence, matchers are given their own
187 /// `matches` vector. Only once we reach the end of a full repetition of the sequence do we add
188 /// all bound matches from the submatcher into the shared top-level `matches` vector. If `sep`
189 /// and `up` are `Some`, then `matches` is _not_ the shared top-level list. Instead, if one
190 /// wants the shared `matches`, one should use `up.matches`.
191 matches: Box<[Rc<NamedMatchVec>]>,
192 /// The position in `matches` corresponding to the first metavar in this matcher's sequence of
193 /// token trees. In other words, the first metavar in the first token of `top_elts` corresponds
194 /// to `matches[match_lo]`.
196 /// The position in `matches` corresponding to the metavar we are currently trying to match
197 /// against the source token stream. `match_lo <= match_cur <= match_hi`.
199 /// Similar to `match_lo` except `match_hi` is the position in `matches` of the _last_ metavar
203 // The following fields are used if we are matching a repetition. If we aren't, they should be
206 /// The KleeneOp of this sequence if we are in a repetition.
207 seq_op: Option<quoted::KleeneOp>,
209 /// The separator if we are in a repetition.
212 /// The "parent" matcher position if we are in a repetition. That is, the matcher position just
213 /// before we enter the sequence.
214 up: Option<MatcherPosHandle<'root, 'tt>>,
216 /// Specifically used to "unzip" token trees. By "unzip", we mean to unwrap the delimiters from
217 /// a delimited token tree (e.g., something wrapped in `(` `)`) or to get the contents of a doc
220 /// When matching against matchers with nested delimited submatchers (e.g., `pat ( pat ( .. )
221 /// pat ) pat`), we need to keep track of the matchers we are descending into. This stack does
222 /// that where the bottom of the stack is the outermost matcher.
223 /// Also, throughout the comments, this "descent" is often referred to as "unzipping"...
224 stack: SmallVec<[MatcherTtFrame<'tt>; 1]>,
227 impl<'root, 'tt> MatcherPos<'root, 'tt> {
228 /// Add `m` as a named match for the `idx`-th metavar.
229 fn push_match(&mut self, idx: usize, m: NamedMatch) {
230 let matches = Rc::make_mut(&mut self.matches[idx]);
235 // Lots of MatcherPos instances are created at runtime. Allocating them on the
236 // heap is slow. Furthermore, using SmallVec<MatcherPos> to allocate them all
237 // on the stack is also slow, because MatcherPos is quite a large type and
238 // instances get moved around a lot between vectors, which requires lots of
239 // slow memcpy calls.
241 // Therefore, the initial MatcherPos is always allocated on the stack,
242 // subsequent ones (of which there aren't that many) are allocated on the heap,
243 // and this type is used to encapsulate both cases.
244 enum MatcherPosHandle<'root, 'tt: 'root> {
245 Ref(&'root mut MatcherPos<'root, 'tt>),
246 Box(Box<MatcherPos<'root, 'tt>>),
249 impl<'root, 'tt> Clone for MatcherPosHandle<'root, 'tt> {
250 // This always produces a new Box.
251 fn clone(&self) -> Self {
252 MatcherPosHandle::Box(match *self {
253 MatcherPosHandle::Ref(ref r) => Box::new((**r).clone()),
254 MatcherPosHandle::Box(ref b) => b.clone(),
259 impl<'root, 'tt> Deref for MatcherPosHandle<'root, 'tt> {
260 type Target = MatcherPos<'root, 'tt>;
261 fn deref(&self) -> &Self::Target {
263 MatcherPosHandle::Ref(ref r) => r,
264 MatcherPosHandle::Box(ref b) => b,
269 impl<'root, 'tt> DerefMut for MatcherPosHandle<'root, 'tt> {
270 fn deref_mut(&mut self) -> &mut MatcherPos<'root, 'tt> {
272 MatcherPosHandle::Ref(ref mut r) => r,
273 MatcherPosHandle::Box(ref mut b) => b,
278 /// Represents the possible results of an attempted parse.
