1 // Copyright 2012 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 // Earley-like parser for macros.
14 use ast::{Matcher, MatchTok, MatchSeq, MatchNonterminal, Ident};
15 use codemap::{BytePos, mk_sp};
17 use parse::lexer::*; //resolve bug?
19 use parse::attr::ParserAttr;
20 use parse::parser::{LifetimeAndTypesWithoutColons, Parser};
21 use parse::token::{Token, EOF, Nonterminal};
24 use collections::HashMap;
27 /* This is an Earley-like parser, without support for in-grammar nonterminals,
28 only by calling out to the main rust parser for named nonterminals (which it
29 commits to fully when it hits one in a grammar). This means that there are no
30 completer or predictor rules, and therefore no need to store one column per
31 token: instead, there's a set of current Earley items and a set of next
32 ones. Instead of NTs, we have a special case for Kleene star. The big-O, in
33 pathological cases, is worse than traditional Earley parsing, but it's an
34 easier fit for Macro-by-Example-style rules, and I think the overhead is
35 lower. (In order to prevent the pathological case, we'd need to lazily
36 construct the resulting `NamedMatch`es at the very end. It'd be a pain,
37 and require more memory to keep around old items, but it would also save
40 /* Quick intro to how the parser works:
42 A 'position' is a dot in the middle of a matcher, usually represented as a
43 dot. For example `· a $( a )* a b` is a position, as is `a $( · a )* a b`.
45 The parser walks through the input a character at a time, maintaining a list
46 of items consistent with the current position in the input string: `cur_eis`.
48 As it processes them, it fills up `eof_eis` with items that would be valid if
49 the macro invocation is now over, `bb_eis` with items that are waiting on
50 a Rust nonterminal like `$e:expr`, and `next_eis` with items that are waiting
51 on the a particular token. Most of the logic concerns moving the · through the
52 repetitions indicated by Kleene stars. It only advances or calls out to the
53 real Rust parser when no `cur_eis` items remain
55 Example: Start parsing `a a a a b` against [· a $( a )* a b].
57 Remaining input: `a a a a b`
58 next_eis: [· a $( a )* a b]
60 - - - Advance over an `a`. - - -
62 Remaining input: `a a a b`
63 cur: [a · $( a )* a b]
64 Descend/Skip (first item).
65 next: [a $( · a )* a b] [a $( a )* · a b].
67 - - - Advance over an `a`. - - -
69 Remaining input: `a a b`
70 cur: [a $( a · )* a b] next: [a $( a )* a · b]
71 Finish/Repeat (first item)
72 next: [a $( a )* · a b] [a $( · a )* a b] [a $( a )* a · b]
74 - - - Advance over an `a`. - - - (this looks exactly like the last step)
76 Remaining input: `a b`
77 cur: [a $( a · )* a b] next: [a $( a )* a · b]
78 Finish/Repeat (first item)
79 next: [a $( a )* · a b] [a $( · a )* a b] [a $( a )* a · b]
81 - - - Advance over an `a`. - - - (this looks exactly like the last step)
84 cur: [a $( a · )* a b] next: [a $( a )* a · b]
85 Finish/Repeat (first item)
86 next: [a $( a )* · a b] [a $( · a )* a b]
88 - - - Advance over a `b`. - - -
91 eof: [a $( a )* a b ·]
96 /* to avoid costly uniqueness checks, we require that `MatchSeq` always has a
101 pub struct MatcherPos {
102 elts: Vec<ast::Matcher> , // maybe should be <'>? Need to understand regions.
105 up: Option<~MatcherPos>,
106 matches: Vec<Vec<@NamedMatch>>,
107 match_lo: uint, match_hi: uint,
111 pub fn count_names(ms: &[Matcher]) -> uint {
112 ms.iter().fold(0, |ct, m| {
115 MatchSeq(ref more_ms, _, _, _, _) => {
116 count_names(more_ms.as_slice())
118 MatchNonterminal(_, _, _) => 1u
122 pub fn initial_matcher_pos(ms: Vec<Matcher> , sep: Option<Token>, lo: BytePos)
124 let mut match_idx_hi = 0u;
125 for elt in ms.iter() {
128 MatchSeq(_,_,_,_,hi) => {
129 match_idx_hi = hi; // it is monotonic...
131 MatchNonterminal(_,_,pos) => {
132 match_idx_hi = pos+1u; // ...so latest is highest
136 let matches = Vec::from_fn(count_names(ms.as_slice()), |_i| Vec::new());
144 match_hi: match_idx_hi,
149 // NamedMatch is a pattern-match result for a single ast::MatchNonterminal:
150 // so it is associated with a single ident in a parse, and all
151 // MatchedNonterminal's in the NamedMatch have the same nonterminal type
152 // (expr, item, etc). All the leaves in a single NamedMatch correspond to a
153 // single matcher_nonterminal in the ast::Matcher that produced it.
