src/libsyntax/ext/tt/macro_rules.rs

   1 // Copyright 2015 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.
   4 //
   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.
  10
  11 use ast::{self, TokenTree};
  12 use codemap::{Span, DUMMY_SP};
  13 use ext::base::{DummyResult, ExtCtxt, MacResult, SyntaxExtension};
  14 use ext::base::{NormalTT, TTMacroExpander};
  15 use ext::tt::macro_parser::{Success, Error, Failure};
  16 use ext::tt::macro_parser::{MatchedSeq, MatchedNonterminal};
  17 use ext::tt::macro_parser::parse;
  18 use parse::lexer::new_tt_reader;
  19 use parse::parser::{Parser, Restrictions};
  20 use parse::token::{self, special_idents, gensym_ident, NtTT, Token};
  21 use parse::token::Token::*;
  22 use print;
  23 use ptr::P;
  24
  25 use util::small_vector::SmallVector;
  26
  27 use std::cell::RefCell;
  28 use std::collections::{HashMap};
  29 use std::collections::hash_map::{Entry};
  30 use std::rc::Rc;
  31
  32 struct ParserAnyMacro<'a> {
  33     parser: RefCell<Parser<'a>>,
  34
  35     /// Span of the expansion site of the macro this parser is for
  36     site_span: Span,
  37     /// The ident of the macro we're parsing
  38     macro_ident: ast::Ident
  39 }
  40
  41 impl<'a> ParserAnyMacro<'a> {
  42     /// Make sure we don't have any tokens left to parse, so we don't
  43     /// silently drop anything. `allow_semi` is so that "optional"
  44     /// semicolons at the end of normal expressions aren't complained
  45     /// about e.g. the semicolon in `macro_rules! kapow { () => {
  46     /// panic!(); } }` doesn't get picked up by .parse_expr(), but it's
  47     /// allowed to be there.
  48     fn ensure_complete_parse(&self, allow_semi: bool, context: &str) {
  49         let mut parser = self.parser.borrow_mut();
  50         if allow_semi && parser.token == token::Semi {
  51             parser.bump();
  52         }
  53         if parser.token != token::Eof {
  54             let token_str = parser.this_token_to_string();
  55             let msg = format!("macro expansion ignores token `{}` and any \
  56                                following",
  57                               token_str);
  58             let span = parser.span;
  59             let mut err = parser.diagnostic().struct_span_err(span, &msg[..]);
  60             let msg = format!("caused by the macro expansion here; the usage \
  61                                of `{}!` is likely invalid in {} context",
  62                                self.macro_ident, context);
  63             err.span_note(self.site_span, &msg[..])
  64                .emit();
  65         }
  66     }
  67 }
  68
  69 impl<'a> MacResult for ParserAnyMacro<'a> {
  70     fn make_expr(self: Box<ParserAnyMacro<'a>>) -> Option<P<ast::Expr>> {
  71         let ret = panictry!(self.parser.borrow_mut().parse_expr());
  72         self.ensure_complete_parse(true, "expression");
  73         Some(ret)
  74     }
  75     fn make_pat(self: Box<ParserAnyMacro<'a>>) -> Option<P<ast::Pat>> {
  76         let ret = panictry!(self.parser.borrow_mut().parse_pat());
  77         self.ensure_complete_parse(false, "pattern");
  78         Some(ret)
  79     }
  80     fn make_items(self: Box<ParserAnyMacro<'a>>) -> Option<SmallVector<P<ast::Item>>> {
  81         let mut ret = SmallVector::zero();
  82         while let Some(item) = panictry!(self.parser.borrow_mut().parse_item()) {
  83             ret.push(item);
  84         }
  85         self.ensure_complete_parse(false, "item");
  86         Some(ret)
  87     }
  88
  89     fn make_impl_items(self: Box<ParserAnyMacro<'a>>)
  90                        -> Option<SmallVector<ast::ImplItem>> {
  91         let mut ret = SmallVector::zero();
  92         loop {
  93             let mut parser = self.parser.borrow_mut();
  94             match parser.token {
  95                 token::Eof => break,
  96                 _ => ret.push(panictry!(parser.parse_impl_item()))
  97             }
  98         }
  99         self.ensure_complete_parse(false, "item");
 100         Some(ret)
 101     }
 102
 103     fn make_stmts(self: Box<ParserAnyMacro<'a>>)
 104                  -> Option<SmallVector<ast::Stmt>> {
 105         let mut ret = SmallVector::zero();
 106         loop {
 107             let mut parser = self.parser.borrow_mut();
 108             match parser.token {
 109                 token::Eof => break,
 110                 _ => match parser.parse_stmt() {
 111                     Ok(maybe_stmt) => match maybe_stmt {
 112                         Some(stmt) => ret.push(stmt),
 113                         None => (),
 114                     },
 115                     Err(mut e) => {
 116                         e.emit();
 117                         break;
 118                     }
 119                 }
 120             }
 121         }
 122         self.ensure_complete_parse(false, "statement");
 123         Some(ret)
 124     }
 125
 126     fn make_ty(self: Box<ParserAnyMacro<'a>>) -> Option<P<ast::Ty>> {
