]> git.proxmox.com Git - rustc.git/blob - src/librustc_parse/lib.rs
New upstream version 1.45.0+dfsg1
[rustc.git] / src / librustc_parse / lib.rs
1 //! The main parser interface.
2
3 #![feature(bool_to_option)]
4 #![feature(crate_visibility_modifier)]
5 #![feature(bindings_after_at)]
6 #![feature(try_blocks)]
7 #![feature(or_patterns)]
8
9 use rustc_ast::ast;
10 use rustc_ast::token::{self, Nonterminal};
11 use rustc_ast::tokenstream::{self, TokenStream, TokenTree};
12 use rustc_ast_pretty::pprust;
13 use rustc_data_structures::sync::Lrc;
14 use rustc_errors::{Diagnostic, FatalError, Level, PResult};
15 use rustc_session::parse::ParseSess;
16 use rustc_span::{FileName, SourceFile, Span};
17
18 use std::path::Path;
19 use std::str;
20
21 use log::{debug, info};
22
23 pub const MACRO_ARGUMENTS: Option<&'static str> = Some("macro arguments");
24
25 #[macro_use]
26 pub mod parser;
27 use parser::{emit_unclosed_delims, make_unclosed_delims_error, Parser};
28 pub mod lexer;
29 pub mod validate_attr;
30
31 // A bunch of utility functions of the form `parse_<thing>_from_<source>`
32 // where <thing> includes crate, expr, item, stmt, tts, and one that
33 // uses a HOF to parse anything, and <source> includes file and
34 // `source_str`.
35
36 /// A variant of 'panictry!' that works on a Vec<Diagnostic> instead of a single DiagnosticBuilder.
37 macro_rules! panictry_buffer {
38 ($handler:expr, $e:expr) => {{
39 use rustc_errors::FatalError;
40 use std::result::Result::{Err, Ok};
41 match $e {
42 Ok(e) => e,
43 Err(errs) => {
44 for e in errs {
45 $handler.emit_diagnostic(&e);
46 }
47 FatalError.raise()
48 }
49 }
50 }};
51 }
52
53 pub fn parse_crate_from_file<'a>(input: &Path, sess: &'a ParseSess) -> PResult<'a, ast::Crate> {
54 let mut parser = new_parser_from_file(sess, input, None);
55 parser.parse_crate_mod()
56 }
57
58 pub fn parse_crate_attrs_from_file<'a>(
59 input: &Path,
60 sess: &'a ParseSess,
61 ) -> PResult<'a, Vec<ast::Attribute>> {
62 let mut parser = new_parser_from_file(sess, input, None);
63 parser.parse_inner_attributes()
64 }
65
66 pub fn parse_crate_from_source_str(
67 name: FileName,
68 source: String,
69 sess: &ParseSess,
70 ) -> PResult<'_, ast::Crate> {
71 new_parser_from_source_str(sess, name, source).parse_crate_mod()
72 }
73
74 pub fn parse_crate_attrs_from_source_str(
75 name: FileName,
76 source: String,
77 sess: &ParseSess,
78 ) -> PResult<'_, Vec<ast::Attribute>> {
79 new_parser_from_source_str(sess, name, source).parse_inner_attributes()
80 }
81
82 pub fn parse_stream_from_source_str(
83 name: FileName,
84 source: String,
85 sess: &ParseSess,
86 override_span: Option<Span>,
87 ) -> TokenStream {
88 let (stream, mut errors) =
89 source_file_to_stream(sess, sess.source_map().new_source_file(name, source), override_span);
90 emit_unclosed_delims(&mut errors, &sess);
91 stream
92 }
93
94 /// Creates a new parser from a source string.
95 pub fn new_parser_from_source_str(sess: &ParseSess, name: FileName, source: String) -> Parser<'_> {
96 panictry_buffer!(&sess.span_diagnostic, maybe_new_parser_from_source_str(sess, name, source))
97 }
98
99 /// Creates a new parser from a source string. Returns any buffered errors from lexing the initial
100 /// token stream.
