1 use crate::base
::{DummyResult, ExtCtxt, MacResult, TTMacroExpander}
;
2 use crate::base
::{SyntaxExtension, SyntaxExtensionKind}
;
3 use crate::expand
::{ensure_complete_parse, parse_ast_fragment, AstFragment, AstFragmentKind}
;
5 use crate::mbe
::macro_check
;
6 use crate::mbe
::macro_parser
::{Error, ErrorReported, Failure, Success, TtParser}
;
7 use crate::mbe
::macro_parser
::{MatchedSeq, MatchedTokenTree, MatcherLoc}
;
8 use crate::mbe
::transcribe
::transcribe
;
11 use rustc_ast
::token
::{self, Delimiter, NonterminalKind, Token, TokenKind, TokenKind::*}
;
12 use rustc_ast
::tokenstream
::{DelimSpan, TokenStream}
;
13 use rustc_ast
::{NodeId, DUMMY_NODE_ID}
;
14 use rustc_ast_pretty
::pprust
;
15 use rustc_attr
::{self as attr, TransparencyError}
;
16 use rustc_data_structures
::fx
::{FxHashMap, FxIndexMap}
;
17 use rustc_errors
::{Applicability, Diagnostic, DiagnosticBuilder, ErrorGuaranteed}
;
18 use rustc_feature
::Features
;
19 use rustc_lint_defs
::builtin
::{
20 RUST_2021_INCOMPATIBLE_OR_PATTERNS
, SEMICOLON_IN_EXPRESSIONS_FROM_MACROS
,
22 use rustc_lint_defs
::BuiltinLintDiagnostics
;
23 use rustc_parse
::parser
::Parser
;
24 use rustc_session
::parse
::ParseSess
;
25 use rustc_session
::Session
;
26 use rustc_span
::edition
::Edition
;
27 use rustc_span
::hygiene
::Transparency
;
28 use rustc_span
::source_map
::SourceMap
;
29 use rustc_span
::symbol
::{kw, sym, Ident, MacroRulesNormalizedIdent}
;
33 use std
::collections
::hash_map
::Entry
;
34 use std
::{mem, slice}
;
37 pub(crate) struct ParserAnyMacro
<'a
> {
40 /// Span of the expansion site of the macro this parser is for
42 /// The ident of the macro we're parsing
45 is_trailing_mac
: bool
,
47 /// Whether or not this macro is defined in the current crate
51 pub(crate) fn annotate_err_with_kind(err
: &mut Diagnostic
, kind
: AstFragmentKind
, span
: Span
) {
53 AstFragmentKind
::Ty
=> {
54 err
.span_label(span
, "this macro call doesn't expand to a type");
56 AstFragmentKind
::Pat
=> {
57 err
.span_label(span
, "this macro call doesn't expand to a pattern");
63 fn emit_frag_parse_err(
64 mut e
: DiagnosticBuilder
<'_
, rustc_errors
::ErrorGuaranteed
>,
66 orig_parser
: &mut Parser
<'_
>,
69 kind
: AstFragmentKind
,
71 // FIXME(davidtwco): avoid depending on the error message text
72 if parser
.token
== token
::Eof
&& e
.message
[0].0.expect_str().ends_with(", found `<eof>`") {
73 if !e
.span
.is_dummy() {
74 // early end of macro arm (#52866)
75 e
.replace_span_with(parser
.sess
.source_map().next_point(parser
.token
.span
));
77 let msg
= &e
.message
[0];
79 rustc_errors
::DiagnosticMessage
::Str(format
!(
80 "macro expansion ends with an incomplete expression: {}",
81 msg
.0.expect_str().replace(", found `<eof>`", ""),
86 if e
.span
.is_dummy() {
87 // Get around lack of span in error (#30128)
88 e
.replace_span_with(site_span
);
89 if !parser
.sess
.source_map().is_imported(arm_span
) {
90 e
.span_label(arm_span
, "in this macro arm");
92 } else if parser
.sess
.source_map().is_imported(parser
.token
.span
) {
93 e
.span_label(site_span
, "in this macro invocation");
96 // Try a statement if an expression is wanted but failed and suggest adding `;` to call.
97 AstFragmentKind
::Expr
=> match parse_ast_fragment(orig_parser
, AstFragmentKind
::Stmts
) {
98 Err(err
) => err
.cancel(),
101 "the macro call doesn't expand to an expression, but it can expand to a statement",
103 e
.span_suggestion_verbose(
104 site_span
.shrink_to_hi(),
105 "add `;` to interpret the expansion as a statement",
107 Applicability
::MaybeIncorrect
,
111 _
=> annotate_err_with_kind(&mut e
, kind
, site_span
),
116 impl<'a
> ParserAnyMacro
<'a
> {
117 pub(crate) fn make(mut self: Box
<ParserAnyMacro
<'a
>>, kind
: AstFragmentKind
) -> AstFragment
{
127 let snapshot
= &mut parser
.create_snapshot_for_diagnostic();
128 let fragment
= match parse_ast_fragment(parser
, kind
) {
131 emit_frag_parse_err(err
, parser
, snapshot
, site_span
, arm_span
, kind
);
132 return kind
.dummy(site_span
);
136 // We allow semicolons at the end of expressions -- e.g., the semicolon in
137 // `macro_rules! m { () => { panic!(); } }` isn't parsed by `.parse_expr()`,
138 // but `m!()` is allowed in expression positions (cf. issue #34706).
139 if kind
== AstFragmentKind
::Expr
&& parser
.token
== token
::Semi
{
141 parser
.sess
.buffer_lint_with_diagnostic(
142 SEMICOLON_IN_EXPRESSIONS_FROM_MACROS
,
145 "trailing semicolon in macro used in expression position",
146 BuiltinLintDiagnostics
::TrailingMacro(is_trailing_mac
, macro_ident
),
152 // Make sure we don't have any tokens left to parse so we don't silently drop anything.
