]> git.proxmox.com Git - rustc.git/blame - compiler/rustc_expand/src/mbe/macro_rules.rs
New upstream version 1.63.0+dfsg1
[rustc.git] / compiler / rustc_expand / src / mbe / macro_rules.rs
CommitLineData
ba9703b0 1use crate::base::{DummyResult, ExtCtxt, MacResult, TTMacroExpander};
e74abb32 2use crate::base::{SyntaxExtension, SyntaxExtensionKind};
dfeec247 3use crate::expand::{ensure_complete_parse, parse_ast_fragment, AstFragment, AstFragmentKind};
e74abb32
XL
4use crate::mbe;
5use crate::mbe::macro_check;
5e7ed085 6use crate::mbe::macro_parser::{Error, ErrorReported, Failure, Success, TtParser};
04454e1e 7use crate::mbe::macro_parser::{MatchedSeq, MatchedTokenTree, MatcherLoc};
e74abb32
XL
8use crate::mbe::transcribe::transcribe;
9
3dfed10e 10use rustc_ast as ast;
04454e1e 11use rustc_ast::token::{self, Delimiter, NonterminalKind, Token, TokenKind, TokenKind::*};
74b04a01 12use rustc_ast::tokenstream::{DelimSpan, TokenStream};
136023e0 13use rustc_ast::{NodeId, DUMMY_NODE_ID};
74b04a01
XL
14use rustc_ast_pretty::pprust;
15use rustc_attr::{self as attr, TransparencyError};
dfeec247 16use rustc_data_structures::fx::FxHashMap;
5e7ed085 17use rustc_errors::{Applicability, Diagnostic, DiagnosticBuilder};
60c5eb7d 18use rustc_feature::Features;
136023e0
XL
19use rustc_lint_defs::builtin::{
20 RUST_2021_INCOMPATIBLE_OR_PATTERNS, SEMICOLON_IN_EXPRESSIONS_FROM_MACROS,
21};
cdc7bbd5 22use rustc_lint_defs::BuiltinLintDiagnostics;
60c5eb7d 23use rustc_parse::parser::Parser;
74b04a01 24use rustc_session::parse::ParseSess;
3dfed10e 25use rustc_session::Session;
dfeec247
XL
26use rustc_span::edition::Edition;
27use rustc_span::hygiene::Transparency;
3dfed10e 28use rustc_span::symbol::{kw, sym, Ident, MacroRulesNormalizedIdent};
dfeec247 29use rustc_span::Span;
9fa01778 30
94b46f34 31use std::borrow::Cow;
7cac9316 32use std::collections::hash_map::Entry;
60c5eb7d 33use std::{mem, slice};
3dfed10e 34use tracing::debug;
1a4d82fc 35
923072b8 36pub(crate) struct ParserAnyMacro<'a> {
9e0c209e 37 parser: Parser<'a>,
9346a6ac
AL
38
39 /// Span of the expansion site of the macro this parser is for
40 site_span: Span,
41 /// The ident of the macro we're parsing
f9f354fc 42 macro_ident: Ident,
5869c6ff 43 lint_node_id: NodeId,
94222f64 44 is_trailing_mac: bool,
a1dfa0c6 45 arm_span: Span,
94222f64
XL
46 /// Whether or not this macro is defined in the current crate
47 is_local: bool,
1a4d82fc
JJ
48}
49
923072b8 50pub(crate) fn annotate_err_with_kind(err: &mut Diagnostic, kind: AstFragmentKind, span: Span) {
416331ca
XL
51 match kind {
52 AstFragmentKind::Ty => {
53 err.span_label(span, "this macro call doesn't expand to a type");
54 }
55 AstFragmentKind::Pat => {
56 err.span_label(span, "this macro call doesn't expand to a pattern");
57 }
58 _ => {}
59 };
60}
61
ba9703b0 62fn emit_frag_parse_err(
5e7ed085 63 mut e: DiagnosticBuilder<'_, rustc_errors::ErrorGuaranteed>,
ba9703b0
XL
64 parser: &Parser<'_>,
65 orig_parser: &mut Parser<'_>,
66 site_span: Span,
ba9703b0
XL
67 arm_span: Span,
68 kind: AstFragmentKind,
69) {
04454e1e
FG
70 // FIXME(davidtwco): avoid depending on the error message text
71 if parser.token == token::Eof && e.message[0].0.expect_str().ends_with(", found `<eof>`") {
ba9703b0
XL
72 if !e.span.is_dummy() {
73 // early end of macro arm (#52866)
74 e.replace_span_with(parser.sess.source_map().next_point(parser.token.span));
75 }
76 let msg = &e.message[0];
77 e.message[0] = (
04454e1e 78 rustc_errors::DiagnosticMessage::Str(format!(
ba9703b0 79 "macro expansion ends with an incomplete expression: {}",
04454e1e
FG
80 msg.0.expect_str().replace(", found `<eof>`", ""),
81 )),
ba9703b0
XL
82 msg.1,
83 );
84 }
85 if e.span.is_dummy() {
86 // Get around lack of span in error (#30128)
87 e.replace_span_with(site_span);
88 if !parser.sess.source_map().is_imported(arm_span) {
89 e.span_label(arm_span, "in this macro arm");
90 }
91 } else if parser.sess.source_map().is_imported(parser.token.span) {
92 e.span_label(site_span, "in this macro invocation");
93 }
94 match kind {
ba9703b0
XL
95 // Try a statement if an expression is wanted but failed and suggest adding `;` to call.
96 AstFragmentKind::Expr => match parse_ast_fragment(orig_parser, AstFragmentKind::Stmts) {
5e7ed085 97 Err(err) => err.cancel(),
ba9703b0
XL
98 Ok(_) => {
99 e.note(
100 "the macro call doesn't expand to an expression, but it can expand to a statement",
101 );
102 e.span_suggestion_verbose(
103 site_span.shrink_to_hi(),
104 "add `;` to interpret the expansion as a statement",
923072b8 105 ";",
ba9703b0
XL
106 Applicability::MaybeIncorrect,
107 );
108 }
109 },
110 _ => annotate_err_with_kind(&mut e, kind, site_span),
111 };
112 e.emit();
113}
114
1a4d82fc 115impl<'a> ParserAnyMacro<'a> {
923072b8 116 pub(crate) fn make(mut self: Box<ParserAnyMacro<'a>>, kind: AstFragmentKind) -> AstFragment {
94222f64
XL
117 let ParserAnyMacro {
118 site_span,
119 macro_ident,
120 ref mut parser,
121 lint_node_id,
122 arm_span,
123 is_trailing_mac,
124 is_local,
125 } = *self;
923072b8 126 let snapshot = &mut parser.create_snapshot_for_diagnostic();
ba9703b0
XL
127 let fragment = match parse_ast_fragment(parser, kind) {
128 Ok(f) => f,
129 Err(err) => {
5869c6ff 130 emit_frag_parse_err(err, parser, snapshot, site_span, arm_span, kind);
ba9703b0 131 return kind.dummy(site_span);
a1dfa0c6 132 }
ba9703b0 133 };
9e0c209e 134
0731742a 135 // We allow semicolons at the end of expressions -- e.g., the semicolon in
9e0c209e 136 // `macro_rules! m { () => { panic!(); } }` isn't parsed by `.parse_expr()`,
0731742a 137 // but `m!()` is allowed in expression positions (cf. issue #34706).
8faf50e0 138 if kind == AstFragmentKind::Expr && parser.token == token::Semi {
94222f64
XL
139 if is_local {
140 parser.sess.buffer_lint_with_diagnostic(
141 SEMICOLON_IN_EXPRESSIONS_FROM_MACROS,
142 parser.token.span,
143 lint_node_id,
144 "trailing semicolon in macro used in expression position",
145 BuiltinLintDiagnostics::TrailingMacro(is_trailing_mac, macro_ident),
146 );
147 }
9cc50fc6 148 parser.bump();
1a4d82fc 149 }
e9174d1e 150
9e0c209e 151 // Make sure we don't have any tokens left to parse so we don't silently drop anything.
