1 //! This is an NFA-based parser, which calls out to the main rust parser for named nonterminals
2 //! (which it commits to fully when it hits one in a grammar). There's a set of current NFA threads
3 //! and a set of next ones. Instead of NTs, we have a special case for Kleene star. The big-O, in
4 //! pathological cases, is worse than traditional use of NFA or Earley parsing, but it's an easier
5 //! fit for Macro-by-Example-style rules.
7 //! (In order to prevent the pathological case, we'd need to lazily construct the resulting
8 //! `NamedMatch`es at the very end. It'd be a pain, and require more memory to keep around old
9 //! items, but it would also save overhead)
11 //! We don't say this parser uses the Earley algorithm, because it's unnecessarily inaccurate.
12 //! The macro parser restricts itself to the features of finite state automata. Earley parsers
13 //! can be described as an extension of NFAs with completion rules, prediction rules, and recursion.
15 //! Quick intro to how the parser works:
17 //! A 'position' is a dot in the middle of a matcher, usually represented as a
18 //! dot. For example `· a $( a )* a b` is a position, as is `a $( · a )* a b`.
20 //! The parser walks through the input a character at a time, maintaining a list
21 //! of threads consistent with the current position in the input string: `cur_items`.
23 //! As it processes them, it fills up `eof_items` with threads that would be valid if
24 //! the macro invocation is now over, `bb_items` with threads that are waiting on
25 //! a Rust nonterminal like `$e:expr`, and `next_items` with threads that are waiting
26 //! on a particular token. Most of the logic concerns moving the · through the
27 //! repetitions indicated by Kleene stars. The rules for moving the · without
28 //! consuming any input are called epsilon transitions. It only advances or calls
29 //! out to the real Rust parser when no `cur_items` threads remain.
34 //! Start parsing a a a a b against [· a $( a )* a b].
36 //! Remaining input: a a a a b
37 //! next: [· a $( a )* a b]
39 //! - - - Advance over an a. - - -
41 //! Remaining input: a a a b
42 //! cur: [a · $( a )* a b]
43 //! Descend/Skip (first item).
44 //! next: [a $( · a )* a b] [a $( a )* · a b].
46 //! - - - Advance over an a. - - -
48 //! Remaining input: a a b
49 //! cur: [a $( a · )* a b] [a $( a )* a · b]
50 //! Follow epsilon transition: Finish/Repeat (first item)
51 //! next: [a $( a )* · a b] [a $( · a )* a b] [a $( a )* a · b]
53 //! - - - Advance over an a. - - - (this looks exactly like the last step)
55 //! Remaining input: a b
56 //! cur: [a $( a · )* a b] [a $( a )* a · b]
57 //! Follow epsilon transition: Finish/Repeat (first item)
58 //! next: [a $( a )* · a b] [a $( · a )* a b] [a $( a )* a · b]
60 //! - - - Advance over an a. - - - (this looks exactly like the last step)
62 //! Remaining input: b
63 //! cur: [a $( a · )* a b] [a $( a )* a · b]
64 //! Follow epsilon transition: Finish/Repeat (first item)
65 //! next: [a $( a )* · a b] [a $( · a )* a b] [a $( a )* a · b]
67 //! - - - Advance over a b. - - -
69 //! Remaining input: ''
70 //! eof: [a $( a )* a b ·]
73 pub use self::NamedMatch::*;
74 pub use self::ParseResult::*;
75 use self::TokenTreeOrTokenTreeSlice::*;
78 use syntax_pos::{self, Span};
79 use errors::FatalError;
80 use ext::tt::quoted::{self, TokenTree};
81 use parse::{Directory, ParseSess};
82 use parse::parser::{Parser, PathStyle};
83 use parse::token::{self, DocComment, Nonterminal, Token};
85 use smallvec::SmallVec;
87 use tokenstream::{DelimSpan, TokenStream};
89 use rustc_data_structures::fx::FxHashMap;
90 use std::collections::hash_map::Entry::{Occupied, Vacant};
92 use std::ops::{Deref, DerefMut};
95 // To avoid costly uniqueness checks, we require that `MatchSeq` always has a nonempty body.
97 /// Either a sequence of token trees or a single one. This is used as the representation of the
98 /// sequence of tokens that make up a matcher.
