]>
Commit | Line | Data |
---|---|---|
e74abb32 | 1 | use crate::mbe::macro_parser; |
dfeec247 | 2 | use crate::mbe::{Delimited, KleeneOp, KleeneToken, SequenceRepetition, TokenTree}; |
9fa01778 | 3 | |
74b04a01 XL |
4 | use rustc_ast::token::{self, Token}; |
5 | use rustc_ast::tokenstream; | |
b9856134 | 6 | use rustc_ast::{NodeId, DUMMY_NODE_ID}; |
74b04a01 | 7 | use rustc_ast_pretty::pprust; |
5869c6ff | 8 | use rustc_feature::Features; |
cdc7bbd5 XL |
9 | use rustc_session::parse::ParseSess; |
10 | use rustc_span::symbol::{kw, Ident}; | |
8bb4bdeb | 11 | |
17df50a5 XL |
12 | use rustc_span::edition::Edition; |
13 | use rustc_span::{Span, SyntaxContext}; | |
416331ca | 14 | |
e74abb32 | 15 | use rustc_data_structures::sync::Lrc; |
8bb4bdeb | 16 | |
3dfed10e XL |
17 | const VALID_FRAGMENT_NAMES_MSG: &str = "valid fragment specifiers are \ |
18 | `ident`, `block`, `stmt`, `expr`, `pat`, `ty`, `lifetime`, \ | |
19 | `literal`, `path`, `meta`, `tt`, `item` and `vis`"; | |
20 | ||
2c00a5a8 XL |
21 | /// Takes a `tokenstream::TokenStream` and returns a `Vec<self::TokenTree>`. Specifically, this |
22 | /// takes a generic `TokenStream`, such as is used in the rest of the compiler, and returns a | |
23 | /// collection of `TokenTree` for use in parsing a macro. | |
24 | /// | |
25 | /// # Parameters | |
26 | /// | |
27 | /// - `input`: a token stream to read from, the contents of which we are parsing. | |
28 | /// - `expect_matchers`: `parse` can be used to parse either the "patterns" or the "body" of a | |
29 | /// macro. Both take roughly the same form _except_ that in a pattern, metavars are declared with | |
30 | /// their "matcher" type. For example `$var:expr` or `$id:ident`. In this example, `expr` and | |
31 | /// `ident` are "matchers". They are not present in the body of a macro rule -- just in the | |
32 | /// pattern, so we pass a parameter to indicate whether to expect them or not. | |
33 | /// - `sess`: the parsing session. Any errors will be emitted to this session. | |
5869c6ff XL |
34 | /// - `node_id`: the NodeId of the macro we are parsing. |
35 | /// - `features`: language features so we can do feature gating. | |
17df50a5 | 36 | /// - `edition`: the edition of the crate defining the macro |
2c00a5a8 XL |
37 | /// |
38 | /// # Returns | |
39 | /// | |
40 | /// A collection of `self::TokenTree`. There may also be some errors emitted to `sess`. | |
e74abb32 | 41 | pub(super) fn parse( |
2c00a5a8 XL |
42 | input: tokenstream::TokenStream, |
43 | expect_matchers: bool, | |
44 | sess: &ParseSess, | |
f035d41b | 45 | node_id: NodeId, |
5869c6ff | 46 | features: &Features, |
17df50a5 | 47 | edition: Edition, |
2c00a5a8 XL |
48 | ) -> Vec<TokenTree> { |
49 | // Will contain the final collection of `self::TokenTree` | |
8bb4bdeb | 50 | let mut result = Vec::new(); |
2c00a5a8 XL |
51 | |
52 | // For each token tree in `input`, parse the token into a `self::TokenTree`, consuming | |
53 | // additional trees if need be. | |
e74abb32 | 54 | let mut trees = input.trees(); |
8bb4bdeb | 55 | while let Some(tree) = trees.next() { |
