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