 127         let ret = panictry!(self.parser.borrow_mut().parse_ty());
 128         self.ensure_complete_parse(false, "type");
 129         Some(ret)
 130     }
 131 }
 132
 133 struct MacroRulesMacroExpander {
 134     name: ast::Ident,
 135     imported_from: Option<ast::Ident>,
 136     lhses: Vec<TokenTree>,
 137     rhses: Vec<TokenTree>,
 138     valid: bool,
 139 }
 140
 141 impl TTMacroExpander for MacroRulesMacroExpander {
 142     fn expand<'cx>(&self,
 143                    cx: &'cx mut ExtCtxt,
 144                    sp: Span,
 145                    arg: &[TokenTree])
 146                    -> Box<MacResult+'cx> {
 147         if !self.valid {
 148             return DummyResult::any(sp);
 149         }
 150         generic_extension(cx,
 151                           sp,
 152                           self.name,
 153                           self.imported_from,
 154                           arg,
 155                           &self.lhses,
 156                           &self.rhses)
 157     }
 158 }
 159
 160 /// Given `lhses` and `rhses`, this is the new macro we create
 161 fn generic_extension<'cx>(cx: &'cx ExtCtxt,
 162                           sp: Span,
 163                           name: ast::Ident,
 164                           imported_from: Option<ast::Ident>,
 165                           arg: &[TokenTree],
 166                           lhses: &[TokenTree],
 167                           rhses: &[TokenTree])
 168                           -> Box<MacResult+'cx> {
 169     if cx.trace_macros() {
 170         println!("{}! {{ {} }}",
 171                  name,
 172                  print::pprust::tts_to_string(arg));
 173     }
 174
 175     // Which arm's failure should we report? (the one furthest along)
 176     let mut best_fail_spot = DUMMY_SP;
 177     let mut best_fail_msg = "internal error: ran no matchers".to_string();
 178
 179     for (i, lhs) in lhses.iter().enumerate() { // try each arm's matchers
 180         let lhs_tt = match *lhs {
 181             TokenTree::Delimited(_, ref delim) => &delim.tts[..],
 182             _ => cx.span_fatal(sp, "malformed macro lhs")
 183         };
 184
 185         match TokenTree::parse(cx, lhs_tt, arg) {
 186             Success(named_matches) => {
 187                 let rhs = match rhses[i] {
 188                     // ignore delimiters
 189                     TokenTree::Delimited(_, ref delimed) => delimed.tts.clone(),
 190                     _ => cx.span_fatal(sp, "malformed macro rhs"),
 191                 };
 192                 // rhs has holes ( `$id` and `$(...)` that need filled)
 193                 let trncbr = new_tt_reader(&cx.parse_sess().span_diagnostic,
 194                                            Some(named_matches),
 195                                            imported_from,
 196                                            rhs);
 197                 let mut p = Parser::new(cx.parse_sess(), cx.cfg(), Box::new(trncbr));
 198                 p.filename = cx.filename.clone();
 199                 p.mod_path_stack = cx.mod_path_stack.clone();
 200                 p.restrictions = match cx.in_block {
 201                     true => Restrictions::NO_NONINLINE_MOD,
 202                     false => Restrictions::empty(),
 203                 };
 204                 p.check_unknown_macro_variable();
 205                 // Let the context choose how to interpret the result.
 206                 // Weird, but useful for X-macros.
 207                 return Box::new(ParserAnyMacro {
 208                     parser: RefCell::new(p),
 209
 210                     // Pass along the original expansion site and the name of the macro
 211                     // so we can print a useful error message if the parse of the expanded
 212                     // macro leaves unparsed tokens.
 213                     site_span: sp,
 214                     macro_ident: name
 215                 })
 216             }
 217             Failure(sp, ref msg) => if sp.lo >= best_fail_spot.lo {
 218                 best_fail_spot = sp;
 219                 best_fail_msg = (*msg).clone();
 220             },
 221             Error(err_sp, ref msg) => {
 222                 cx.span_fatal(err_sp.substitute_dummy(sp), &msg[..])
 223             }
 224         }
 225     }
 226
 227      cx.span_fatal(best_fail_spot.substitute_dummy(sp), &best_fail_msg[..]);
 228 }
 229
 230 // Note that macro-by-example's input is also matched against a token tree:
 231 //                   $( $lhs:tt => $rhs:tt );+
 232 //
 233 // Holy self-referential!
 234
 235 /// Converts a `macro_rules!` invocation into a syntax extension.
 236 pub fn compile<'cx>(cx: &'cx mut ExtCtxt,
 237                     def: &ast::MacroDef) -> SyntaxExtension {
 238
 239     let lhs_nm =  gensym_ident("lhs");
 240     let rhs_nm =  gensym_ident("rhs");
 241
 242     // The pattern that macro_rules matches.
 243     // The grammar for macro_rules! is:
 244     // $( $lhs:tt => $rhs:tt );+
 245     // ...quasiquoting this would be nice.