2c00a5a8 | 56 | // Given the parsed tree, if there is a metavar and we are expecting matchers, actually |
0731742a | 57 | // parse out the matcher (i.e., in `$id:ident` this would parse the `:` and `ident`). |
17df50a5 | 58 | let tree = parse_tree(tree, &mut trees, expect_matchers, sess, node_id, features, edition); |
8bb4bdeb | 59 | match tree { |
041b39d2 | 60 | TokenTree::MetaVar(start_sp, ident) if expect_matchers => { |
8bb4bdeb | 61 | let span = match trees.next() { |
dc9dc135 XL |
62 | Some(tokenstream::TokenTree::Token(Token { kind: token::Colon, span })) => { |
63 | match trees.next() { | |
64 | Some(tokenstream::TokenTree::Token(token)) => match token.ident() { | |
3dfed10e | 65 | Some((frag, _)) => { |
dc9dc135 | 66 | let span = token.span.with_lo(start_sp.lo()); |
5869c6ff | 67 | |
5869c6ff XL |
68 | let kind = |
69 | token::NonterminalKind::from_symbol(frag.name, || { | |
17df50a5 XL |
70 | // FIXME(#85708) - once we properly decode a foreign |
71 | // crate's `SyntaxContext::root`, then we can replace | |
72 | // this with just `span.edition()`. A | |
73 | // `SyntaxContext::root()` from the current crate will | |
74 | // have the edition of the current crate, and a | |
75 | // `SyntaxxContext::root()` from a foreign crate will | |
76 | // have the edition of that crate (which we manually | |
77 | // retrieve via the `edition` parameter). | |
78 | if span.ctxt() == SyntaxContext::root() { | |
79 | edition | |
80 | } else { | |
81 | span.edition() | |
82 | } | |
5869c6ff XL |
83 | }) |
84 | .unwrap_or_else( | |
85 | || { | |
86 | let msg = format!( | |
87 | "invalid fragment specifier `{}`", | |
88 | frag.name | |
89 | ); | |
90 | sess.span_diagnostic | |
91 | .struct_span_err(span, &msg) | |
92 | .help(VALID_FRAGMENT_NAMES_MSG) | |
93 | .emit(); | |
94 | token::NonterminalKind::Ident | |
95 | }, | |
96 | ); | |
b9856134 | 97 | result.push(TokenTree::MetaVarDecl(span, ident, Some(kind))); |
dc9dc135 XL |
98 | continue; |
99 | } | |
100 | _ => token.span, | |
101 | }, | |
5869c6ff | 102 | tree => tree.as_ref().map_or(span, tokenstream::TokenTree::span), |
dc9dc135 XL |
103 | } |
104 | } | |
5869c6ff | 105 | tree => tree.as_ref().map_or(start_sp, tokenstream::TokenTree::span), |
8bb4bdeb | 106 | }; |
b9856134 XL |
107 | if node_id != DUMMY_NODE_ID { |
108 | // Macros loaded from other crates have dummy node ids. | |
109 | sess.missing_fragment_specifiers.borrow_mut().insert(span, node_id); | |
110 | } | |
111 | result.push(TokenTree::MetaVarDecl(span, ident, None)); | |
8bb4bdeb | 112 | } |
2c00a5a8 XL |
113 | |
114 | // Not a metavar or no matchers allowed, so just return the tree | |
8bb4bdeb XL |
115 | _ => result.push(tree), |
116 | } | |
117 | } | |
118 | result | |
119 | } | |
120 | ||
2c00a5a8 XL |
121 | /// Takes a `tokenstream::TokenTree` and returns a `self::TokenTree`. Specifically, this takes a |
122 | /// generic `TokenTree`, such as is used in the rest of the compiler, and returns a `TokenTree` | |
123 | /// for use in parsing a macro. | |
124 | /// | |
125 | /// Converting the given tree may involve reading more tokens. | |
126 | /// | |
127 | /// # Parameters | |
