1 // Copyright 2012-2014 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 Earley-like parser, without support for in-grammar nonterminals,
12 //! only by calling out to the main rust parser for named nonterminals (which it
13 //! commits to fully when it hits one in a grammar). This means that there are no
14 //! completer or predictor rules, and therefore no need to store one column per
15 //! token: instead, there's a set of current Earley items and a set of next
16 //! ones. Instead of NTs, we have a special case for Kleene star. The big-O, in
17 //! pathological cases, is worse than traditional Earley parsing, but it's an
18 //! easier fit for Macro-by-Example-style rules, and I think the overhead is
19 //! lower. (In order to prevent the pathological case, we'd need to lazily
20 //! construct the resulting `NamedMatch`es at the very end. It'd be a pain,
21 //! and require more memory to keep around old items, but it would also save
24 //! Quick intro to how the parser works:
26 //! A 'position' is a dot in the middle of a matcher, usually represented as a
27 //! dot. For example `· a $( a )* a b` is a position, as is `a $( · a )* a b`.
29 //! The parser walks through the input a character at a time, maintaining a list
30 //! of items consistent with the current position in the input string: `cur_eis`.
32 //! As it processes them, it fills up `eof_eis` with items that would be valid if
33 //! the macro invocation is now over, `bb_eis` with items that are waiting on
34 //! a Rust nonterminal like `$e:expr`, and `next_eis` with items that are waiting
35 //! on a particular token. Most of the logic concerns moving the · through the
36 //! repetitions indicated by Kleene stars. It only advances or calls out to the
37 //! real Rust parser when no `cur_eis` items remain
39 //! Example: Start parsing `a a a a b` against [· a $( a )* a b].
41 //! Remaining input: `a a a a b`
42 //! next_eis: [· a $( a )* a b]
44 //! - - - Advance over an `a`. - - -
46 //! Remaining input: `a a a b`
47 //! cur: [a · $( a )* a b]
48 //! Descend/Skip (first item).
49 //! next: [a $( · a )* a b] [a $( a )* · a b].
51 //! - - - Advance over an `a`. - - -
53 //! Remaining input: `a a b`
54 //! cur: [a $( a · )* a b] next: [a $( a )* a · b]
55 //! Finish/Repeat (first item)
56 //! next: [a $( a )* · a b] [a $( · a )* a b] [a $( a )* a · b]
58 //! - - - Advance over an `a`. - - - (this looks exactly like the last step)
60 //! Remaining input: `a b`
61 //! cur: [a $( a · )* a b] next: [a $( a )* a · b]
62 //! Finish/Repeat (first item)
63 //! next: [a $( a )* · a b] [a $( · a )* a b] [a $( a )* a · b]
65 //! - - - Advance over an `a`. - - - (this looks exactly like the last step)
67 //! Remaining input: `b`
68 //! cur: [a $( a · )* a b] next: [a $( a )* a · b]
69 //! Finish/Repeat (first item)
70 //! next: [a $( a )* · a b] [a $( · a )* a b]
72 //! - - - Advance over a `b`. - - -
74 //! Remaining input: ``
75 //! eof: [a $( a )* a b ·]
77 pub use self::NamedMatch::*;
78 pub use self::ParseResult::*;
79 use self::TokenTreeOrTokenTreeVec::*;
82 use syntax_pos::{self, BytePos, Span};
84 use errors::FatalError;
85 use ext::tt::quoted::{self, TokenTree};
86 use parse::{Directory, ParseSess};
87 use parse::parser::{PathStyle, Parser};
88 use parse::token::{self, DocComment, Token, Nonterminal};
91 use tokenstream::TokenStream;
92 use util::small_vector::SmallVector;
96 use std::collections::HashMap;
97 use std::collections::hash_map::Entry::{Vacant, Occupied};
99 // To avoid costly uniqueness checks, we require that `MatchSeq` always has
