rustc_expand/mbe/quoted.rs
1use rustc_ast::token::{self, Delimiter, IdentIsRaw, NonterminalKind, Token};
2use rustc_ast::tokenstream::TokenStreamIter;
3use rustc_ast::{NodeId, tokenstream};
4use rustc_ast_pretty::pprust;
5use rustc_feature::Features;
6use rustc_session::Session;
7use rustc_session::parse::feature_err;
8use rustc_span::edition::Edition;
9use rustc_span::{Ident, Span, kw, sym};
10
11use crate::errors;
12use crate::mbe::macro_parser::count_metavar_decls;
13use crate::mbe::{Delimited, KleeneOp, KleeneToken, MetaVarExpr, SequenceRepetition, TokenTree};
14
15pub(crate) const VALID_FRAGMENT_NAMES_MSG: &str = "valid fragment specifiers are \
16 `ident`, `block`, `stmt`, `expr`, `pat`, `ty`, `lifetime`, `literal`, `path`, \
17 `meta`, `tt`, `item` and `vis`, along with `expr_2021` and `pat_param` for edition compatibility";
18
19/// Which part of a macro rule we're parsing
20#[derive(Copy, Clone)]
21pub(crate) enum RulePart {
22 /// The left-hand side, with patterns and metavar definitions with types
23 Pattern,
24 /// The right-hand side body, with metavar references and metavar expressions
25 Body,
26}
27
28impl RulePart {
29 #[inline(always)]
30 fn is_pattern(&self) -> bool {
31 matches!(self, Self::Pattern)
32 }
33
34 #[inline(always)]
35 fn is_body(&self) -> bool {
36 matches!(self, Self::Body)
37 }
38}
39
40/// Takes a `tokenstream::TokenStream` and returns a `Vec<self::TokenTree>`. Specifically, this
41/// takes a generic `TokenStream`, such as is used in the rest of the compiler, and returns a
42/// collection of `TokenTree` for use in parsing a macro.
43///
44/// # Parameters
45///
46/// - `input`: a token stream to read from, the contents of which we are parsing.
47/// - `part`: whether we're parsing the patterns or the body of a macro. Both take roughly the same
48/// form _except_ that:
49/// - In a pattern, metavars are declared with their "matcher" type. For example `$var:expr` or
50/// `$id:ident`. In this example, `expr` and `ident` are "matchers". They are not present in the
51/// body of a macro rule -- just in the pattern.
52/// - Metavariable expressions are only valid in the "body", not the "pattern".
53/// - `sess`: the parsing session. Any errors will be emitted to this session.
54/// - `node_id`: the NodeId of the macro we are parsing.
55/// - `features`: language features so we can do feature gating.
56///
57/// # Returns
58///
59/// A collection of `self::TokenTree`. There may also be some errors emitted to `sess`.
60fn parse(
61 input: &tokenstream::TokenStream,
62 part: RulePart,
63 sess: &Session,
64 node_id: NodeId,
65 features: &Features,
66 edition: Edition,
67) -> Vec<TokenTree> {
68 // Will contain the final collection of `self::TokenTree`
69 let mut result = Vec::new();
70
71 // For each token tree in `input`, parse the token into a `self::TokenTree`, consuming
72 // additional trees if need be.
73 let mut iter = input.iter();
74 while let Some(tree) = iter.next() {
75 // Given the parsed tree, if there is a metavar and we are expecting matchers, actually
76 // parse out the matcher (i.e., in `$id:ident` this would parse the `:` and `ident`).
77 let tree = parse_tree(tree, &mut iter, part, sess, node_id, features, edition);
78
79 if part.is_body() {
80 // No matchers allowed, nothing to process here
81 result.push(tree);
82 continue;
83 }
84
85 let TokenTree::MetaVar(start_sp, ident) = tree else {
86 // Not a metavariable, just return the tree
87 result.push(tree);
88 continue;
89 };
90
91 // Push a metavariable with no fragment specifier at the given span
92 let mut missing_fragment_specifier = |span| {
93 sess.dcx().emit_err(errors::MissingFragmentSpecifier {
94 span,
95 add_span: span.shrink_to_hi(),
96 valid: VALID_FRAGMENT_NAMES_MSG,
97 });
98
99 // Fall back to a `TokenTree` since that will match anything if we continue expanding.
100 result.push(TokenTree::MetaVarDecl { span, name: ident, kind: NonterminalKind::TT });
101 };
102
103 // Not consuming the next token immediately, as it may not be a colon
104 if let Some(peek) = iter.peek()
105 && let tokenstream::TokenTree::Token(token, _spacing) = peek
106 && let Token { kind: token::Colon, span: colon_span } = token
107 {
108 // Next token is a colon; consume it
109 iter.next();
110
111 // It's ok to consume the next tree no matter how,
112 // since if it's not a token then it will be an invalid declaration.