83c7162d 152 let path = ast::Path::from_ident(macro_ident.with_span_pos(site_span));
e74abb32 153 ensure_complete_parse(parser, &path, kind.name(), site_span);
8faf50e0 154 fragment
e9174d1e 155 }
1a4d82fc
JJ
156}
157
158struct MacroRulesMacroExpander {
04454e1e 159 node_id: NodeId,
f9f354fc 160 name: Ident,
416331ca 161 span: Span,
e1599b0c 162 transparency: Transparency,
04454e1e 163 lhses: Vec<Vec<MatcherLoc>>,
e74abb32 164 rhses: Vec<mbe::TokenTree>,
92a42be0 165 valid: bool,
1a4d82fc
JJ
166}
167
168impl TTMacroExpander for MacroRulesMacroExpander {
a1dfa0c6
XL
169 fn expand<'cx>(
170 &self,
9fa01778 171 cx: &'cx mut ExtCtxt<'_>,
a1dfa0c6
XL
172 sp: Span,
173 input: TokenStream,
dc9dc135 174 ) -> Box<dyn MacResult + 'cx> {
92a42be0
SL
175 if !self.valid {
176 return DummyResult::any(sp);
177 }
04454e1e 178 expand_macro(
dfeec247
XL
179 cx,
180 sp,
181 self.span,
04454e1e 182 self.node_id,
dfeec247
XL
183 self.name,
184 self.transparency,
185 input,
186 &self.lhses,
187 &self.rhses,
e1599b0c 188 )
1a4d82fc
JJ
189 }
190}
191
ba9703b0
XL
192fn macro_rules_dummy_expander<'cx>(
193 _: &'cx mut ExtCtxt<'_>,
194 span: Span,
195 _: TokenStream,
196) -> Box<dyn MacResult + 'cx> {
197 DummyResult::any(span)
198}
199
74b04a01 200fn trace_macros_note(cx_expansions: &mut FxHashMap<Span, Vec<String>>, sp: Span, message: String) {
5869c6ff 201 let sp = sp.macro_backtrace().last().map_or(sp, |trace| trace.call_site);
74b04a01 202 cx_expansions.entry(sp).or_default().push(message);
7cac9316
XL
203}
204
04454e1e
FG
205/// Expands the rules based macro defined by `lhses` and `rhses` for a given
206/// input `arg`.
923072b8 207fn expand_macro<'cx>(
dc9dc135
XL
208 cx: &'cx mut ExtCtxt<'_>,
209 sp: Span,
416331ca 210 def_span: Span,
04454e1e 211 node_id: NodeId,
f9f354fc 212 name: Ident,
e1599b0c 213 transparency: Transparency,
dc9dc135 214 arg: TokenStream,
923072b8
FG
215 lhses: &[Vec<MatcherLoc>],
216 rhses: &[mbe::TokenTree],
dc9dc135 217) -> Box<dyn MacResult + 'cx> {
3dfed10e 218 let sess = &cx.sess.parse_sess;
04454e1e
FG
219 // Macros defined in the current crate have a real node id,
220 // whereas macros from an external crate have a dummy id.
221 let is_local = node_id != DUMMY_NODE_ID;
ba9703b0 222
1a4d82fc 223 if cx.trace_macros() {
f035d41b 224 let msg = format!("expanding `{}! {{ {} }}`", name, pprust::tts_to_string(&arg));
74b04a01 225 trace_macros_note(&mut cx.expansions, sp, msg);
1a4d82fc
JJ
226 }
227
228 // Which arm's failure should we report? (the one furthest along)
dc9dc135 229 let mut best_failure: Option<(Token, &str)> = None;
74b04a01
XL
230
231 // We create a base parser that can be used for the "black box" parts.
232 // Every iteration needs a fresh copy of that parser. However, the parser
233 // is not mutated on many of the iterations, particularly when dealing with
234 // macros like this:
235 //
236 // macro_rules! foo {
237 // ("a") => (A);
238 // ("b") => (B);
239 // ("c") => (C);
240 // // ... etc. (maybe hundreds more)
241 // }
242 //
243 // as seen in the `html5ever` benchmark. We use a `Cow` so that the base
244 // parser is only cloned when necessary (upon mutation). Furthermore, we
245 // reinitialize the `Cow` with the base parser at the start of every
246 // iteration, so that any mutated parsers are not reused. This is all quite
247 // hacky, but speeds up the `html5ever` benchmark significantly. (Issue
248 // 68836 suggests a more comprehensive but more complex change to deal with
249 // this situation.)
ba9703b0 250 let parser = parser_from_cx(sess, arg.clone());
74b04a01 251
5e7ed085
FG
252 // Try each arm's matchers.
253 let mut tt_parser = TtParser::new(name);
254 for (i, lhs) in lhses.iter().enumerate() {
60c5eb7d
XL
255 // Take a snapshot of the state of pre-expansion gating at this point.
256 // This is used so that if a matcher is not `Success(..)`ful,
257 // then the spans which became gated when parsing the unsuccessful matcher
258 // are not recorded. On the first `Success(..)`ful matcher, the spans are merged.
ba9703b0 259 let mut gated_spans_snapshot = mem::take(&mut *sess.gated_spans.spans.borrow_mut());
60c5eb7d 260
04454e1e 261 match tt_parser.parse_tt(&mut Cow::Borrowed(&parser), lhs) {
92a42be0 262 Success(named_matches) => {
60c5eb7d
XL
263 // The matcher was `Success(..)`ful.
264 // Merge the gated spans from parsing the matcher with the pre-existing ones.
ba9703b0 265 sess.gated_spans.merge(gated_spans_snapshot);
60c5eb7d 266
04454e1e
FG
267 let (rhs, rhs_span): (&mbe::Delimited, DelimSpan) = match &rhses[i] {
268 mbe::TokenTree::Delimited(span, delimited) => (&delimited, *span),
269 _ => cx.span_bug(sp, "malformed macro rhs"),
270 };
a1dfa0c6 271 let arm_span = rhses[i].span();
3b2f2976 272
04454e1e 273 let rhs_spans = rhs.tts.iter().map(|t| t.span()).collect::<Vec<_>>();
1a4d82fc 274 // rhs has holes ( `$id` and `$(...)` that need filled)
04454e1e 275 let mut tts = match transcribe(cx, &named_matches, &rhs, rhs_span, transparency) {
ba9703b0
XL
276 Ok(tts) => tts,
277 Err(mut err) => {
278 err.emit();
279 return DummyResult::any(arm_span);
280 }
281 };
3b2f2976
XL
282
283 // Replace all the tokens for the corresponding positions in the macro, to maintain
284 // proper positions in error reporting, while maintaining the macro_backtrace.
285 if rhs_spans.len() == tts.len() {
29967ef6
XL
286 tts = tts.map_enumerated(|i, tt| {
287 let mut tt = tt.clone();
3b2f2976 288 let mut sp = rhs_spans[i];
ea8adc8c 289 sp = sp.with_ctxt(tt.span().ctxt());
3b2f2976
XL
290 tt.set_span(sp);
291 tt
292 });
293 }
7cac9316
XL
294
295 if cx.trace_macros() {
f035d41b 296 let msg = format!("to `{}`", pprust::tts_to_string(&tts));
74b04a01 297 trace_macros_note(&mut cx.expansions, sp, msg);
7cac9316
XL
298 }
299
ba9703b0 300 let mut p = Parser::new(sess, tts, false, None);
416331ca 301 p.last_type_ascription = cx.current_expansion.prior_type_ascription;
476ff2be 302
04454e1e
FG
303 if is_local {
304 cx.resolver.record_macro_rule_usage(node_id, i);
305 }
306
1a4d82fc
JJ
307 // Let the context choose how to interpret the result.
308 // Weird, but useful for X-macros.
d9579d0f 309 return Box::new(ParserAnyMacro {
9e0c209e 310 parser: p,
9346a6ac
AL
311
312 // Pass along the original expansion site and the name of the macro
313 // so we can print a useful error message if the parse of the expanded
314 // macro leaves unparsed tokens.
315 site_span: sp,
a1dfa0c6 316 macro_ident: name,
136023e0 317 lint_node_id: cx.current_expansion.lint_node_id,
94222f64 318 is_trailing_mac: cx.current_expansion.is_trailing_mac,
a1dfa0c6 319 arm_span,
94222f64 320 is_local,
dc9dc135 321 });
92a42be0 322 }
dc9dc135
XL
323 Failure(token, msg) => match best_failure {
324 Some((ref best_token, _)) if best_token.span.lo() >= token.span.lo() => {}
325 _ => best_failure = Some((token, msg)),
92a42be0 326 },
ba9703b0
XL
327 Error(err_sp, ref msg) => {
328 let span = err_sp.substitute_dummy(sp);
329 cx.struct_span_err(span, &msg).emit();
330 return DummyResult::any(span);
331 }
332 ErrorReported => return DummyResult::any(sp),
1a4d82fc 333 }
60c5eb7d
XL
334
335 // The matcher was not `Success(..)`ful.
