1 use crate::base::ExtCtxt;
2 use crate::mbe::macro_parser::{MatchedNonterminal, MatchedSeq, MatchedTokenTree, NamedMatch};
3 use crate::mbe::{self, MetaVarExpr};
4 use rustc_ast::mut_visit::{self, MutVisitor};
5 use rustc_ast::token::{self, Delimiter, Token, TokenKind};
6 use rustc_ast::tokenstream::{DelimSpan, TokenStream, TokenTree, TreeAndSpacing};
7 use rustc_data_structures::fx::FxHashMap;
8 use rustc_errors::{pluralize, PResult};
9 use rustc_errors::{DiagnosticBuilder, ErrorGuaranteed};
10 use rustc_span::hygiene::{LocalExpnId, Transparency};
11 use rustc_span::symbol::{sym, Ident, MacroRulesNormalizedIdent};
14 use smallvec::{smallvec, SmallVec};
17 // A Marker adds the given mark to the syntax context.
18 struct Marker(LocalExpnId, Transparency);
20 impl MutVisitor for Marker {
21 const VISIT_TOKENS: bool = true;
23 fn visit_span(&mut self, span: &mut Span) {
24 *span = span.apply_mark(self.0.to_expn_id(), self.1)
28 /// An iterator over the token trees in a delimited token tree (`{ ... }`) or a sequence (`$(...)`).
30 Delimited { tts: &'a [mbe::TokenTree], idx: usize, delim: Delimiter, span: DelimSpan },
31 Sequence { tts: &'a [mbe::TokenTree], idx: usize, sep: Option<Token> },
35 /// Construct a new frame around the delimited set of tokens.
36 fn new(src: &'a mbe::Delimited, span: DelimSpan) -> Frame<'a> {
37 Frame::Delimited { tts: &src.tts, idx: 0, delim: src.delim, span }
41 impl<'a> Iterator for Frame<'a> {
42 type Item = &'a mbe::TokenTree;
44 fn next(&mut self) -> Option<&'a mbe::TokenTree> {
46 Frame::Delimited { tts, ref mut idx, .. }
47 | Frame::Sequence { tts, ref mut idx, .. } => {
48 let res = tts.get(*idx);
56 /// This can do Macro-By-Example transcription.
57 /// - `interp` is a map of meta-variables to the tokens (non-terminals) they matched in the
58 /// invocation. We are assuming we already know there is a match.
59 /// - `src` is the RHS of the MBE, that is, the "example" we are filling in.
64 /// macro_rules! foo {
65 /// ($id:ident) => { println!("{}", stringify!($id)); }
71 /// `interp` would contain `$id => bar` and `src` would contain `println!("{}", stringify!($id));`.
73 /// `transcribe` would return a `TokenStream` containing `println!("{}", stringify!(bar));`.
75 /// Along the way, we do some additional error checking.
76 pub(super) fn transcribe<'a>(
78 interp: &FxHashMap<MacroRulesNormalizedIdent, NamedMatch>,
81 transparency: Transparency,
82 ) -> PResult<'a, TokenStream> {
83 // Nothing for us to transcribe...
84 if src.tts.is_empty() {
85 return Ok(TokenStream::default());
88 // We descend into the RHS (`src`), expanding things as we go. This stack contains the things
89 // we have yet to expand/are still expanding. We start the stack off with the whole RHS.
90 let mut stack: SmallVec<[Frame<'_>; 1]> = smallvec![Frame::new(&src, src_span)];
92 // As we descend in the RHS, we will need to be able to match nested sequences of matchers.
93 // `repeats` keeps track of where we are in matching at each level, with the last element being
94 // the most deeply nested sequence. This is used as a stack.
95 let mut repeats = Vec::new();
97 // `result` contains resulting token stream from the TokenTree we just finished processing. At
98 // the end, this will contain the full result of transcription, but at arbitrary points during
99 // `transcribe`, `result` will contain subsets of the final result.
101 // Specifically, as we descend into each TokenTree, we will push the existing results onto the
102 // `result_stack` and clear `results`. We will then produce the results of transcribing the
103 // TokenTree into `results`. Then, as we unwind back out of the `TokenTree`, we will pop the
104 // `result_stack` and append `results` too it to produce the new `results` up to that point.