279 pub enum ParseResult<T> {
280 /// Parsed successfully.
282 /// Arm failed to match. If the second parameter is `token::Eof`, it indicates an unexpected
283 /// end of macro invocation. Otherwise, it indicates that no rules expected the given token.
284 Failure(syntax_pos::Span, Token, String),
285 /// Fatal error (malformed macro?). Abort compilation.
286 Error(syntax_pos::Span, String),
289 /// A `ParseResult` where the `Success` variant contains a mapping of `Ident`s to `NamedMatch`es.
290 /// This represents the mapping of metavars to the token trees they bind to.
291 pub type NamedParseResult = ParseResult<FxHashMap<Ident, Rc<NamedMatch>>>;
293 /// Count how many metavars are named in the given matcher `ms`.
294 pub fn count_names(ms: &[TokenTree]) -> usize {
295 ms.iter().fold(0, |count, elt| {
297 TokenTree::Sequence(_, ref seq) => seq.num_captures,
298 TokenTree::Delimited(_, ref delim) => count_names(&delim.tts),
299 TokenTree::MetaVar(..) => 0,
300 TokenTree::MetaVarDecl(..) => 1,
301 TokenTree::Token(..) => 0,
306 /// `len` `Vec`s (initially shared and empty) that will store matches of metavars.
307 fn create_matches(len: usize) -> Box<[Rc<NamedMatchVec>]> {
311 let empty_matches = Rc::new(SmallVec::new());
312 vec![empty_matches; len]
316 /// Generate the top-level matcher position in which the "dot" is before the first token of the
317 /// matcher `ms` and we are going to start matching at the span `open` in the source.
318 fn initial_matcher_pos<'root, 'tt>(ms: &'tt [TokenTree], open: Span) -> MatcherPos<'root, 'tt> {
319 let match_idx_hi = count_names(ms);
320 let matches = create_matches(match_idx_hi);
322 // Start with the top level matcher given to us
323 top_elts: TtSeq(ms), // "elts" is an abbr. for "elements"
324 // The "dot" is before the first token of the matcher
326 // We start matching at the span `open` in the source code
329 // Initialize `matches` to a bunch of empty `Vec`s -- one for each metavar in `top_elts`.
330 // `match_lo` for `top_elts` is 0 and `match_hi` is `matches.len()`. `match_cur` is 0 since
331 // we haven't actually matched anything yet.
335 match_hi: match_idx_hi,
337 // Haven't descended into any delimiters, so empty stack
340 // Haven't descended into any sequences, so both of these are `None`.
347 /// `NamedMatch` is a pattern-match result for a single `token::MATCH_NONTERMINAL`:
348 /// so it is associated with a single ident in a parse, and all
349 /// `MatchedNonterminal`s in the `NamedMatch` have the same nonterminal type
350 /// (expr, item, etc). Each leaf in a single `NamedMatch` corresponds to a
351 /// single `token::MATCH_NONTERMINAL` in the `TokenTree` that produced it.
353 /// The in-memory structure of a particular `NamedMatch` represents the match
354 /// that occurred when a particular subset of a matcher was applied to a
355 /// particular token tree.
357 /// The width of each `MatchedSeq` in the `NamedMatch`, and the identity of
358 /// the `MatchedNonterminal`s, will depend on the token tree it was applied
359 /// to: each `MatchedSeq` corresponds to a single `TTSeq` in the originating
360 /// token tree. The depth of the `NamedMatch` structure will therefore depend
361 /// only on the nesting depth of `ast::TTSeq`s in the originating
362 /// token tree it was derived from.
363 #[derive(Debug, Clone)]
364 pub enum NamedMatch {
365 MatchedSeq(Rc<NamedMatchVec>, DelimSpan),
366 MatchedNonterminal(Rc<Nonterminal>),
369 /// Takes a sequence of token trees `ms` representing a matcher which successfully matched input
370 /// and an iterator of items that matched input and produces a `NamedParseResult`.