336 // Restore to the state before snapshotting and maybe try again.
ba9703b0 337 mem::swap(&mut gated_spans_snapshot, &mut sess.gated_spans.spans.borrow_mut());
1a4d82fc 338 }
74b04a01 339 drop(parser);
e9174d1e 340
dc9dc135
XL
341 let (token, label) = best_failure.expect("ran no matchers");
342 let span = token.span.substitute_dummy(sp);
343 let mut err = cx.struct_span_err(span, &parse_failure_msg(&token));
344 err.span_label(span, label);
ba9703b0
XL
345 if !def_span.is_dummy() && !cx.source_map().is_imported(def_span) {
346 err.span_label(cx.source_map().guess_head_span(def_span), "when calling this macro");
a1dfa0c6 347 }
b7449926
XL
348
349 // Check whether there's a missing comma in this macro call, like `println!("{}" a);`
350 if let Some((arg, comma_span)) = arg.add_comma() {
dc9dc135 351 for lhs in lhses {
04454e1e
FG
352 let parser = parser_from_cx(sess, arg.clone());
353 if let Success(_) = tt_parser.parse_tt(&mut Cow::Borrowed(&parser), lhs) {
ba9703b0
XL
354 if comma_span.is_dummy() {
355 err.note("you might be missing a comma");
356 } else {
357 err.span_suggestion_short(
358 comma_span,
359 "missing comma here",
923072b8 360 ", ",
ba9703b0
XL
361 Applicability::MachineApplicable,
362 );
b7449926 363 }
b7449926
XL
364 }
365 }
366 }
367 err.emit();
ea8adc8c
XL
368 cx.trace_macros_diag();
369 DummyResult::any(sp)
1a4d82fc
JJ
370}
371
372// Note that macro-by-example's input is also matched against a token tree:
373// $( $lhs:tt => $rhs:tt );+
374//
375// Holy self-referential!
376
e74abb32
XL
377/// Converts a macro item into a syntax extension.
378pub fn compile_declarative_macro(
3dfed10e 379 sess: &Session,
9fa01778
XL
380 features: &Features,
381 def: &ast::Item,
dc9dc135 382 edition: Edition,
923072b8 383) -> (SyntaxExtension, Vec<(usize, Span)>) {
f035d41b 384 debug!("compile_declarative_macro: {:?}", def);
ba9703b0
XL
385 let mk_syn_ext = |expander| {
386 SyntaxExtension::new(
387 sess,
388 SyntaxExtensionKind::LegacyBang(expander),
389 def.span,
390 Vec::new(),
391 edition,
392 def.ident.name,
393 &def.attrs,
394 )
395 };
04454e1e 396 let dummy_syn_ext = || (mk_syn_ext(Box::new(macro_rules_dummy_expander)), Vec::new());
ba9703b0 397
3dfed10e 398 let diag = &sess.parse_sess.span_diagnostic;
f9f354fc
XL
399 let lhs_nm = Ident::new(sym::lhs, def.span);
400 let rhs_nm = Ident::new(sym::rhs, def.span);
b9856134 401 let tt_spec = Some(NonterminalKind::TT);
223e47cc 402
7cac9316 403 // Parse the macro_rules! invocation
ba9703b0
XL
404 let (macro_rules, body) = match &def.kind {
405 ast::ItemKind::MacroDef(def) => (def.macro_rules, def.body.inner_tokens()),
7cac9316
XL
406 _ => unreachable!(),
407 };
408
1a4d82fc 409 // The pattern that macro_rules matches.
223e47cc 410 // The grammar for macro_rules! is:
1a4d82fc 411 // $( $lhs:tt => $rhs:tt );+
223e47cc 412 // ...quasiquoting this would be nice.
1a4d82fc 413 // These spans won't matter, anyways
3157f602 414 let argument_gram = vec![
e74abb32 415 mbe::TokenTree::Sequence(
dc9dc135 416 DelimSpan::dummy(),
04454e1e 417 mbe::SequenceRepetition {
dc9dc135 418 tts: vec![
e74abb32
XL
419 mbe::TokenTree::MetaVarDecl(def.span, lhs_nm, tt_spec),
420 mbe::TokenTree::token(token::FatArrow, def.span),
421 mbe::TokenTree::MetaVarDecl(def.span, rhs_nm, tt_spec),
dc9dc135
XL
422 ],
423 separator: Some(Token::new(
ba9703b0 424 if macro_rules { token::Semi } else { token::Comma },
dc9dc135
XL
425 def.span,
426 )),
e74abb32 427 kleene: mbe::KleeneToken::new(mbe::KleeneOp::OneOrMore, def.span),
dc9dc135 428 num_captures: 2,
04454e1e 429 },
dc9dc135 430 ),
3157f602 431 // to phase into semicolon-termination instead of semicolon-separation
e74abb32 432 mbe::TokenTree::Sequence(
dc9dc135 433 DelimSpan::dummy(),
04454e1e 434 mbe::SequenceRepetition {
e74abb32 435 tts: vec![mbe::TokenTree::token(
ba9703b0 436 if macro_rules { token::Semi } else { token::Comma },
416331ca
XL
437 def.span,
438 )],
dc9dc135 439 separator: None,
e74abb32 440 kleene: mbe::KleeneToken::new(mbe::KleeneOp::ZeroOrMore, def.span),
dc9dc135 441 num_captures: 0,
04454e1e 442 },
dc9dc135 443 ),
3157f602 444 ];
04454e1e
FG
445 // Convert it into `MatcherLoc` form.
