1 use crate::base::ExtCtxt;
3 CountRepetitionMisplaced, MetaVarExprUnrecognizedVar, MetaVarsDifSeqMatchers, MustRepeatOnce,
4 NoSyntaxVarsExprRepeat, VarStillRepeating,
6 use crate::mbe::macro_parser::{MatchedNonterminal, MatchedSeq, MatchedTokenTree, NamedMatch};
7 use crate::mbe::{self, MetaVarExpr};
8 use rustc_ast::mut_visit::{self, MutVisitor};
9 use rustc_ast::token::{self, Delimiter, Token, TokenKind};
10 use rustc_ast::tokenstream::{DelimSpan, Spacing, TokenStream, TokenTree};
11 use rustc_data_structures::fx::FxHashMap;
12 use rustc_errors::{pluralize, PResult};
13 use rustc_errors::{DiagnosticBuilder, ErrorGuaranteed};
14 use rustc_span::hygiene::{LocalExpnId, Transparency};
15 use rustc_span::symbol::{sym, Ident, MacroRulesNormalizedIdent};
18 use smallvec::{smallvec, SmallVec};
21 // A Marker adds the given mark to the syntax context.
22 struct Marker(LocalExpnId, Transparency);
24 impl MutVisitor for Marker {
25 const VISIT_TOKENS: bool = true;
27 fn visit_span(&mut self, span: &mut Span) {
28 *span = span.apply_mark(self.0.to_expn_id(), self.1)
32 /// An iterator over the token trees in a delimited token tree (`{ ... }`) or a sequence (`$(...)`).
34 Delimited { tts: &'a [mbe::TokenTree], idx: usize, delim: Delimiter, span: DelimSpan },
35 Sequence { tts: &'a [mbe::TokenTree], idx: usize, sep: Option<Token> },
39 /// Construct a new frame around the delimited set of tokens.
40 fn new(src: &'a mbe::Delimited, span: DelimSpan) -> Frame<'a> {
41 Frame::Delimited { tts: &src.tts, idx: 0, delim: src.delim, span }
45 impl<'a> Iterator for Frame<'a> {
46 type Item = &'a mbe::TokenTree;
48 fn next(&mut self) -> Option<&'a mbe::TokenTree> {
50 Frame::Delimited { tts, idx, .. } | Frame::Sequence { tts, idx, .. } => {
51 let res = tts.get(*idx);
59 /// This can do Macro-By-Example transcription.
60 /// - `interp` is a map of meta-variables to the tokens (non-terminals) they matched in the
61 /// invocation. We are assuming we already know there is a match.
62 /// - `src` is the RHS of the MBE, that is, the "example" we are filling in.
67 /// macro_rules! foo {
68 /// ($id:ident) => { println!("{}", stringify!($id)); }
74 /// `interp` would contain `$id => bar` and `src` would contain `println!("{}", stringify!($id));`.
76 /// `transcribe` would return a `TokenStream` containing `println!("{}", stringify!(bar));`.
78 /// Along the way, we do some additional error checking.
79 pub(super) fn transcribe<'a>(
81 interp: &FxHashMap<MacroRulesNormalizedIdent, NamedMatch>,
84 transparency: Transparency,
85 ) -> PResult<'a, TokenStream> {
86 // Nothing for us to transcribe...
87 if src.tts.is_empty() {
88 return Ok(TokenStream::default());
91 // We descend into the RHS (`src`), expanding things as we go. This stack contains the things
92 // we have yet to expand/are still expanding. We start the stack off with the whole RHS.
93 let mut stack: SmallVec<[Frame<'_>; 1]> = smallvec![Frame::new(&src, src_span)];
95 // As we descend in the RHS, we will need to be able to match nested sequences of matchers.
96 // `repeats` keeps track of where we are in matching at each level, with the last element being
97 // the most deeply nested sequence. This is used as a stack.
98 let mut repeats = Vec::new();
100 // `result` contains resulting token stream from the TokenTree we just finished processing. At
101 // the end, this will contain the full result of transcription, but at arbitrary points during
102 // `transcribe`, `result` will contain subsets of the final result.
104 // Specifically, as we descend into each TokenTree, we will push the existing results onto the
105 // `result_stack` and clear `results`. We will then produce the results of transcribing the
106 // TokenTree into `results`. Then, as we unwind back out of the `TokenTree`, we will pop the
107 // `result_stack` and append `results` too it to produce the new `results` up to that point.