101 pub fn maybe_new_parser_from_source_str(
102 sess: &ParseSess,
103 name: FileName,
104 source: String,
105 ) -> Result<Parser<'_>, Vec<Diagnostic>> {
106 maybe_source_file_to_parser(sess, sess.source_map().new_source_file(name, source))
107 }
108
109 /// Creates a new parser, handling errors as appropriate if the file doesn't exist.
110 /// If a span is given, that is used on an error as the as the source of the problem.
111 pub fn new_parser_from_file<'a>(sess: &'a ParseSess, path: &Path, sp: Option<Span>) -> Parser<'a> {
112 source_file_to_parser(sess, file_to_source_file(sess, path, sp))
113 }
114
115 /// Creates a new parser, returning buffered diagnostics if the file doesn't exist,
116 /// or from lexing the initial token stream.
117 pub fn maybe_new_parser_from_file<'a>(
118 sess: &'a ParseSess,
119 path: &Path,
120 ) -> Result<Parser<'a>, Vec<Diagnostic>> {
121 let file = try_file_to_source_file(sess, path, None).map_err(|db| vec![db])?;
122 maybe_source_file_to_parser(sess, file)
123 }
124
125 /// Given a `source_file` and config, returns a parser.
126 fn source_file_to_parser(sess: &ParseSess, source_file: Lrc<SourceFile>) -> Parser<'_> {
127 panictry_buffer!(&sess.span_diagnostic, maybe_source_file_to_parser(sess, source_file))
128 }
129
130 /// Given a `source_file` and config, return a parser. Returns any buffered errors from lexing the
131 /// initial token stream.
132 fn maybe_source_file_to_parser(
133 sess: &ParseSess,
134 source_file: Lrc<SourceFile>,
135 ) -> Result<Parser<'_>, Vec<Diagnostic>> {
136 let end_pos = source_file.end_pos;
137 let (stream, unclosed_delims) = maybe_file_to_stream(sess, source_file, None)?;
138 let mut parser = stream_to_parser(sess, stream, None);
139 parser.unclosed_delims = unclosed_delims;
140 if parser.token == token::Eof {
141 parser.token.span = Span::new(end_pos, end_pos, parser.token.span.ctxt());
142 }
143
144 Ok(parser)
145 }
146
147 // Must preserve old name for now, because `quote!` from the *existing*
148 // compiler expands into it.
149 pub fn new_parser_from_tts(sess: &ParseSess, tts: Vec<TokenTree>) -> Parser<'_> {
150 stream_to_parser(sess, tts.into_iter().collect(), crate::MACRO_ARGUMENTS)
151 }
152
153 // Base abstractions
154
155 /// Given a session and a path and an optional span (for error reporting),
156 /// add the path to the session's source_map and return the new source_file or
157 /// error when a file can't be read.
158 fn try_file_to_source_file(
159 sess: &ParseSess,
160 path: &Path,
161 spanopt: Option<Span>,
162 ) -> Result<Lrc<SourceFile>, Diagnostic> {
163 sess.source_map().load_file(path).map_err(|e| {
164 let msg = format!("couldn't read {}: {}", path.display(), e);
165 let mut diag = Diagnostic::new(Level::Fatal, &msg);
166 if let Some(sp) = spanopt {
167 diag.set_span(sp);
168 }
169 diag
170 })
171 }
172
173 /// Given a session and a path and an optional span (for error reporting),
174 /// adds the path to the session's `source_map` and returns the new `source_file`.
175 fn file_to_source_file(sess: &ParseSess, path: &Path, spanopt: Option<Span>) -> Lrc<SourceFile> {
176 match try_file_to_source_file(sess, path, spanopt) {
177 Ok(source_file) => source_file,
178 Err(d) => {
179 sess.span_diagnostic.emit_diagnostic(&d);
180 FatalError.raise();
181 }
182 }
183 }
184
185 /// Given a `source_file`, produces a sequence of token trees.