446 let argument_gram = mbe::macro_parser::compute_locs(&argument_gram);
223e47cc 447
3dfed10e 448 let parser = Parser::new(&sess.parse_sess, body, true, rustc_parse::MACRO_ARGUMENTS);
04454e1e
FG
449 let mut tt_parser =
450 TtParser::new(Ident::with_dummy_span(if macro_rules { kw::MacroRules } else { kw::Macro }));
5e7ed085 451 let argument_map = match tt_parser.parse_tt(&mut Cow::Borrowed(&parser), &argument_gram) {
e9174d1e 452 Success(m) => m,
dc9dc135
XL
453 Failure(token, msg) => {
454 let s = parse_failure_msg(&token);
455 let sp = token.span.substitute_dummy(def.span);
3dfed10e 456 sess.parse_sess.span_diagnostic.struct_span_err(sp, &s).span_label(sp, msg).emit();
04454e1e 457 return dummy_syn_ext();
c30ab7b3 458 }
ba9703b0 459 Error(sp, msg) => {
3dfed10e
XL
460 sess.parse_sess
461 .span_diagnostic
462 .struct_span_err(sp.substitute_dummy(def.span), &msg)
463 .emit();
04454e1e 464 return dummy_syn_ext();
ba9703b0
XL
465 }
466 ErrorReported => {
04454e1e 467 return dummy_syn_ext();
e9174d1e
SL
468 }
469 };
223e47cc 470
92a42be0
SL
471 let mut valid = true;
472
223e47cc 473 // Extract the arguments:
ba9703b0 474 let lhses = match argument_map[&MacroRulesNormalizedIdent::new(lhs_nm)] {
60c5eb7d 475 MatchedSeq(ref s) => s
dc9dc135
XL
476 .iter()
477 .map(|m| {
5e7ed085 478 if let MatchedTokenTree(ref tt) = *m {
04454e1e 479 let tt = mbe::quoted::parse(
5e7ed085
FG
480 tt.clone().into(),
481 true,
482 &sess.parse_sess,
483 def.id,
484 features,
485 edition,
04454e1e
FG
486 )
487 .pop()
488 .unwrap();
489 valid &= check_lhs_nt_follows(&sess.parse_sess, &def, &tt);
5e7ed085 490 return tt;
3157f602 491 }
3dfed10e 492 sess.parse_sess.span_diagnostic.span_bug(def.span, "wrong-structured lhs")
dc9dc135 493 })
e74abb32 494 .collect::<Vec<mbe::TokenTree>>(),
3dfed10e 495 _ => sess.parse_sess.span_diagnostic.span_bug(def.span, "wrong-structured lhs"),
223e47cc
LB
496 };
497
ba9703b0 498 let rhses = match argument_map[&MacroRulesNormalizedIdent::new(rhs_nm)] {
60c5eb7d 499 MatchedSeq(ref s) => s
dc9dc135
XL
500 .iter()
501 .map(|m| {
5e7ed085 502 if let MatchedTokenTree(ref tt) = *m {
04454e1e 503 return mbe::quoted::parse(
5e7ed085
FG
504 tt.clone().into(),
505 false,
506 &sess.parse_sess,
507 def.id,
508 features,
509 edition,
04454e1e
FG
510 )
511 .pop()
512 .unwrap();
c30ab7b3 513 }
3dfed10e 514 sess.parse_sess.span_diagnostic.span_bug(def.span, "wrong-structured lhs")
dc9dc135 515 })
e74abb32 516 .collect::<Vec<mbe::TokenTree>>(),
3dfed10e 517 _ => sess.parse_sess.span_diagnostic.span_bug(def.span, "wrong-structured rhs"),
223e47cc
LB
518 };
519
92a42be0 520 for rhs in &rhses {
3dfed10e 521 valid &= check_rhs(&sess.parse_sess, rhs);
9e0c209e
SL
522 }
523
524 // don't abort iteration early, so that errors for multiple lhses can be reported
525 for lhs in &lhses {
3dfed10e 526 valid &= check_lhs_no_empty_seq(&sess.parse_sess, slice::from_ref(lhs));
92a42be0
SL
527 }
528
3dfed10e 529 valid &= macro_check::check_meta_variables(&sess.parse_sess, def.id, def.span, &lhses, &rhses);
dc9dc135 530
94222f64 531 let (transparency, transparency_error) = attr::find_transparency(&def.attrs, macro_rules);
416331ca 532 match transparency_error {
dfeec247 533 Some(TransparencyError::UnknownTransparency(value, span)) => {
5e7ed085 534 diag.span_err(span, &format!("unknown macro transparency: `{}`", value));
dfeec247
XL
535 }
536 Some(TransparencyError::MultipleTransparencyAttrs(old_span, new_span)) => {
5e7ed085 537 diag.span_err(vec![old_span, new_span], "multiple macro transparency attributes");
dfeec247 538 }
416331ca
XL
539 None => {}
540 }
dc9dc135 541
923072b8
FG
542 // Compute the spans of the macro rules for unused rule linting.
543 // To avoid warning noise, only consider the rules of this
544 // macro for the lint, if all rules are valid.
545 // Also, we are only interested in non-foreign macros.
546 let rule_spans = if valid && def.id != DUMMY_NODE_ID {
547 lhses
548 .iter()
549 .zip(rhses.iter())
550 .enumerate()
551 // If the rhs contains an invocation like compile_error!,
552 // don't consider the rule for the unused rule lint.
553 .filter(|(_idx, (_lhs, rhs))| !has_compile_error_macro(rhs))
554 // We only take the span of the lhs here,
555 // so that the spans of created warnings are smaller.
556 .map(|(idx, (lhs, _rhs))| (idx, lhs.span()))
557 .collect::<Vec<_>>()
04454e1e
FG
558 } else {
559 Vec::new()
560 };
561
562 // Convert the lhses into `MatcherLoc` form, which is better for doing the
563 // actual matching. Unless the matcher is invalid.
564 let lhses = if valid {
565 lhses
566 .iter()
567 .map(|lhs| {
568 // Ignore the delimiters around the matcher.
569 match lhs {
570 mbe::TokenTree::Delimited(_, delimited) => {
571 mbe::macro_parser::compute_locs(&delimited.tts)
572 }
573 _ => sess.parse_sess.span_diagnostic.span_bug(def.span, "malformed macro lhs"),
574 }
575 })
576 .collect()
577 } else {
578 vec![]
579 };
580
581 let expander = Box::new(MacroRulesMacroExpander {
dfeec247
XL
582 name: def.ident,
583 span: def.span,
04454e1e 584 node_id: def.id,
dfeec247
XL
585 transparency,
586 lhses,
587 rhses,
588 valid,
04454e1e
FG
589 });
590 (mk_syn_ext(expander), rule_spans)
1a4d82fc
JJ
591}
592
04454e1e 593fn check_lhs_nt_follows(sess: &ParseSess, def: &ast::Item, lhs: &mbe::TokenTree) -> bool {
92a42be0
SL
594 // lhs is going to be like TokenTree::Delimited(...), where the
595 // entire lhs is those tts. Or, it can be a "bare sequence", not wrapped in parens.
5e7ed085 596 if let mbe::TokenTree::Delimited(_, delimited) = lhs {
04454e1e 597 check_matcher(sess, def, &delimited.tts)
7cac9316
XL
598 } else {
599 let msg = "invalid macro matcher; matchers must be contained in balanced delimiters";
600 sess.span_diagnostic.span_err(lhs.span(), msg);
601 false
3157f602 602 }
1a4d82fc
JJ
603 // we don't abort on errors on rejection, the driver will do that for us
604 // after parsing/expansion. we can report every error in every macro this way.
605}
606
9fa01778 607/// Checks that the lhs contains no repetition which could match an empty token
9e0c209e 608/// tree, because then the matcher would hang indefinitely.
e74abb32
XL
609fn check_lhs_no_empty_seq(sess: &ParseSess, tts: &[mbe::TokenTree]) -> bool {
610 use mbe::TokenTree;
9e0c209e
SL
611 for tt in tts {
612 match *tt {
5e7ed085
FG
613 TokenTree::Token(..)
614 | TokenTree::MetaVar(..)
615 | TokenTree::MetaVarDecl(..)
616 | TokenTree::MetaVarExpr(..) => (),
dc9dc135 617 TokenTree::Delimited(_, ref del) => {
04454e1e 618 if !check_lhs_no_empty_seq(sess, &del.tts) {
dc9dc135
XL
619 return false;
620 }
621 }
9e0c209e 622 TokenTree::Sequence(span, ref seq) => {
dc9dc135
XL
623 if seq.separator.is_none()
624 && seq.tts.iter().all(|seq_tt| match *seq_tt {
b9856134 625 TokenTree::MetaVarDecl(_, _, Some(NonterminalKind::Vis)) => true,
dc9dc135 626 TokenTree::Sequence(_, ref sub_seq) => {
e74abb32
XL
627 sub_seq.kleene.op == mbe::KleeneOp::ZeroOrMore
628 || sub_seq.kleene.op == mbe::KleeneOp::ZeroOrOne
dc9dc135 629 }
7cac9316 630 _ => false,
dc9dc135
XL
631 })
632 {
b7449926
XL
633 let sp = span.entire();
634 sess.span_diagnostic.span_err(sp, "repetition matches empty token tree");
7cac9316 635 return false;
9e0c209e
SL
636 }
637 if !check_lhs_no_empty_seq(sess, &seq.tts) {
638 return false;
639 }
640 }
641 }
642 }
643
644 true
645}
646
e74abb32 647fn check_rhs(sess: &ParseSess, rhs: &mbe::TokenTree) -> bool {
92a42be0 648 match *rhs {
e74abb32 649 mbe::TokenTree::Delimited(..) => return true,
5e7ed085
FG
650 _ => {
651 sess.span_diagnostic.span_err(rhs.span(), "macro rhs must be delimited");
652 }
92a42be0
SL
653 }
654 false
655}
656
04454e1e 657fn check_matcher(sess: &ParseSess, def: &ast::Item, matcher: &[mbe::TokenTree]) -> bool {
9cc50fc6
SL
658 let first_sets = FirstSets::new(matcher);
659 let empty_suffix = TokenSet::empty();
9e0c209e 660 let err = sess.span_diagnostic.err_count();
04454e1e 661 check_matcher_core(sess, def, &first_sets, matcher, &empty_suffix);
9e0c209e 662 err == sess.span_diagnostic.err_count()
9cc50fc6
SL
663}
664
923072b8
FG
665fn has_compile_error_macro(rhs: &mbe::TokenTree) -> bool {
666 match rhs {
667 mbe::TokenTree::Delimited(_sp, d) => {
668 let has_compile_error = d.tts.array_windows::<3>().any(|[ident, bang, args]| {
669 if let mbe::TokenTree::Token(ident) = ident &&
670 let TokenKind::Ident(ident, _) = ident.kind &&
671 ident == sym::compile_error &&
672 let mbe::TokenTree::Token(bang) = bang &&
673 let TokenKind::Not = bang.kind &&
674 let mbe::TokenTree::Delimited(_, del) = args &&
675 del.delim != Delimiter::Invisible
676 {
677 true
678 } else {
679 false
680 }
681 });
682 if has_compile_error { true } else { d.tts.iter().any(has_compile_error_macro) }
683 }
684 _ => false,
685 }
686}
687
0731742a 688// `The FirstSets` for a matcher is a mapping from subsequences in the
9cc50fc6
SL
689// matcher to the FIRST set for that subsequence.