 246     // These spans won't matter, anyways
 247     let match_lhs_tok = MatchNt(lhs_nm, special_idents::tt, token::Plain, token::Plain);
 248     let match_rhs_tok = MatchNt(rhs_nm, special_idents::tt, token::Plain, token::Plain);
 249     let argument_gram = vec!(
 250         TokenTree::Sequence(DUMMY_SP,
 251                    Rc::new(ast::SequenceRepetition {
 252                        tts: vec![
 253                            TokenTree::Token(DUMMY_SP, match_lhs_tok),
 254                            TokenTree::Token(DUMMY_SP, token::FatArrow),
 255                            TokenTree::Token(DUMMY_SP, match_rhs_tok)],
 256                        separator: Some(token::Semi),
 257                        op: ast::KleeneOp::OneOrMore,
 258                        num_captures: 2
 259                    })),
 260         //to phase into semicolon-termination instead of
 261         //semicolon-separation
 262         TokenTree::Sequence(DUMMY_SP,
 263                    Rc::new(ast::SequenceRepetition {
 264                        tts: vec![TokenTree::Token(DUMMY_SP, token::Semi)],
 265                        separator: None,
 266                        op: ast::KleeneOp::ZeroOrMore,
 267                        num_captures: 0
 268                    })));
 269
 270
 271     // Parse the macro_rules! invocation (`none` is for no interpolations):
 272     let arg_reader = new_tt_reader(&cx.parse_sess().span_diagnostic,
 273                                    None,
 274                                    None,
 275                                    def.body.clone());
 276
 277     let argument_map = match parse(cx.parse_sess(),
 278                                    cx.cfg(),
 279                                    arg_reader,
 280                                    &argument_gram) {
 281         Success(m) => m,
 282         Failure(sp, str) | Error(sp, str) => {
 283             panic!(cx.parse_sess().span_diagnostic
 284                      .span_fatal(sp.substitute_dummy(def.span), &str[..]));
 285         }
 286     };
 287
 288     let mut valid = true;
 289
 290     // Extract the arguments:
 291     let lhses = match **argument_map.get(&lhs_nm.name).unwrap() {
 292         MatchedSeq(ref s, _) => {
 293             s.iter().map(|m| match **m {
 294                 MatchedNonterminal(NtTT(ref tt)) => (**tt).clone(),
 295                 _ => cx.span_bug(def.span, "wrong-structured lhs")
 296             }).collect()
 297         }
 298         _ => cx.span_bug(def.span, "wrong-structured lhs")
 299     };
 300
 301     for lhs in &lhses {
 302         check_lhs_nt_follows(cx, lhs, def.span);
 303     }
 304
 305     let rhses = match **argument_map.get(&rhs_nm.name).unwrap() {
 306         MatchedSeq(ref s, _) => {
 307             s.iter().map(|m| match **m {
 308                 MatchedNonterminal(NtTT(ref tt)) => (**tt).clone(),
 309                 _ => cx.span_bug(def.span, "wrong-structured rhs")
 310             }).collect()
 311         }
 312         _ => cx.span_bug(def.span, "wrong-structured rhs")
 313     };
 314
 315     for rhs in &rhses {
 316         valid &= check_rhs(cx, rhs);
 317     }
 318
 319     let exp: Box<_> = Box::new(MacroRulesMacroExpander {
 320         name: def.ident,
 321         imported_from: def.imported_from,
 322         lhses: lhses,
 323         rhses: rhses,
 324         valid: valid,
 325     });
 326
 327     NormalTT(exp, Some(def.span), def.allow_internal_unstable)
 328 }
 329
 330 // why is this here? because of https://github.com/rust-lang/rust/issues/27774
 331 fn ref_slice<A>(s: &A) -> &[A] { use std::slice::from_raw_parts; unsafe { from_raw_parts(s, 1) } }
 332
 333 fn check_lhs_nt_follows(cx: &mut ExtCtxt, lhs: &TokenTree, sp: Span) {
 334     // lhs is going to be like TokenTree::Delimited(...), where the
 335     // entire lhs is those tts. Or, it can be a "bare sequence", not wrapped in parens.
 336     match lhs {
 337         &TokenTree::Delimited(_, ref tts) => {
 338             check_matcher(cx, &tts.tts);
 339         },
 340         tt @ &TokenTree::Sequence(..) => {
 341             check_matcher(cx, ref_slice(tt));
 342         },
 343         _ => cx.span_err(sp, "invalid macro matcher; matchers must be contained \
 344                               in balanced delimiters or a repetition indicator")
 345     };
 346     // we don't abort on errors on rejection, the driver will do that for us
 347     // after parsing/expansion. we can report every error in every macro this way.
 348 }
 349
 350 fn check_rhs(cx: &mut ExtCtxt, rhs: &TokenTree) -> bool {
 351     match *rhs {
 352         TokenTree::Delimited(..) => return true,
 353         _ => cx.span_err(rhs.get_span(), "macro rhs must be delimited")
 354     }
 355     false
 356 }
 357
 358 // Issue 30450: when we are through a warning cycle, we can just error
 359 // on all failure conditions and remove this struct and enum.
 360
 361 #[derive(Debug)]
 362 struct OnFail {
 363     saw_failure: bool,
 364     action: OnFailAction,
 365 }
 366
 367 #[derive(Copy, Clone, Debug)]
 368 enum OnFailAction { Warn, Error, DoNothing }
 369
 370 impl OnFail {
 371     fn warn() -> OnFail { OnFail { saw_failure: false, action: OnFailAction::Warn } }
 372     fn error() -> OnFail { OnFail { saw_failure: false, action: OnFailAction::Error } }
 373     fn do_nothing() -> OnFail { OnFail { saw_failure: false, action: OnFailAction::DoNothing } }
 374     fn react(&mut self, cx: &mut ExtCtxt, sp: Span, msg: &str) {
 375         match self.action {
 376             OnFailAction::DoNothing => {}
 377             OnFailAction::Error => cx.span_err(sp, msg),
 378             OnFailAction::Warn => {
 379                 cx.struct_span_warn(sp, msg)
 380                     .span_note(sp, "The above warning will be a hard error in the next release.")
 381                     .emit();
 382             }
 383         };
 384         self.saw_failure = true;
 385     }
 386 }
 387
 388 fn check_matcher(cx: &mut ExtCtxt, matcher: &[TokenTree]) {
 389     // Issue 30450: when we are through a warning cycle, we can just
 390     // error on all failure conditions (and remove check_matcher_old).