186 pub fn source_file_to_stream(
187 sess: &ParseSess,
188 source_file: Lrc<SourceFile>,
189 override_span: Option<Span>,
190 ) -> (TokenStream, Vec<lexer::UnmatchedBrace>) {
191 panictry_buffer!(&sess.span_diagnostic, maybe_file_to_stream(sess, source_file, override_span))
192 }
193
194 /// Given a source file, produces a sequence of token trees. Returns any buffered errors from
195 /// parsing the token stream.
196 pub fn maybe_file_to_stream(
197 sess: &ParseSess,
198 source_file: Lrc<SourceFile>,
199 override_span: Option<Span>,
200 ) -> Result<(TokenStream, Vec<lexer::UnmatchedBrace>), Vec<Diagnostic>> {
201 let srdr = lexer::StringReader::new(sess, source_file, override_span);
202 let (token_trees, unmatched_braces) = srdr.into_token_trees();
203
204 match token_trees {
205 Ok(stream) => Ok((stream, unmatched_braces)),
206 Err(err) => {
207 let mut buffer = Vec::with_capacity(1);
208 err.buffer(&mut buffer);
209 // Not using `emit_unclosed_delims` to use `db.buffer`
210 for unmatched in unmatched_braces {
211 if let Some(err) = make_unclosed_delims_error(unmatched, &sess) {
212 err.buffer(&mut buffer);
213 }
214 }
215 Err(buffer)
216 }
217 }
218 }
219
220 /// Given a stream and the `ParseSess`, produces a parser.
221 pub fn stream_to_parser<'a>(
222 sess: &'a ParseSess,
223 stream: TokenStream,
224 subparser_name: Option<&'static str>,
225 ) -> Parser<'a> {
226 Parser::new(sess, stream, false, subparser_name)
227 }
228
229 /// Runs the given subparser `f` on the tokens of the given `attr`'s item.
230 pub fn parse_in<'a, T>(
231 sess: &'a ParseSess,
232 tts: TokenStream,
233 name: &'static str,
234 mut f: impl FnMut(&mut Parser<'a>) -> PResult<'a, T>,
235 ) -> PResult<'a, T> {
236 let mut parser = Parser::new(sess, tts, false, Some(name));
237 let result = f(&mut parser)?;
238 if parser.token != token::Eof {
239 parser.unexpected()?;
240 }
241 Ok(result)
242 }
243
244 // NOTE(Centril): The following probably shouldn't be here but it acknowledges the
245 // fact that architecturally, we are using parsing (read on below to understand why).
246
247 pub fn nt_to_tokenstream(nt: &Nonterminal, sess: &ParseSess, span: Span) -> TokenStream {
248 // A `Nonterminal` is often a parsed AST item. At this point we now
249 // need to convert the parsed AST to an actual token stream, e.g.
250 // un-parse it basically.
251 //
252 // Unfortunately there's not really a great way to do that in a
253 // guaranteed lossless fashion right now. The fallback here is to just
254 // stringify the AST node and reparse it, but this loses all span
255 // information.
256 //
257 // As a result, some AST nodes are annotated with the token stream they
258 // came from. Here we attempt to extract these lossless token streams
259 // before we fall back to the stringification.
260 let tokens = match *nt {
261 Nonterminal::NtItem(ref item) => {
262 prepend_attrs(sess, &item.attrs, item.tokens.as_ref(), span)
263 }
264 Nonterminal::NtIdent(ident, is_raw) => {
265 Some(tokenstream::TokenTree::token(token::Ident(ident.name, is_raw), ident.span).into())
266 }
267 Nonterminal::NtLifetime(ident) => {
268 Some(tokenstream::TokenTree::token(token::Lifetime(ident.name), ident.span).into())
269 }
270 Nonterminal::NtTT(ref tt) => Some(tt.clone().into()),
271 Nonterminal::NtExpr(ref expr) => {
272 if expr.tokens.is_none() {
273 debug!("missing tokens for expr {:?}", expr);
274 }
275 prepend_attrs(sess, &expr.attrs, expr.tokens.as_ref(), span)
276 }
277 _ => None,
278 };
279
280 // FIXME(#43081): Avoid this pretty-print + reparse hack
281 let source = pprust::nonterminal_to_string(nt);
282 let filename = FileName::macro_expansion_source_code(&source);
283 let tokens_for_real = parse_stream_from_source_str(filename, source, sess, Some(span));
284
285 // During early phases of the compiler the AST could get modified
286 // directly (e.g., attributes added or removed) and the internal cache
287 // of tokens my not be invalidated or updated. Consequently if the
288 // "lossless" token stream disagrees with our actual stringification
289 // (which has historically been much more battle-tested) then we go
290 // with the lossy stream anyway (losing span information).