100 enum TokenTreeOrTokenTreeSlice<'tt> {
102 TtSeq(&'tt [TokenTree]),
105 impl<'tt> TokenTreeOrTokenTreeSlice<'tt> {
106 /// Returns the number of constituent top-level token trees of `self` (top-level in that it
107 /// will not recursively descend into subtrees).
108 fn len(&self) -> usize {
110 TtSeq(ref v) => v.len(),
111 Tt(ref tt) => tt.len(),
115 /// The `index`-th token tree of `self`.
116 fn get_tt(&self, index: usize) -> TokenTree {
118 TtSeq(ref v) => v[index].clone(),
119 Tt(ref tt) => tt.get_tt(index),
124 /// An unzipping of `TokenTree`s... see the `stack` field of `MatcherPos`.
126 /// This is used by `inner_parse_loop` to keep track of delimited submatchers that we have
129 struct MatcherTtFrame<'tt> {
130 /// The "parent" matcher that we are descending into.
131 elts: TokenTreeOrTokenTreeSlice<'tt>,
132 /// The position of the "dot" in `elts` at the time we descended.
136 type NamedMatchVec = SmallVec<[NamedMatch; 4]>;
138 /// Represents a single "position" (aka "matcher position", aka "item"), as
139 /// described in the module documentation.
143 /// - `'root` represents the lifetime of the stack slot that holds the root
144 /// `MatcherPos`. As described in `MatcherPosHandle`, the root `MatcherPos`
145 /// structure is stored on the stack, but subsequent instances are put into
147 /// - `'tt` represents the lifetime of the token trees that this matcher
148 /// position refers to.
150 /// It is important to distinguish these two lifetimes because we have a
151 /// `SmallVec<TokenTreeOrTokenTreeSlice<'tt>>` below, and the destructor of
152 /// that is considered to possibly access the data from its elements (it lacks
153 /// a `#[may_dangle]` attribute). As a result, the compiler needs to know that
154 /// all the elements in that `SmallVec` strictly outlive the root stack slot
155 /// lifetime. By separating `'tt` from `'root`, we can show that.
157 struct MatcherPos<'root, 'tt: 'root> {
158 /// The token or sequence of tokens that make up the matcher
159 top_elts: TokenTreeOrTokenTreeSlice<'tt>,
161 /// The position of the "dot" in this matcher
164 /// The first span of source that the beginning of this matcher corresponds to. In other
165 /// words, the token in the source whose span is `sp_open` is matched against the first token of
169 /// For each named metavar in the matcher, we keep track of token trees matched against the
170 /// metavar by the black box parser. In particular, there may be more than one match per
171 /// metavar if we are in a repetition (each repetition matches each of the variables).
172 /// Moreover, matchers and repetitions can be nested; the `matches` field is shared (hence the
173 /// `Rc`) among all "nested" matchers. `match_lo`, `match_cur`, and `match_hi` keep track of
174 /// the current position of the `self` matcher position in the shared `matches` list.
176 /// Also, note that while we are descending into a sequence, matchers are given their own
177 /// `matches` vector. Only once we reach the end of a full repetition of the sequence do we add
178 /// all bound matches from the submatcher into the shared top-level `matches` vector. If `sep`
179 /// and `up` are `Some`, then `matches` is _not_ the shared top-level list. Instead, if one
180 /// wants the shared `matches`, one should use `up.matches`.
181 matches: Box<[Rc<NamedMatchVec>]>,
182 /// The position in `matches` corresponding to the first metavar in this matcher's sequence of
183 /// token trees. In other words, the first metavar in the first token of `top_elts` corresponds
184 /// to `matches[match_lo]`.
186 /// The position in `matches` corresponding to the metavar we are currently trying to match
187 /// against the source token stream. `match_lo <= match_cur <= match_hi`.
189 /// Similar to `match_lo` except `match_hi` is the position in `matches` of the _last_ metavar
193 // The following fields are used if we are matching a repetition. If we aren't, they should be
196 /// The KleeneOp of this sequence if we are in a repetition.
197 seq_op: Option<quoted::KleeneOp>,
199 /// The separator if we are in a repetition.
202 /// The "parent" matcher position if we are in a repetition. That is, the matcher position just
203 /// before we enter the sequence.