153 let path
= ast
::Path
::from_ident(macro_ident
.with_span_pos(site_span
));
154 ensure_complete_parse(parser
, &path
, kind
.name(), site_span
);
159 struct MacroRulesMacroExpander
{
163 transparency
: Transparency
,
164 lhses
: Vec
<Vec
<MatcherLoc
>>,
165 rhses
: Vec
<mbe
::TokenTree
>,
169 impl TTMacroExpander
for MacroRulesMacroExpander
{
172 cx
: &'cx
mut ExtCtxt
<'_
>,
175 ) -> Box
<dyn MacResult
+ 'cx
> {
177 return DummyResult
::any(sp
);
193 fn macro_rules_dummy_expander
<'cx
>(
194 _
: &'cx
mut ExtCtxt
<'_
>,
197 ) -> Box
<dyn MacResult
+ 'cx
> {
198 DummyResult
::any(span
)
201 fn trace_macros_note(cx_expansions
: &mut FxIndexMap
<Span
, Vec
<String
>>, sp
: Span
, message
: String
) {
202 let sp
= sp
.macro_backtrace().last().map_or(sp
, |trace
| trace
.call_site
);
203 cx_expansions
.entry(sp
).or_default().push(message
);
206 /// Expands the rules based macro defined by `lhses` and `rhses` for a given
208 fn expand_macro
<'cx
>(
209 cx
: &'cx
mut ExtCtxt
<'_
>,
214 transparency
: Transparency
,
216 lhses
: &[Vec
<MatcherLoc
>],
217 rhses
: &[mbe
::TokenTree
],
218 ) -> Box
<dyn MacResult
+ 'cx
> {
219 let sess
= &cx
.sess
.parse_sess
;
220 // Macros defined in the current crate have a real node id,
221 // whereas macros from an external crate have a dummy id.
222 let is_local
= node_id
!= DUMMY_NODE_ID
;
224 if cx
.trace_macros() {
225 let msg
= format
!("expanding `{}! {{ {} }}`", name
, pprust
::tts_to_string(&arg
));
226 trace_macros_note(&mut cx
.expansions
, sp
, msg
);
229 // Which arm's failure should we report? (the one furthest along)
230 let mut best_failure
: Option
<(Token
, &str)> = None
;
232 // We create a base parser that can be used for the "black box" parts.
233 // Every iteration needs a fresh copy of that parser. However, the parser
234 // is not mutated on many of the iterations, particularly when dealing with
237 // macro_rules! foo {
241 // // ... etc. (maybe hundreds more)
244 // as seen in the `html5ever` benchmark. We use a `Cow` so that the base
245 // parser is only cloned when necessary (upon mutation). Furthermore, we
246 // reinitialize the `Cow` with the base parser at the start of every
247 // iteration, so that any mutated parsers are not reused. This is all quite
248 // hacky, but speeds up the `html5ever` benchmark significantly. (Issue
249 // 68836 suggests a more comprehensive but more complex change to deal with
251 let parser
= parser_from_cx(sess
, arg
.clone());
253 // Try each arm's matchers.
254 let mut tt_parser
= TtParser
::new(name
);
255 for (i
, lhs
) in lhses
.iter().enumerate() {
256 // Take a snapshot of the state of pre-expansion gating at this point.
257 // This is used so that if a matcher is not `Success(..)`ful,
258 // then the spans which became gated when parsing the unsuccessful matcher
259 // are not recorded. On the first `Success(..)`ful matcher, the spans are merged.
260 let mut gated_spans_snapshot
= mem
::take(&mut *sess
.gated_spans
.spans
.borrow_mut());
262 match tt_parser
.parse_tt(&mut Cow
::Borrowed(&parser
), lhs
) {
263 Success(named_matches
) => {
264 // The matcher was `Success(..)`ful.
265 // Merge the gated spans from parsing the matcher with the pre-existing ones.
266 sess
.gated_spans
.merge(gated_spans_snapshot
);
268 let (rhs
, rhs_span
): (&mbe
::Delimited
, DelimSpan
) = match &rhses
[i
] {
269 mbe
::TokenTree
::Delimited(span
, delimited
) => (&delimited
, *span
),
270 _
=> cx
.span_bug(sp
, "malformed macro rhs"),
272 let arm_span
= rhses
[i
].span();
274 let rhs_spans
= rhs
.tts
.iter().map(|t
| t
.span()).collect
::<Vec
<_
>>();
275 // rhs has holes ( `$id` and `$(...)` that need filled)
276 let mut tts
= match transcribe(cx
, &named_matches
, &rhs
, rhs_span
, transparency
) {
280 return DummyResult
::any(arm_span
);
284 // Replace all the tokens for the corresponding positions in the macro, to maintain
285 // proper positions in error reporting, while maintaining the macro_backtrace.
286 if rhs_spans
.len() == tts
.len() {
287 tts
= tts
.map_enumerated(|i
, tt
| {
288 let mut tt
= tt
.clone();
289 let mut sp
= rhs_spans
[i
];
290 sp
= sp
.with_ctxt(tt
.span().ctxt());
296 if cx
.trace_macros() {
297 let msg
= format
!("to `{}`", pprust
::tts_to_string(&tts
));
298 trace_macros_note(&mut cx
.expansions
, sp
, msg
);
301 let mut p
= Parser
::new(sess
, tts
, false, None
);
302 p
.last_type_ascription
= cx
.current_expansion
.prior_type_ascription
;
305 cx
.resolver
.record_macro_rule_usage(node_id
, i
);
308 // Let the context choose how to interpret the result.
309 // Weird, but useful for X-macros.
310 return Box
::new(ParserAnyMacro
{
313 // Pass along the original expansion site and the name of the macro
314 // so we can print a useful error message if the parse of the expanded
315 // macro leaves unparsed tokens.
318 lint_node_id
: cx
.current_expansion
.lint_node_id
,
319 is_trailing_mac
: cx
.current_expansion
.is_trailing_mac
,
324 Failure(token
, msg
) => match best_failure
{
325 Some((ref best_token
, _
)) if best_token
.span
.lo() >= token
.span
.lo() => {}
326 _
=> best_failure
= Some((token
, msg
)),
328 Error(err_sp
, ref msg
) => {
329 let span
= err_sp
.substitute_dummy(sp
);
330 cx
.struct_span_err(span
, &msg
).emit();
331 return DummyResult
::any(span
);
333 ErrorReported
=> return DummyResult
::any(sp
),
336 // The matcher was not `Success(..)`ful.
337 // Restore to the state before snapshotting and maybe try again.