 391
 392     // First run the old-pass, but *only* to find out if it would have failed.
 393     let mut on_fail = OnFail::do_nothing();
 394     check_matcher_old(cx, matcher.iter(), &Eof, &mut on_fail);
 395     // Then run the new pass, but merely warn if the old pass accepts and new pass rejects.
 396     // (Note this silently accepts code if new pass accepts.)
 397     let mut on_fail = if on_fail.saw_failure {
 398         OnFail::error()
 399     } else {
 400         OnFail::warn()
 401     };
 402     check_matcher_new(cx, matcher, &mut on_fail);
 403 }
 404
 405 // returns the last token that was checked, for TokenTree::Sequence.
 406 // return value is used by recursive calls.
 407 fn check_matcher_old<'a, I>(cx: &mut ExtCtxt, matcher: I, follow: &Token, on_fail: &mut OnFail)
 408 -> Option<(Span, Token)> where I: Iterator<Item=&'a TokenTree> {
 409     use print::pprust::token_to_string;
 410     use std::iter::once;
 411
 412     let mut last = None;
 413
 414     // 2. For each token T in M:
 415     let mut tokens = matcher.peekable();
 416     while let Some(token) = tokens.next() {
 417         last = match *token {
 418             TokenTree::Token(sp, MatchNt(ref name, ref frag_spec, _, _)) => {
 419                 // ii. If T is a simple NT, look ahead to the next token T' in
 420                 // M. If T' is in the set FOLLOW(NT), continue. Else; reject.
 421                 if can_be_followed_by_any(&frag_spec.name.as_str()) {
 422                     continue
 423                 } else {
 424                     let next_token = match tokens.peek() {
 425                         // If T' closes a complex NT, replace T' with F
 426                         Some(&&TokenTree::Token(_, CloseDelim(_))) => follow.clone(),
 427                         Some(&&TokenTree::Token(_, ref tok)) => tok.clone(),
 428                         Some(&&TokenTree::Sequence(sp, _)) => {
 429                             // Be conservative around sequences: to be
 430                             // more specific, we would need to
 431                             // consider FIRST sets, but also the
 432                             // possibility that the sequence occurred
 433                             // zero times (in which case we need to
 434                             // look at the token that follows the
 435                             // sequence, which may itself be a sequence,
 436                             // and so on).
 437                             on_fail.react(cx, sp,
 438                                         &format!("`${0}:{1}` is followed by a \
 439                                                   sequence repetition, which is not \
 440                                                   allowed for `{1}` fragments",
 441                                                  name, frag_spec)
 442                                         );
 443                             Eof
 444                         },
 445                         // die next iteration
 446                         Some(&&TokenTree::Delimited(_, ref delim)) => delim.close_token(),
 447                         // else, we're at the end of the macro or sequence
 448                         None => follow.clone()
 449                     };
 450
 451                     let tok = if let TokenTree::Token(_, ref tok) = *token {
 452                         tok
 453                     } else {
 454                         unreachable!()
 455                     };
 456
 457                     // If T' is in the set FOLLOW(NT), continue. Else, reject.
 458                     match (&next_token, is_in_follow(cx, &next_token, &frag_spec.name.as_str())) {
 459                         (_, Err(msg)) => {
 460                             on_fail.react(cx, sp, &msg);
 461                             continue
 462                         }
 463                         (&Eof, _) => return Some((sp, tok.clone())),
 464                         (_, Ok(true)) => continue,
 465                         (next, Ok(false)) => {
 466                             on_fail.react(cx, sp, &format!("`${0}:{1}` is followed by `{2}`, which \
 467                                                       is not allowed for `{1}` fragments",
 468                                                      name, frag_spec,
 469                                                      token_to_string(next)));
 470                             continue
 471                         },
 472                     }
 473                 }
 474             },
 475             TokenTree::Sequence(sp, ref seq) => {
 476                 // iii. Else, T is a complex NT.
 477                 match seq.separator {
 478                     // If T has the form $(...)U+ or $(...)U* for some token U,
 479                     // run the algorithm on the contents with F set to U. If it
 480                     // accepts, continue, else, reject.
 481                     Some(ref u) => {
 482                         let last = check_matcher_old(cx, seq.tts.iter(), u, on_fail);
 483                         match last {
 484                             // Since the delimiter isn't required after the last
 485                             // repetition, make sure that the *next* token is
 486                             // sane. This doesn't actually compute the FIRST of
 487                             // the rest of the matcher yet, it only considers
 488                             // single tokens and simple NTs. This is imprecise,
 489                             // but conservatively correct.
 490                             Some((span, tok)) => {
 491                                 let fol = match tokens.peek() {
 492                                     Some(&&TokenTree::Token(_, ref tok)) => tok.clone(),
 493                                     Some(&&TokenTree::Delimited(_, ref delim)) =>
 494                                         delim.close_token(),
 495                                     Some(_) => {
 496                                         on_fail.react(cx, sp, "sequence repetition followed by \
 497                                                 another sequence repetition, which is not allowed");
 498                                         Eof
 499                                     },
 500                                     None => Eof
 501                                 };
 502                                 check_matcher_old(cx, once(&TokenTree::Token(span, tok.clone())),
 503                                                   &fol, on_fail)
 504                             },
 505                             None => last,
 506                         }
 507                     },
 508                     // If T has the form $(...)+ or $(...)*, run the algorithm
 509                     // on the contents with F set to the token following the
 510                     // sequence. If it accepts, continue, else, reject.