204 up: Option<MatcherPosHandle<'root, 'tt>>,
206 /// Specifically used to "unzip" token trees. By "unzip", we mean to unwrap the delimiters from
207 /// a delimited token tree (e.g., something wrapped in `(` `)`) or to get the contents of a doc
210 /// When matching against matchers with nested delimited submatchers (e.g., `pat ( pat ( .. )
211 /// pat ) pat`), we need to keep track of the matchers we are descending into. This stack does
212 /// that where the bottom of the stack is the outermost matcher.
213 /// Also, throughout the comments, this "descent" is often referred to as "unzipping"...
214 stack: SmallVec<[MatcherTtFrame<'tt>; 1]>,
217 impl<'root, 'tt> MatcherPos<'root, 'tt> {
218 /// Add `m` as a named match for the `idx`-th metavar.
219 fn push_match(&mut self, idx: usize, m: NamedMatch) {
220 let matches = Rc::make_mut(&mut self.matches[idx]);
225 // Lots of MatcherPos instances are created at runtime. Allocating them on the
226 // heap is slow. Furthermore, using SmallVec<MatcherPos> to allocate them all
227 // on the stack is also slow, because MatcherPos is quite a large type and
228 // instances get moved around a lot between vectors, which requires lots of
229 // slow memcpy calls.
231 // Therefore, the initial MatcherPos is always allocated on the stack,
232 // subsequent ones (of which there aren't that many) are allocated on the heap,
233 // and this type is used to encapsulate both cases.
234 enum MatcherPosHandle<'root, 'tt: 'root> {
235 Ref(&'root mut MatcherPos<'root, 'tt>),
236 Box(Box<MatcherPos<'root, 'tt>>),
239 impl<'root, 'tt> Clone for MatcherPosHandle<'root, 'tt> {
240 // This always produces a new Box.
241 fn clone(&self) -> Self {
242 MatcherPosHandle::Box(match *self {
243 MatcherPosHandle::Ref(ref r) => Box::new((**r).clone()),
244 MatcherPosHandle::Box(ref b) => b.clone(),
249 impl<'root, 'tt> Deref for MatcherPosHandle<'root, 'tt> {
250 type Target = MatcherPos<'root, 'tt>;
251 fn deref(&self) -> &Self::Target {
253 MatcherPosHandle::Ref(ref r) => r,
254 MatcherPosHandle::Box(ref b) => b,
259 impl<'root, 'tt> DerefMut for MatcherPosHandle<'root, 'tt> {
260 fn deref_mut(&mut self) -> &mut MatcherPos<'root, 'tt> {
262 MatcherPosHandle::Ref(ref mut r) => r,
263 MatcherPosHandle::Box(ref mut b) => b,
268 /// Represents the possible results of an attempted parse.
269 pub enum ParseResult<T> {
270 /// Parsed successfully.
272 /// Arm failed to match. If the second parameter is `token::Eof`, it indicates an unexpected
273 /// end of macro invocation. Otherwise, it indicates that no rules expected the given token.
274 Failure(syntax_pos::Span, Token, &'static str),
275 /// Fatal error (malformed macro?). Abort compilation.
276 Error(syntax_pos::Span, String),
279 /// A `ParseResult` where the `Success` variant contains a mapping of `Ident`s to `NamedMatch`es.
280 /// This represents the mapping of metavars to the token trees they bind to.
281 pub type NamedParseResult = ParseResult<FxHashMap<Ident, Rc<NamedMatch>>>;
283 /// Count how many metavars are named in the given matcher `ms`.
284 pub fn count_names(ms: &[TokenTree]) -> usize {
285 ms.iter().fold(0, |count, elt| {
287 TokenTree::Sequence(_, ref seq) => seq.num_captures,
288 TokenTree::Delimited(_, ref delim) => count_names(&delim.tts),
289 TokenTree::MetaVar(..) => 0,
290 TokenTree::MetaVarDecl(..) => 1,
291 TokenTree::Token(..) => 0,
296 /// `len` `Vec`s (initially shared and empty) that will store matches of metavars.
297 fn create_matches(len: usize) -> Box<[Rc<NamedMatchVec>]> {
301 let empty_matches = Rc::new(SmallVec::new());
302 vec![empty_matches; len]
306 /// Generate the top-level matcher position in which the "dot" is before the first token of the
307 /// matcher `ms` and we are going to start matching at the span `open` in the source.