371 fn nameize<I: Iterator<Item = NamedMatch>>(
375 ) -> NamedParseResult {
376 // Recursively descend into each type of matcher (e.g., sequences, delimited, metavars) and make
377 // sure that each metavar has _exactly one_ binding. If a metavar does not have exactly one
378 // binding, then there is an error. If it does, then we insert the binding into the
379 // `NamedParseResult`.
380 fn n_rec<I: Iterator<Item = NamedMatch>>(
384 ret_val: &mut FxHashMap<Ident, Rc<NamedMatch>>,
385 ) -> Result<(), (syntax_pos::Span, String)> {
387 TokenTree::Sequence(_, ref seq) => for next_m in &seq.tts {
388 n_rec(sess, next_m, res.by_ref(), ret_val)?
390 TokenTree::Delimited(_, ref delim) => for next_m in &delim.tts {
391 n_rec(sess, next_m, res.by_ref(), ret_val)?;
393 TokenTree::MetaVarDecl(span, _, id) if id.name == keywords::Invalid.name() => {
394 if sess.missing_fragment_specifiers.borrow_mut().remove(&span) {
395 return Err((span, "missing fragment specifier".to_string()));
398 TokenTree::MetaVarDecl(sp, bind_name, _) => {
399 match ret_val.entry(bind_name) {
401 // FIXME(simulacrum): Don't construct Rc here
402 spot.insert(Rc::new(res.next().unwrap()));
405 return Err((sp, format!("duplicated bind name: {}", bind_name)))
409 TokenTree::MetaVar(..) | TokenTree::Token(..) => (),
415 let mut ret_val = FxHashMap::default();
417 match n_rec(sess, m, res.by_ref(), &mut ret_val) {
419 Err((sp, msg)) => return Error(sp, msg),
426 /// Generate an appropriate parsing failure message. For EOF, this is "unexpected end...". For
427 /// other tokens, this is "unexpected token...".
428 pub fn parse_failure_msg(tok: Token) -> String {
430 token::Eof => "unexpected end of macro invocation".to_string(),
432 "no rules expected the token `{}`",
433 pprust::token_to_string(&tok)
438 /// Perform a token equality check, ignoring syntax context (that is, an unhygienic comparison)
439 fn token_name_eq(t1: &Token, t2: &Token) -> bool {
440 if let (Some((id1, is_raw1)), Some((id2, is_raw2))) = (t1.ident(), t2.ident()) {
441 id1.name == id2.name && is_raw1 == is_raw2
442 } else if let (Some(id1), Some(id2)) = (t1.lifetime(), t2.lifetime()) {
449 /// Process the matcher positions of `cur_items` until it is empty. In the process, this will
450 /// produce more items in `next_items`, `eof_items`, and `bb_items`.
452 /// For more info about the how this happens, see the module-level doc comments and the inline
453 /// comments of this function.
457 /// - `sess`: the parsing session into which errors are emitted.
458 /// - `cur_items`: the set of current items to be processed. This should be empty by the end of a
459 /// successful execution of this function.
460 /// - `next_items`: the set of newly generated items. These are used to replenish `cur_items` in
461 /// the function `parse`.
462 /// - `eof_items`: the set of items that would be valid if this was the EOF.
463 /// - `bb_items`: the set of items that are waiting for the black-box parser.
464 /// - `token`: the current token of the parser.
465 /// - `span`: the `Span` in the source code corresponding to the token trees we are trying to match
466 /// against the matcher positions in `cur_items`.
470 /// A `ParseResult`. Note that matches are kept track of through the items generated.
471 fn inner_parse_loop<'root, 'tt>(
473 cur_items: &mut SmallVec<[MatcherPosHandle<'root, 'tt>; 1]>,
474 next_items: &mut Vec<MatcherPosHandle<'root, 'tt>>,
475 eof_items: &mut SmallVec<[MatcherPosHandle<'root, 'tt>; 1]>,
476 bb_items: &mut SmallVec<[MatcherPosHandle<'root, 'tt>; 1]>,
478 span: syntax_pos::Span,
479 ) -> ParseResult<()> {
480 // Pop items from `cur_items` until it is empty.