 511                     None => {
 512                         let fol = match tokens.peek() {
 513                             Some(&&TokenTree::Token(_, ref tok)) => tok.clone(),
 514                             Some(&&TokenTree::Delimited(_, ref delim)) => delim.close_token(),
 515                             Some(_) => {
 516                                 on_fail.react(cx, sp, "sequence repetition followed by another \
 517                                              sequence repetition, which is not allowed");
 518                                 Eof
 519                             },
 520                             None => Eof
 521                         };
 522                         check_matcher_old(cx, seq.tts.iter(), &fol, on_fail)
 523                     }
 524                 }
 525             },
 526             TokenTree::Token(..) => {
 527                 // i. If T is not an NT, continue.
 528                 continue
 529             },
 530             TokenTree::Delimited(_, ref tts) => {
 531                 // if we don't pass in that close delimiter, we'll incorrectly consider the matcher
 532                 // `{ $foo:ty }` as having a follow that isn't `RBrace`
 533                 check_matcher_old(cx, tts.tts.iter(), &tts.close_token(), on_fail)
 534             }
 535         }
 536     }
 537     last
 538 }
 539
 540 fn check_matcher_new(cx: &mut ExtCtxt, matcher: &[TokenTree], on_fail: &mut OnFail) {
 541     let first_sets = FirstSets::new(matcher);
 542     let empty_suffix = TokenSet::empty();
 543     check_matcher_core(cx, &first_sets, matcher, &empty_suffix, on_fail);
 544 }
 545
 546 // The FirstSets for a matcher is a mapping from subsequences in the
 547 // matcher to the FIRST set for that subsequence.
 548 //
 549 // This mapping is partially precomputed via a backwards scan over the
 550 // token trees of the matcher, which provides a mapping from each
 551 // repetition sequence to its FIRST set.
 552 //
 553 // (Hypothetically sequences should be uniquely identifiable via their
 554 // spans, though perhaps that is false e.g. for macro-generated macros
 555 // that do not try to inject artificial span information. My plan is
 556 // to try to catch such cases ahead of time and not include them in
 557 // the precomputed mapping.)
 558 struct FirstSets {
 559     // this maps each TokenTree::Sequence `$(tt ...) SEP OP` that is uniquely identified by its
 560     // span in the original matcher to the First set for the inner sequence `tt ...`.
 561     //
 562     // If two sequences have the same span in a matcher, then map that
 563     // span to None (invalidating the mapping here and forcing the code to
 564     // use a slow path).
 565     first: HashMap<Span, Option<TokenSet>>,
 566 }
 567
 568 impl FirstSets {
 569     fn new(tts: &[TokenTree]) -> FirstSets {
 570         let mut sets = FirstSets { first: HashMap::new() };
 571         build_recur(&mut sets, tts);
 572         return sets;
 573
 574         // walks backward over `tts`, returning the FIRST for `tts`
 575         // and updating `sets` at the same time for all sequence
 576         // substructure we find within `tts`.
 577         fn build_recur(sets: &mut FirstSets, tts: &[TokenTree]) -> TokenSet {
 578             let mut first = TokenSet::empty();
 579             for tt in tts.iter().rev() {
 580                 match *tt {
 581                     TokenTree::Token(sp, ref tok) => {
 582                         first.replace_with((sp, tok.clone()));
 583                     }
 584                     TokenTree::Delimited(_, ref delimited) => {
 585                         build_recur(sets, &delimited.tts[..]);
 586                         first.replace_with((delimited.open_span,
 587                                             Token::OpenDelim(delimited.delim)));
 588                     }
 589                     TokenTree::Sequence(sp, ref seq_rep) => {
 590                         let subfirst = build_recur(sets, &seq_rep.tts[..]);
 591
 592                         match sets.first.entry(sp) {
 593                             Entry::Vacant(vac) => {
 594                                 vac.insert(Some(subfirst.clone()));
 595                             }
 596                             Entry::Occupied(mut occ) => {
 597                                 // if there is already an entry, then a span must have collided.
 598                                 // This should not happen with typical macro_rules macros,
 599                                 // but syntax extensions need not maintain distinct spans,
 600                                 // so distinct syntax trees can be assigned the same span.
 601                                 // In such a case, the map cannot be trusted; so mark this
 602                                 // entry as unusable.
 603                                 occ.insert(None);
 604                             }
 605                         }
 606
 607                         // If the sequence contents can be empty, then the first
 608                         // token could be the separator token itself.
 609
 610                         if let (Some(ref sep), true) = (seq_rep.separator.clone(),
 611                                                         subfirst.maybe_empty) {
 612                             first.add_one_maybe((sp, sep.clone()));
 613                         }
 614
 615                         // Reverse scan: Sequence comes before `first`.
 616                         if subfirst.maybe_empty || seq_rep.op == ast::KleeneOp::ZeroOrMore {
 617                             // If sequence is potentially empty, then
 618                             // union them (preserving first emptiness).
 619                             first.add_all(&TokenSet { maybe_empty: true, ..subfirst });
 620                         } else {
 621                             // Otherwise, sequence guaranteed
 622                             // non-empty; replace first.
 623                             first = subfirst;
 624                         }
 625                     }
 626                 }
 627             }
 628
 629             return first;
 630         }
 631     }
 632
 633     // walks forward over `tts` until all potential FIRST tokens are
 634     // identified.