109 // Thus, if we try to pop the `result_stack` and it is empty, we have reached the top-level
110 // again, and we are done transcribing.
111 let mut result: Vec<TokenTree> = Vec::new();
112 let mut result_stack = Vec::new();
113 let mut marker = Marker(cx.current_expansion.id, transparency);
116 // Look at the last frame on the stack.
117 // If it still has a TokenTree we have not looked at yet, use that tree.
118 let Some(tree) = stack.last_mut().unwrap().next() else {
119 // This else-case never produces a value for `tree` (it `continue`s or `return`s).
121 // Otherwise, if we have just reached the end of a sequence and we can keep repeating,
122 // go back to the beginning of the sequence.
123 if let Frame::Sequence { idx, sep, .. } = stack.last_mut().unwrap() {
124 let (repeat_idx, repeat_len) = repeats.last_mut().unwrap();
126 if repeat_idx < repeat_len {
128 if let Some(sep) = sep {
129 result.push(TokenTree::Token(sep.clone(), Spacing::Alone));
135 // We are done with the top of the stack. Pop it. Depending on what it was, we do
136 // different things. Note that the outermost item must be the delimited, wrapped RHS
137 // that was passed in originally to `transcribe`.
138 match stack.pop().unwrap() {
139 // Done with a sequence. Pop from repeats.
140 Frame::Sequence { .. } => {
144 // We are done processing a Delimited. If this is the top-level delimited, we are
145 // done. Otherwise, we unwind the result_stack to append what we have produced to
146 // any previous results.
147 Frame::Delimited { delim, span, .. } => {
148 if result_stack.is_empty() {
149 // No results left to compute! We are back at the top-level.
150 return Ok(TokenStream::new(result));
153 // Step back into the parent Delimited.
154 let tree = TokenTree::Delimited(span, delim, TokenStream::new(result));
155 result = result_stack.pop().unwrap();
162 // At this point, we know we are in the middle of a TokenTree (the last one on `stack`).
163 // `tree` contains the next `TokenTree` to be processed.
165 // We are descending into a sequence. We first make sure that the matchers in the RHS
166 // and the matches in `interp` have the same shape. Otherwise, either the caller or the
167 // macro writer has made a mistake.
168 seq @ mbe::TokenTree::Sequence(_, delimited) => {
169 match lockstep_iter_size(&seq, interp, &repeats) {
170 LockstepIterSize::Unconstrained => {
171 return Err(cx.create_err(NoSyntaxVarsExprRepeat { span: seq.span() }));
174 LockstepIterSize::Contradiction(msg) => {
175 // FIXME: this really ought to be caught at macro definition time... It
176 // happens when two meta-variables are used in the same repetition in a
177 // sequence, but they come from different sequence matchers and repeat
178 // different amounts.
179 return Err(cx.create_err(MetaVarsDifSeqMatchers { span: seq.span(), msg }));
182 LockstepIterSize::Constraint(len, _) => {
183 // We do this to avoid an extra clone above. We know that this is a
185 let mbe::TokenTree::Sequence(sp, seq) = seq else {
189 // Is the repetition empty?
191 if seq.kleene.op == mbe::KleeneOp::OneOrMore {
192 // FIXME: this really ought to be caught at macro definition
193 // time... It happens when the Kleene operator in the matcher and
194 // the body for the same meta-variable do not match.
195 return Err(cx.create_err(MustRepeatOnce { span: sp.entire() }));
198 // 0 is the initial counter (we have done 0 repetitions so far). `len`
199 // is the total number of repetitions we should generate.
200 repeats.push((0, len));
202 // The first time we encounter the sequence we push it to the stack. It
203 // then gets reused (see the beginning of the loop) until we are done
205 stack.push(Frame::Sequence {
207 sep: seq.separator.clone(),
215 // Replace the meta-var with the matched token tree from the invocation.
216 mbe::TokenTree::MetaVar(mut sp, mut original_ident) => {
217 // Find the matched nonterminal from the macro invocation, and use it to replace
219 let ident = MacroRulesNormalizedIdent::new(original_ident);
220 if let Some(cur_matched) = lookup_cur_matched(ident, interp, &repeats) {
222 MatchedTokenTree(tt) => {
223 // `tt`s are emitted into the output stream directly as "raw tokens",
224 // without wrapping them into groups.