481 while let Some(mut item) = cur_items.pop() {
482 // When unzipped trees end, remove them. This corresponds to backtracking out of a
483 // delimited submatcher into which we already descended. In backtracking out again, we need
484 // to advance the "dot" past the delimiters in the outer matcher.
485 while item.idx >= item.top_elts.len() {
486 match item.stack.pop() {
487 Some(MatcherTtFrame { elts, idx }) => {
488 item.top_elts = elts;
495 // Get the current position of the "dot" (`idx`) in `item` and the number of token trees in
496 // the matcher (`len`).
498 let len = item.top_elts.len();
500 // If `idx >= len`, then we are at or past the end of the matcher of `item`.
502 // We are repeating iff there is a parent. If the matcher is inside of a repetition,
503 // then we could be at the end of a sequence or at the beginning of the next
505 if item.up.is_some() {
506 // At this point, regardless of whether there is a separator, we should add all
507 // matches from the complete repetition of the sequence to the shared, top-level
508 // `matches` list (actually, `up.matches`, which could itself not be the top-level,
509 // but anyway...). Moreover, we add another item to `cur_items` in which the "dot"
510 // is at the end of the `up` matcher. This ensures that the "dot" in the `up`
511 // matcher is also advanced sufficiently.
513 // NOTE: removing the condition `idx == len` allows trailing separators.
515 // Get the `up` matcher
516 let mut new_pos = item.up.clone().unwrap();
518 // Add matches from this repetition to the `matches` of `up`
519 for idx in item.match_lo..item.match_hi {
520 let sub = item.matches[idx].clone();
521 let span = DelimSpan::from_pair(item.sp_open, span);
522 new_pos.push_match(idx, MatchedSeq(sub, span));
525 // Move the "dot" past the repetition in `up`
526 new_pos.match_cur = item.match_hi;
528 cur_items.push(new_pos);
531 // Check if we need a separator.
532 if idx == len && item.sep.is_some() {
533 // We have a separator, and it is the current token. We can advance past the
537 .map(|sep| token_name_eq(token, sep))
541 next_items.push(item);
544 // We don't need a separator. Move the "dot" back to the beginning of the matcher
545 // and try to match again UNLESS we are only allowed to have _one_ repetition.
546 else if item.seq_op != Some(quoted::KleeneOp::ZeroOrOne) {
547 item.match_cur = item.match_lo;
549 cur_items.push(item);
552 // If we are not in a repetition, then being at the end of a matcher means that we have
553 // reached the potential end of the input.
555 eof_items.push(item);
558 // We are in the middle of a matcher.
560 // Look at what token in the matcher we are trying to match the current token (`token`)
561 // against. Depending on that, we may generate new items.
562 match item.top_elts.get_tt(idx) {
563 // Need to descend into a sequence
564 TokenTree::Sequence(sp, seq) => {
565 // Examine the case where there are 0 matches of this sequence
566 if seq.op == quoted::KleeneOp::ZeroOrMore
567 || seq.op == quoted::KleeneOp::ZeroOrOne
569 let mut new_item = item.clone();
570 new_item.match_cur += seq.num_captures;
572 for idx in item.match_cur..item.match_cur + seq.num_captures {
573 new_item.push_match(idx, MatchedSeq(Rc::new(smallvec![]), sp));
575 cur_items.push(new_item);
578 let matches = create_matches(item.matches.len());
579 cur_items.push(MatcherPosHandle::Box(Box::new(MatcherPos {
581 sep: seq.separator.clone(),
582 seq_op: Some(seq.op),
585 match_lo: item.match_cur,
586 match_cur: item.match_cur,
587 match_hi: item.match_cur + seq.num_captures,
590 top_elts: Tt(TokenTree::Sequence(sp, seq)),
594 // We need to match a metavar (but the identifier is invalid)... this is an error
595 TokenTree::MetaVarDecl(span, _, id) if id.name == keywords::Invalid.name() => {
596 if sess.missing_fragment_specifiers.borrow_mut().remove(&span) {
597 return Error(span, "missing fragment specifier".to_string());
601 // We need to match a metavar with a valid ident... call out to the black-box
602 // parser by adding an item to `bb_items`.