308 fn initial_matcher_pos<'root, 'tt>(ms: &'tt [TokenTree], open: Span) -> MatcherPos<'root, 'tt> {
309 let match_idx_hi = count_names(ms);
310 let matches = create_matches(match_idx_hi);
312 // Start with the top level matcher given to us
313 top_elts: TtSeq(ms), // "elts" is an abbr. for "elements"
314 // The "dot" is before the first token of the matcher
316 // We start matching at the span `open` in the source code
319 // Initialize `matches` to a bunch of empty `Vec`s -- one for each metavar in `top_elts`.
320 // `match_lo` for `top_elts` is 0 and `match_hi` is `matches.len()`. `match_cur` is 0 since
321 // we haven't actually matched anything yet.
325 match_hi: match_idx_hi,
327 // Haven't descended into any delimiters, so empty stack
330 // Haven't descended into any sequences, so both of these are `None`.
337 /// `NamedMatch` is a pattern-match result for a single `token::MATCH_NONTERMINAL`:
338 /// so it is associated with a single ident in a parse, and all
339 /// `MatchedNonterminal`s in the `NamedMatch` have the same nonterminal type
340 /// (expr, item, etc). Each leaf in a single `NamedMatch` corresponds to a
341 /// single `token::MATCH_NONTERMINAL` in the `TokenTree` that produced it.
343 /// The in-memory structure of a particular `NamedMatch` represents the match
344 /// that occurred when a particular subset of a matcher was applied to a
345 /// particular token tree.
347 /// The width of each `MatchedSeq` in the `NamedMatch`, and the identity of
348 /// the `MatchedNonterminal`s, will depend on the token tree it was applied
349 /// to: each `MatchedSeq` corresponds to a single `TTSeq` in the originating
350 /// token tree. The depth of the `NamedMatch` structure will therefore depend
351 /// only on the nesting depth of `ast::TTSeq`s in the originating
352 /// token tree it was derived from.
353 #[derive(Debug, Clone)]
354 pub enum NamedMatch {
355 MatchedSeq(Rc<NamedMatchVec>, DelimSpan),
356 MatchedNonterminal(Rc<Nonterminal>),
359 /// Takes a sequence of token trees `ms` representing a matcher which successfully matched input
360 /// and an iterator of items that matched input and produces a `NamedParseResult`.
361 fn nameize<I: Iterator<Item = NamedMatch>>(
365 ) -> NamedParseResult {
366 // Recursively descend into each type of matcher (e.g., sequences, delimited, metavars) and make
367 // sure that each metavar has _exactly one_ binding. If a metavar does not have exactly one
368 // binding, then there is an error. If it does, then we insert the binding into the
369 // `NamedParseResult`.
370 fn n_rec<I: Iterator<Item = NamedMatch>>(
374 ret_val: &mut FxHashMap<Ident, Rc<NamedMatch>>,
375 ) -> Result<(), (syntax_pos::Span, String)> {
377 TokenTree::Sequence(_, ref seq) => for next_m in &seq.tts {
378 n_rec(sess, next_m, res.by_ref(), ret_val)?
380 TokenTree::Delimited(_, ref delim) => for next_m in &delim.tts {
381 n_rec(sess, next_m, res.by_ref(), ret_val)?;
383 TokenTree::MetaVarDecl(span, _, id) if id.name == keywords::Invalid.name() => {
384 if sess.missing_fragment_specifiers.borrow_mut().remove(&span) {
385 return Err((span, "missing fragment specifier".to_string()));
388 TokenTree::MetaVarDecl(sp, bind_name, _) => {
389 match ret_val.entry(bind_name) {
391 // FIXME(simulacrum): Don't construct Rc here
392 spot.insert(Rc::new(res.next().unwrap()));
395 return Err((sp, format!("duplicated bind name: {}", bind_name)))
399 TokenTree::MetaVar(..) | TokenTree::Token(..) => (),
405 let mut ret_val = FxHashMap::default();
407 match n_rec(sess, m, res.by_ref(), &mut ret_val) {
409 Err((sp, msg)) => return Error(sp, msg),
416 /// Generate an appropriate parsing failure message. For EOF, this is "unexpected end...". For
417 /// other tokens, this is "unexpected token...".