225 let token = tt.clone();
228 MatchedNonterminal(nt) => {
229 // Other variables are emitted into the output stream as groups with
230 // `Delimiter::Invisible` to maintain parsing priorities.
231 // `Interpolated` is currently used for such groups in rustc parser.
232 marker.visit_span(&mut sp);
233 let token = TokenTree::token_alone(token::Interpolated(nt.clone()), sp);
237 // We were unable to descend far enough. This is an error.
238 return Err(cx.create_err(VarStillRepeating { span: sp, ident }));
242 // If we aren't able to match the meta-var, we push it back into the result but
243 // with modified syntax context. (I believe this supports nested macros).
244 marker.visit_span(&mut sp);
245 marker.visit_ident(&mut original_ident);
246 result.push(TokenTree::token_alone(token::Dollar, sp));
247 result.push(TokenTree::Token(
248 Token::from_ast_ident(original_ident),
254 // Replace meta-variable expressions with the result of their expansion.
255 mbe::TokenTree::MetaVarExpr(sp, expr) => {
256 transcribe_metavar_expr(cx, expr, interp, &mut marker, &repeats, &mut result, &sp)?;
259 // If we are entering a new delimiter, we push its contents to the `stack` to be
260 // processed, and we push all of the currently produced results to the `result_stack`.
261 // We will produce all of the results of the inside of the `Delimited` and then we will
262 // jump back out of the Delimited, pop the result_stack and add the new results back to
263 // the previous results (from outside the Delimited).
264 mbe::TokenTree::Delimited(mut span, delimited) => {
265 mut_visit::visit_delim_span(&mut span, &mut marker);
266 stack.push(Frame::Delimited {
268 delim: delimited.delim,
272 result_stack.push(mem::take(&mut result));
275 // Nothing much to do here. Just push the token to the result, being careful to
276 // preserve syntax context.
277 mbe::TokenTree::Token(token) => {
278 let mut token = token.clone();
279 mut_visit::visit_token(&mut token, &mut marker);
280 let tt = TokenTree::Token(token, Spacing::Alone);
284 // There should be no meta-var declarations in the invocation of a macro.
285 mbe::TokenTree::MetaVarDecl(..) => panic!("unexpected `TokenTree::MetaVarDecl"),
290 /// Lookup the meta-var named `ident` and return the matched token tree from the invocation using
291 /// the set of matches `interpolations`.
293 /// See the definition of `repeats` in the `transcribe` function. `repeats` is used to descend
294 /// into the right place in nested matchers. If we attempt to descend too far, the macro writer has
295 /// made a mistake, and we return `None`.
296 fn lookup_cur_matched<'a>(
297 ident: MacroRulesNormalizedIdent,
298 interpolations: &'a FxHashMap<MacroRulesNormalizedIdent, NamedMatch>,
299 repeats: &[(usize, usize)],
300 ) -> Option<&'a NamedMatch> {
301 interpolations.get(&ident).map(|mut matched| {
302 for &(idx, _) in repeats {
304 MatchedTokenTree(_) | MatchedNonterminal(_) => break,
305 MatchedSeq(ads) => matched = ads.get(idx).unwrap(),
313 /// An accumulator over a TokenTree to be used with `fold`. During transcription, we need to make
314 /// sure that the size of each sequence and all of its nested sequences are the same as the sizes
315 /// of all the matched (nested) sequences in the macro invocation. If they don't match, somebody
316 /// has made a mistake (either the macro writer or caller).
318 enum LockstepIterSize {
319 /// No constraints on length of matcher. This is true for any TokenTree variants except a
320 /// `MetaVar` with an actual `MatchedSeq` (as opposed to a `MatchedNonterminal`).
323 /// A `MetaVar` with an actual `MatchedSeq`. The length of the match and the name of the
324 /// meta-var are returned.
325 Constraint(usize, MacroRulesNormalizedIdent),
327 /// Two `Constraint`s on the same sequence had different lengths. This is an error.
328 Contradiction(String),
331 impl LockstepIterSize {
332 /// Find incompatibilities in matcher/invocation sizes.