128 | /// | |
129 | /// - `tree`: the tree we wish to convert. | |
f035d41b | 130 | /// - `outer_trees`: an iterator over trees. We may need to read more tokens from it in order to finish |
2c00a5a8 XL |
131 | /// converting `tree` |
132 | /// - `expect_matchers`: same as for `parse` (see above). | |
133 | /// - `sess`: the parsing session. Any errors will be emitted to this session. | |
5869c6ff | 134 | /// - `features`: language features so we can do feature gating. |
17df50a5 | 135 | /// - `edition` - the edition of the crate defining the macro |
dc9dc135 | 136 | fn parse_tree( |
2c00a5a8 | 137 | tree: tokenstream::TokenTree, |
f035d41b | 138 | outer_trees: &mut impl Iterator<Item = tokenstream::TokenTree>, |
2c00a5a8 XL |
139 | expect_matchers: bool, |
140 | sess: &ParseSess, | |
f035d41b | 141 | node_id: NodeId, |
5869c6ff | 142 | features: &Features, |
17df50a5 | 143 | edition: Edition, |
dc9dc135 | 144 | ) -> TokenTree { |
2c00a5a8 | 145 | // Depending on what `tree` is, we could be parsing different parts of a macro |
8bb4bdeb | 146 | match tree { |
2c00a5a8 | 147 | // `tree` is a `$` token. Look at the next token in `trees` |
f035d41b XL |
148 | tokenstream::TokenTree::Token(Token { kind: token::Dollar, span }) => { |
149 | // FIXME: Handle `None`-delimited groups in a more systematic way | |
150 | // during parsing. | |
151 | let mut next = outer_trees.next(); | |
152 | let mut trees: Box<dyn Iterator<Item = tokenstream::TokenTree>>; | |
153 | if let Some(tokenstream::TokenTree::Delimited(_, token::NoDelim, tts)) = next { | |
154 | trees = Box::new(tts.into_trees()); | |
155 | next = trees.next(); | |
156 | } else { | |
157 | trees = Box::new(outer_trees); | |
8bb4bdeb | 158 | } |
2c00a5a8 | 159 | |
f035d41b XL |
160 | match next { |
161 | // `tree` is followed by a delimited set of token trees. This indicates the beginning | |
162 | // of a repetition sequence in the macro (e.g. `$(pat)*`). | |
163 | Some(tokenstream::TokenTree::Delimited(span, delim, tts)) => { | |
164 | // Must have `(` not `{` or `[` | |
165 | if delim != token::Paren { | |
166 | let tok = pprust::token_kind_to_string(&token::OpenDelim(delim)); | |
167 | let msg = format!("expected `(`, found `{}`", tok); | |
168 | sess.span_diagnostic.span_err(span.entire(), &msg); | |
169 | } | |
170 | // Parse the contents of the sequence itself | |
17df50a5 | 171 | let sequence = parse(tts, expect_matchers, sess, node_id, features, edition); |
f035d41b XL |
172 | // Get the Kleene operator and optional separator |
173 | let (separator, kleene) = | |
174 | parse_sep_and_kleene_op(&mut trees, span.entire(), sess); | |
175 | // Count the number of captured "names" (i.e., named metavars) | |
176 | let name_captures = macro_parser::count_names(&sequence); | |
177 | TokenTree::Sequence( | |
178 | span, | |
179 | Lrc::new(SequenceRepetition { | |
180 | tts: sequence, | |
181 | separator, | |
182 | kleene, | |
183 | num_captures: name_captures, | |
184 | }), | |
185 | ) | |
8bb4bdeb | 186 | } |
2c00a5a8 | 187 | |
f035d41b XL |
188 | // `tree` is followed by an `ident`. This could be `$meta_var` or the `$crate` special |