103 enum TokenTreeOrTokenTreeVec {
105 TtSeq(Vec<TokenTree>),
108 impl TokenTreeOrTokenTreeVec {
109 fn len(&self) -> usize {
111 TtSeq(ref v) => v.len(),
112 Tt(ref tt) => tt.len(),
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
126 struct MatcherTtFrame {
127 elts: TokenTreeOrTokenTreeVec,
133 stack: Vec<MatcherTtFrame>,
134 top_elts: TokenTreeOrTokenTreeVec,
137 up: Option<Box<MatcherPos>>,
138 matches: Vec<Vec<Rc<NamedMatch>>>,
145 pub type NamedParseResult = ParseResult<HashMap<Ident, Rc<NamedMatch>>>;
147 pub fn count_names(ms: &[TokenTree]) -> usize {
148 ms.iter().fold(0, |count, elt| {
150 TokenTree::Sequence(_, ref seq) => {
153 TokenTree::Delimited(_, ref delim) => {
154 count_names(&delim.tts)
156 TokenTree::MetaVarDecl(..) => {
159 TokenTree::Token(..) => 0,
164 fn initial_matcher_pos(ms: Vec<TokenTree>, lo: BytePos) -> Box<MatcherPos> {
165 let match_idx_hi = count_names(&ms[..]);
166 let matches = create_matches(match_idx_hi);
167 Box::new(MatcherPos {
176 match_hi: match_idx_hi,
181 /// NamedMatch is a pattern-match result for a single token::MATCH_NONTERMINAL:
182 /// so it is associated with a single ident in a parse, and all
183 /// `MatchedNonterminal`s in the NamedMatch have the same nonterminal type
184 /// (expr, item, etc). Each leaf in a single NamedMatch corresponds to a
185 /// single token::MATCH_NONTERMINAL in the TokenTree that produced it.
187 /// The in-memory structure of a particular NamedMatch represents the match
188 /// that occurred when a particular subset of a matcher was applied to a
189 /// particular token tree.
191 /// The width of each MatchedSeq in the NamedMatch, and the identity of the
192 /// `MatchedNonterminal`s, will depend on the token tree it was applied to:
193 /// each MatchedSeq corresponds to a single TTSeq in the originating
194 /// token tree. The depth of the NamedMatch structure will therefore depend
195 /// only on the nesting depth of `ast::TTSeq`s in the originating
196 /// token tree it was derived from.
198 pub enum NamedMatch {
199 MatchedSeq(Vec<Rc<NamedMatch>>, syntax_pos::Span),
200 MatchedNonterminal(Rc<Nonterminal>)
203 fn nameize<I: Iterator<Item=Rc<NamedMatch>>>(sess: &ParseSess, ms: &[TokenTree], mut res: I)
204 -> NamedParseResult {
205 fn n_rec<I: Iterator<Item=Rc<NamedMatch>>>(sess: &ParseSess, m: &TokenTree, mut res: &mut I,
206 ret_val: &mut HashMap<Ident, Rc<NamedMatch>>)
207 -> Result<(), (syntax_pos::Span, String)> {
209 TokenTree::Sequence(_, ref seq) => {
210 for next_m in &seq.tts {
211 n_rec(sess, next_m, res.by_ref(), ret_val)?
214 TokenTree::Delimited(_, ref delim) => {
215 for next_m in &delim.tts {
216 n_rec(sess, next_m, res.by_ref(), ret_val)?;
219 TokenTree::MetaVarDecl(span, _, id) if id.name == keywords::Invalid.name() => {
220 if sess.missing_fragment_specifiers.borrow_mut().remove(&span) {
221 return Err((span, "missing fragment specifier".to_string()));
224 TokenTree::MetaVarDecl(sp, bind_name, _) => {
225 match ret_val.entry(bind_name) {
227 spot.insert(res.next().unwrap());
230 return Err((sp, format!("duplicated bind name: {}", bind_name)))
234 TokenTree::Token(..) => (),
240 let mut ret_val = HashMap::new();
242 match n_rec(sess, m, res.by_ref(), &mut ret_val) {
244 Err((sp, msg)) => return Error(sp, msg),
251 pub enum ParseResult<T> {
253 /// Arm failed to match. If the second parameter is `token::Eof`, it
254 /// indicates an unexpected end of macro invocation. Otherwise, it
255 /// indicates that no rules expected the given token.
256 Failure(syntax_pos::Span, Token),
257 /// Fatal error (malformed macro?). Abort compilation.