690//
691// This mapping is partially precomputed via a backwards scan over the
692// token trees of the matcher, which provides a mapping from each
0731742a 693// repetition sequence to its *first* set.
9cc50fc6 694//
0731742a
XL
695// (Hypothetically, sequences should be uniquely identifiable via their
696// spans, though perhaps that is false, e.g., for macro-generated macros
9cc50fc6
SL
697// that do not try to inject artificial span information. My plan is
698// to try to catch such cases ahead of time and not include them in
699// the precomputed mapping.)
04454e1e 700struct FirstSets<'tt> {
9cc50fc6
SL
701 // this maps each TokenTree::Sequence `$(tt ...) SEP OP` that is uniquely identified by its
702 // span in the original matcher to the First set for the inner sequence `tt ...`.
703 //
704 // If two sequences have the same span in a matcher, then map that
705 // span to None (invalidating the mapping here and forcing the code to
706 // use a slow path).
04454e1e 707 first: FxHashMap<Span, Option<TokenSet<'tt>>>,
9cc50fc6
SL
708}
709
04454e1e
FG
710impl<'tt> FirstSets<'tt> {
711 fn new(tts: &'tt [mbe::TokenTree]) -> FirstSets<'tt> {
e74abb32 712 use mbe::TokenTree;
8bb4bdeb 713
b7449926 714 let mut sets = FirstSets { first: FxHashMap::default() };
9cc50fc6
SL
715 build_recur(&mut sets, tts);
716 return sets;
717
718 // walks backward over `tts`, returning the FIRST for `tts`
719 // and updating `sets` at the same time for all sequence
720 // substructure we find within `tts`.
04454e1e 721 fn build_recur<'tt>(sets: &mut FirstSets<'tt>, tts: &'tt [TokenTree]) -> TokenSet<'tt> {
9cc50fc6
SL
722 let mut first = TokenSet::empty();
723 for tt in tts.iter().rev() {
724 match *tt {
5e7ed085
FG
725 TokenTree::Token(..)
726 | TokenTree::MetaVar(..)
727 | TokenTree::MetaVarDecl(..)
728 | TokenTree::MetaVarExpr(..) => {
04454e1e 729 first.replace_with(TtHandle::TtRef(tt));
9cc50fc6 730 }
04454e1e
FG
731 TokenTree::Delimited(span, ref delimited) => {
732 build_recur(sets, &delimited.tts);
733 first.replace_with(TtHandle::from_token_kind(
734 token::OpenDelim(delimited.delim),
735 span.open,
736 ));
9cc50fc6
SL
737 }
738 TokenTree::Sequence(sp, ref seq_rep) => {
a2a8927a 739 let subfirst = build_recur(sets, &seq_rep.tts);
9cc50fc6 740
b7449926 741 match sets.first.entry(sp.entire()) {
9cc50fc6
SL
742 Entry::Vacant(vac) => {
743 vac.insert(Some(subfirst.clone()));
744 }
745 Entry::Occupied(mut occ) => {
746 // if there is already an entry, then a span must have collided.
747 // This should not happen with typical macro_rules macros,
748 // but syntax extensions need not maintain distinct spans,
749 // so distinct syntax trees can be assigned the same span.
750 // In such a case, the map cannot be trusted; so mark this
751 // entry as unusable.
752 occ.insert(None);
753 }
754 }
755
756 // If the sequence contents can be empty, then the first
757 // token could be the separator token itself.
758
dc9dc135 759 if let (Some(sep), true) = (&seq_rep.separator, subfirst.maybe_empty) {
04454e1e 760 first.add_one_maybe(TtHandle::from_token(sep.clone()));
9cc50fc6
SL
761 }
762
763 // Reverse scan: Sequence comes before `first`.
9fa01778 764 if subfirst.maybe_empty
e74abb32
XL
765 || seq_rep.kleene.op == mbe::KleeneOp::ZeroOrMore
766 || seq_rep.kleene.op == mbe::KleeneOp::ZeroOrOne
9fa01778 767 {
9cc50fc6
SL
768 // If sequence is potentially empty, then
769 // union them (preserving first emptiness).
770 first.add_all(&TokenSet { maybe_empty: true, ..subfirst });
771 } else {
772 // Otherwise, sequence guaranteed
773 // non-empty; replace first.
774 first = subfirst;
775 }
776 }
777 }
778 }
779
7cac9316 780 first
9cc50fc6
SL
781 }
782 }
783
784 // walks forward over `tts` until all potential FIRST tokens are
785 // identified.
04454e1e 786 fn first(&self, tts: &'tt [mbe::TokenTree]) -> TokenSet<'tt> {
e74abb32 787 use mbe::TokenTree;
8bb4bdeb 788
9cc50fc6
SL
789 let mut first = TokenSet::empty();
790 for tt in tts.iter() {
791 assert!(first.maybe_empty);
792 match *tt {
5e7ed085
FG
793 TokenTree::Token(..)
794 | TokenTree::MetaVar(..)
795 | TokenTree::MetaVarDecl(..)
796 | TokenTree::MetaVarExpr(..) => {
04454e1e 797 first.add_one(TtHandle::TtRef(tt));
9cc50fc6
SL
798 return first;
799 }
04454e1e
FG
800 TokenTree::Delimited(span, ref delimited) => {
801 first.add_one(TtHandle::from_token_kind(
802 token::OpenDelim(delimited.delim),
803 span.open,
804 ));
9cc50fc6
SL
805 return first;
806 }
807 TokenTree::Sequence(sp, ref seq_rep) => {
416331ca
XL
808 let subfirst_owned;
809 let subfirst = match self.first.get(&sp.entire()) {
810 Some(&Some(ref subfirst)) => subfirst,
9cc50fc6 811 Some(&None) => {
a2a8927a 812 subfirst_owned = self.first(&seq_rep.tts);
416331ca 813 &subfirst_owned
9cc50fc6 814 }
9cc50fc6
SL
815 None => {
816 panic!("We missed a sequence during FirstSets construction");
817 }
416331ca
XL
818 };
819
820 // If the sequence contents can be empty, then the first
821 // token could be the separator token itself.
822 if let (Some(sep), true) = (&seq_rep.separator, subfirst.maybe_empty) {
04454e1e 823 first.add_one_maybe(TtHandle::from_token(sep.clone()));
416331ca
XL
824 }
825
826 assert!(first.maybe_empty);
827 first.add_all(subfirst);
828 if subfirst.maybe_empty
e74abb32
XL
829 || seq_rep.kleene.op == mbe::KleeneOp::ZeroOrMore
830 || seq_rep.kleene.op == mbe::KleeneOp::ZeroOrOne
416331ca
XL
831 {
832 // Continue scanning for more first
833 // tokens, but also make sure we
834 // restore empty-tracking state.
835 first.maybe_empty = true;
836 continue;
837 } else {
838 return first;
9cc50fc6
SL
839 }
840 }
841 }
842 }
843
844 // we only exit the loop if `tts` was empty or if every
845 // element of `tts` matches the empty sequence.
846 assert!(first.maybe_empty);
7cac9316 847 first
9cc50fc6
SL
848 }
849}
850
04454e1e
FG
851// Most `mbe::TokenTree`s are pre-existing in the matcher, but some are defined
852// implicitly, such as opening/closing delimiters and sequence repetition ops.
853// This type encapsulates both kinds. It implements `Clone` while avoiding the
854// need for `mbe::TokenTree` to implement `Clone`.
855#[derive(Debug)]
856enum TtHandle<'tt> {
857 /// This is used in most cases.