333 /// - `Unconstrained` is compatible with everything.
334 /// - `Contradiction` is incompatible with everything.
335 /// - `Constraint(len)` is only compatible with other constraints of the same length.
336 fn with(self, other: LockstepIterSize) -> LockstepIterSize {
338 LockstepIterSize::Unconstrained => other,
339 LockstepIterSize::Contradiction(_) => self,
340 LockstepIterSize::Constraint(l_len, l_id) => match other {
341 LockstepIterSize::Unconstrained => self,
342 LockstepIterSize::Contradiction(_) => other,
343 LockstepIterSize::Constraint(r_len, _) if l_len == r_len => self,
344 LockstepIterSize::Constraint(r_len, r_id) => {
346 "meta-variable `{}` repeats {} time{}, but `{}` repeats {} time{}",
354 LockstepIterSize::Contradiction(msg)
361 /// Given a `tree`, make sure that all sequences have the same length as the matches for the
362 /// appropriate meta-vars in `interpolations`.
364 /// Note that if `repeats` does not match the exact correct depth of a meta-var,
365 /// `lookup_cur_matched` will return `None`, which is why this still works even in the presence of
366 /// multiple nested matcher sequences.
368 /// Example: `$($($x $y)+*);+` -- we need to make sure that `x` and `y` repeat the same amount as
369 /// each other at the given depth when the macro was invoked. If they don't it might mean they were
370 /// declared at unequal depths or there was a compile bug. For example, if we have 3 repetitions of
371 /// the outer sequence and 4 repetitions of the inner sequence for `x`, we should have the same for
372 /// `y`; otherwise, we can't transcribe them both at the given depth.
373 fn lockstep_iter_size(
374 tree: &mbe::TokenTree,
375 interpolations: &FxHashMap<MacroRulesNormalizedIdent, NamedMatch>,
376 repeats: &[(usize, usize)],
377 ) -> LockstepIterSize {
380 TokenTree::Delimited(_, delimited) => {
381 delimited.tts.iter().fold(LockstepIterSize::Unconstrained, |size, tt| {
382 size.with(lockstep_iter_size(tt, interpolations, repeats))
385 TokenTree::Sequence(_, seq) => {
386 seq.tts.iter().fold(LockstepIterSize::Unconstrained, |size, tt| {
387 size.with(lockstep_iter_size(tt, interpolations, repeats))
390 TokenTree::MetaVar(_, name) | TokenTree::MetaVarDecl(_, name, _) => {
391 let name = MacroRulesNormalizedIdent::new(*name);
392 match lookup_cur_matched(name, interpolations, repeats) {
393 Some(matched) => match matched {
394 MatchedTokenTree(_) | MatchedNonterminal(_) => LockstepIterSize::Unconstrained,
395 MatchedSeq(ads) => LockstepIterSize::Constraint(ads.len(), name),
397 _ => LockstepIterSize::Unconstrained,
400 TokenTree::MetaVarExpr(_, expr) => {
401 let default_rslt = LockstepIterSize::Unconstrained;
402 let Some(ident) = expr.ident() else { return default_rslt; };
403 let name = MacroRulesNormalizedIdent::new(ident);
404 match lookup_cur_matched(name, interpolations, repeats) {
405 Some(MatchedSeq(ads)) => {
406 default_rslt.with(LockstepIterSize::Constraint(ads.len(), name))
411 TokenTree::Token(..) => LockstepIterSize::Unconstrained,
415 /// Used solely by the `count` meta-variable expression, counts the outer-most repetitions at a
416 /// given optional nested depth.
418 /// For example, a macro parameter of `$( { $( $foo:ident ),* } )*` called with `{ a, b } { c }`:
420 /// * `[ $( ${count(foo)} ),* ]` will return [2, 1] with a, b = 2 and c = 1
421 /// * `[ $( ${count(foo, 0)} ),* ]` will be the same as `[ $( ${count(foo)} ),* ]`
422 /// * `[ $( ${count(foo, 1)} ),* ]` will return an error because `${count(foo, 1)}` is
423 /// declared inside a single repetition and the index `1` implies two nested repetitions.
424 fn count_repetitions<'a>(
426 depth_opt: Option<usize>,
427 mut matched: &NamedMatch,
428 repeats: &[(usize, usize)],
430 ) -> PResult<'a, usize> {
431 // Recursively count the number of matches in `matched` at given depth
432 // (or at the top-level of `matched` if no depth is given).