418 pub fn parse_failure_msg(tok: Token) -> String {
420 token::Eof => "unexpected end of macro invocation".to_string(),
422 "no rules expected the token `{}`",
423 pprust::token_to_string(&tok)
428 /// Perform a token equality check, ignoring syntax context (that is, an unhygienic comparison)
429 fn token_name_eq(t1: &Token, t2: &Token) -> bool {
430 if let (Some((id1, is_raw1)), Some((id2, is_raw2))) = (t1.ident(), t2.ident()) {
431 id1.name == id2.name && is_raw1 == is_raw2
432 } else if let (Some(id1), Some(id2)) = (t1.lifetime(), t2.lifetime()) {
439 /// Process the matcher positions of `cur_items` until it is empty. In the process, this will
440 /// produce more items in `next_items`, `eof_items`, and `bb_items`.
442 /// For more info about the how this happens, see the module-level doc comments and the inline
443 /// comments of this function.
447 /// - `sess`: the parsing session into which errors are emitted.
448 /// - `cur_items`: the set of current items to be processed. This should be empty by the end of a
449 /// successful execution of this function.
450 /// - `next_items`: the set of newly generated items. These are used to replenish `cur_items` in
451 /// the function `parse`.
452 /// - `eof_items`: the set of items that would be valid if this was the EOF.
453 /// - `bb_items`: the set of items that are waiting for the black-box parser.
454 /// - `token`: the current token of the parser.
455 /// - `span`: the `Span` in the source code corresponding to the token trees we are trying to match
456 /// against the matcher positions in `cur_items`.
460 /// A `ParseResult`. Note that matches are kept track of through the items generated.
461 fn inner_parse_loop<'root, 'tt>(
463 cur_items: &mut SmallVec<[MatcherPosHandle<'root, 'tt>; 1]>,
464 next_items: &mut Vec<MatcherPosHandle<'root, 'tt>>,
465 eof_items: &mut SmallVec<[MatcherPosHandle<'root, 'tt>; 1]>,
466 bb_items: &mut SmallVec<[MatcherPosHandle<'root, 'tt>; 1]>,
468 span: syntax_pos::Span,
469 ) -> ParseResult<()> {
470 // Pop items from `cur_items` until it is empty.
471 while let Some(mut item) = cur_items.pop() {
472 // When unzipped trees end, remove them. This corresponds to backtracking out of a
473 // delimited submatcher into which we already descended. In backtracking out again, we need
474 // to advance the "dot" past the delimiters in the outer matcher.
475 while item.idx >= item.top_elts.len() {
476 match item.stack.pop() {
477 Some(MatcherTtFrame { elts, idx }) => {
478 item.top_elts = elts;
485 // Get the current position of the "dot" (`idx`) in `item` and the number of token trees in
486 // the matcher (`len`).
488 let len = item.top_elts.len();
490 // If `idx >= len`, then we are at or past the end of the matcher of `item`.
492 // We are repeating iff there is a parent. If the matcher is inside of a repetition,
493 // then we could be at the end of a sequence or at the beginning of the next
495 if item.up.is_some() {
496 // At this point, regardless of whether there is a separator, we should add all
497 // matches from the complete repetition of the sequence to the shared, top-level
498 // `matches` list (actually, `up.matches`, which could itself not be the top-level,
499 // but anyway...). Moreover, we add another item to `cur_items` in which the "dot"
500 // is at the end of the `up` matcher. This ensures that the "dot" in the `up`
501 // matcher is also advanced sufficiently.
503 // NOTE: removing the condition `idx == len` allows trailing separators.
505 // Get the `up` matcher
506 let mut new_pos = item.up.clone().unwrap();
508 // Add matches from this repetition to the `matches` of `up`
509 for idx in item.match_lo..item.match_hi {
510 let sub = item.matches[idx].clone();
511 let span = DelimSpan::from_pair(item.sp_open, span);
512 new_pos.push_match(idx, MatchedSeq(sub, span));
515 // Move the "dot" past the repetition in `up`
516 new_pos.match_cur = item.match_hi;
518 cur_items.push(new_pos);
521 // Check if we need a separator.
522 if idx == len && item.sep.is_some() {
523 // We have a separator, and it is the current token. We can advance past the
527 .map(|sep| token_name_eq(token, sep))
531 next_items.push(item);
534 // We don't need a separator. Move the "dot" back to the beginning of the matcher
535 // and try to match again UNLESS we are only allowed to have _one_ repetition.
536 else if item.seq_op != Some(quoted::KleeneOp::ZeroOrOne) {
537 item.match_cur = item.match_lo;
539 cur_items.push(item);
542 // If we are not in a repetition, then being at the end of a matcher means that we have
543 // reached the potential end of the input.