155 // It should probably be renamed, it has more or less exact correspondence to
156 // ast::match nodes, and the in-memory structure of a particular NamedMatch
157 // represents the match that occurred when a particular subset of an
158 // ast::match -- those ast::Matcher nodes leading to a single
159 // MatchNonterminal -- was applied to a particular token tree.
161 // The width of each MatchedSeq in the NamedMatch, and the identity of the
162 // MatchedNonterminal's, will depend on the token tree it was applied to: each
163 // MatchedSeq corresponds to a single MatchSeq in the originating
164 // ast::Matcher. The depth of the NamedMatch structure will therefore depend
165 // only on the nesting depth of ast::MatchSeq's in the originating
166 // ast::Matcher it was derived from.
168 pub enum NamedMatch {
169 MatchedSeq(Vec<@NamedMatch> , codemap::Span),
170 MatchedNonterminal(Nonterminal)
173 pub fn nameize(p_s: @ParseSess, ms: &[Matcher], res: &[@NamedMatch])
174 -> HashMap<Ident, @NamedMatch> {
175 fn n_rec(p_s: @ParseSess, m: &Matcher, res: &[@NamedMatch],
176 ret_val: &mut HashMap<Ident, @NamedMatch>) {
178 codemap::Spanned {node: MatchTok(_), .. } => (),
179 codemap::Spanned {node: MatchSeq(ref more_ms, _, _, _, _), .. } => {
180 for next_m in more_ms.iter() {
181 n_rec(p_s, next_m, res, ret_val)
185 node: MatchNonterminal(bind_name, _, idx),
188 if ret_val.contains_key(&bind_name) {
189 let string = token::get_ident(bind_name);
191 .span_fatal(span, "duplicated bind name: " + string.get())
193 ret_val.insert(bind_name, res[idx]);
197 let mut ret_val = HashMap::new();
198 for m in ms.iter() { n_rec(p_s, m, res, &mut ret_val) }
202 pub enum ParseResult {
203 Success(HashMap<Ident, @NamedMatch>),
204 Failure(codemap::Span, ~str),
205 Error(codemap::Span, ~str)
208 pub fn parse_or_else<R: Reader>(sess: @ParseSess,
209 cfg: ast::CrateConfig,
212 -> HashMap<Ident, @NamedMatch> {
213 match parse(sess, cfg, rdr, ms.as_slice()) {
215 Failure(sp, str) => sess.span_diagnostic.span_fatal(sp, str),
216 Error(sp, str) => sess.span_diagnostic.span_fatal(sp, str)
220 // perform a token equality check, ignoring syntax context (that is, an unhygienic comparison)
221 pub fn token_name_eq(t1 : &Token, t2 : &Token) -> bool {
223 (&token::IDENT(id1,_),&token::IDENT(id2,_))
224 | (&token::LIFETIME(id1),&token::LIFETIME(id2)) =>
225 id1.name == id2.name,
230 pub fn parse<R: Reader>(sess: @ParseSess,
231 cfg: ast::CrateConfig,
235 let mut cur_eis = Vec::new();
236 cur_eis.push(initial_matcher_pos(ms.iter()
237 .map(|x| (*x).clone())
243 let mut bb_eis = Vec::new(); // black-box parsed by parser.rs
244 let mut next_eis = Vec::new(); // or proceed normally
245 let mut eof_eis = Vec::new();
247 let TokenAndSpan {tok: tok, sp: sp} = rdr.peek();
249 /* we append new items to this while we go */
251 let ei = match cur_eis.pop() {
252 None => break, /* for each Earley Item */
257 let len = ei.elts.len();
259 /* at end of sequence */
261 // can't move out of `match`es, so:
263 // hack: a matcher sequence is repeating iff it has a
264 // parent (the top level is just a container)
267 // disregard separator, try to go up
268 // (remove this condition to make trailing seps ok)
270 // pop from the matcher position
272 let mut new_pos = ei.up.clone().unwrap();
274 // update matches (the MBE "parse tree") by appending
275 // each tree as a subtree.
277 // I bet this is a perf problem: we're preemptively
278 // doing a lot of array work that will get thrown away
281 // Only touch the binders we have actually bound
282 for idx in range(ei.match_lo, ei.match_hi) {
283 let sub = (*ei.matches.get(idx)).clone();
286 .push(@MatchedSeq(sub, mk_sp(ei.sp_lo,
291 cur_eis.push(new_pos);
294 // can we go around again?