435 declared_lhs_depth: usize,
436 depth_opt: Option<usize>,
437 matched: &NamedMatch,
439 ) -> PResult<'a, usize> {
441 MatchedTokenTree(_) | MatchedNonterminal(_) => {
442 if declared_lhs_depth == 0 {
443 return Err(cx.create_err(CountRepetitionMisplaced { span: sp.entire() }));
447 Some(_) => Err(out_of_bounds_err(cx, declared_lhs_depth, sp.entire(), "count")),
450 MatchedSeq(named_matches) => {
451 let new_declared_lhs_depth = declared_lhs_depth + 1;
453 None => named_matches
455 .map(|elem| count(cx, new_declared_lhs_depth, None, elem, sp))
457 Some(0) => Ok(named_matches.len()),
458 Some(depth) => named_matches
460 .map(|elem| count(cx, new_declared_lhs_depth, Some(depth - 1), elem, sp))
466 // `repeats` records all of the nested levels at which we are currently
467 // matching meta-variables. The meta-var-expr `count($x)` only counts
468 // matches that occur in this "subtree" of the `NamedMatch` where we
469 // are currently transcribing, so we need to descend to that subtree
470 // before we start counting. `matched` contains the various levels of the
471 // tree as we descend, and its final value is the subtree we are currently at.
472 for &(idx, _) in repeats {
473 if let MatchedSeq(ads) = matched {
477 count(cx, 0, depth_opt, matched, sp)
480 /// Returns a `NamedMatch` item declared on the LHS given an arbitrary [Ident]
481 fn matched_from_ident<'ctx, 'interp, 'rslt>(
484 interp: &'interp FxHashMap<MacroRulesNormalizedIdent, NamedMatch>,
485 ) -> PResult<'ctx, &'rslt NamedMatch>
489 let span = ident.span;
490 let key = MacroRulesNormalizedIdent::new(ident);
491 interp.get(&key).ok_or_else(|| cx.create_err(MetaVarExprUnrecognizedVar { span, key }))
494 /// Used by meta-variable expressions when an user input is out of the actual declared bounds. For
495 /// example, index(999999) in an repetition of only three elements.
496 fn out_of_bounds_err<'a>(
501 ) -> DiagnosticBuilder<'a, ErrorGuaranteed> {
502 let msg = if max == 0 {
504 "meta-variable expression `{ty}` with depth parameter \
505 must be called inside of a macro repetition"
509 "depth parameter on meta-variable expression `{ty}` \
510 must be less than {max}"
513 cx.struct_span_err(span, &msg)
516 fn transcribe_metavar_expr<'a>(
519 interp: &FxHashMap<MacroRulesNormalizedIdent, NamedMatch>,
521 repeats: &[(usize, usize)],
522 result: &mut Vec<TokenTree>,
524 ) -> PResult<'a, ()> {
525 let mut visited_span = || {
526 let mut span = sp.entire();
527 marker.visit_span(&mut span);
531 MetaVarExpr::Count(original_ident, depth_opt) => {
532 let matched = matched_from_ident(cx, original_ident, interp)?;
533 let count = count_repetitions(cx, depth_opt, matched, &repeats, sp)?;
534 let tt = TokenTree::token_alone(
535 TokenKind::lit(token::Integer, sym::integer(count), None),
540 MetaVarExpr::Ignore(original_ident) => {
541 // Used to ensure that `original_ident` is present in the LHS
542 let _ = matched_from_ident(cx, original_ident, interp)?;
544 MetaVarExpr::Index(depth) => match repeats.iter().nth_back(depth) {
545 Some((index, _)) => {
546 result.push(TokenTree::token_alone(
547 TokenKind::lit(token::Integer, sym::integer(*index), None),
551 None => return Err(out_of_bounds_err(cx, repeats.len(), sp.entire(), "index")),
553 MetaVarExpr::Length(depth) => match repeats.iter().nth_back(depth) {
554 Some((_, length)) => {
555 result.push(TokenTree::token_alone(
556 TokenKind::lit(token::Integer, sym::integer(*length), None),
560 None => return Err(out_of_bounds_err(cx, repeats.len(), sp.entire(), "length")),