189 | // metavariable that names the crate of the invocation. | |
190 | Some(tokenstream::TokenTree::Token(token)) if token.is_ident() => { | |
191 | let (ident, is_raw) = token.ident().unwrap(); | |
192 | let span = ident.span.with_lo(span.lo()); | |
193 | if ident.name == kw::Crate && !is_raw { | |
194 | TokenTree::token(token::Ident(kw::DollarCrate, is_raw), span) | |
195 | } else { | |
196 | TokenTree::MetaVar(span, ident) | |
197 | } | |
198 | } | |
199 | ||
200 | // `tree` is followed by a random token. This is an error. | |
201 | Some(tokenstream::TokenTree::Token(token)) => { | |
202 | let msg = format!( | |
203 | "expected identifier, found `{}`", | |
204 | pprust::token_to_string(&token), | |
205 | ); | |
206 | sess.span_diagnostic.span_err(token.span, &msg); | |
207 | TokenTree::MetaVar(token.span, Ident::invalid()) | |
208 | } | |
2c00a5a8 | 209 | |
f035d41b XL |
210 | // There are no more tokens. Just return the `$` we already have. |
211 | None => TokenTree::token(token::Dollar, span), | |
212 | } | |
213 | } | |
2c00a5a8 XL |
214 | |
215 | // `tree` is an arbitrary token. Keep it. | |
dc9dc135 | 216 | tokenstream::TokenTree::Token(token) => TokenTree::Token(token), |
2c00a5a8 | 217 | |
0731742a | 218 | // `tree` is the beginning of a delimited set of tokens (e.g., `(` or `{`). We need to |
2c00a5a8 | 219 | // descend into the delimited set and further parse it. |
0731742a | 220 | tokenstream::TokenTree::Delimited(span, delim, tts) => TokenTree::Delimited( |
2c00a5a8 | 221 | span, |
5869c6ff XL |
222 | Lrc::new(Delimited { |
223 | delim, | |
17df50a5 | 224 | tts: parse(tts, expect_matchers, sess, node_id, features, edition), |
5869c6ff | 225 | }), |
2c00a5a8 | 226 | ), |
8bb4bdeb XL |
227 | } |
228 | } | |
229 | ||
2c00a5a8 XL |
230 | /// Takes a token and returns `Some(KleeneOp)` if the token is `+` `*` or `?`. Otherwise, return |
231 | /// `None`. | |
dc9dc135 XL |
232 | fn kleene_op(token: &Token) -> Option<KleeneOp> { |
233 | match token.kind { | |
2c00a5a8 XL |
234 | token::BinOp(token::Star) => Some(KleeneOp::ZeroOrMore), |
235 | token::BinOp(token::Plus) => Some(KleeneOp::OneOrMore), | |
236 | token::Question => Some(KleeneOp::ZeroOrOne), | |
237 | _ => None, | |
8bb4bdeb | 238 | } |
2c00a5a8 | 239 | } |
8bb4bdeb | 240 | |
2c00a5a8 XL |
241 | /// Parse the next token tree of the input looking for a KleeneOp. Returns |
242 | /// | |
8faf50e0 | 243 | /// - Ok(Ok((op, span))) if the next token tree is a KleeneOp |
2c00a5a8 XL |
244 | /// - Ok(Err(tok, span)) if the next token tree is a token but not a KleeneOp |
245 | /// - Err(span) if the next token tree is not a token | |
dc9dc135 XL |
246 | fn parse_kleene_op( |
247 | input: &mut impl Iterator<Item = tokenstream::TokenTree>, | |
2c00a5a8 | 248 | span: Span, |
dc9dc135 | 249 | ) -> Result<Result<(KleeneOp, Span), Token>, Span> { |
2c00a5a8 | 250 | match input.next() { |
dc9dc135 XL |
251 | Some(tokenstream::TokenTree::Token(token)) => match kleene_op(&token) { |
252 | Some(op) => Ok(Ok((op, token.span))), | |
253 | None => Ok(Err(token)), | |
2c00a5a8 | 254 | }, |