338 mem
::swap(&mut gated_spans_snapshot
, &mut sess
.gated_spans
.spans
.borrow_mut());
342 let (token
, label
) = best_failure
.expect("ran no matchers");
343 let span
= token
.span
.substitute_dummy(sp
);
344 let mut err
= cx
.struct_span_err(span
, &parse_failure_msg(&token
));
345 err
.span_label(span
, label
);
346 if !def_span
.is_dummy() && !cx
.source_map().is_imported(def_span
) {
347 err
.span_label(cx
.source_map().guess_head_span(def_span
), "when calling this macro");
349 annotate_doc_comment(&mut err
, sess
.source_map(), span
);
350 // Check whether there's a missing comma in this macro call, like `println!("{}" a);`
351 if let Some((arg
, comma_span
)) = arg
.add_comma() {
353 let parser
= parser_from_cx(sess
, arg
.clone());
354 if let Success(_
) = tt_parser
.parse_tt(&mut Cow
::Borrowed(&parser
), lhs
) {
355 if comma_span
.is_dummy() {
356 err
.note("you might be missing a comma");
358 err
.span_suggestion_short(
360 "missing comma here",
362 Applicability
::MachineApplicable
,
369 cx
.trace_macros_diag();
373 // Note that macro-by-example's input is also matched against a token tree:
374 // $( $lhs:tt => $rhs:tt );+
376 // Holy self-referential!
378 /// Converts a macro item into a syntax extension.
379 pub fn compile_declarative_macro(
384 ) -> (SyntaxExtension
, Vec
<(usize, Span
)>) {
385 debug
!("compile_declarative_macro: {:?}", def
);
386 let mk_syn_ext
= |expander
| {
387 SyntaxExtension
::new(
389 SyntaxExtensionKind
::LegacyBang(expander
),
397 let dummy_syn_ext
= || (mk_syn_ext(Box
::new(macro_rules_dummy_expander
)), Vec
::new());
399 let diag
= &sess
.parse_sess
.span_diagnostic
;
400 let lhs_nm
= Ident
::new(sym
::lhs
, def
.span
);
401 let rhs_nm
= Ident
::new(sym
::rhs
, def
.span
);
402 let tt_spec
= Some(NonterminalKind
::TT
);
404 // Parse the macro_rules! invocation
405 let (macro_rules
, body
) = match &def
.kind
{
406 ast
::ItemKind
::MacroDef(def
) => (def
.macro_rules
, def
.body
.inner_tokens()),
410 // The pattern that macro_rules matches.
411 // The grammar for macro_rules! is:
412 // $( $lhs:tt => $rhs:tt );+
413 // ...quasiquoting this would be nice.
414 // These spans won't matter, anyways
415 let argument_gram
= vec
![
416 mbe
::TokenTree
::Sequence(
418 mbe
::SequenceRepetition
{
420 mbe
::TokenTree
::MetaVarDecl(def
.span
, lhs_nm
, tt_spec
),
421 mbe
::TokenTree
::token(token
::FatArrow
, def
.span
),
422 mbe
::TokenTree
::MetaVarDecl(def
.span
, rhs_nm
, tt_spec
),
424 separator
: Some(Token
::new(
425 if macro_rules { token::Semi }
else { token::Comma }
,
428 kleene
: mbe
::KleeneToken
::new(mbe
::KleeneOp
::OneOrMore
, def
.span
),
432 // to phase into semicolon-termination instead of semicolon-separation
433 mbe
::TokenTree
::Sequence(
435 mbe
::SequenceRepetition
{
436 tts
: vec
![mbe
::TokenTree
::token(
437 if macro_rules { token::Semi }
else { token::Comma }
,
441 kleene
: mbe
::KleeneToken
::new(mbe
::KleeneOp
::ZeroOrMore
, def
.span
),
446 // Convert it into `MatcherLoc` form.
447 let argument_gram
= mbe
::macro_parser
::compute_locs(&argument_gram
);
449 let parser
= Parser
::new(&sess
.parse_sess
, body
, true, rustc_parse
::MACRO_ARGUMENTS
);
451 TtParser
::new(Ident
::with_dummy_span(if macro_rules { kw::MacroRules }
else { kw::Macro }
));
452 let argument_map
= match tt_parser
.parse_tt(&mut Cow
::Borrowed(&parser
), &argument_gram
) {
454 Failure(token
, msg
) => {
455 let s
= parse_failure_msg(&token
);
456 let sp
= token
.span
.substitute_dummy(def
.span
);
457 let mut err
= sess
.parse_sess
.span_diagnostic
.struct_span_err(sp
, &s
);
458 err
.span_label(sp
, msg
);
459 annotate_doc_comment(&mut err
, sess
.source_map(), sp
);
461 return dummy_syn_ext();
466 .struct_span_err(sp
.substitute_dummy(def
.span
), &msg
)
468 return dummy_syn_ext();
471 return dummy_syn_ext();
475 let mut valid
= true;
477 // Extract the arguments:
478 let lhses
= match argument_map
[&MacroRulesNormalizedIdent
::new(lhs_nm
)] {
479 MatchedSeq(ref s
) => s
482 if let MatchedTokenTree(ref tt
) = *m
{
483 let tt
= mbe
::quoted
::parse(
484 TokenStream
::new(vec
![tt
.clone()]),
493 valid
&= check_lhs_nt_follows(&sess
.parse_sess
, &def
, &tt
);
496 sess
.parse_sess
.span_diagnostic
.span_bug(def
.span
, "wrong-structured lhs")
498 .collect
::<Vec
<mbe
::TokenTree
>>(),
499 _
=> sess
.parse_sess
.span_diagnostic
.span_bug(def
.span
, "wrong-structured lhs"),
502 let rhses
= match argument_map
[&MacroRulesNormalizedIdent
::new(rhs_nm
)] {
503 MatchedSeq(ref s
) => s
506 if let MatchedTokenTree(ref tt
) = *m
{
507 return mbe
::quoted
::parse(
508 TokenStream
::new(vec
![tt
.clone()]),
518 sess
.parse_sess
.span_diagnostic
.span_bug(def
.span
, "wrong-structured lhs")
520 .collect
::<Vec
<mbe
::TokenTree
>>(),
521 _
=> sess
.parse_sess
.span_diagnostic
.span_bug(def
.span
, "wrong-structured rhs"),
525 valid
&= check_rhs(&sess
.parse_sess
, rhs
);
528 // don't abort iteration early, so that errors for multiple lhses can be reported
530 valid
&= check_lhs_no_empty_seq(&sess
.parse_sess
, slice
::from_ref(lhs
));
533 valid
&= macro_check
::check_meta_variables(&sess
.parse_sess
, def
.id
, def
.span
, &lhses
, &rhses
);
535 let (transparency
, transparency_error
) = attr
::find_transparency(&def
.attrs
, macro_rules
);
536 match transparency_error
{
537 Some(TransparencyError
::UnknownTransparency(value
, span
)) => {
538 diag
.span_err(span
, &format
!("unknown macro transparency: `{}`", value
));
540 Some(TransparencyError
::MultipleTransparencyAttrs(old_span
, new_span
)) => {
541 diag
.span_err(vec
![old_span
, new_span
], "multiple macro transparency attributes");
546 // Compute the spans of the macro rules for unused rule linting.