106 // Thus, if we try to pop the `result_stack` and it is empty, we have reached the top-level
107 // again, and we are done transcribing.
108 let mut result: Vec<TreeAndSpacing> = Vec::new();
109 let mut result_stack = Vec::new();
110 let mut marker = Marker(cx.current_expansion.id, transparency);
113 // Look at the last frame on the stack.
114 // If it still has a TokenTree we have not looked at yet, use that tree.
115 let Some(tree) = stack.last_mut().unwrap().next() else {
116 // This else-case never produces a value for `tree` (it `continue`s or `return`s).
118 // Otherwise, if we have just reached the end of a sequence and we can keep repeating,
119 // go back to the beginning of the sequence.
120 if let Frame::Sequence { idx, sep, .. } = stack.last_mut().unwrap() {
121 let (repeat_idx, repeat_len) = repeats.last_mut().unwrap();
123 if repeat_idx < repeat_len {
125 if let Some(sep) = sep {
126 result.push(TokenTree::Token(sep.clone()).into());
132 // We are done with the top of the stack. Pop it. Depending on what it was, we do
133 // different things. Note that the outermost item must be the delimited, wrapped RHS
134 // that was passed in originally to `transcribe`.
135 match stack.pop().unwrap() {
136 // Done with a sequence. Pop from repeats.
137 Frame::Sequence { .. } => {
141 // We are done processing a Delimited. If this is the top-level delimited, we are
142 // done. Otherwise, we unwind the result_stack to append what we have produced to
143 // any previous results.
144 Frame::Delimited { delim, span, .. } => {
145 if result_stack.is_empty() {
146 // No results left to compute! We are back at the top-level.
147 return Ok(TokenStream::new(result));
150 // Step back into the parent Delimited.
151 let tree = TokenTree::Delimited(span, delim, TokenStream::new(result));
152 result = result_stack.pop().unwrap();
153 result.push(tree.into());
159 // At this point, we know we are in the middle of a TokenTree (the last one on `stack`).
160 // `tree` contains the next `TokenTree` to be processed.
162 // We are descending into a sequence. We first make sure that the matchers in the RHS
163 // and the matches in `interp` have the same shape. Otherwise, either the caller or the
164 // macro writer has made a mistake.
165 seq @ mbe::TokenTree::Sequence(_, delimited) => {
166 match lockstep_iter_size(&seq, interp, &repeats) {
167 LockstepIterSize::Unconstrained => {
168 return Err(cx.struct_span_err(
169 seq.span(), /* blame macro writer */
170 "attempted to repeat an expression containing no syntax variables \
171 matched as repeating at this depth",
175 LockstepIterSize::Contradiction(msg) => {
176 // FIXME: this really ought to be caught at macro definition time... It
177 // happens when two meta-variables are used in the same repetition in a
178 // sequence, but they come from different sequence matchers and repeat
179 // different amounts.
180 return Err(cx.struct_span_err(seq.span(), &msg));
183 LockstepIterSize::Constraint(len, _) => {
184 // We do this to avoid an extra clone above. We know that this is a
186 let mbe::TokenTree::Sequence(sp, seq) = seq else {
190 // Is the repetition empty?
192 if seq.kleene.op == mbe::KleeneOp::OneOrMore {
193 // FIXME: this really ought to be caught at macro definition
194 // time... It happens when the Kleene operator in the matcher and
195 // the body for the same meta-variable do not match.
196 return Err(cx.struct_span_err(
198 "this must repeat at least once",
202 // 0 is the initial counter (we have done 0 repetitions so far). `len`
203 // is the total number of repetitions we should generate.
204 repeats.push((0, len));
206 // The first time we encounter the sequence we push it to the stack. It
207 // then gets reused (see the beginning of the loop) until we are done
209 stack.push(Frame::Sequence {
211 sep: seq.separator.clone(),
219 // Replace the meta-var with the matched token tree from the invocation.
220 mbe::TokenTree::MetaVar(mut sp, mut original_ident) => {
221 // Find the matched nonterminal from the macro invocation, and use it to replace
223 let ident = MacroRulesNormalizedIdent::new(original_ident);
224 if let Some(cur_matched) = lookup_cur_matched(ident, interp, &repeats) {
226 MatchedTokenTree(ref tt) => {
227 // `tt`s are emitted into the output stream directly as "raw tokens",
228 // without wrapping them into groups.