296 // the *_t vars are workarounds for the lack of unary move
298 Some(ref t) if idx == len => { // we need a separator
299 // i'm conflicted about whether this should be hygienic....
300 // though in this case, if the separators are never legal
301 // idents, it shouldn't matter.
302 if token_name_eq(&tok, t) { //pass the separator
303 let mut ei_t = ei.clone();
308 _ => { // we don't need a separator
318 match ei.elts.get(idx).node.clone() {
319 /* need to descend into sequence */
320 MatchSeq(ref matchers, ref sep, zero_ok,
321 match_idx_lo, match_idx_hi) => {
323 let mut new_ei = ei.clone();
325 //we specifically matched zero repeats.
326 for idx in range(match_idx_lo, match_idx_hi) {
329 .push(@MatchedSeq(Vec::new(), sp));
332 cur_eis.push(new_ei);
335 let matches = Vec::from_elem(ei.matches.len(), Vec::new());
337 cur_eis.push(~MatcherPos {
338 elts: (*matchers).clone(),
343 match_lo: match_idx_lo, match_hi: match_idx_hi,
347 MatchNonterminal(_,_,_) => { bb_eis.push(ei) }
349 let mut ei_t = ei.clone();
350 //if (token_name_eq(t,&tok)) {
351 if token::mtwt_token_eq(t,&tok) {
360 /* error messages here could be improved with links to orig. rules */
361 if token_name_eq(&tok, &EOF) {
362 if eof_eis.len() == 1u {
363 let mut v = Vec::new();
364 for dv in eof_eis.get_mut(0).matches.mut_iter() {
365 v.push(dv.pop().unwrap());
367 return Success(nameize(sess, ms, v.as_slice()));
368 } else if eof_eis.len() > 1u {
369 return Error(sp, ~"ambiguity: multiple successful parses");
371 return Failure(sp, ~"unexpected end of macro invocation");
374 if (bb_eis.len() > 0u && next_eis.len() > 0u)
375 || bb_eis.len() > 1u {
376 let nts = bb_eis.map(|ei| {
377 match ei.elts.get(ei.idx).node {
378 MatchNonterminal(bind, name, _) => {
380 token::get_ident(name),
381 token::get_ident(bind))
384 } }).connect(" or ");
385 return Error(sp, format!(
386 "local ambiguity: multiple parsing options: \
387 built-in NTs {} or {} other options.",
388 nts, next_eis.len()));
389 } else if bb_eis.len() == 0u && next_eis.len() == 0u {
390 return Failure(sp, format!("no rules expected the token `{}`",
391 token::to_str(&tok)));
392 } else if next_eis.len() > 0u {
393 /* Now process the next token */
394 while next_eis.len() > 0u {
395 cur_eis.push(next_eis.pop().unwrap());
398 } else /* bb_eis.len() == 1 */ {
399 let mut rust_parser = Parser(sess, cfg.clone(), rdr.dup());
401 let mut ei = bb_eis.pop().unwrap();
402 match ei.elts.get(ei.idx).node {
403 MatchNonterminal(_, name, idx) => {
404 let name_string = token::get_ident(name);
405 ei.matches.get_mut(idx).push(@MatchedNonterminal(
406 parse_nt(&mut rust_parser, name_string.get())));
413 for _ in range(0, rust_parser.tokens_consumed) {
414 let _ = rdr.next_token();
419 assert!(cur_eis.len() > 0u);
423 pub fn parse_nt(p: &mut Parser, name: &str) -> Nonterminal {
425 "item" => match p.parse_item(Vec::new()) {
426 Some(i) => token::NtItem(i),
427 None => p.fatal("expected an item keyword")
429 "block" => token::NtBlock(p.parse_block()),
430 "stmt" => token::NtStmt(p.parse_stmt(Vec::new())),
431 "pat" => token::NtPat(p.parse_pat()),
432 "expr" => token::NtExpr(p.parse_expr()),
433 "ty" => token::NtTy(p.parse_ty(false /* no need to disambiguate*/)),
434 // this could be handled like a token, since it is one
435 "ident" => match p.token {
436 token::IDENT(sn,b) => { p.bump(); token::NtIdent(~sn,b) }
438 let token_str = token::to_str(&p.token);
439 p.fatal(~"expected ident, found " + token_str)
443 token::NtPath(~p.parse_path(LifetimeAndTypesWithoutColons).path)
445 "attr" => token::NtAttr(@p.parse_attribute(false)),
447 p.quote_depth += 1u; //but in theory, non-quoted tts might be useful
448 let res = token::NtTT(@p.parse_token_tree());
452 "matchers" => token::NtMatchers(p.parse_matchers()),
453 _ => p.fatal(~"unsupported builtin nonterminal parser: " + name)