547 // To avoid warning noise, only consider the rules of this
548 // macro for the lint, if all rules are valid.
549 // Also, we are only interested in non-foreign macros.
550 let rule_spans
= if valid
&& def
.id
!= DUMMY_NODE_ID
{
555 // If the rhs contains an invocation like compile_error!,
556 // don't consider the rule for the unused rule lint.
557 .filter(|(_idx
, (_lhs
, rhs
))| !has_compile_error_macro(rhs
))
558 // We only take the span of the lhs here,
559 // so that the spans of created warnings are smaller.
560 .map(|(idx
, (lhs
, _rhs
))| (idx
, lhs
.span()))
566 // Convert the lhses into `MatcherLoc` form, which is better for doing the
567 // actual matching. Unless the matcher is invalid.
568 let lhses
= if valid
{
572 // Ignore the delimiters around the matcher.
574 mbe
::TokenTree
::Delimited(_
, delimited
) => {
575 mbe
::macro_parser
::compute_locs(&delimited
.tts
)
577 _
=> sess
.parse_sess
.span_diagnostic
.span_bug(def
.span
, "malformed macro lhs"),
585 let expander
= Box
::new(MacroRulesMacroExpander
{
594 (mk_syn_ext(expander
), rule_spans
)
597 #[derive(SessionSubdiagnostic)]
598 enum ExplainDocComment
{
599 #[label(expand::explain_doc_comment_inner)]
604 #[label(expand::explain_doc_comment_outer)]
611 fn annotate_doc_comment(
612 err
: &mut DiagnosticBuilder
<'_
, ErrorGuaranteed
>,
616 if let Ok(src
) = sm
.span_to_snippet(span
) {
617 if src
.starts_with("///") || src
.starts_with("/**") {
618 err
.subdiagnostic(ExplainDocComment
::Outer { span }
);
619 } else if src
.starts_with("//!") || src
.starts_with("/*!") {
620 err
.subdiagnostic(ExplainDocComment
::Inner { span }
);
625 fn check_lhs_nt_follows(sess
: &ParseSess
, def
: &ast
::Item
, lhs
: &mbe
::TokenTree
) -> bool
{
626 // lhs is going to be like TokenTree::Delimited(...), where the
627 // entire lhs is those tts. Or, it can be a "bare sequence", not wrapped in parens.
628 if let mbe
::TokenTree
::Delimited(_
, delimited
) = lhs
{
629 check_matcher(sess
, def
, &delimited
.tts
)
631 let msg
= "invalid macro matcher; matchers must be contained in balanced delimiters";
632 sess
.span_diagnostic
.span_err(lhs
.span(), msg
);
635 // we don't abort on errors on rejection, the driver will do that for us
636 // after parsing/expansion. we can report every error in every macro this way.
639 /// Checks that the lhs contains no repetition which could match an empty token
640 /// tree, because then the matcher would hang indefinitely.
641 fn check_lhs_no_empty_seq(sess
: &ParseSess
, tts
: &[mbe
::TokenTree
]) -> bool
{
646 | TokenTree
::MetaVar(..)
647 | TokenTree
::MetaVarDecl(..)
648 | TokenTree
::MetaVarExpr(..) => (),
649 TokenTree
::Delimited(_
, ref del
) => {
650 if !check_lhs_no_empty_seq(sess
, &del
.tts
) {
654 TokenTree
::Sequence(span
, ref seq
) => {
655 if seq
.separator
.is_none()
656 && seq
.tts
.iter().all(|seq_tt
| match *seq_tt
{
657 TokenTree
::MetaVarDecl(_
, _
, Some(NonterminalKind
::Vis
)) => true,
658 TokenTree
::Sequence(_
, ref sub_seq
) => {
659 sub_seq
.kleene
.op
== mbe
::KleeneOp
::ZeroOrMore
660 || sub_seq
.kleene
.op
== mbe
::KleeneOp
::ZeroOrOne
665 let sp
= span
.entire();
666 sess
.span_diagnostic
.span_err(sp
, "repetition matches empty token tree");
669 if !check_lhs_no_empty_seq(sess
, &seq
.tts
) {
679 fn check_rhs(sess
: &ParseSess
, rhs
: &mbe
::TokenTree
) -> bool
{
681 mbe
::TokenTree
::Delimited(..) => return true,
683 sess
.span_diagnostic
.span_err(rhs
.span(), "macro rhs must be delimited");
689 fn check_matcher(sess
: &ParseSess
, def
: &ast
::Item
, matcher
: &[mbe
::TokenTree
]) -> bool
{
690 let first_sets
= FirstSets
::new(matcher
);
691 let empty_suffix
= TokenSet
::empty();
692 let err
= sess
.span_diagnostic
.err_count();
693 check_matcher_core(sess
, def
, &first_sets
, matcher
, &empty_suffix
);
694 err
== sess
.span_diagnostic
.err_count()
697 fn has_compile_error_macro(rhs
: &mbe
::TokenTree
) -> bool
{
699 mbe
::TokenTree
::Delimited(_sp
, d
) => {
700 let has_compile_error
= d
.tts
.array_windows
::<3>().any(|[ident
, bang
, args
]| {
701 if let mbe
::TokenTree
::Token(ident
) = ident
&&
702 let TokenKind
::Ident(ident
, _
) = ident
.kind
&&
703 ident
== sym
::compile_error
&&
704 let mbe
::TokenTree
::Token(bang
) = bang
&&
705 let TokenKind
::Not
= bang
.kind
&&
706 let mbe
::TokenTree
::Delimited(_
, del
) = args
&&
707 del
.delim
!= Delimiter
::Invisible
714 if has_compile_error { true }
else { d.tts.iter().any(has_compile_error_macro) }
720 // `The FirstSets` for a matcher is a mapping from subsequences in the
721 // matcher to the FIRST set for that subsequence.