858 TtRef(&'tt mbe::TokenTree),
859
860 /// This is only used for implicit token trees. The `mbe::TokenTree` *must*
861 /// be `mbe::TokenTree::Token`. No other variants are allowed. We store an
862 /// `mbe::TokenTree` rather than a `Token` so that `get()` can return a
863 /// `&mbe::TokenTree`.
864 Token(mbe::TokenTree),
865}
866
867impl<'tt> TtHandle<'tt> {
868 fn from_token(tok: Token) -> Self {
869 TtHandle::Token(mbe::TokenTree::Token(tok))
870 }
871
872 fn from_token_kind(kind: TokenKind, span: Span) -> Self {
873 TtHandle::from_token(Token::new(kind, span))
874 }
875
876 // Get a reference to a token tree.
877 fn get(&'tt self) -> &'tt mbe::TokenTree {
878 match self {
879 TtHandle::TtRef(tt) => tt,
880 TtHandle::Token(token_tt) => &token_tt,
881 }
882 }
883}
884
885impl<'tt> PartialEq for TtHandle<'tt> {
886 fn eq(&self, other: &TtHandle<'tt>) -> bool {
887 self.get() == other.get()
888 }
889}
890
891impl<'tt> Clone for TtHandle<'tt> {
892 fn clone(&self) -> Self {
893 match self {
894 TtHandle::TtRef(tt) => TtHandle::TtRef(tt),
895
896 // This variant *must* contain a `mbe::TokenTree::Token`, and not
897 // any other variant of `mbe::TokenTree`.
898 TtHandle::Token(mbe::TokenTree::Token(tok)) => {
899 TtHandle::Token(mbe::TokenTree::Token(tok.clone()))
900 }
901
902 _ => unreachable!(),
903 }
904 }
905}
906
e74abb32 907// A set of `mbe::TokenTree`s, which may include `TokenTree::Match`s
8bb4bdeb 908// (for macro-by-example syntactic variables). It also carries the
9cc50fc6
SL
909// `maybe_empty` flag; that is true if and only if the matcher can
910// match an empty token sequence.
911//
912// The First set is computed on submatchers like `$($a:expr b),* $(c)* d`,
913// which has corresponding FIRST = {$a:expr, c, d}.
914// Likewise, `$($a:expr b),* $(c)+ d` has FIRST = {$a:expr, c}.
915//
916// (Notably, we must allow for *-op to occur zero times.)
917#[derive(Clone, Debug)]
04454e1e
FG
918struct TokenSet<'tt> {
919 tokens: Vec<TtHandle<'tt>>,
9cc50fc6
SL
920 maybe_empty: bool,
921}
922
04454e1e 923impl<'tt> TokenSet<'tt> {
9cc50fc6 924 // Returns a set for the empty sequence.
dc9dc135
XL
925 fn empty() -> Self {
926 TokenSet { tokens: Vec::new(), maybe_empty: true }
927 }
9cc50fc6
SL
928
929 // Returns the set `{ tok }` for the single-token (and thus
930 // non-empty) sequence [tok].
04454e1e
FG
931 fn singleton(tt: TtHandle<'tt>) -> Self {
932 TokenSet { tokens: vec![tt], maybe_empty: false }
9cc50fc6
SL
933 }
934
935 // Changes self to be the set `{ tok }`.
936 // Since `tok` is always present, marks self as non-empty.
04454e1e 937 fn replace_with(&mut self, tt: TtHandle<'tt>) {
9cc50fc6 938 self.tokens.clear();
04454e1e 939 self.tokens.push(tt);
9cc50fc6
SL
940 self.maybe_empty = false;
941 }
942
943 // Changes self to be the empty set `{}`; meant for use when
944 // the particular token does not matter, but we want to
945 // record that it occurs.
946 fn replace_with_irrelevant(&mut self) {
947 self.tokens.clear();
948 self.maybe_empty = false;
949 }
950
951 // Adds `tok` to the set for `self`, marking sequence as non-empy.
04454e1e
FG
952 fn add_one(&mut self, tt: TtHandle<'tt>) {
953 if !self.tokens.contains(&tt) {
954 self.tokens.push(tt);
9cc50fc6
SL
955 }
956 self.maybe_empty = false;
957 }
958
959 // Adds `tok` to the set for `self`. (Leaves `maybe_empty` flag alone.)
04454e1e
FG
960 fn add_one_maybe(&mut self, tt: TtHandle<'tt>) {
961 if !self.tokens.contains(&tt) {
962 self.tokens.push(tt);
9cc50fc6
SL
963 }
964 }
965
966 // Adds all elements of `other` to this.
967 //
968 // (Since this is a set, we filter out duplicates.)
969 //
970 // If `other` is potentially empty, then preserves the previous
971 // setting of the empty flag of `self`. If `other` is guaranteed
972 // non-empty, then `self` is marked non-empty.
973 fn add_all(&mut self, other: &Self) {
04454e1e
FG
974 for tt in &other.tokens {
975 if !self.tokens.contains(tt) {
976 self.tokens.push(tt.clone());
9cc50fc6
SL
977 }
978 }
979 if !other.maybe_empty {
980 self.maybe_empty = false;
981 }
982 }
983}
984
985// Checks that `matcher` is internally consistent and that it
416331ca 986// can legally be followed by a token `N`, for all `N` in `follow`.
9cc50fc6
SL
987// (If `follow` is empty, then it imposes no constraint on
988// the `matcher`.)
989//
990// Returns the set of NT tokens that could possibly come last in
991// `matcher`. (If `matcher` matches the empty sequence, then
992// `maybe_empty` will be set to true.)
993//
994// Requires that `first_sets` is pre-computed for `matcher`;
995// see `FirstSets::new`.
04454e1e 996fn check_matcher_core<'tt>(
dc9dc135 997 sess: &ParseSess,
136023e0 998 def: &ast::Item,
04454e1e
FG
999 first_sets: &FirstSets<'tt>,
1000 matcher: &'tt [mbe::TokenTree],
1001 follow: &TokenSet<'tt>,
1002) -> TokenSet<'tt> {
e74abb32 1003 use mbe::TokenTree;
9cc50fc6
SL
1004
1005 let mut last = TokenSet::empty();
1006
1007 // 2. For each token and suffix [T, SUFFIX] in M:
1008 // ensure that T can be followed by SUFFIX, and if SUFFIX may be empty,
1009 // then ensure T can also be followed by any element of FOLLOW.
1010 'each_token: for i in 0..matcher.len() {
1011 let token = &matcher[i];
dc9dc135 1012 let suffix = &matcher[i + 1..];
9cc50fc6
SL
1013
1014 let build_suffix_first = || {
1015 let mut s = first_sets.first(suffix);
dc9dc135
XL
1016 if s.maybe_empty {
1017 s.add_all(follow);
1018 }
7cac9316 1019 s
9cc50fc6
SL
1020 };
1021
1022 // (we build `suffix_first` on demand below; you can tell
1023 // which cases are supposed to fall through by looking for the
1024 // initialization of this variable.)
1025 let suffix_first;
1026
1027 // First, update `last` so that it corresponds to the set
1028 // of NT tokens that might end the sequence `... token`.
1029 match *token {
5e7ed085
FG
1030 TokenTree::Token(..)
1031 | TokenTree::MetaVar(..)
1032 | TokenTree::MetaVarDecl(..)
1033 | TokenTree::MetaVarExpr(..) => {
3dfed10e 1034 if token_can_be_followed_by_any(token) {
9cc50fc6
SL
1035 // don't need to track tokens that work with any,
1036 last.replace_with_irrelevant();
1037 // ... and don't need to check tokens that can be
1038 // followed by anything against SUFFIX.
1039 continue 'each_token;
1040 } else {
04454e1e 1041 last.replace_with(TtHandle::TtRef(token));
9cc50fc6
SL
1042 suffix_first = build_suffix_first();
1043 }
1044 }
04454e1e
FG
1045 TokenTree::Delimited(span, ref d) => {
1046 let my_suffix = TokenSet::singleton(TtHandle::from_token_kind(
1047 token::CloseDelim(d.delim),
1048 span.close,
1049 ));
1050 check_matcher_core(sess, def, first_sets, &d.tts, &my_suffix);
9cc50fc6
SL
1051 // don't track non NT tokens
1052 last.replace_with_irrelevant();
1053
1054 // also, we don't need to check delimited sequences
1055 // against SUFFIX
1056 continue 'each_token;
1057 }
dc9dc135 1058 TokenTree::Sequence(_, ref seq_rep) => {
9cc50fc6
SL
1059 suffix_first = build_suffix_first();
1060 // The trick here: when we check the interior, we want
1061 // to include the separator (if any) as a potential
1062 // (but not guaranteed) element of FOLLOW. So in that
1063 // case, we make a temp copy of suffix and stuff
1064 // delimiter in there.