113 let Some(tokenstream::TokenTree::Token(token, _)) = iter.next() else {
114 // Invalid, return a nice source location as `var:`
115 missing_fragment_specifier(colon_span.with_lo(start_sp.lo()));
116 continue;
117 };
118
119 let Some((fragment, _)) = token.ident() else {
120 // No identifier for the fragment specifier;
121 missing_fragment_specifier(token.span);
122 continue;
123 };
124
125 let span = token.span.with_lo(start_sp.lo());
126 let edition = || {
127 // FIXME(#85708) - once we properly decode a foreign
128 // crate's `SyntaxContext::root`, then we can replace
129 // this with just `span.edition()`. A
130 // `SyntaxContext::root()` from the current crate will
131 // have the edition of the current crate, and a
132 // `SyntaxContext::root()` from a foreign crate will
133 // have the edition of that crate (which we manually
134 // retrieve via the `edition` parameter).
135 if !span.from_expansion() { edition } else { span.edition() }
136 };
137 let kind = NonterminalKind::from_symbol(fragment.name, edition).unwrap_or_else(|| {
138 sess.dcx().emit_err(errors::InvalidFragmentSpecifier {
139 span,
140 fragment,
141 help: VALID_FRAGMENT_NAMES_MSG,
142 });
143 NonterminalKind::TT
144 });
145 result.push(TokenTree::MetaVarDecl { span, name: ident, kind });
146 } else {
147 // Whether it's none or some other tree, it doesn't belong to
148 // the current meta variable, returning the original span.
149 missing_fragment_specifier(start_sp);
150 }
151 }
152 result
153}
154
155/// Takes a `tokenstream::TokenTree` and returns a `self::TokenTree`. Like `parse`, but for a
156/// single token tree. Emits errors to `sess` if needed.
157#[inline]
158pub(super) fn parse_one_tt(
159 input: tokenstream::TokenTree,
160 part: RulePart,
161 sess: &Session,
162 node_id: NodeId,
163 features: &Features,
164 edition: Edition,
165) -> TokenTree {
166 parse(&tokenstream::TokenStream::new(vec![input]), part, sess, node_id, features, edition)
167 .pop()
168 .unwrap()
169}
170
171/// Asks for the `macro_metavar_expr` feature if it is not enabled
172fn maybe_emit_macro_metavar_expr_feature(features: &Features, sess: &Session, span: Span) {
173 if !features.macro_metavar_expr() {
174 let msg = "meta-variable expressions are unstable";
175 feature_err(sess, sym::macro_metavar_expr, span, msg).emit();
176 }
177}
178
179fn maybe_emit_macro_metavar_expr_concat_feature(features: &Features, sess: &Session, span: Span) {
180 if !features.macro_metavar_expr_concat() {
181 let msg = "the `concat` meta-variable expression is unstable";
182 feature_err(sess, sym::macro_metavar_expr_concat, span, msg).emit();
183 }
184}
185
186/// Takes a `tokenstream::TokenTree` and returns a `self::TokenTree`. Specifically, this takes a
187/// generic `TokenTree`, such as is used in the rest of the compiler, and returns a `TokenTree`
188/// for use in parsing a macro.
189///
190/// Converting the given tree may involve reading more tokens.
191///
192/// # Parameters
193///
194/// - `tree`: the tree we wish to convert.
195/// - `outer_iter`: an iterator over trees. We may need to read more tokens from it in order to finish
196/// converting `tree`
197/// - `part`: same as [parse].
198/// - `sess`: the parsing session. Any errors will be emitted to this session.
199/// - `features`: language features so we can do feature gating.
200fn parse_tree<'a>(
201 tree: &'a tokenstream::TokenTree,
202 outer_iter: &mut TokenStreamIter<'a>,
203 part: RulePart,
204 sess: &Session,
205 node_id: NodeId,
206 features: &Features,
207 edition: Edition,
208) -> TokenTree {
209 // Depending on what `tree` is, we could be parsing different parts of a macro
210 match tree {
211 // `tree` is a `$` token. Look at the next token in `trees`
212 &tokenstream::TokenTree::Token(Token { kind: token::Dollar, span: dollar_span }, _) => {
213 // FIXME: Handle `Invisible`-delimited groups in a more systematic way
214 // during parsing.
215 let mut next = outer_iter.next();
216 let mut iter_storage;
217 let mut iter: &mut TokenStreamIter<'_> = match next {
218 Some(tokenstream::TokenTree::Delimited(.., delim, tts)) if delim.skip() => {
219 iter_storage = tts.iter();
220 next = iter_storage.next();
221 &mut iter_storage
222 }
223 _ => outer_iter,
224 };
225
226 match next {
227 // `tree` is followed by a delimited set of token trees.