723 // This mapping is partially precomputed via a backwards scan over the
724 // token trees of the matcher, which provides a mapping from each
725 // repetition sequence to its *first* set.
727 // (Hypothetically, sequences should be uniquely identifiable via their
728 // spans, though perhaps that is false, e.g., for macro-generated macros
729 // that do not try to inject artificial span information. My plan is
730 // to try to catch such cases ahead of time and not include them in
731 // the precomputed mapping.)
732 struct FirstSets
<'tt
> {
733 // this maps each TokenTree::Sequence `$(tt ...) SEP OP` that is uniquely identified by its
734 // span in the original matcher to the First set for the inner sequence `tt ...`.
736 // If two sequences have the same span in a matcher, then map that
737 // span to None (invalidating the mapping here and forcing the code to
739 first
: FxHashMap
<Span
, Option
<TokenSet
<'tt
>>>,
742 impl<'tt
> FirstSets
<'tt
> {
743 fn new(tts
: &'tt
[mbe
::TokenTree
]) -> FirstSets
<'tt
> {
746 let mut sets
= FirstSets { first: FxHashMap::default() }
;
747 build_recur(&mut sets
, tts
);
750 // walks backward over `tts`, returning the FIRST for `tts`
751 // and updating `sets` at the same time for all sequence
752 // substructure we find within `tts`.
753 fn build_recur
<'tt
>(sets
: &mut FirstSets
<'tt
>, tts
: &'tt
[TokenTree
]) -> TokenSet
<'tt
> {
754 let mut first
= TokenSet
::empty();
755 for tt
in tts
.iter().rev() {
758 | TokenTree
::MetaVar(..)
759 | TokenTree
::MetaVarDecl(..)
760 | TokenTree
::MetaVarExpr(..) => {
761 first
.replace_with(TtHandle
::TtRef(tt
));
763 TokenTree
::Delimited(span
, ref delimited
) => {
764 build_recur(sets
, &delimited
.tts
);
765 first
.replace_with(TtHandle
::from_token_kind(
766 token
::OpenDelim(delimited
.delim
),
770 TokenTree
::Sequence(sp
, ref seq_rep
) => {
771 let subfirst
= build_recur(sets
, &seq_rep
.tts
);
773 match sets
.first
.entry(sp
.entire()) {
774 Entry
::Vacant(vac
) => {
775 vac
.insert(Some(subfirst
.clone()));
777 Entry
::Occupied(mut occ
) => {
778 // if there is already an entry, then a span must have collided.
779 // This should not happen with typical macro_rules macros,
780 // but syntax extensions need not maintain distinct spans,
781 // so distinct syntax trees can be assigned the same span.
782 // In such a case, the map cannot be trusted; so mark this
783 // entry as unusable.
788 // If the sequence contents can be empty, then the first
789 // token could be the separator token itself.
791 if let (Some(sep
), true) = (&seq_rep
.separator
, subfirst
.maybe_empty
) {
792 first
.add_one_maybe(TtHandle
::from_token(sep
.clone()));
795 // Reverse scan: Sequence comes before `first`.
796 if subfirst
.maybe_empty
797 || seq_rep
.kleene
.op
== mbe
::KleeneOp
::ZeroOrMore
798 || seq_rep
.kleene
.op
== mbe
::KleeneOp
::ZeroOrOne
800 // If sequence is potentially empty, then
801 // union them (preserving first emptiness).
802 first
.add_all(&TokenSet { maybe_empty: true, ..subfirst }
);
804 // Otherwise, sequence guaranteed
805 // non-empty; replace first.
816 // walks forward over `tts` until all potential FIRST tokens are
818 fn first(&self, tts
: &'tt
[mbe
::TokenTree
]) -> TokenSet
<'tt
> {
821 let mut first
= TokenSet
::empty();
822 for tt
in tts
.iter() {
823 assert
!(first
.maybe_empty
);
826 | TokenTree
::MetaVar(..)
827 | TokenTree
::MetaVarDecl(..)
828 | TokenTree
::MetaVarExpr(..) => {
829 first
.add_one(TtHandle
::TtRef(tt
));
832 TokenTree
::Delimited(span
, ref delimited
) => {
833 first
.add_one(TtHandle
::from_token_kind(
834 token
::OpenDelim(delimited
.delim
),
839 TokenTree
::Sequence(sp
, ref seq_rep
) => {
841 let subfirst
= match self.first
.get(&sp
.entire()) {
842 Some(&Some(ref subfirst
)) => subfirst
,
844 subfirst_owned
= self.first(&seq_rep
.tts
);
848 panic
!("We missed a sequence during FirstSets construction");
852 // If the sequence contents can be empty, then the first
853 // token could be the separator token itself.
854 if let (Some(sep
), true) = (&seq_rep
.separator
, subfirst
.maybe_empty
) {
855 first
.add_one_maybe(TtHandle
::from_token(sep
.clone()));
858 assert
!(first
.maybe_empty
);
859 first
.add_all(subfirst
);
860 if subfirst
.maybe_empty
861 || seq_rep
.kleene
.op
== mbe
::KleeneOp
::ZeroOrMore
862 || seq_rep
.kleene
.op
== mbe
::KleeneOp
::ZeroOrOne
864 // Continue scanning for more first
865 // tokens, but also make sure we
866 // restore empty-tracking state.
867 first
.maybe_empty
= true;
876 // we only exit the loop if `tts` was empty or if every
877 // element of `tts` matches the empty sequence.
878 assert
!(first
.maybe_empty
);
883 // Most `mbe::TokenTree`s are pre-existing in the matcher, but some are defined
884 // implicitly, such as opening/closing delimiters and sequence repetition ops.
885 // This type encapsulates both kinds. It implements `Clone` while avoiding the
886 // need for `mbe::TokenTree` to implement `Clone`.
889 /// This is used in most cases.