291 //
292 // Note that the comparison isn't `==` here to avoid comparing spans,
293 // but it *also* is a "probable" equality which is a pretty weird
294 // definition. We mostly want to catch actual changes to the AST
295 // like a `#[cfg]` being processed or some weird `macro_rules!`
296 // expansion.
297 //
298 // What we *don't* want to catch is the fact that a user-defined
299 // literal like `0xf` is stringified as `15`, causing the cached token
300 // stream to not be literal `==` token-wise (ignoring spans) to the
301 // token stream we got from stringification.
302 //
303 // Instead the "probably equal" check here is "does each token
304 // recursively have the same discriminant?" We basically don't look at
305 // the token values here and assume that such fine grained token stream
306 // modifications, including adding/removing typically non-semantic
307 // tokens such as extra braces and commas, don't happen.
308 if let Some(tokens) = tokens {
309 if tokens.probably_equal_for_proc_macro(&tokens_for_real) {
310 return tokens;
311 }
312 info!(
313 "cached tokens found, but they're not \"probably equal\", \
314 going with stringified version"
315 );
316 info!("cached tokens: {:?}", tokens);
317 info!("reparsed tokens: {:?}", tokens_for_real);
318 }
319 tokens_for_real
320 }
321
322 fn prepend_attrs(
323 sess: &ParseSess,
324 attrs: &[ast::Attribute],
325 tokens: Option<&tokenstream::TokenStream>,
326 span: rustc_span::Span,
327 ) -> Option<tokenstream::TokenStream> {
328 let tokens = tokens?;
329 if attrs.is_empty() {
330 return Some(tokens.clone());
331 }
332 let mut builder = tokenstream::TokenStreamBuilder::new();
333 for attr in attrs {
334 assert_eq!(
335 attr.style,
336 ast::AttrStyle::Outer,
337 "inner attributes should prevent cached tokens from existing"
338 );
339
340 let source = pprust::attribute_to_string(attr);
341 let macro_filename = FileName::macro_expansion_source_code(&source);
342
343 let item = match attr.kind {
344 ast::AttrKind::Normal(ref item) => item,
345 ast::AttrKind::DocComment(_) => {
346 let stream = parse_stream_from_source_str(macro_filename, source, sess, Some(span));
347 builder.push(stream);
348 continue;
349 }
350 };
351
352 // synthesize # [ $path $tokens ] manually here
353 let mut brackets = tokenstream::TokenStreamBuilder::new();
354
355 // For simple paths, push the identifier directly
356 if item.path.segments.len() == 1 && item.path.segments[0].args.is_none() {
357 let ident = item.path.segments[0].ident;
358 let token = token::Ident(ident.name, ident.as_str().starts_with("r#"));
359 brackets.push(tokenstream::TokenTree::token(token, ident.span));
360
361 // ... and for more complicated paths, fall back to a reparse hack that
362 // should eventually be removed.
363 } else {
364 let stream = parse_stream_from_source_str(macro_filename, source, sess, Some(span));
365 brackets.push(stream);
366 }
367
368 brackets.push(item.args.outer_tokens());
369
370 // The span we list here for `#` and for `[ ... ]` are both wrong in
371 // that it encompasses more than each token, but it hopefully is "good
372 // enough" for now at least.
373 builder.push(tokenstream::TokenTree::token(token::Pound, attr.span));
374 let delim_span = tokenstream::DelimSpan::from_single(attr.span);
375 builder.push(tokenstream::TokenTree::Delimited(
376 delim_span,
377 token::DelimToken::Bracket,
378 brackets.build(),
379 ));
380 }
381 builder.push(tokens.clone());
382 Some(builder.build())
383 }