228 Some(&tokenstream::TokenTree::Delimited(delim_span, _, delim, ref tts)) => {
229 if part.is_pattern() {
230 if delim != Delimiter::Parenthesis {
231 span_dollar_dollar_or_metavar_in_the_lhs_err(
232 sess,
233 &Token {
234 kind: delim.as_open_token_kind(),
235 span: delim_span.entire(),
236 },
237 );
238 }
239 } else {
240 match delim {
241 Delimiter::Brace => {
242 // The delimiter is `{`. This indicates the beginning
243 // of a meta-variable expression (e.g. `${count(ident)}`).
244 // Try to parse the meta-variable expression.
245 match MetaVarExpr::parse(tts, delim_span.entire(), &sess.psess) {
246 Err(err) => {
247 err.emit();
248 // Returns early the same read `$` to avoid spanning
249 // unrelated diagnostics that could be performed afterwards
250 return TokenTree::token(token::Dollar, dollar_span);
251 }
252 Ok(elem) => {
253 if let MetaVarExpr::Concat(_) = elem {
254 maybe_emit_macro_metavar_expr_concat_feature(
255 features,
256 sess,
257 delim_span.entire(),
258 );
259 } else {
260 maybe_emit_macro_metavar_expr_feature(
261 features,
262 sess,
263 delim_span.entire(),
264 );
265 }
266 return TokenTree::MetaVarExpr(delim_span, elem);
267 }
268 }
269 }
270 Delimiter::Parenthesis => {}
271 _ => {
272 let token =
273 pprust::token_kind_to_string(&delim.as_open_token_kind());
274 sess.dcx().emit_err(errors::ExpectedParenOrBrace {
275 span: delim_span.entire(),
276 token,
277 });
278 }
279 }
280 }
281 // If we didn't find a metavar expression above, then we must have a
282 // repetition sequence in the macro (e.g. `$(pat)*`). Parse the
283 // contents of the sequence itself
284 let sequence = parse(tts, part, sess, node_id, features, edition);
285 // Get the Kleene operator and optional separator
286 let (separator, kleene) =
287 parse_sep_and_kleene_op(&mut iter, delim_span.entire(), sess);
288 // Count the number of captured "names" (i.e., named metavars)
289 let num_captures =
290 if part.is_pattern() { count_metavar_decls(&sequence) } else { 0 };
291 TokenTree::Sequence(
292 delim_span,
293 SequenceRepetition { tts: sequence, separator, kleene, num_captures },
294 )
295 }
296
297 // `tree` is followed by an `ident`. This could be `$meta_var` or the `$crate`
298 // special metavariable that names the crate of the invocation.
299 Some(tokenstream::TokenTree::Token(token, _)) if token.is_ident() => {
300 let (ident, is_raw) = token.ident().unwrap();
301 let span = ident.span.with_lo(dollar_span.lo());
302 if ident.name == kw::Crate && matches!(is_raw, IdentIsRaw::No) {
303 TokenTree::token(token::Ident(kw::DollarCrate, is_raw), span)
304 } else {
305 TokenTree::MetaVar(span, ident)
306 }
307 }
308
309 // `tree` is followed by another `$`. This is an escaped `$`.
310 Some(&tokenstream::TokenTree::Token(
311 Token { kind: token::Dollar, span: dollar_span2 },
312 _,
313 )) => {
314 if part.is_pattern() {
315 span_dollar_dollar_or_metavar_in_the_lhs_err(
316 sess,
317 &Token { kind: token::Dollar, span: dollar_span2 },
318 );
319 } else {
320 maybe_emit_macro_metavar_expr_feature(features, sess, dollar_span2);
321 }
322 TokenTree::token(token::Dollar, dollar_span2)
323 }
324
325 // `tree` is followed by some other token. This is an error.
326 Some(tokenstream::TokenTree::Token(token, _)) => {
327 let msg =
328 format!("expected identifier, found `{}`", pprust::token_to_string(token),);
329 sess.dcx().span_err(token.span, msg);
330 TokenTree::MetaVar(token.span, Ident::dummy())
331 }
332
333 // There are no more tokens. Just return the `$` we already have.
334 None => TokenTree::token(token::Dollar, dollar_span),
335 }
336 }
337
338 // `tree` is an arbitrary token. Keep it.
339 tokenstream::TokenTree::Token(token, _) => TokenTree::Token(*token),
340
341 // `tree` is the beginning of a delimited set of tokens (e.g., `(` or `{`). We need to
342 // descend into the delimited set and further parse it.