258 Error(syntax_pos::Span, String)
261 pub fn parse_failure_msg(tok: Token) -> String {
263 token::Eof => "unexpected end of macro invocation".to_string(),
264 _ => format!("no rules expected the token `{}`", pprust::token_to_string(&tok)),
268 /// Perform a token equality check, ignoring syntax context (that is, an unhygienic comparison)
269 fn token_name_eq(t1 : &Token, t2 : &Token) -> bool {
271 (&token::Ident(id1),&token::Ident(id2))
272 | (&token::Lifetime(id1),&token::Lifetime(id2)) =>
273 id1.name == id2.name,
278 fn create_matches(len: usize) -> Vec<Vec<Rc<NamedMatch>>> {
279 (0..len).into_iter().map(|_| Vec::new()).collect()
282 fn inner_parse_loop(sess: &ParseSess,
283 cur_eis: &mut SmallVector<Box<MatcherPos>>,
284 next_eis: &mut Vec<Box<MatcherPos>>,
285 eof_eis: &mut SmallVector<Box<MatcherPos>>,
286 bb_eis: &mut SmallVector<Box<MatcherPos>>,
288 span: syntax_pos::Span)
290 while let Some(mut ei) = cur_eis.pop() {
291 // When unzipped trees end, remove them
292 while ei.idx >= ei.top_elts.len() {
293 match ei.stack.pop() {
294 Some(MatcherTtFrame { elts, idx }) => {
303 let len = ei.top_elts.len();
305 // at end of sequence
307 // We are repeating iff there is a parent
309 // Disregarding the separator, add the "up" case to the tokens that should be
311 // (remove this condition to make trailing seps ok)
313 let mut new_pos = ei.up.clone().unwrap();
315 // update matches (the MBE "parse tree") by appending
316 // each tree as a subtree.
318 // I bet this is a perf problem: we're preemptively
319 // doing a lot of array work that will get thrown away
322 // Only touch the binders we have actually bound
323 for idx in ei.match_lo..ei.match_hi {
324 let sub = ei.matches[idx].clone();
326 .push(Rc::new(MatchedSeq(sub, Span { lo: ei.sp_lo, ..span })));
329 new_pos.match_cur = ei.match_hi;
331 cur_eis.push(new_pos);
334 // Check if we need a separator
335 if idx == len && ei.sep.is_some() {
336 // We have a separator, and it is the current token.
337 if ei.sep.as_ref().map(|ref sep| token_name_eq(&token, sep)).unwrap_or(false) {
341 } else { // we don't need a separator
342 ei.match_cur = ei.match_lo;
347 // We aren't repeating, so we must be potentially at the end of the input.
351 match ei.top_elts.get_tt(idx) {
352 /* need to descend into sequence */
353 TokenTree::Sequence(sp, seq) => {
354 if seq.op == quoted::KleeneOp::ZeroOrMore {
355 // Examine the case where there are 0 matches of this sequence
356 let mut new_ei = ei.clone();
357 new_ei.match_cur += seq.num_captures;
359 for idx in ei.match_cur..ei.match_cur + seq.num_captures {
360 new_ei.matches[idx].push(Rc::new(MatchedSeq(vec![], sp)));
362 cur_eis.push(new_ei);
365 // Examine the case where there is at least one match of this sequence
366 let matches = create_matches(ei.matches.len());
367 cur_eis.push(Box::new(MatcherPos {
369 sep: seq.separator.clone(),
372 match_lo: ei.match_cur,
373 match_cur: ei.match_cur,
374 match_hi: ei.match_cur + seq.num_captures,
377 top_elts: Tt(TokenTree::Sequence(sp, seq)),
380 TokenTree::MetaVarDecl(span, _, id) if id.name == keywords::Invalid.name() => {
381 if sess.missing_fragment_specifiers.borrow_mut().remove(&span) {
382 return Error(span, "missing fragment specifier".to_string());
385 TokenTree::MetaVarDecl(..) => {
386 // Built-in nonterminals never start with these tokens,
387 // so we can eliminate them from consideration.