 635     fn first(&self, tts: &[TokenTree]) -> TokenSet {
 636         let mut first = TokenSet::empty();
 637         for tt in tts.iter() {
 638             assert!(first.maybe_empty);
 639             match *tt {
 640                 TokenTree::Token(sp, ref tok) => {
 641                     first.add_one((sp, tok.clone()));
 642                     return first;
 643                 }
 644                 TokenTree::Delimited(_, ref delimited) => {
 645                     first.add_one((delimited.open_span,
 646                                    Token::OpenDelim(delimited.delim)));
 647                     return first;
 648                 }
 649                 TokenTree::Sequence(sp, ref seq_rep) => {
 650                     match self.first.get(&sp) {
 651                         Some(&Some(ref subfirst)) => {
 652
 653                             // If the sequence contents can be empty, then the first
 654                             // token could be the separator token itself.
 655
 656                             if let (Some(ref sep), true) = (seq_rep.separator.clone(),
 657                                                             subfirst.maybe_empty) {
 658                                 first.add_one_maybe((sp, sep.clone()));
 659                             }
 660
 661                             assert!(first.maybe_empty);
 662                             first.add_all(subfirst);
 663                             if subfirst.maybe_empty || seq_rep.op == ast::KleeneOp::ZeroOrMore {
 664                                 // continue scanning for more first
 665                                 // tokens, but also make sure we
 666                                 // restore empty-tracking state
 667                                 first.maybe_empty = true;
 668                                 continue;
 669                             } else {
 670                                 return first;
 671                             }
 672                         }
 673
 674                         Some(&None) => {
 675                             panic!("assume all sequences have (unique) spans for now");
 676                         }
 677
 678                         None => {
 679                             panic!("We missed a sequence during FirstSets construction");
 680                         }
 681                     }
 682                 }
 683             }
 684         }
 685
 686         // we only exit the loop if `tts` was empty or if every
 687         // element of `tts` matches the empty sequence.
 688         assert!(first.maybe_empty);
 689         return first;
 690     }
 691 }
 692
 693 // A set of Tokens, which may include MatchNt tokens (for
 694 // macro-by-example syntactic variables). It also carries the
 695 // `maybe_empty` flag; that is true if and only if the matcher can
 696 // match an empty token sequence.
 697 //
 698 // The First set is computed on submatchers like `$($a:expr b),* $(c)* d`,
 699 // which has corresponding FIRST = {$a:expr, c, d}.
 700 // Likewise, `$($a:expr b),* $(c)+ d` has FIRST = {$a:expr, c}.
 701 //
 702 // (Notably, we must allow for *-op to occur zero times.)
 703 #[derive(Clone, Debug)]
 704 struct TokenSet {
 705     tokens: Vec<(Span, Token)>,
 706     maybe_empty: bool,
 707 }
 708
 709 impl TokenSet {
 710     // Returns a set for the empty sequence.
 711     fn empty() -> Self { TokenSet { tokens: Vec::new(), maybe_empty: true } }
 712
 713     // Returns the set `{ tok }` for the single-token (and thus
 714     // non-empty) sequence [tok].
 715     fn singleton(tok: (Span, Token)) -> Self {
 716         TokenSet { tokens: vec![tok], maybe_empty: false }
 717     }
 718
 719     // Changes self to be the set `{ tok }`.
 720     // Since `tok` is always present, marks self as non-empty.
 721     fn replace_with(&mut self, tok: (Span, Token)) {
 722         self.tokens.clear();
 723         self.tokens.push(tok);
 724         self.maybe_empty = false;
 725     }
 726
 727     // Changes self to be the empty set `{}`; meant for use when
 728     // the particular token does not matter, but we want to
 729     // record that it occurs.
 730     fn replace_with_irrelevant(&mut self) {
 731         self.tokens.clear();
 732         self.maybe_empty = false;
 733     }
 734
 735     // Adds `tok` to the set for `self`, marking sequence as non-empy.
 736     fn add_one(&mut self, tok: (Span, Token)) {
 737         if !self.tokens.contains(&tok) {
 738             self.tokens.push(tok);
 739         }
 740         self.maybe_empty = false;
 741     }
 742
 743     // Adds `tok` to the set for `self`. (Leaves `maybe_empty` flag alone.)
 744     fn add_one_maybe(&mut self, tok: (Span, Token)) {
 745         if !self.tokens.contains(&tok) {
 746             self.tokens.push(tok);
 747         }
 748     }
 749
 750     // Adds all elements of `other` to this.
 751     //
 752     // (Since this is a set, we filter out duplicates.)
 753     //
 754     // If `other` is potentially empty, then preserves the previous
 755     // setting of the empty flag of `self`. If `other` is guaranteed
 756     // non-empty, then `self` is marked non-empty.
 757     fn add_all(&mut self, other: &Self) {
 758         for tok in &other.tokens {
 759             if !self.tokens.contains(tok) {
 760                 self.tokens.push(tok.clone());
 761             }
 762         }
 763         if !other.maybe_empty {
 764             self.maybe_empty = false;
 765         }
 766     }
 767 }
 768
 769 // Checks that `matcher` is internally consistent and that it
 770 // can legally by followed by a token N, for all N in `follow`.
 771 // (If `follow` is empty, then it imposes no constraint on
 772 // the `matcher`.)
 773 //
 774 // Returns the set of NT tokens that could possibly come last in
 775 // `matcher`. (If `matcher` matches the empty sequence, then
 776 // `maybe_empty` will be set to true.)
 777 //
 778 // Requires that `first_sets` is pre-computed for `matcher`;
 779 // see `FirstSets::new`.