603 TokenTree::MetaVarDecl(_, _, id) => {
604 // Built-in nonterminals never start with these tokens,
605 // so we can eliminate them from consideration.
606 if may_begin_with(&*id.as_str(), token) {
611 // We need to descend into a delimited submatcher or a doc comment. To do this, we
612 // push the current matcher onto a stack and push a new item containing the
613 // submatcher onto `cur_items`.
615 // At the beginning of the loop, if we reach the end of the delimited submatcher,
616 // we pop the stack to backtrack out of the descent.
617 seq @ TokenTree::Delimited(..) | seq @ TokenTree::Token(_, DocComment(..)) => {
618 let lower_elts = mem::replace(&mut item.top_elts, Tt(seq));
620 item.stack.push(MatcherTtFrame {
625 cur_items.push(item);
628 // We just matched a normal token. We can just advance the parser.
629 TokenTree::Token(_, ref t) if token_name_eq(t, token) => {
631 next_items.push(item);
634 // There was another token that was not `token`... This means we can't add any
635 // rules. NOTE that this is not necessarily an error unless _all_ items in
636 // `cur_items` end up doing this. There may still be some other matchers that do
637 // end up working out.
638 TokenTree::Token(..) | TokenTree::MetaVar(..) => {}
643 // Yay a successful parse (so far)!
647 /// Use the given sequence of token trees (`ms`) as a matcher. Match the given token stream `tts`
648 /// against it and return the match.
652 /// - `sess`: The session into which errors are emitted
653 /// - `tts`: The tokenstream we are matching against the pattern `ms`
654 /// - `ms`: A sequence of token trees representing a pattern against which we are matching
655 /// - `directory`: Information about the file locations (needed for the black-box parser)
656 /// - `recurse_into_modules`: Whether or not to recurse into modules (needed for the black-box
662 directory: Option<Directory>,
663 recurse_into_modules: bool,
664 ) -> NamedParseResult {
665 // Create a parser that can be used for the "black box" parts.
666 let mut parser = Parser::new(sess, tts, directory, recurse_into_modules, true);
668 // A queue of possible matcher positions. We initialize it with the matcher position in which
669 // the "dot" is before the first token of the first token tree in `ms`. `inner_parse_loop` then
670 // processes all of these possible matcher positions and produces possible next positions into
671 // `next_items`. After some post-processing, the contents of `next_items` replenish `cur_items`
672 // and we start over again.
674 // This MatcherPos instance is allocated on the stack. All others -- and
675 // there are frequently *no* others! -- are allocated on the heap.
676 let mut initial = initial_matcher_pos(ms, parser.span);
677 let mut cur_items = smallvec![MatcherPosHandle::Ref(&mut initial)];
678 let mut next_items = Vec::new();
681 // Matcher positions black-box parsed by parser.rs (`parser`)
682 let mut bb_items = SmallVec::new();
684 // Matcher positions that would be valid if the macro invocation was over now
685 let mut eof_items = SmallVec::new();
686 assert!(next_items.is_empty());
688 // Process `cur_items` until either we have finished the input or we need to get some
689 // parsing from the black-box parser done. The result is that `next_items` will contain a
690 // bunch of possible next matcher positions in `next_items`.
691 match inner_parse_loop(
701 Failure(sp, tok, t) => return Failure(sp, tok, t),
702 Error(sp, msg) => return Error(sp, msg),
705 // inner parse loop handled all cur_items, so it's empty
706 assert!(cur_items.is_empty());
708 // We need to do some post processing after the `inner_parser_loop`.