890 TtRef(&'tt mbe
::TokenTree
),
892 /// This is only used for implicit token trees. The `mbe::TokenTree` *must*
893 /// be `mbe::TokenTree::Token`. No other variants are allowed. We store an
894 /// `mbe::TokenTree` rather than a `Token` so that `get()` can return a
895 /// `&mbe::TokenTree`.
896 Token(mbe
::TokenTree
),
899 impl<'tt
> TtHandle
<'tt
> {
900 fn from_token(tok
: Token
) -> Self {
901 TtHandle
::Token(mbe
::TokenTree
::Token(tok
))
904 fn from_token_kind(kind
: TokenKind
, span
: Span
) -> Self {
905 TtHandle
::from_token(Token
::new(kind
, span
))
908 // Get a reference to a token tree.
909 fn get(&'tt
self) -> &'tt mbe
::TokenTree
{
911 TtHandle
::TtRef(tt
) => tt
,
912 TtHandle
::Token(token_tt
) => &token_tt
,
917 impl<'tt
> PartialEq
for TtHandle
<'tt
> {
918 fn eq(&self, other
: &TtHandle
<'tt
>) -> bool
{
919 self.get() == other
.get()
923 impl<'tt
> Clone
for TtHandle
<'tt
> {
924 fn clone(&self) -> Self {
926 TtHandle
::TtRef(tt
) => TtHandle
::TtRef(tt
),
928 // This variant *must* contain a `mbe::TokenTree::Token`, and not
929 // any other variant of `mbe::TokenTree`.
930 TtHandle
::Token(mbe
::TokenTree
::Token(tok
)) => {
931 TtHandle
::Token(mbe
::TokenTree
::Token(tok
.clone()))
939 // A set of `mbe::TokenTree`s, which may include `TokenTree::Match`s
940 // (for macro-by-example syntactic variables). It also carries the
941 // `maybe_empty` flag; that is true if and only if the matcher can
942 // match an empty token sequence.
944 // The First set is computed on submatchers like `$($a:expr b),* $(c)* d`,
945 // which has corresponding FIRST = {$a:expr, c, d}.
946 // Likewise, `$($a:expr b),* $(c)+ d` has FIRST = {$a:expr, c}.
948 // (Notably, we must allow for *-op to occur zero times.)
949 #[derive(Clone, Debug)]
950 struct TokenSet
<'tt
> {
951 tokens
: Vec
<TtHandle
<'tt
>>,
955 impl<'tt
> TokenSet
<'tt
> {
956 // Returns a set for the empty sequence.
958 TokenSet { tokens: Vec::new(), maybe_empty: true }
961 // Returns the set `{ tok }` for the single-token (and thus
962 // non-empty) sequence [tok].
963 fn singleton(tt
: TtHandle
<'tt
>) -> Self {
964 TokenSet { tokens: vec![tt], maybe_empty: false }
967 // Changes self to be the set `{ tok }`.
968 // Since `tok` is always present, marks self as non-empty.
969 fn replace_with(&mut self, tt
: TtHandle
<'tt
>) {
971 self.tokens
.push(tt
);
972 self.maybe_empty
= false;
975 // Changes self to be the empty set `{}`; meant for use when
976 // the particular token does not matter, but we want to
977 // record that it occurs.
978 fn replace_with_irrelevant(&mut self) {
980 self.maybe_empty
= false;
983 // Adds `tok` to the set for `self`, marking sequence as non-empy.
984 fn add_one(&mut self, tt
: TtHandle
<'tt
>) {
985 if !self.tokens
.contains(&tt
) {
986 self.tokens
.push(tt
);
988 self.maybe_empty
= false;
991 // Adds `tok` to the set for `self`. (Leaves `maybe_empty` flag alone.)
992 fn add_one_maybe(&mut self, tt
: TtHandle
<'tt
>) {
993 if !self.tokens
.contains(&tt
) {
994 self.tokens
.push(tt
);
998 // Adds all elements of `other` to this.
1000 // (Since this is a set, we filter out duplicates.)
1002 // If `other` is potentially empty, then preserves the previous
1003 // setting of the empty flag of `self`. If `other` is guaranteed
1004 // non-empty, then `self` is marked non-empty.
1005 fn add_all(&mut self, other
: &Self) {
1006 for tt
in &other
.tokens
{
1007 if !self.tokens
.contains(tt
) {
1008 self.tokens
.push(tt
.clone());
1011 if !other
.maybe_empty
{
1012 self.maybe_empty
= false;
1017 // Checks that `matcher` is internally consistent and that it
1018 // can legally be followed by a token `N`, for all `N` in `follow`.
1019 // (If `follow` is empty, then it imposes no constraint on
1022 // Returns the set of NT tokens that could possibly come last in
1023 // `matcher`. (If `matcher` matches the empty sequence, then
1024 // `maybe_empty` will be set to true.)
1026 // Requires that `first_sets` is pre-computed for `matcher`;
1027 // see `FirstSets::new`.
1028 fn check_matcher_core
<'tt
>(
1031 first_sets
: &FirstSets
<'tt
>,
1032 matcher
: &'tt
[mbe
::TokenTree
],
1033 follow
: &TokenSet
<'tt
>,
1034 ) -> TokenSet
<'tt
> {
1037 let mut last
= TokenSet
::empty();
1039 // 2. For each token and suffix [T, SUFFIX] in M:
1040 // ensure that T can be followed by SUFFIX, and if SUFFIX may be empty,
1041 // then ensure T can also be followed by any element of FOLLOW.
1042 'each_token
: for i
in 0..matcher
.len() {
1043 let token
= &matcher
[i
];
1044 let suffix
= &matcher
[i
+ 1..];
1046 let build_suffix_first
= || {
1047 let mut s
= first_sets
.first(suffix
);
1054 // (we build `suffix_first` on demand below; you can tell
1055 // which cases are supposed to fall through by looking for the
1056 // initialization of this variable.)
1059 // First, update `last` so that it corresponds to the set
1060 // of NT tokens that might end the sequence `... token`.
1062 TokenTree
::Token(..)
1063 | TokenTree
::MetaVar(..)
1064 | TokenTree
::MetaVarDecl(..)
1065 | TokenTree
::MetaVarExpr(..) => {
1066 if token_can_be_followed_by_any(token
) {
1067 // don't need to track tokens that work with any,
1068 last
.replace_with_irrelevant();
1069 // ... and don't need to check tokens that can be
1070 // followed by anything against SUFFIX.