343 &tokenstream::TokenTree::Delimited(span, spacing, delim, ref tts) => TokenTree::Delimited(
344 span,
345 spacing,
346 Delimited { delim, tts: parse(tts, part, sess, node_id, features, edition) },
347 ),
348 }
349}
350
351/// Takes a token and returns `Some(KleeneOp)` if the token is `+` `*` or `?`. Otherwise, return
352/// `None`.
353fn kleene_op(token: &Token) -> Option<KleeneOp> {
354 match token.kind {
355 token::Star => Some(KleeneOp::ZeroOrMore),
356 token::Plus => Some(KleeneOp::OneOrMore),
357 token::Question => Some(KleeneOp::ZeroOrOne),
358 _ => None,
359 }
360}
361
362/// Parse the next token tree of the input looking for a KleeneOp. Returns
363///
364/// - Ok(Ok((op, span))) if the next token tree is a KleeneOp
365/// - Ok(Err(tok, span)) if the next token tree is a token but not a KleeneOp
366/// - Err(span) if the next token tree is not a token
367fn parse_kleene_op(
368 iter: &mut TokenStreamIter<'_>,
369 span: Span,
370) -> Result<Result<(KleeneOp, Span), Token>, Span> {
371 match iter.next() {
372 Some(tokenstream::TokenTree::Token(token, _)) => match kleene_op(token) {
373 Some(op) => Ok(Ok((op, token.span))),
374 None => Ok(Err(*token)),
375 },
376 tree => Err(tree.map_or(span, tokenstream::TokenTree::span)),
377 }
378}
379
380/// Attempt to parse a single Kleene star, possibly with a separator.
381///
382/// For example, in a pattern such as `$(a),*`, `a` is the pattern to be repeated, `,` is the
383/// separator, and `*` is the Kleene operator. This function is specifically concerned with parsing
384/// the last two tokens of such a pattern: namely, the optional separator and the Kleene operator
385/// itself. Note that here we are parsing the _macro_ itself, rather than trying to match some
386/// stream of tokens in an invocation of a macro.
387///
388/// This function will take some input iterator `iter` corresponding to `span` and a parsing
389/// session `sess`. If the next one (or possibly two) tokens in `iter` correspond to a Kleene
390/// operator and separator, then a tuple with `(separator, KleeneOp)` is returned. Otherwise, an
391/// error with the appropriate span is emitted to `sess` and a dummy value is returned.
392fn parse_sep_and_kleene_op(
393 iter: &mut TokenStreamIter<'_>,
394 span: Span,
395 sess: &Session,
396) -> (Option<Token>, KleeneToken) {
397 // We basically look at two token trees here, denoted as #1 and #2 below
398 let span = match parse_kleene_op(iter, span) {
399 // #1 is a `?`, `+`, or `*` KleeneOp
400 Ok(Ok((op, span))) => return (None, KleeneToken::new(op, span)),
401
402 // #1 is a separator followed by #2, a KleeneOp
403 Ok(Err(token)) => match parse_kleene_op(iter, token.span) {
404 // #2 is the `?` Kleene op, which does not take a separator (error)
405 Ok(Ok((KleeneOp::ZeroOrOne, span))) => {
406 // Error!
407 sess.dcx().span_err(
408 token.span,
409 "the `?` macro repetition operator does not take a separator",
410 );
411
412 // Return a dummy
413 return (None, KleeneToken::new(KleeneOp::ZeroOrMore, span));
414 }
415
416 // #2 is a KleeneOp :D
417 Ok(Ok((op, span))) => return (Some(token), KleeneToken::new(op, span)),
418
419 // #2 is a random token or not a token at all :(
420 Ok(Err(Token { span, .. })) | Err(span) => span,
421 },
422
423 // #1 is not a token
424 Err(span) => span,
425 };
426
427 // If we ever get to this point, we have experienced an "unexpected token" error
428 sess.dcx().span_err(span, "expected one of: `*`, `+`, or `?`");
429
430 // Return a dummy
431 (None, KleeneToken::new(KleeneOp::ZeroOrMore, span))
432}
433
434// `$$` or a meta-variable is the lhs of a macro but shouldn't.
435//
436// For example, `macro_rules! foo { ( ${len()} ) => {} }`
437fn span_dollar_dollar_or_metavar_in_the_lhs_err(sess: &Session, token: &Token) {
438 sess.dcx()
439 .span_err(token.span, format!("unexpected token: {}", pprust::token_to_string(token)));
440 sess.dcx().span_note(
441 token.span,
442 "`$$` and meta-variable expressions are not allowed inside macro parameter definitions",
443 );
444}