229 let token = tt.clone();
230 result.push(token.into());
232 MatchedNonterminal(ref nt) => {
233 // Other variables are emitted into the output stream as groups with
234 // `Delimiter::Invisible` to maintain parsing priorities.
235 // `Interpolated` is currently used for such groups in rustc parser.
236 marker.visit_span(&mut sp);
237 let token = TokenTree::token(token::Interpolated(nt.clone()), sp);
238 result.push(token.into());
241 // We were unable to descend far enough. This is an error.
242 return Err(cx.struct_span_err(
243 sp, /* blame the macro writer */
244 &format!("variable '{}' is still repeating at this depth", ident),
249 // If we aren't able to match the meta-var, we push it back into the result but
250 // with modified syntax context. (I believe this supports nested macros).
251 marker.visit_span(&mut sp);
252 marker.visit_ident(&mut original_ident);
253 result.push(TokenTree::token(token::Dollar, sp).into());
254 result.push(TokenTree::Token(Token::from_ast_ident(original_ident)).into());
258 // Replace meta-variable expressions with the result of their expansion.
259 mbe::TokenTree::MetaVarExpr(sp, expr) => {
260 transcribe_metavar_expr(cx, expr, interp, &mut marker, &repeats, &mut result, &sp)?;
263 // If we are entering a new delimiter, we push its contents to the `stack` to be
264 // processed, and we push all of the currently produced results to the `result_stack`.
265 // We will produce all of the results of the inside of the `Delimited` and then we will
266 // jump back out of the Delimited, pop the result_stack and add the new results back to
267 // the previous results (from outside the Delimited).
268 mbe::TokenTree::Delimited(mut span, delimited) => {
269 mut_visit::visit_delim_span(&mut span, &mut marker);
270 stack.push(Frame::Delimited {
272 delim: delimited.delim,
276 result_stack.push(mem::take(&mut result));
279 // Nothing much to do here. Just push the token to the result, being careful to
280 // preserve syntax context.
281 mbe::TokenTree::Token(token) => {
282 let mut token = token.clone();
283 mut_visit::visit_token(&mut token, &mut marker);
284 let tt = TokenTree::Token(token);
285 result.push(tt.into());
288 // There should be no meta-var declarations in the invocation of a macro.
289 mbe::TokenTree::MetaVarDecl(..) => panic!("unexpected `TokenTree::MetaVarDecl"),
294 /// Lookup the meta-var named `ident` and return the matched token tree from the invocation using
295 /// the set of matches `interpolations`.
297 /// See the definition of `repeats` in the `transcribe` function. `repeats` is used to descend
298 /// into the right place in nested matchers. If we attempt to descend too far, the macro writer has
299 /// made a mistake, and we return `None`.
300 fn lookup_cur_matched<'a>(
301 ident: MacroRulesNormalizedIdent,
302 interpolations: &'a FxHashMap<MacroRulesNormalizedIdent, NamedMatch>,
303 repeats: &[(usize, usize)],
304 ) -> Option<&'a NamedMatch> {
305 interpolations.get(&ident).map(|matched| {
306 let mut matched = matched;
307 for &(idx, _) in repeats {
309 MatchedTokenTree(_) | MatchedNonterminal(_) => break,
310 MatchedSeq(ref ads) => matched = ads.get(idx).unwrap(),
318 /// An accumulator over a TokenTree to be used with `fold`. During transcription, we need to make
319 /// sure that the size of each sequence and all of its nested sequences are the same as the sizes
320 /// of all the matched (nested) sequences in the macro invocation. If they don't match, somebody
321 /// has made a mistake (either the macro writer or caller).
323 enum LockstepIterSize {
324 /// No constraints on length of matcher. This is true for any TokenTree variants except a
325 /// `MetaVar` with an actual `MatchedSeq` (as opposed to a `MatchedNonterminal`).
328 /// A `MetaVar` with an actual `MatchedSeq`. The length of the match and the name of the
329 /// meta-var are returned.
330 Constraint(usize, MacroRulesNormalizedIdent),
332 /// Two `Constraint`s on the same sequence had different lengths. This is an error.