1065 //
1066 // FIXME: Should I first scan suffix_first to see if
1067 // delimiter is already in it before I go through the
1068 // work of cloning it? But then again, this way I may
1069 // get a "tighter" span?
1070 let mut new;
dc9dc135 1071 let my_suffix = if let Some(sep) = &seq_rep.separator {
9cc50fc6 1072 new = suffix_first.clone();
04454e1e 1073 new.add_one_maybe(TtHandle::from_token(sep.clone()));
9cc50fc6
SL
1074 &new
1075 } else {
1076 &suffix_first
1077 };
1078
1079 // At this point, `suffix_first` is built, and
1080 // `my_suffix` is some TokenSet that we can use
1081 // for checking the interior of `seq_rep`.
04454e1e 1082 let next = check_matcher_core(sess, def, first_sets, &seq_rep.tts, my_suffix);
9cc50fc6
SL
1083 if next.maybe_empty {
1084 last.add_all(&next);
1085 } else {
1086 last = next;
1087 }
1088
1089 // the recursive call to check_matcher_core already ran the 'each_last
1090 // check below, so we can just keep going forward here.
1091 continue 'each_token;
1092 }
1093 }
1094
1095 // (`suffix_first` guaranteed initialized once reaching here.)
1096
1097 // Now `last` holds the complete set of NT tokens that could
1098 // end the sequence before SUFFIX. Check that every one works with `suffix`.
04454e1e
FG
1099 for tt in &last.tokens {
1100 if let &TokenTree::MetaVarDecl(span, name, Some(kind)) = tt.get() {
8bb4bdeb 1101 for next_token in &suffix_first.tokens {
04454e1e
FG
1102 let next_token = next_token.get();
1103
136023e0
XL
1104 // Check if the old pat is used and the next token is `|`
1105 // to warn about incompatibility with Rust 2021.
1106 // We only emit this lint if we're parsing the original
1107 // definition of this macro_rules, not while (re)parsing
1108 // the macro when compiling another crate that is using the
1109 // macro. (See #86567.)
1110 // Macros defined in the current crate have a real node id,
1111 // whereas macros from an external crate have a dummy id.
1112 if def.id != DUMMY_NODE_ID
1113 && matches!(kind, NonterminalKind::PatParam { inferred: true })
1114 && matches!(next_token, TokenTree::Token(token) if token.kind == BinOp(token::BinOpToken::Or))
1115 {
1116 // It is suggestion to use pat_param, for example: $x:pat -> $x:pat_param.
1117 let suggestion = quoted_tt_to_string(&TokenTree::MetaVarDecl(
1118 span,
1119 name,
1120 Some(NonterminalKind::PatParam { inferred: false }),
1121 ));
1122 sess.buffer_lint_with_diagnostic(
1123 &RUST_2021_INCOMPATIBLE_OR_PATTERNS,
1124 span,
1125 ast::CRATE_NODE_ID,
1126 "the meaning of the `pat` fragment specifier is changing in Rust 2021, which may affect this macro",
1127 BuiltinLintDiagnostics::OrPatternsBackCompat(span, suggestion),
1128 );
cdc7bbd5 1129 }
3dfed10e 1130 match is_in_follow(next_token, kind) {
a1dfa0c6 1131 IsInFollow::Yes => {}
dc9dc135
XL
1132 IsInFollow::No(possible) => {
1133 let may_be = if last.tokens.len() == 1 && suffix_first.tokens.len() == 1
9cc50fc6
SL
1134 {
1135 "is"
1136 } else {
1137 "may be"
1138 };
1139
a1dfa0c6
XL
1140 let sp = next_token.span();
1141 let mut err = sess.span_diagnostic.struct_span_err(
1142 sp,
dc9dc135
XL
1143 &format!(
1144 "`${name}:{frag}` {may_be} followed by `{next}`, which \
1145 is not allowed for `{frag}` fragments",
1146 name = name,
3dfed10e 1147 frag = kind,
dc9dc135
XL
1148 next = quoted_tt_to_string(next_token),
1149 may_be = may_be
1150 ),
3157f602 1151 );
3dfed10e 1152 err.span_label(sp, format!("not allowed after `{}` fragments", kind));
a2a8927a
XL
1153
1154 if kind == NonterminalKind::PatWithOr
04454e1e 1155 && sess.edition.rust_2021()
a2a8927a
XL
1156 && next_token.is_token(&BinOp(token::BinOpToken::Or))
1157 {
1158 let suggestion = quoted_tt_to_string(&TokenTree::MetaVarDecl(
1159 span,
1160 name,
1161 Some(NonterminalKind::PatParam { inferred: false }),
1162 ));
1163 err.span_suggestion(
1164 span,
5099ac24 1165 "try a `pat_param` fragment specifier instead",
a2a8927a
XL
1166 suggestion,
1167 Applicability::MaybeIncorrect,
1168 );
1169 }
1170
a1dfa0c6 1171 let msg = "allowed there are: ";
dc9dc135 1172 match possible {
a1dfa0c6
XL
1173 &[] => {}
1174 &[t] => {
1175 err.note(&format!(
1176 "only {} is allowed after `{}` fragments",
3dfed10e 1177 t, kind,
a1dfa0c6
XL
1178 ));
1179 }
1180 ts => {
1181 err.note(&format!(
1182 "{}{} or {}",
1183 msg,
dc9dc135
XL
1184 ts[..ts.len() - 1]
1185 .iter()
74b04a01 1186 .copied()
dc9dc135
XL
1187 .collect::<Vec<_>>()
1188 .join(", "),
a1dfa0c6
XL
1189 ts[ts.len() - 1],
1190 ));
1191 }
1192 }
1193 err.emit();
9cc50fc6
SL
1194 }
1195 }
1196 }
1197 }
1198 }
1199 }
1200 last
1201}
1202
e74abb32 1203fn token_can_be_followed_by_any(tok: &mbe::TokenTree) -> bool {
b9856134 1204 if let mbe::TokenTree::MetaVarDecl(_, _, Some(kind)) = *tok {
3dfed10e 1205 frag_can_be_followed_by_any(kind)
9cc50fc6 1206 } else {
74b04a01 1207 // (Non NT's can always be followed by anything in matchers.)
9cc50fc6
SL
1208 true
1209 }
1210}
1211
9fa01778
XL
1212/// Returns `true` if a fragment of type `frag` can be followed by any sort of
1213/// token. We use this (among other things) as a useful approximation
9cc50fc6
SL
1214/// for when `frag` can be followed by a repetition like `$(...)*` or
1215/// `$(...)+`. In general, these can be a bit tricky to reason about,
1216/// so we adopt a conservative position that says that any fragment
1217/// specifier which consumes at most one token tree can be followed by
1218/// a fragment specifier (indeed, these fragments can be followed by
1219/// ANYTHING without fear of future compatibility hazards).
3dfed10e 1220fn frag_can_be_followed_by_any(kind: NonterminalKind) -> bool {
29967ef6
XL
1221 matches!(
1222 kind,
3dfed10e
XL
1223 NonterminalKind::Item // always terminated by `}` or `;`
1224 | NonterminalKind::Block // exactly one token tree
1225 | NonterminalKind::Ident // exactly one token tree
1226 | NonterminalKind::Literal // exactly one token tree
1227 | NonterminalKind::Meta // exactly one token tree
1228 | NonterminalKind::Lifetime // exactly one token tree
29967ef6
XL
1229 | NonterminalKind::TT // exactly one token tree
1230 )
62682a34
SL
1231}
1232
a1dfa0c6
XL
1233enum IsInFollow {
1234 Yes,
dc9dc135 1235 No(&'static [&'static str]),
a1dfa0c6
XL
1236}
1237
9fa01778 1238/// Returns `true` if `frag` can legally be followed by the token `tok`. For
9cc50fc6 1239/// fragments that can consume an unbounded number of tokens, `tok`
62682a34
SL
1240/// must be within a well-defined follow set. This is intended to
1241/// guarantee future compatibility: for example, without this rule, if
1242/// we expanded `expr` to include a new binary operator, we might
1243/// break macros that were relying on that binary operator as a
1244/// separator.