710 // Error messages here could be improved with links to original rules.
712 // If we reached the EOF, check that there is EXACTLY ONE possible matcher. Otherwise,
713 // either the parse is ambiguous (which should never happen) or there is a syntax error.
714 if token_name_eq(&parser.token, &token::Eof) {
715 if eof_items.len() == 1 {
716 let matches = eof_items[0]
719 .map(|dv| Rc::make_mut(dv).pop().unwrap());
720 return nameize(sess, ms, matches);
721 } else if eof_items.len() > 1 {
724 "ambiguity: multiple successful parses".to_string(),
728 if parser.span.is_dummy() {
731 sess.source_map().next_point(parser.span)
734 "missing tokens in macro arguments".to_string(),
738 // Performance hack: eof_items may share matchers via Rc with other things that we want
739 // to modify. Dropping eof_items now may drop these refcounts to 1, preventing an
740 // unnecessary implicit clone later in Rc::make_mut.
743 // Another possibility is that we need to call out to parse some rust nonterminal
744 // (black-box) parser. However, if there is not EXACTLY ONE of these, something is wrong.
745 if (!bb_items.is_empty() && !next_items.is_empty()) || bb_items.len() > 1 {
748 .map(|item| match item.top_elts.get_tt(item.idx) {
749 TokenTree::MetaVarDecl(_, bind, name) => format!("{} ('{}')", name, bind),
752 .collect::<Vec<String>>()
758 "local ambiguity: multiple parsing options: {}",
759 match next_items.len() {
760 0 => format!("built-in NTs {}.", nts),
761 1 => format!("built-in NTs {} or 1 other option.", nts),
762 n => format!("built-in NTs {} or {} other options.", nts, n),
767 // If there are no possible next positions AND we aren't waiting for the black-box parser,
768 // then there is a syntax error.
769 else if bb_items.is_empty() && next_items.is_empty() {
773 "no rules expected this token in macro call".to_string(),
776 // Dump all possible `next_items` into `cur_items` for the next iteration.
777 else if !next_items.is_empty() {
778 // Now process the next token
779 cur_items.extend(next_items.drain(..));
782 // Finally, we have the case where we need to call the black-box parser to get some
785 assert_eq!(bb_items.len(), 1);
787 let mut item = bb_items.pop().unwrap();
788 if let TokenTree::MetaVarDecl(span, _, ident) = item.top_elts.get_tt(item.idx) {
789 let match_cur = item.match_cur;
792 MatchedNonterminal(Rc::new(parse_nt(&mut parser, span, &ident.as_str()))),
799 cur_items.push(item);
802 assert!(!cur_items.is_empty());
806 /// The token is an identifier, but not `_`.
807 /// We prohibit passing `_` to macros expecting `ident` for now.
808 fn get_macro_ident(token: &Token) -> Option<(Ident, bool)> {
810 token::Ident(ident, is_raw) if ident.name != keywords::Underscore.name() =>
811 Some((ident, is_raw)),
816 /// Checks whether a non-terminal may begin with a particular token.
818 /// Returning `false` is a *stability guarantee* that such a matcher will *never* begin with that
819 /// token. Be conservative (return true) if not sure.
820 fn may_begin_with(name: &str, token: &Token) -> bool {
821 /// Checks whether the non-terminal may contain a single (non-keyword) identifier.
822 fn may_be_ident(nt: &token::Nonterminal) -> bool {
824 token::NtItem(_) | token::NtBlock(_) | token::NtVis(_) => false,
830 "expr" => token.can_begin_expr(),
831 "ty" => token.can_begin_type(),
832 "ident" => get_macro_ident(token).is_some(),
833 "literal" => token.can_begin_literal_or_bool(),
834 "vis" => match *token {
835 // The follow-set of :vis + "priv" keyword + interpolated
836 Token::Comma | Token::Ident(..) | Token::Interpolated(_) => true,
837 _ => token.can_begin_type(),
839 "block" => match *token {
840 Token::OpenDelim(token::Brace) => true,
841 Token::Interpolated(ref nt) => match nt.0 {
848 | token::NtVis(_) => false, // none of these may start with '{'.