333 Contradiction(String),
336 impl LockstepIterSize {
337 /// Find incompatibilities in matcher/invocation sizes.
338 /// - `Unconstrained` is compatible with everything.
339 /// - `Contradiction` is incompatible with everything.
340 /// - `Constraint(len)` is only compatible with other constraints of the same length.
341 fn with(self, other: LockstepIterSize) -> LockstepIterSize {
343 LockstepIterSize::Unconstrained => other,
344 LockstepIterSize::Contradiction(_) => self,
345 LockstepIterSize::Constraint(l_len, ref l_id) => match other {
346 LockstepIterSize::Unconstrained => self,
347 LockstepIterSize::Contradiction(_) => other,
348 LockstepIterSize::Constraint(r_len, _) if l_len == r_len => self,
349 LockstepIterSize::Constraint(r_len, r_id) => {
351 "meta-variable `{}` repeats {} time{}, but `{}` repeats {} time{}",
359 LockstepIterSize::Contradiction(msg)
366 /// Given a `tree`, make sure that all sequences have the same length as the matches for the
367 /// appropriate meta-vars in `interpolations`.
369 /// Note that if `repeats` does not match the exact correct depth of a meta-var,
370 /// `lookup_cur_matched` will return `None`, which is why this still works even in the presence of
371 /// multiple nested matcher sequences.
373 /// Example: `$($($x $y)+*);+` -- we need to make sure that `x` and `y` repeat the same amount as
374 /// each other at the given depth when the macro was invoked. If they don't it might mean they were
375 /// declared at unequal depths or there was a compile bug. For example, if we have 3 repetitions of
376 /// the outer sequence and 4 repetitions of the inner sequence for `x`, we should have the same for
377 /// `y`; otherwise, we can't transcribe them both at the given depth.
378 fn lockstep_iter_size(
379 tree: &mbe::TokenTree,
380 interpolations: &FxHashMap<MacroRulesNormalizedIdent, NamedMatch>,
381 repeats: &[(usize, usize)],
382 ) -> LockstepIterSize {
385 TokenTree::Delimited(_, ref delimited) => {
386 delimited.tts.iter().fold(LockstepIterSize::Unconstrained, |size, tt| {
387 size.with(lockstep_iter_size(tt, interpolations, repeats))
390 TokenTree::Sequence(_, ref seq) => {
391 seq.tts.iter().fold(LockstepIterSize::Unconstrained, |size, tt| {
392 size.with(lockstep_iter_size(tt, interpolations, repeats))
395 TokenTree::MetaVar(_, name) | TokenTree::MetaVarDecl(_, name, _) => {
396 let name = MacroRulesNormalizedIdent::new(name);
397 match lookup_cur_matched(name, interpolations, repeats) {
398 Some(matched) => match matched {
399 MatchedTokenTree(_) | MatchedNonterminal(_) => LockstepIterSize::Unconstrained,
400 MatchedSeq(ref ads) => LockstepIterSize::Constraint(ads.len(), name),
402 _ => LockstepIterSize::Unconstrained,
405 TokenTree::MetaVarExpr(_, ref expr) => {
406 let default_rslt = LockstepIterSize::Unconstrained;
407 let Some(ident) = expr.ident() else { return default_rslt; };
408 let name = MacroRulesNormalizedIdent::new(ident);
409 match lookup_cur_matched(name, interpolations, repeats) {
410 Some(MatchedSeq(ref ads)) => {
411 default_rslt.with(LockstepIterSize::Constraint(ads.len(), name))
416 TokenTree::Token(..) => LockstepIterSize::Unconstrained,
420 /// Used solely by the `count` meta-variable expression, counts the outer-most repetitions at a
421 /// given optional nested depth.
423 /// For example, a macro parameter of `$( { $( $foo:ident ),* } )*` called with `{ a, b } { c }`:
425 /// * `[ $( ${count(foo)} ),* ]` will return [2, 1] with a, b = 2 and c = 1
426 /// * `[ $( ${count(foo, 0)} ),* ]` will be the same as `[ $( ${count(foo)} ),* ]`
427 /// * `[ $( ${count(foo, 1)} ),* ]` will return an error because `${count(foo, 1)}` is
428 /// declared inside a single repetition and the index `1` implies two nested repetitions.