1071 continue 'each_token
;
1073 last
.replace_with(TtHandle
::TtRef(token
));
1074 suffix_first
= build_suffix_first();
1077 TokenTree
::Delimited(span
, ref d
) => {
1078 let my_suffix
= TokenSet
::singleton(TtHandle
::from_token_kind(
1079 token
::CloseDelim(d
.delim
),
1082 check_matcher_core(sess
, def
, first_sets
, &d
.tts
, &my_suffix
);
1083 // don't track non NT tokens
1084 last
.replace_with_irrelevant();
1086 // also, we don't need to check delimited sequences
1088 continue 'each_token
;
1090 TokenTree
::Sequence(_
, ref seq_rep
) => {
1091 suffix_first
= build_suffix_first();
1092 // The trick here: when we check the interior, we want
1093 // to include the separator (if any) as a potential
1094 // (but not guaranteed) element of FOLLOW. So in that
1095 // case, we make a temp copy of suffix and stuff
1096 // delimiter in there.
1098 // FIXME: Should I first scan suffix_first to see if
1099 // delimiter is already in it before I go through the
1100 // work of cloning it? But then again, this way I may
1101 // get a "tighter" span?
1103 let my_suffix
= if let Some(sep
) = &seq_rep
.separator
{
1104 new
= suffix_first
.clone();
1105 new
.add_one_maybe(TtHandle
::from_token(sep
.clone()));
1111 // At this point, `suffix_first` is built, and
1112 // `my_suffix` is some TokenSet that we can use
1113 // for checking the interior of `seq_rep`.
1114 let next
= check_matcher_core(sess
, def
, first_sets
, &seq_rep
.tts
, my_suffix
);
1115 if next
.maybe_empty
{
1116 last
.add_all(&next
);
1121 // the recursive call to check_matcher_core already ran the 'each_last
1122 // check below, so we can just keep going forward here.
1123 continue 'each_token
;
1127 // (`suffix_first` guaranteed initialized once reaching here.)
1129 // Now `last` holds the complete set of NT tokens that could
1130 // end the sequence before SUFFIX. Check that every one works with `suffix`.
1131 for tt
in &last
.tokens
{
1132 if let &TokenTree
::MetaVarDecl(span
, name
, Some(kind
)) = tt
.get() {
1133 for next_token
in &suffix_first
.tokens
{
1134 let next_token
= next_token
.get();
1136 // Check if the old pat is used and the next token is `|`
1137 // to warn about incompatibility with Rust 2021.
1138 // We only emit this lint if we're parsing the original
1139 // definition of this macro_rules, not while (re)parsing
1140 // the macro when compiling another crate that is using the
1141 // macro. (See #86567.)
1142 // Macros defined in the current crate have a real node id,
1143 // whereas macros from an external crate have a dummy id.
1144 if def
.id
!= DUMMY_NODE_ID
1145 && matches
!(kind
, NonterminalKind
::PatParam { inferred: true }
)
1146 && matches
!(next_token
, TokenTree
::Token(token
) if token
.kind
== BinOp(token
::BinOpToken
::Or
))
1148 // It is suggestion to use pat_param, for example: $x:pat -> $x:pat_param.
1149 let suggestion
= quoted_tt_to_string(&TokenTree
::MetaVarDecl(
1152 Some(NonterminalKind
::PatParam { inferred: false }
),
1154 sess
.buffer_lint_with_diagnostic(
1155 &RUST_2021_INCOMPATIBLE_OR_PATTERNS
,
1158 "the meaning of the `pat` fragment specifier is changing in Rust 2021, which may affect this macro",
1159 BuiltinLintDiagnostics
::OrPatternsBackCompat(span
, suggestion
),
1162 match is_in_follow(next_token
, kind
) {
1163 IsInFollow
::Yes
=> {}
1164 IsInFollow
::No(possible
) => {
1165 let may_be
= if last
.tokens
.len() == 1 && suffix_first
.tokens
.len() == 1
1172 let sp
= next_token
.span();
1173 let mut err
= sess
.span_diagnostic
.struct_span_err(
1176 "`${name}:{frag}` {may_be} followed by `{next}`, which \
1177 is not allowed for `{frag}` fragments",
1180 next
= quoted_tt_to_string(next_token
),
1184 err
.span_label(sp
, format
!("not allowed after `{}` fragments", kind
));
1186 if kind
== NonterminalKind
::PatWithOr
1187 && sess
.edition
.rust_2021()
1188 && next_token
.is_token(&BinOp(token
::BinOpToken
::Or
))
1190 let suggestion
= quoted_tt_to_string(&TokenTree
::MetaVarDecl(
1193 Some(NonterminalKind
::PatParam { inferred: false }
),
1195 err
.span_suggestion(
1197 "try a `pat_param` fragment specifier instead",
1199 Applicability
::MaybeIncorrect
,
1203 let msg
= "allowed there are: ";
1208 "only {} is allowed after `{}` fragments",
1219 .collect
::<Vec
<_
>>()
1235 fn token_can_be_followed_by_any(tok
: &mbe
::TokenTree
) -> bool
{
1236 if let mbe
::TokenTree
::MetaVarDecl(_
, _
, Some(kind
)) = *tok
{
1237 frag_can_be_followed_by_any(kind
)
1239 // (Non NT's can always be followed by anything in matchers.)
1244 /// Returns `true` if a fragment of type `frag` can be followed by any sort of
1245 /// token. We use this (among other things) as a useful approximation
1246 /// for when `frag` can be followed by a repetition like `$(...)*` or
1247 /// `$(...)+`. In general, these can be a bit tricky to reason about,
1248 /// so we adopt a conservative position that says that any fragment
1249 /// specifier which consumes at most one token tree can be followed by
1250 /// a fragment specifier (indeed, these fragments can be followed by
1251 /// ANYTHING without fear of future compatibility hazards).