5869c6ff | 255 | tree => Err(tree.as_ref().map_or(span, tokenstream::TokenTree::span)), |
2c00a5a8 XL |
256 | } |
257 | } | |
258 | ||
259 | /// Attempt to parse a single Kleene star, possibly with a separator. | |
260 | /// | |
261 | /// For example, in a pattern such as `$(a),*`, `a` is the pattern to be repeated, `,` is the | |
262 | /// separator, and `*` is the Kleene operator. This function is specifically concerned with parsing | |
263 | /// the last two tokens of such a pattern: namely, the optional separator and the Kleene operator | |
264 | /// itself. Note that here we are parsing the _macro_ itself, rather than trying to match some | |
265 | /// stream of tokens in an invocation of a macro. | |
266 | /// | |
267 | /// This function will take some input iterator `input` corresponding to `span` and a parsing | |
268 | /// session `sess`. If the next one (or possibly two) tokens in `input` correspond to a Kleene | |
269 | /// operator and separator, then a tuple with `(separator, KleeneOp)` is returned. Otherwise, an | |
270 | /// error with the appropriate span is emitted to `sess` and a dummy value is returned. | |
dc9dc135 | 271 | fn parse_sep_and_kleene_op( |
e74abb32 | 272 | input: &mut impl Iterator<Item = tokenstream::TokenTree>, |
8faf50e0 XL |
273 | span: Span, |
274 | sess: &ParseSess, | |
416331ca | 275 | ) -> (Option<Token>, KleeneToken) { |
8faf50e0 XL |
276 | // We basically look at two token trees here, denoted as #1 and #2 below |
277 | let span = match parse_kleene_op(input, span) { | |
dc9dc135 | 278 | // #1 is a `?`, `+`, or `*` KleeneOp |
416331ca | 279 | Ok(Ok((op, span))) => return (None, KleeneToken::new(op, span)), |
8faf50e0 | 280 | |
2c00a5a8 | 281 | // #1 is a separator followed by #2, a KleeneOp |
dc9dc135 | 282 | Ok(Err(token)) => match parse_kleene_op(input, token.span) { |
8faf50e0 | 283 | // #2 is the `?` Kleene op, which does not take a separator (error) |
416331ca | 284 | Ok(Ok((KleeneOp::ZeroOrOne, span))) => { |
8faf50e0 | 285 | // Error! |
a1dfa0c6 | 286 | sess.span_diagnostic.span_err( |
dc9dc135 | 287 | token.span, |
a1dfa0c6 XL |
288 | "the `?` macro repetition operator does not take a separator", |
289 | ); | |
8faf50e0 XL |
290 | |
291 | // Return a dummy | |
416331ca | 292 | return (None, KleeneToken::new(KleeneOp::ZeroOrMore, span)); |
2c00a5a8 | 293 | } |
8faf50e0 XL |
294 | |
295 | // #2 is a KleeneOp :D | |
416331ca | 296 | Ok(Ok((op, span))) => return (Some(token), KleeneToken::new(op, span)), |
2c00a5a8 | 297 | |
dc9dc135 XL |
298 | // #2 is a random token or not a token at all :( |
299 | Ok(Err(Token { span, .. })) | Err(span) => span, | |
8bb4bdeb | 300 | }, |
2c00a5a8 XL |
301 | |
302 | // #1 is not a token | |
303 | Err(span) => span, | |
8bb4bdeb XL |
304 | }; |
305 | ||
8faf50e0 | 306 | // If we ever get to this point, we have experienced an "unexpected token" error |
dc9dc135 | 307 | sess.span_diagnostic.span_err(span, "expected one of: `*`, `+`, or `?`"); |
8faf50e0 XL |
308 | |
309 | // Return a dummy | |
416331ca | 310 | (None, KleeneToken::new(KleeneOp::ZeroOrMore, span)) |
8bb4bdeb | 311 | } |