389 token::CloseDelim(_) => {},
390 _ => bb_eis.push(ei),
393 seq @ TokenTree::Delimited(..) | seq @ TokenTree::Token(_, DocComment(..)) => {
394 let lower_elts = mem::replace(&mut ei.top_elts, Tt(seq));
396 ei.stack.push(MatcherTtFrame {
403 TokenTree::Token(_, ref t) => {
404 if token_name_eq(t, &token) {
416 pub fn parse(sess: &ParseSess, tts: TokenStream, ms: &[TokenTree], directory: Option<Directory>)
417 -> NamedParseResult {
418 let mut parser = Parser::new(sess, tts, directory, true);
419 let mut cur_eis = SmallVector::one(initial_matcher_pos(ms.to_owned(), parser.span.lo));
420 let mut next_eis = Vec::new(); // or proceed normally
423 let mut bb_eis = SmallVector::new(); // black-box parsed by parser.rs
424 let mut eof_eis = SmallVector::new();
425 assert!(next_eis.is_empty());
427 match inner_parse_loop(sess, &mut cur_eis, &mut next_eis, &mut eof_eis, &mut bb_eis,
428 &parser.token, parser.span) {
430 Failure(sp, tok) => return Failure(sp, tok),
431 Error(sp, msg) => return Error(sp, msg),
434 // inner parse loop handled all cur_eis, so it's empty
435 assert!(cur_eis.is_empty());
437 /* error messages here could be improved with links to orig. rules */
438 if token_name_eq(&parser.token, &token::Eof) {
439 if eof_eis.len() == 1 {
440 let matches = eof_eis[0].matches.iter_mut().map(|mut dv| dv.pop().unwrap());
441 return nameize(sess, ms, matches);
442 } else if eof_eis.len() > 1 {
443 return Error(parser.span, "ambiguity: multiple successful parses".to_string());
445 return Failure(parser.span, token::Eof);
447 } else if (!bb_eis.is_empty() && !next_eis.is_empty()) || bb_eis.len() > 1 {
448 let nts = bb_eis.iter().map(|ei| match ei.top_elts.get_tt(ei.idx) {
449 TokenTree::MetaVarDecl(_, bind, name) => {
450 format!("{} ('{}')", name, bind)
453 }).collect::<Vec<String>>().join(" or ");
455 return Error(parser.span, format!(
456 "local ambiguity: multiple parsing options: {}",
457 match next_eis.len() {
458 0 => format!("built-in NTs {}.", nts),
459 1 => format!("built-in NTs {} or 1 other option.", nts),
460 n => format!("built-in NTs {} or {} other options.", nts, n),
463 } else if bb_eis.is_empty() && next_eis.is_empty() {
464 return Failure(parser.span, parser.token);
465 } else if !next_eis.is_empty() {
466 /* Now process the next token */
467 cur_eis.extend(next_eis.drain(..));
469 } else /* bb_eis.len() == 1 */ {
470 let mut ei = bb_eis.pop().unwrap();
471 if let TokenTree::MetaVarDecl(span, _, ident) = ei.top_elts.get_tt(ei.idx) {
472 let match_cur = ei.match_cur;
473 ei.matches[match_cur].push(Rc::new(MatchedNonterminal(
474 Rc::new(parse_nt(&mut parser, span, &ident.name.as_str())))));
483 assert!(!cur_eis.is_empty());
487 fn parse_nt<'a>(p: &mut Parser<'a>, sp: Span, name: &str) -> Nonterminal {
490 return token::NtTT(p.parse_token_tree());
494 // check at the beginning and the parser checks after each bump
495 p.process_potential_macro_variable();
497 "item" => match panictry!(p.parse_item()) {
498 Some(i) => token::NtItem(i),
500 p.fatal("expected an item keyword").emit();
504 "block" => token::NtBlock(panictry!(p.parse_block())),
505 "stmt" => match panictry!(p.parse_stmt()) {
506 Some(s) => token::NtStmt(s),
508 p.fatal("expected a statement").emit();
512 "pat" => token::NtPat(panictry!(p.parse_pat())),
513 "expr" => token::NtExpr(panictry!(p.parse_expr())),
514 "ty" => token::NtTy(panictry!(p.parse_ty())),
515 // this could be handled like a token, since it is one
516 "ident" => match p.token {
517 token::Ident(sn) => {
519 token::NtIdent(Spanned::<Ident>{node: sn, span: p.prev_span})
522 let token_str = pprust::token_to_string(&p.token);
523 p.fatal(&format!("expected ident, found {}",
524 &token_str[..])).emit();
529 token::NtPath(panictry!(p.parse_path(PathStyle::Type)))
531 "meta" => token::NtMeta(panictry!(p.parse_meta_item())),
532 "vis" => token::NtVis(panictry!(p.parse_visibility(true))),
533 // this is not supposed to happen, since it has been checked
534 // when compiling the macro.
535 _ => p.span_bug(sp, "invalid fragment specifier")