1252 fn frag_can_be_followed_by_any(kind
: NonterminalKind
) -> bool
{
1255 NonterminalKind
::Item
// always terminated by `}` or `;`
1256 | NonterminalKind
::Block
// exactly one token tree
1257 | NonterminalKind
::Ident
// exactly one token tree
1258 | NonterminalKind
::Literal
// exactly one token tree
1259 | NonterminalKind
::Meta
// exactly one token tree
1260 | NonterminalKind
::Lifetime
// exactly one token tree
1261 | NonterminalKind
::TT
// exactly one token tree
1267 No(&'
static [&'
static str]),
1270 /// Returns `true` if `frag` can legally be followed by the token `tok`. For
1271 /// fragments that can consume an unbounded number of tokens, `tok`
1272 /// must be within a well-defined follow set. This is intended to
1273 /// guarantee future compatibility: for example, without this rule, if
1274 /// we expanded `expr` to include a new binary operator, we might
1275 /// break macros that were relying on that binary operator as a
1277 // when changing this do not forget to update doc/book/macros.md!
1278 fn is_in_follow(tok
: &mbe
::TokenTree
, kind
: NonterminalKind
) -> IsInFollow
{
1281 if let TokenTree
::Token(Token { kind: token::CloseDelim(_), .. }
) = *tok
{
1282 // closing a token tree can never be matched by any fragment;
1283 // iow, we always require that `(` and `)` match, etc.
1287 NonterminalKind
::Item
=> {
1288 // since items *must* be followed by either a `;` or a `}`, we can
1289 // accept anything after them
1292 NonterminalKind
::Block
=> {
1293 // anything can follow block, the braces provide an easy boundary to
1297 NonterminalKind
::Stmt
| NonterminalKind
::Expr
=> {
1298 const TOKENS
: &[&str] = &["`=>`", "`,`", "`;`"];
1300 TokenTree
::Token(token
) => match token
.kind
{
1301 FatArrow
| Comma
| Semi
=> IsInFollow
::Yes
,
1302 _
=> IsInFollow
::No(TOKENS
),
1304 _
=> IsInFollow
::No(TOKENS
),
1307 NonterminalKind
::PatParam { .. }
=> {
1308 const TOKENS
: &[&str] = &["`=>`", "`,`", "`=`", "`|`", "`if`", "`in`"];
1310 TokenTree
::Token(token
) => match token
.kind
{
1311 FatArrow
| Comma
| Eq
| BinOp(token
::Or
) => IsInFollow
::Yes
,
1312 Ident(name
, false) if name
== kw
::If
|| name
== kw
::In
=> IsInFollow
::Yes
,
1313 _
=> IsInFollow
::No(TOKENS
),
1315 _
=> IsInFollow
::No(TOKENS
),
1318 NonterminalKind
::PatWithOr { .. }
=> {
1319 const TOKENS
: &[&str] = &["`=>`", "`,`", "`=`", "`if`", "`in`"];
1321 TokenTree
::Token(token
) => match token
.kind
{
1322 FatArrow
| Comma
| Eq
=> IsInFollow
::Yes
,
1323 Ident(name
, false) if name
== kw
::If
|| name
== kw
::In
=> IsInFollow
::Yes
,
1324 _
=> IsInFollow
::No(TOKENS
),
1326 _
=> IsInFollow
::No(TOKENS
),
1329 NonterminalKind
::Path
| NonterminalKind
::Ty
=> {
1330 const TOKENS
: &[&str] = &[
1331 "`{`", "`[`", "`=>`", "`,`", "`>`", "`=`", "`:`", "`;`", "`|`", "`as`",
1335 TokenTree
::Token(token
) => match token
.kind
{
1336 OpenDelim(Delimiter
::Brace
)
1337 | OpenDelim(Delimiter
::Bracket
)
1345 | BinOp(token
::Or
) => IsInFollow
::Yes
,
1346 Ident(name
, false) if name
== kw
::As
|| name
== kw
::Where
=> {
1349 _
=> IsInFollow
::No(TOKENS
),
1351 TokenTree
::MetaVarDecl(_
, _
, Some(NonterminalKind
::Block
)) => IsInFollow
::Yes
,
1352 _
=> IsInFollow
::No(TOKENS
),
1355 NonterminalKind
::Ident
| NonterminalKind
::Lifetime
=> {
1356 // being a single token, idents and lifetimes are harmless
1359 NonterminalKind
::Literal
=> {
1360 // literals may be of a single token, or two tokens (negative numbers)
1363 NonterminalKind
::Meta
| NonterminalKind
::TT
=> {
1364 // being either a single token or a delimited sequence, tt is
1368 NonterminalKind
::Vis
=> {
1369 // Explicitly disallow `priv`, on the off chance it comes back.
1370 const TOKENS
: &[&str] = &["`,`", "an ident", "a type"];
1372 TokenTree
::Token(token
) => match token
.kind
{
1373 Comma
=> IsInFollow
::Yes
,
1374 Ident(name
, is_raw
) if is_raw
|| name
!= kw
::Priv
=> IsInFollow
::Yes
,
1376 if token
.can_begin_type() {
1379 IsInFollow
::No(TOKENS
)
1383 TokenTree
::MetaVarDecl(
1386 Some(NonterminalKind
::Ident
| NonterminalKind
::Ty
| NonterminalKind
::Path
),
1387 ) => IsInFollow
::Yes
,
1388 _
=> IsInFollow
::No(TOKENS
),
1395 fn quoted_tt_to_string(tt
: &mbe
::TokenTree
) -> String
{
1397 mbe
::TokenTree
::Token(ref token
) => pprust
::token_to_string(&token
).into(),
1398 mbe
::TokenTree
::MetaVar(_
, name
) => format
!("${}", name
),
1399 mbe
::TokenTree
::MetaVarDecl(_
, name
, Some(kind
)) => format
!("${}:{}", name
, kind
),
1400 mbe
::TokenTree
::MetaVarDecl(_
, name
, None
) => format
!("${}:", name
),
1403 "unexpected mbe::TokenTree::{Sequence or Delimited} \
1404 in follow set checker"
1409 fn parser_from_cx(sess
: &ParseSess
, tts
: TokenStream
) -> Parser
<'_
> {
1410 Parser
::new(sess
, tts
, true, rustc_parse
::MACRO_ARGUMENTS
)
1413 /// Generates an appropriate parsing failure message. For EOF, this is "unexpected end...". For
1414 /// other tokens, this is "unexpected token...".
1415 fn parse_failure_msg(tok
: &Token
) -> String
{
1417 token
::Eof
=> "unexpected end of macro invocation".to_string(),
1418 _
=> format
!("no rules expected the token `{}`", pprust
::token_to_string(tok
),),