 780 fn check_matcher_core(cx: &mut ExtCtxt,
 781                       first_sets: &FirstSets,
 782                       matcher: &[TokenTree],
 783                       follow: &TokenSet,
 784                       on_fail: &mut OnFail) -> TokenSet {
 785     use print::pprust::token_to_string;
 786
 787     let mut last = TokenSet::empty();
 788
 789     // 2. For each token and suffix  [T, SUFFIX] in M:
 790     // ensure that T can be followed by SUFFIX, and if SUFFIX may be empty,
 791     // then ensure T can also be followed by any element of FOLLOW.
 792     'each_token: for i in 0..matcher.len() {
 793         let token = &matcher[i];
 794         let suffix = &matcher[i+1..];
 795
 796         let build_suffix_first = || {
 797             let mut s = first_sets.first(suffix);
 798             if s.maybe_empty { s.add_all(follow); }
 799             return s;
 800         };
 801
 802         // (we build `suffix_first` on demand below; you can tell
 803         // which cases are supposed to fall through by looking for the
 804         // initialization of this variable.)
 805         let suffix_first;
 806
 807         // First, update `last` so that it corresponds to the set
 808         // of NT tokens that might end the sequence `... token`.
 809         match *token {
 810             TokenTree::Token(sp, ref tok) => {
 811                 let can_be_followed_by_any;
 812                 if let Err(bad_frag) = has_legal_fragment_specifier(tok) {
 813                     on_fail.react(cx, sp, &format!("invalid fragment specifier `{}`", bad_frag));
 814                     // (This eliminates false positives and duplicates
 815                     // from error messages.)
 816                     can_be_followed_by_any = true;
 817                 } else {
 818                     can_be_followed_by_any = token_can_be_followed_by_any(tok);
 819                 }
 820
 821                 if can_be_followed_by_any {
 822                     // don't need to track tokens that work with any,
 823                     last.replace_with_irrelevant();
 824                     // ... and don't need to check tokens that can be
 825                     // followed by anything against SUFFIX.
 826                     continue 'each_token;
 827                 } else {
 828                     last.replace_with((sp, tok.clone()));
 829                     suffix_first = build_suffix_first();
 830                 }
 831             }
 832             TokenTree::Delimited(_, ref d) => {
 833                 let my_suffix = TokenSet::singleton((d.close_span, Token::CloseDelim(d.delim)));
 834                 check_matcher_core(cx, first_sets, &d.tts, &my_suffix, on_fail);
 835                 // don't track non NT tokens
 836                 last.replace_with_irrelevant();
 837
 838                 // also, we don't need to check delimited sequences
 839                 // against SUFFIX
 840                 continue 'each_token;
 841             }
 842             TokenTree::Sequence(sp, ref seq_rep) => {
 843                 suffix_first = build_suffix_first();
 844                 // The trick here: when we check the interior, we want
 845                 // to include the separator (if any) as a potential
 846                 // (but not guaranteed) element of FOLLOW. So in that
 847                 // case, we make a temp copy of suffix and stuff
 848                 // delimiter in there.
 849                 //
 850                 // FIXME: Should I first scan suffix_first to see if
 851                 // delimiter is already in it before I go through the
 852                 // work of cloning it? But then again, this way I may
 853                 // get a "tighter" span?
 854                 let mut new;
 855                 let my_suffix = if let Some(ref u) = seq_rep.separator {
 856                     new = suffix_first.clone();
 857                     new.add_one_maybe((sp, u.clone()));
 858                     &new
 859                 } else {
 860                     &suffix_first
 861                 };
 862
 863                 // At this point, `suffix_first` is built, and
 864                 // `my_suffix` is some TokenSet that we can use
 865                 // for checking the interior of `seq_rep`.
 866                 let next = check_matcher_core(cx, first_sets, &seq_rep.tts, my_suffix, on_fail);
 867                 if next.maybe_empty {
 868                     last.add_all(&next);
 869                 } else {
 870                     last = next;
 871                 }
 872
 873                 // the recursive call to check_matcher_core already ran the 'each_last
 874                 // check below, so we can just keep going forward here.
 875                 continue 'each_token;
 876             }
 877         }
 878
 879         // (`suffix_first` guaranteed initialized once reaching here.)
 880
 881         // Now `last` holds the complete set of NT tokens that could
 882         // end the sequence before SUFFIX. Check that every one works with `suffix`.
 883         'each_last: for &(_sp, ref t) in &last.tokens {
 884             if let MatchNt(ref name, ref frag_spec, _, _) = *t {
 885                 for &(sp, ref next_token) in &suffix_first.tokens {
 886                     match is_in_follow(cx, next_token, &frag_spec.name.as_str()) {
 887                         Err(msg) => {
 888                             on_fail.react(cx, sp, &msg);
 889                             // don't bother reporting every source of
 890                             // conflict for a particular element of `last`.
 891                             continue 'each_last;
 892                         }
 893                         Ok(true) => {}
 894                         Ok(false) => {
 895                             let may_be = if last.tokens.len() == 1 &&
 896                                 suffix_first.tokens.len() == 1
 897                             {
 898                                 "is"
 899                             } else {
 900                                 "may be"
 901                             };
 902
 903                             on_fail.react(
 904                                 cx, sp,
 905                                 &format!("`${name}:{frag}` {may_be} followed by `{next}`, which \
 906                                           is not allowed for `{frag}` fragments",
 907                                          name=name,
 908                                          frag=frag_spec,
 909                                          next=token_to_string(next_token),
 910                                          may_be=may_be));
 911                         }
 912                     }
 913                 }
 914             }
 915         }
 916     }
 917     last
 918 }
 919
 920
 921 fn token_can_be_followed_by_any(tok: &Token) -> bool {
 922     if let &MatchNt(_, ref frag_spec, _, _) = tok {
 923         frag_can_be_followed_by_any(&frag_spec.name.as_str())
 924     } else {
 925         // (Non NT's can always be followed by anthing in matchers.)