545 eof_items.push(item);
548 // We are in the middle of a matcher.
550 // Look at what token in the matcher we are trying to match the current token (`token`)
551 // against. Depending on that, we may generate new items.
552 match item.top_elts.get_tt(idx) {
553 // Need to descend into a sequence
554 TokenTree::Sequence(sp, seq) => {
555 // Examine the case where there are 0 matches of this sequence
556 if seq.op == quoted::KleeneOp::ZeroOrMore
557 || seq.op == quoted::KleeneOp::ZeroOrOne
559 let mut new_item = item.clone();
560 new_item.match_cur += seq.num_captures;
562 for idx in item.match_cur..item.match_cur + seq.num_captures {
563 new_item.push_match(idx, MatchedSeq(Rc::new(smallvec![]), sp));
565 cur_items.push(new_item);
568 let matches = create_matches(item.matches.len());
569 cur_items.push(MatcherPosHandle::Box(Box::new(MatcherPos {
571 sep: seq.separator.clone(),
572 seq_op: Some(seq.op),
575 match_lo: item.match_cur,
576 match_cur: item.match_cur,
577 match_hi: item.match_cur + seq.num_captures,
580 top_elts: Tt(TokenTree::Sequence(sp, seq)),
584 // We need to match a metavar (but the identifier is invalid)... this is an error
585 TokenTree::MetaVarDecl(span, _, id) if id.name == keywords::Invalid.name() => {
586 if sess.missing_fragment_specifiers.borrow_mut().remove(&span) {
587 return Error(span, "missing fragment specifier".to_string());
591 // We need to match a metavar with a valid ident... call out to the black-box
592 // parser by adding an item to `bb_items`.
593 TokenTree::MetaVarDecl(_, _, id) => {
594 // Built-in nonterminals never start with these tokens,
595 // so we can eliminate them from consideration.
596 if may_begin_with(&*id.as_str(), token) {
601 // We need to descend into a delimited submatcher or a doc comment. To do this, we
602 // push the current matcher onto a stack and push a new item containing the
603 // submatcher onto `cur_items`.
605 // At the beginning of the loop, if we reach the end of the delimited submatcher,
606 // we pop the stack to backtrack out of the descent.
607 seq @ TokenTree::Delimited(..) | seq @ TokenTree::Token(_, DocComment(..)) => {
608 let lower_elts = mem::replace(&mut item.top_elts, Tt(seq));
610 item.stack.push(MatcherTtFrame {
615 cur_items.push(item);
618 // We just matched a normal token. We can just advance the parser.
619 TokenTree::Token(_, ref t) if token_name_eq(t, token) => {
621 next_items.push(item);
624 // There was another token that was not `token`... This means we can't add any
625 // rules. NOTE that this is not necessarily an error unless _all_ items in
626 // `cur_items` end up doing this. There may still be some other matchers that do
627 // end up working out.
628 TokenTree::Token(..) | TokenTree::MetaVar(..) => {}
633 // Yay a successful parse (so far)!
637 /// Use the given sequence of token trees (`ms`) as a matcher. Match the given token stream `tts`
638 /// against it and return the match.
642 /// - `sess`: The session into which errors are emitted
643 /// - `tts`: The tokenstream we are matching against the pattern `ms`
644 /// - `ms`: A sequence of token trees representing a pattern against which we are matching
645 /// - `directory`: Information about the file locations (needed for the black-box parser)
646 /// - `recurse_into_modules`: Whether or not to recurse into modules (needed for the black-box
652 directory: Option<Directory>,
653 recurse_into_modules: bool,
654 ) -> NamedParseResult {
655 // Create a parser that can be used for the "black box" parts.
656 let mut parser = Parser::new(sess, tts, directory, recurse_into_modules, true);
658 // A queue of possible matcher positions. We initialize it with the matcher position in which
659 // the "dot" is before the first token of the first token tree in `ms`. `inner_parse_loop` then
660 // processes all of these possible matcher positions and produces possible next positions into
661 // `next_items`. After some post-processing, the contents of `next_items` replenish `cur_items`
662 // and we start over again.
664 // This MatcherPos instance is allocated on the stack. All others -- and
665 // there are frequently *no* others! -- are allocated on the heap.