9cc50fc6 1245// when changing this do not forget to update doc/book/macros.md!
3dfed10e 1246fn is_in_follow(tok: &mbe::TokenTree, kind: NonterminalKind) -> IsInFollow {
e74abb32 1247 use mbe::TokenTree;
8bb4bdeb 1248
dc9dc135 1249 if let TokenTree::Token(Token { kind: token::CloseDelim(_), .. }) = *tok {
62682a34
SL
1250 // closing a token tree can never be matched by any fragment;
1251 // iow, we always require that `(` and `)` match, etc.
a1dfa0c6 1252 IsInFollow::Yes
85aaf69f 1253 } else {
3dfed10e
XL
1254 match kind {
1255 NonterminalKind::Item => {
85aaf69f
SL
1256 // since items *must* be followed by either a `;` or a `}`, we can
1257 // accept anything after them
a1dfa0c6 1258 IsInFollow::Yes
dc9dc135 1259 }
3dfed10e 1260 NonterminalKind::Block => {
b039eaaf 1261 // anything can follow block, the braces provide an easy boundary to
85aaf69f 1262 // maintain
a1dfa0c6 1263 IsInFollow::Yes
dc9dc135 1264 }
3dfed10e 1265 NonterminalKind::Stmt | NonterminalKind::Expr => {
dc9dc135
XL
1266 const TOKENS: &[&str] = &["`=>`", "`,`", "`;`"];
1267 match tok {
1268 TokenTree::Token(token) => match token.kind {
a1dfa0c6 1269 FatArrow | Comma | Semi => IsInFollow::Yes,
dc9dc135 1270 _ => IsInFollow::No(TOKENS),
a1dfa0c6 1271 },
dc9dc135 1272 _ => IsInFollow::No(TOKENS),
a1dfa0c6 1273 }
dc9dc135 1274 }
cdc7bbd5 1275 NonterminalKind::PatParam { .. } => {
dc9dc135
XL
1276 const TOKENS: &[&str] = &["`=>`", "`,`", "`=`", "`|`", "`if`", "`in`"];
1277 match tok {
1278 TokenTree::Token(token) => match token.kind {
a1dfa0c6 1279 FatArrow | Comma | Eq | BinOp(token::Or) => IsInFollow::Yes,
dc9dc135 1280 Ident(name, false) if name == kw::If || name == kw::In => IsInFollow::Yes,
cdc7bbd5
XL
1281 _ => IsInFollow::No(TOKENS),
1282 },
1283 _ => IsInFollow::No(TOKENS),
1284 }
1285 }
1286 NonterminalKind::PatWithOr { .. } => {
1287 const TOKENS: &[&str] = &["`=>`", "`,`", "`=`", "`if`", "`in`"];
1288 match tok {
1289 TokenTree::Token(token) => match token.kind {
1290 FatArrow | Comma | Eq => IsInFollow::Yes,
1291 Ident(name, false) if name == kw::If || name == kw::In => IsInFollow::Yes,
dc9dc135 1292 _ => IsInFollow::No(TOKENS),
a1dfa0c6 1293 },
dc9dc135 1294 _ => IsInFollow::No(TOKENS),
a1dfa0c6 1295 }
dc9dc135 1296 }
3dfed10e 1297 NonterminalKind::Path | NonterminalKind::Ty => {
dc9dc135
XL
1298 const TOKENS: &[&str] = &[
1299 "`{`", "`[`", "`=>`", "`,`", "`>`", "`=`", "`:`", "`;`", "`|`", "`as`",
a1dfa0c6
XL
1300 "`where`",
1301 ];
dc9dc135
XL
1302 match tok {
1303 TokenTree::Token(token) => match token.kind {
04454e1e
FG
1304 OpenDelim(Delimiter::Brace)
1305 | OpenDelim(Delimiter::Bracket)
dc9dc135
XL
1306 | Comma
1307 | FatArrow
1308 | Colon
1309 | Eq
1310 | Gt
1311 | BinOp(token::Shr)
1312 | Semi
1313 | BinOp(token::Or) => IsInFollow::Yes,
1314 Ident(name, false) if name == kw::As || name == kw::Where => {
1315 IsInFollow::Yes
1316 }
1317 _ => IsInFollow::No(TOKENS),
a1dfa0c6 1318 },
b9856134 1319 TokenTree::MetaVarDecl(_, _, Some(NonterminalKind::Block)) => IsInFollow::Yes,
dc9dc135 1320 _ => IsInFollow::No(TOKENS),
a1dfa0c6 1321 }
dc9dc135 1322 }
3dfed10e 1323 NonterminalKind::Ident | NonterminalKind::Lifetime => {
ff7c6d11 1324 // being a single token, idents and lifetimes are harmless
a1dfa0c6 1325 IsInFollow::Yes
dc9dc135 1326 }
3dfed10e 1327 NonterminalKind::Literal => {
94b46f34 1328 // literals may be of a single token, or two tokens (negative numbers)
a1dfa0c6 1329 IsInFollow::Yes
dc9dc135 1330 }
3dfed10e 1331 NonterminalKind::Meta | NonterminalKind::TT => {
85aaf69f
SL
1332 // being either a single token or a delimited sequence, tt is
1333 // harmless
a1dfa0c6 1334 IsInFollow::Yes
dc9dc135 1335 }
3dfed10e 1336 NonterminalKind::Vis => {
cc61c64b 1337 // Explicitly disallow `priv`, on the off chance it comes back.
dc9dc135
XL
1338 const TOKENS: &[&str] = &["`,`", "an ident", "a type"];
1339 match tok {
1340 TokenTree::Token(token) => match token.kind {
a1dfa0c6 1341 Comma => IsInFollow::Yes,
dc9dc135
XL
1342 Ident(name, is_raw) if is_raw || name != kw::Priv => IsInFollow::Yes,
1343 _ => {
1344 if token.can_begin_type() {
1345 IsInFollow::Yes
1346 } else {
1347 IsInFollow::No(TOKENS)
1348 }
a1dfa0c6 1349 }
cc61c64b 1350 },
3dfed10e
XL
1351 TokenTree::MetaVarDecl(
1352 _,
1353 _,
b9856134 1354 Some(NonterminalKind::Ident | NonterminalKind::Ty | NonterminalKind::Path),
3dfed10e 1355 ) => IsInFollow::Yes,
dc9dc135 1356 _ => IsInFollow::No(TOKENS),
cc61c64b 1357 }
dc9dc135 1358 }
85aaf69f 1359 }
1a4d82fc 1360 }
223e47cc 1361}
9cc50fc6 1362
e74abb32 1363fn quoted_tt_to_string(tt: &mbe::TokenTree) -> String {
8bb4bdeb 1364 match *tt {
94222f64 1365 mbe::TokenTree::Token(ref token) => pprust::token_to_string(&token).into(),
e74abb32 1366 mbe::TokenTree::MetaVar(_, name) => format!("${}", name),
b9856134
XL
1367 mbe::TokenTree::MetaVarDecl(_, name, Some(kind)) => format!("${}:{}", name, kind),
1368 mbe::TokenTree::MetaVarDecl(_, name, None) => format!("${}:", name),
dc9dc135 1369 _ => panic!(
fc512014
XL
1370 "{}",
1371 "unexpected mbe::TokenTree::{Sequence or Delimited} \
dc9dc135
XL
1372 in follow set checker"
1373 ),
8bb4bdeb
XL
1374 }
1375}
e74abb32 1376
ba9703b0
XL
1377fn parser_from_cx(sess: &ParseSess, tts: TokenStream) -> Parser<'_> {
1378 Parser::new(sess, tts, true, rustc_parse::MACRO_ARGUMENTS)
e74abb32
XL
1379}
1380
1381/// Generates an appropriate parsing failure message. For EOF, this is "unexpected end...". For
1382/// other tokens, this is "unexpected token...".
1383fn parse_failure_msg(tok: &Token) -> String {
1384 match tok.kind {
1385 token::Eof => "unexpected end of macro invocation".to_string(),
dfeec247 1386 _ => format!("no rules expected the token `{}`", pprust::token_to_string(tok),),
e74abb32
XL
1387 }
1388}