 926         true
 927     }
 928 }
 929
 930 /// True if a fragment of type `frag` can be followed by any sort of
 931 /// token.  We use this (among other things) as a useful approximation
 932 /// for when `frag` can be followed by a repetition like `$(...)*` or
 933 /// `$(...)+`. In general, these can be a bit tricky to reason about,
 934 /// so we adopt a conservative position that says that any fragment
 935 /// specifier which consumes at most one token tree can be followed by
 936 /// a fragment specifier (indeed, these fragments can be followed by
 937 /// ANYTHING without fear of future compatibility hazards).
 938 fn frag_can_be_followed_by_any(frag: &str) -> bool {
 939     match frag {
 940         "item" |  // always terminated by `}` or `;`
 941         "block" | // exactly one token tree
 942         "ident" | // exactly one token tree
 943         "meta" |  // exactly one token tree
 944         "tt" =>    // exactly one token tree
 945             true,
 946
 947         _ =>
 948             false,
 949     }
 950 }
 951
 952 /// True if a fragment of type `frag` can be followed by any sort of
 953 /// token.  We use this (among other things) as a useful approximation
 954 /// for when `frag` can be followed by a repetition like `$(...)*` or
 955 /// `$(...)+`. In general, these can be a bit tricky to reason about,
 956 /// so we adopt a conservative position that says that any fragment
 957 /// specifier which consumes at most one token tree can be followed by
 958 /// a fragment specifier (indeed, these fragments can be followed by
 959 /// ANYTHING without fear of future compatibility hazards).
 960 fn can_be_followed_by_any(frag: &str) -> bool {
 961     match frag {
 962         "item" |  // always terminated by `}` or `;`
 963         "block" | // exactly one token tree
 964         "ident" | // exactly one token tree
 965         "meta" |  // exactly one token tree
 966         "tt" =>    // exactly one token tree
 967             true,
 968
 969         _ =>
 970             false,
 971     }
 972 }
 973
 974 /// True if `frag` can legally be followed by the token `tok`. For
 975 /// fragments that can consume an unbounded number of tokens, `tok`
 976 /// must be within a well-defined follow set. This is intended to
 977 /// guarantee future compatibility: for example, without this rule, if
 978 /// we expanded `expr` to include a new binary operator, we might
 979 /// break macros that were relying on that binary operator as a
 980 /// separator.
 981 // when changing this do not forget to update doc/book/macros.md!
 982 fn is_in_follow(_: &ExtCtxt, tok: &Token, frag: &str) -> Result<bool, String> {
 983     if let &CloseDelim(_) = tok {
 984         // closing a token tree can never be matched by any fragment;
 985         // iow, we always require that `(` and `)` match, etc.
 986         Ok(true)
 987     } else {
 988         match frag {
 989             "item" => {
 990                 // since items *must* be followed by either a `;` or a `}`, we can
 991                 // accept anything after them
 992                 Ok(true)
 993             },
 994             "block" => {
 995                 // anything can follow block, the braces provide an easy boundary to
 996                 // maintain
 997                 Ok(true)
 998             },
 999             "stmt" | "expr"  => {
1000                 match *tok {
1001                     FatArrow | Comma | Semi => Ok(true),
1002                     _ => Ok(false)
1003                 }
1004             },
1005             "pat" => {
1006                 match *tok {
1007                     FatArrow | Comma | Eq | BinOp(token::Or) => Ok(true),
1008                     Ident(i, _) if (i.name.as_str() == "if" ||
1009                                     i.name.as_str() == "in") => Ok(true),
1010                     _ => Ok(false)
1011                 }
1012             },
1013             "path" | "ty" => {
1014                 match *tok {
1015                     OpenDelim(token::DelimToken::Brace) | OpenDelim(token::DelimToken::Bracket) |
1016                     Comma | FatArrow | Colon | Eq | Gt | Semi | BinOp(token::Or) => Ok(true),
1017                     Ident(i, _) if (i.name.as_str() == "as" ||
1018                                     i.name.as_str() == "where") => Ok(true),
1019                     _ => Ok(false)
1020                 }
1021             },
1022             "ident" => {
1023                 // being a single token, idents are harmless
1024                 Ok(true)
1025             },
1026             "meta" | "tt" => {
1027                 // being either a single token or a delimited sequence, tt is
1028                 // harmless
1029                 Ok(true)
1030             },
1031             _ => Err(format!("invalid fragment specifier `{}`", frag))
1032         }
1033     }
1034 }
1035
1036 fn has_legal_fragment_specifier(tok: &Token) -> Result<(), String> {
1037     debug!("has_legal_fragment_specifier({:?})", tok);
1038     if let &MatchNt(_, ref frag_spec, _, _) = tok {
1039         let s = &frag_spec.name.as_str();
1040         if !is_legal_fragment_specifier(s) {
1041             return Err(s.to_string());
1042         }
1043     }
1044     Ok(())
1045 }
1046
1047 fn is_legal_fragment_specifier(frag: &str) -> bool {
1048     match frag {
1049         "item" | "block" | "stmt" | "expr" | "pat" |
1050         "path" | "ty" | "ident" | "meta" | "tt" => true,
1051         _ => false,
1052     }
1053 }