666 let mut initial = initial_matcher_pos(ms, parser.span);
667 let mut cur_items = smallvec![MatcherPosHandle::Ref(&mut initial)];
668 let mut next_items = Vec::new();
671 // Matcher positions black-box parsed by parser.rs (`parser`)
672 let mut bb_items = SmallVec::new();
674 // Matcher positions that would be valid if the macro invocation was over now
675 let mut eof_items = SmallVec::new();
676 assert!(next_items.is_empty());
678 // Process `cur_items` until either we have finished the input or we need to get some
679 // parsing from the black-box parser done. The result is that `next_items` will contain a
680 // bunch of possible next matcher positions in `next_items`.
681 match inner_parse_loop(
691 Failure(sp, tok, t) => return Failure(sp, tok, t),
692 Error(sp, msg) => return Error(sp, msg),
695 // inner parse loop handled all cur_items, so it's empty
696 assert!(cur_items.is_empty());
698 // We need to do some post processing after the `inner_parser_loop`.
700 // Error messages here could be improved with links to original rules.
702 // If we reached the EOF, check that there is EXACTLY ONE possible matcher. Otherwise,
703 // either the parse is ambiguous (which should never happen) or there is a syntax error.
704 if token_name_eq(&parser.token, &token::Eof) {
705 if eof_items.len() == 1 {
706 let matches = eof_items[0]
709 .map(|dv| Rc::make_mut(dv).pop().unwrap());
710 return nameize(sess, ms, matches);
711 } else if eof_items.len() > 1 {
714 "ambiguity: multiple successful parses".to_string(),
718 if parser.span.is_dummy() {
721 sess.source_map().next_point(parser.span)
724 "missing tokens in macro arguments",
728 // Performance hack: eof_items may share matchers via Rc with other things that we want
729 // to modify. Dropping eof_items now may drop these refcounts to 1, preventing an
730 // unnecessary implicit clone later in Rc::make_mut.
733 // Another possibility is that we need to call out to parse some rust nonterminal
734 // (black-box) parser. However, if there is not EXACTLY ONE of these, something is wrong.
735 if (!bb_items.is_empty() && !next_items.is_empty()) || bb_items.len() > 1 {
738 .map(|item| match item.top_elts.get_tt(item.idx) {
739 TokenTree::MetaVarDecl(_, bind, name) => format!("{} ('{}')", name, bind),
742 .collect::<Vec<String>>()
748 "local ambiguity: multiple parsing options: {}",
749 match next_items.len() {
750 0 => format!("built-in NTs {}.", nts),
751 1 => format!("built-in NTs {} or 1 other option.", nts),
752 n => format!("built-in NTs {} or {} other options.", nts, n),
757 // If there are no possible next positions AND we aren't waiting for the black-box parser,
758 // then there is a syntax error.
759 else if bb_items.is_empty() && next_items.is_empty() {
763 "no rules expected this token in macro call",
766 // Dump all possible `next_items` into `cur_items` for the next iteration.
767 else if !next_items.is_empty() {
768 // Now process the next token
769 cur_items.extend(next_items.drain(..));
772 // Finally, we have the case where we need to call the black-box parser to get some
775 assert_eq!(bb_items.len(), 1);
777 let mut item = bb_items.pop().unwrap();
778 if let TokenTree::MetaVarDecl(span, _, ident) = item.top_elts.get_tt(item.idx) {
779 let match_cur = item.match_cur;
782 MatchedNonterminal(Rc::new(parse_nt(&mut parser, span, &ident.as_str()))),
789 cur_items.push(item);
792 assert!(!cur_items.is_empty());
796 /// The token is an identifier, but not `_`.
797 /// We prohibit passing `_` to macros expecting `ident` for now.
798 fn get_macro_ident(token: &Token) -> Option<(Ident, bool)> {
800 token::Ident(ident, is_raw) if ident.name != keywords::Underscore.name() =>
801 Some((ident, is_raw)),
806 /// Checks whether a non-terminal may begin with a particular token.
808 /// Returning `false` is a *stability guarantee* that such a matcher will *never* begin with that
809 /// token. Be conservative (return true) if not sure.
810 fn may_begin_with(name: &str, token: &Token) -> bool {
811 /// Checks whether the non-terminal may contain a single (non-keyword) identifier.