429 fn count_repetitions<'a>(
431 depth_opt: Option<usize>,
432 mut matched: &NamedMatch,
433 repeats: &[(usize, usize)],
435 ) -> PResult<'a, usize> {
436 // Recursively count the number of matches in `matched` at given depth
437 // (or at the top-level of `matched` if no depth is given).
440 declared_lhs_depth: usize,
441 depth_opt: Option<usize>,
442 matched: &NamedMatch,
444 ) -> PResult<'a, usize> {
446 MatchedTokenTree(_) | MatchedNonterminal(_) => {
447 if declared_lhs_depth == 0 {
448 return Err(cx.struct_span_err(
450 "`count` can not be placed inside the inner-most repetition",
455 Some(_) => Err(out_of_bounds_err(cx, declared_lhs_depth, sp.entire(), "count")),
458 MatchedSeq(ref named_matches) => {
459 let new_declared_lhs_depth = declared_lhs_depth + 1;
461 None => named_matches
463 .map(|elem| count(cx, new_declared_lhs_depth, None, elem, sp))
465 Some(0) => Ok(named_matches.len()),
466 Some(depth) => named_matches
468 .map(|elem| count(cx, new_declared_lhs_depth, Some(depth - 1), elem, sp))
474 // `repeats` records all of the nested levels at which we are currently
475 // matching meta-variables. The meta-var-expr `count($x)` only counts
476 // matches that occur in this "subtree" of the `NamedMatch` where we
477 // are currently transcribing, so we need to descend to that subtree
478 // before we start counting. `matched` contains the various levels of the
479 // tree as we descend, and its final value is the subtree we are currently at.
480 for &(idx, _) in repeats {
481 if let MatchedSeq(ref ads) = matched {
485 count(cx, 0, depth_opt, matched, sp)
488 /// Returns a `NamedMatch` item declared on the LHS given an arbitrary [Ident]
489 fn matched_from_ident<'ctx, 'interp, 'rslt>(
492 interp: &'interp FxHashMap<MacroRulesNormalizedIdent, NamedMatch>,
493 ) -> PResult<'ctx, &'rslt NamedMatch>
497 let span = ident.span;
498 let key = MacroRulesNormalizedIdent::new(ident);
499 interp.get(&key).ok_or_else(|| {
502 &format!("variable `{}` is not recognized in meta-variable expression", key),
507 /// Used by meta-variable expressions when an user input is out of the actual declared bounds. For
508 /// example, index(999999) in an repetition of only three elements.
509 fn out_of_bounds_err<'a>(
514 ) -> DiagnosticBuilder<'a, ErrorGuaranteed> {
515 cx.struct_span_err(span, &format!("{ty} depth must be less than {max}"))
518 fn transcribe_metavar_expr<'a>(
521 interp: &FxHashMap<MacroRulesNormalizedIdent, NamedMatch>,
523 repeats: &[(usize, usize)],
524 result: &mut Vec<TreeAndSpacing>,
526 ) -> PResult<'a, ()> {
527 let mut visited_span = || {
528 let mut span = sp.entire();
529 marker.visit_span(&mut span);
533 MetaVarExpr::Count(original_ident, depth_opt) => {
534 let matched = matched_from_ident(cx, original_ident, interp)?;
535 let count = count_repetitions(cx, depth_opt, matched, &repeats, sp)?;
536 let tt = TokenTree::token(
537 TokenKind::lit(token::Integer, sym::integer(count), None),
540 result.push(tt.into());
542 MetaVarExpr::Ignore(original_ident) => {
543 // Used to ensure that `original_ident` is present in the LHS
544 let _ = matched_from_ident(cx, original_ident, interp)?;
546 MetaVarExpr::Index(depth) => match repeats.iter().nth_back(depth) {
547 Some((index, _)) => {
550 TokenKind::lit(token::Integer, sym::integer(*index), None),
556 None => return Err(out_of_bounds_err(cx, repeats.len(), sp.entire(), "index")),
558 MetaVarExpr::Length(depth) => match repeats.iter().nth_back(depth) {
559 Some((_, length)) => {
562 TokenKind::lit(token::Integer, sym::integer(*length), None),
568 None => return Err(out_of_bounds_err(cx, repeats.len(), sp.entire(), "length")),