853 "path" | "meta" => match *token {
854 Token::ModSep | Token::Ident(..) => true,
855 Token::Interpolated(ref nt) => match nt.0 {
856 token::NtPath(_) | token::NtMeta(_) => true,
857 _ => may_be_ident(&nt.0),
861 "pat" => match *token {
862 Token::Ident(..) | // box, ref, mut, and other identifiers (can stricten)
863 Token::OpenDelim(token::Paren) | // tuple pattern
864 Token::OpenDelim(token::Bracket) | // slice pattern
865 Token::BinOp(token::And) | // reference
866 Token::BinOp(token::Minus) | // negative literal
867 Token::AndAnd | // double reference
868 Token::Literal(..) | // literal
869 Token::DotDot | // range pattern (future compat)
870 Token::DotDotDot | // range pattern (future compat)
871 Token::ModSep | // path
872 Token::Lt | // path (UFCS constant)
873 Token::BinOp(token::Shl) => true, // path (double UFCS)
874 Token::Interpolated(ref nt) => may_be_ident(&nt.0),
877 "lifetime" => match *token {
878 Token::Lifetime(_) => true,
879 Token::Interpolated(ref nt) => match nt.0 {
880 token::NtLifetime(_) | token::NtTT(_) => true,
886 token::CloseDelim(_) => false,
892 /// A call to the "black-box" parser to parse some rust nonterminal.
896 /// - `p`: the "black-box" parser to use
897 /// - `sp`: the `Span` we want to parse
898 /// - `name`: the name of the metavar _matcher_ we want to match (e.g., `tt`, `ident`, `block`,
903 /// The parsed nonterminal.
904 fn parse_nt<'a>(p: &mut Parser<'a>, sp: Span, name: &str) -> Nonterminal {
906 return token::NtTT(p.parse_token_tree());
908 // check at the beginning and the parser checks after each bump
909 p.process_potential_macro_variable();
911 "item" => match panictry!(p.parse_item()) {
912 Some(i) => token::NtItem(i),
914 p.fatal("expected an item keyword").emit();
918 "block" => token::NtBlock(panictry!(p.parse_block())),
919 "stmt" => match panictry!(p.parse_stmt()) {
920 Some(s) => token::NtStmt(s),
922 p.fatal("expected a statement").emit();
926 "pat" => token::NtPat(panictry!(p.parse_pat(None))),
927 "expr" => token::NtExpr(panictry!(p.parse_expr())),
928 "literal" => token::NtLiteral(panictry!(p.parse_literal_maybe_minus())),
929 "ty" => token::NtTy(panictry!(p.parse_ty())),
930 // this could be handled like a token, since it is one
931 "ident" => if let Some((ident, is_raw)) = get_macro_ident(&p.token) {
934 token::NtIdent(Ident::new(ident.name, span), is_raw)
936 let token_str = pprust::token_to_string(&p.token);
937 p.fatal(&format!("expected ident, found {}", &token_str)).emit();
940 "path" => token::NtPath(panictry!(p.parse_path_common(PathStyle::Type, false))),
941 "meta" => token::NtMeta(panictry!(p.parse_meta_item())),
942 "vis" => token::NtVis(panictry!(p.parse_visibility(true))),
943 "lifetime" => if p.check_lifetime() {
944 token::NtLifetime(p.expect_lifetime().ident)
946 let token_str = pprust::token_to_string(&p.token);
947 p.fatal(&format!("expected a lifetime, found `{}`", &token_str)).emit();
950 // this is not supposed to happen, since it has been checked
951 // when compiling the macro.
952 _ => p.span_bug(sp, "invalid fragment specifier"),