812 fn may_be_ident(nt: &token::Nonterminal) -> bool {
814 token::NtItem(_) | token::NtBlock(_) | token::NtVis(_) => false,
820 "expr" => token.can_begin_expr(),
821 "ty" => token.can_begin_type(),
822 "ident" => get_macro_ident(token).is_some(),
823 "literal" => token.can_begin_literal_or_bool(),
824 "vis" => match *token {
825 // The follow-set of :vis + "priv" keyword + interpolated
826 Token::Comma | Token::Ident(..) | Token::Interpolated(_) => true,
827 _ => token.can_begin_type(),
829 "block" => match *token {
830 Token::OpenDelim(token::Brace) => true,
831 Token::Interpolated(ref nt) => match nt.0 {
838 | token::NtVis(_) => false, // none of these may start with '{'.
843 "path" | "meta" => match *token {
844 Token::ModSep | Token::Ident(..) => true,
845 Token::Interpolated(ref nt) => match nt.0 {
846 token::NtPath(_) | token::NtMeta(_) => true,
847 _ => may_be_ident(&nt.0),
851 "pat" => match *token {
852 Token::Ident(..) | // box, ref, mut, and other identifiers (can stricten)
853 Token::OpenDelim(token::Paren) | // tuple pattern
854 Token::OpenDelim(token::Bracket) | // slice pattern
855 Token::BinOp(token::And) | // reference
856 Token::BinOp(token::Minus) | // negative literal
857 Token::AndAnd | // double reference
858 Token::Literal(..) | // literal
859 Token::DotDot | // range pattern (future compat)
860 Token::DotDotDot | // range pattern (future compat)
861 Token::ModSep | // path
862 Token::Lt | // path (UFCS constant)
863 Token::BinOp(token::Shl) => true, // path (double UFCS)
864 Token::Interpolated(ref nt) => may_be_ident(&nt.0),
867 "lifetime" => match *token {
868 Token::Lifetime(_) => true,
869 Token::Interpolated(ref nt) => match nt.0 {
870 token::NtLifetime(_) | token::NtTT(_) => true,
876 token::CloseDelim(_) => false,
882 /// A call to the "black-box" parser to parse some rust nonterminal.
886 /// - `p`: the "black-box" parser to use
887 /// - `sp`: the `Span` we want to parse
888 /// - `name`: the name of the metavar _matcher_ we want to match (e.g., `tt`, `ident`, `block`,
893 /// The parsed nonterminal.
894 fn parse_nt<'a>(p: &mut Parser<'a>, sp: Span, name: &str) -> Nonterminal {
896 return token::NtTT(p.parse_token_tree());
898 // check at the beginning and the parser checks after each bump
899 p.process_potential_macro_variable();
901 "item" => match panictry!(p.parse_item()) {
902 Some(i) => token::NtItem(i),
904 p.fatal("expected an item keyword").emit();
908 "block" => token::NtBlock(panictry!(p.parse_block())),
909 "stmt" => match panictry!(p.parse_stmt()) {
910 Some(s) => token::NtStmt(s),
912 p.fatal("expected a statement").emit();
916 "pat" => token::NtPat(panictry!(p.parse_pat(None))),
917 "expr" => token::NtExpr(panictry!(p.parse_expr())),
918 "literal" => token::NtLiteral(panictry!(p.parse_literal_maybe_minus())),
919 "ty" => token::NtTy(panictry!(p.parse_ty())),
920 // this could be handled like a token, since it is one
921 "ident" => if let Some((ident, is_raw)) = get_macro_ident(&p.token) {
924 token::NtIdent(Ident::new(ident.name, span), is_raw)
926 let token_str = pprust::token_to_string(&p.token);
927 p.fatal(&format!("expected ident, found {}", &token_str)).emit();
930 "path" => token::NtPath(panictry!(p.parse_path_common(PathStyle::Type, false))),
931 "meta" => token::NtMeta(panictry!(p.parse_meta_item())),
932 "vis" => token::NtVis(panictry!(p.parse_visibility(true))),
933 "lifetime" => if p.check_lifetime() {
934 token::NtLifetime(p.expect_lifetime().ident)
936 let token_str = pprust::token_to_string(&p.token);
937 p.fatal(&format!("expected a lifetime, found `{}`", &token_str)).emit();
940 // this is not supposed to happen, since it has been checked
941 // when compiling the macro.
942 _ => p.span_bug(sp, "invalid fragment specifier"),