1 //! The main parser interface.
4 use crate::parse::parser::{Parser, emit_unclosed_delims, make_unclosed_delims_error};
5 use crate::parse::token::Nonterminal;
6 use crate::tokenstream::{self, TokenStream, TokenTree};
7 use crate::print::pprust;
8 use crate::sess::ParseSess;
10 use errors::{FatalError, Level, Diagnostic, DiagnosticBuilder};
11 #[cfg(target_arch = "x86_64")]
12 use rustc_data_structures::static_assert_size;
13 use rustc_data_structures::sync::Lrc;
14 use syntax_pos::{Span, SourceFile, FileName};
32 crate mod unescape_error_reporting;
34 pub type PResult<'a, T> = Result<T, DiagnosticBuilder<'a>>;
36 // `PResult` is used a lot. Make sure it doesn't unintentionally get bigger.
37 // (See also the comment on `DiagnosticBuilderInner`.)
38 #[cfg(target_arch = "x86_64")]
39 static_assert_size!(PResult<'_, bool>, 16);
42 pub struct Directory<'a> {
43 pub path: Cow<'a, Path>,
44 pub ownership: DirectoryOwnership,
47 #[derive(Copy, Clone)]
48 pub enum DirectoryOwnership {
50 // None if `mod.rs`, `Some("foo")` if we're in `foo.rs`.
51 relative: Option<ast::Ident>,
54 UnownedViaMod(bool /* legacy warnings? */),
57 // A bunch of utility functions of the form `parse_<thing>_from_<source>`
58 // where <thing> includes crate, expr, item, stmt, tts, and one that
59 // uses a HOF to parse anything, and <source> includes file and
62 /// A variant of 'panictry!' that works on a Vec<Diagnostic> instead of a single DiagnosticBuilder.
63 macro_rules! panictry_buffer {
64 ($handler:expr, $e:expr) => ({
65 use std::result::Result::{Ok, Err};
66 use errors::FatalError;
71 $handler.emit_diagnostic(&e);
79 pub fn parse_crate_from_file<'a>(input: &Path, sess: &'a ParseSess) -> PResult<'a, ast::Crate> {
80 let mut parser = new_parser_from_file(sess, input);
81 parser.parse_crate_mod()
84 pub fn parse_crate_attrs_from_file<'a>(input: &Path, sess: &'a ParseSess)
85 -> PResult<'a, Vec<ast::Attribute>> {
86 let mut parser = new_parser_from_file(sess, input);
87 parser.parse_inner_attributes()
90 pub fn parse_crate_from_source_str(name: FileName, source: String, sess: &ParseSess)
91 -> PResult<'_, ast::Crate> {
92 new_parser_from_source_str(sess, name, source).parse_crate_mod()
95 pub fn parse_crate_attrs_from_source_str(name: FileName, source: String, sess: &ParseSess)
96 -> PResult<'_, Vec<ast::Attribute>> {
97 new_parser_from_source_str(sess, name, source).parse_inner_attributes()
100 pub fn parse_stream_from_source_str(
104 override_span: Option<Span>,
106 let (stream, mut errors) = source_file_to_stream(
108 sess.source_map().new_source_file(name, source),
111 emit_unclosed_delims(&mut errors, &sess);
115 /// Creates a new parser from a source string.
116 pub fn new_parser_from_source_str(sess: &ParseSess, name: FileName, source: String) -> Parser<'_> {
117 panictry_buffer!(&sess.span_diagnostic, maybe_new_parser_from_source_str(sess, name, source))
120 /// Creates a new parser from a source string. Returns any buffered errors from lexing the initial
122 pub fn maybe_new_parser_from_source_str(sess: &ParseSess, name: FileName, source: String)
123 -> Result<Parser<'_>, Vec<Diagnostic>>
125 let mut parser = maybe_source_file_to_parser(sess,
126 sess.source_map().new_source_file(name, source))?;
127 parser.recurse_into_file_modules = false;
131 /// Creates a new parser, handling errors as appropriate if the file doesn't exist.
132 pub fn new_parser_from_file<'a>(sess: &'a ParseSess, path: &Path) -> Parser<'a> {
133 source_file_to_parser(sess, file_to_source_file(sess, path, None))
136 /// Creates a new parser, returning buffered diagnostics if the file doesn't exist,
137 /// or from lexing the initial token stream.
138 pub fn maybe_new_parser_from_file<'a>(sess: &'a ParseSess, path: &Path)
139 -> Result<Parser<'a>, Vec<Diagnostic>> {
140 let file = try_file_to_source_file(sess, path, None).map_err(|db| vec![db])?;
141 maybe_source_file_to_parser(sess, file)
144 /// Given a session, a crate config, a path, and a span, add
145 /// the file at the given path to the `source_map`, and returns a parser.
146 /// On an error, uses the given span as the source of the problem.
147 pub fn new_sub_parser_from_file<'a>(sess: &'a ParseSess,
149 directory_ownership: DirectoryOwnership,
150 module_name: Option<String>,
151 sp: Span) -> Parser<'a> {
152 let mut p = source_file_to_parser(sess, file_to_source_file(sess, path, Some(sp)));
153 p.directory.ownership = directory_ownership;
154 p.root_module_name = module_name;
158 /// Given a `source_file` and config, returns a parser.
159 fn source_file_to_parser(sess: &ParseSess, source_file: Lrc<SourceFile>) -> Parser<'_> {
160 panictry_buffer!(&sess.span_diagnostic,
161 maybe_source_file_to_parser(sess, source_file))
164 /// Given a `source_file` and config, return a parser. Returns any buffered errors from lexing the
165 /// initial token stream.
166 fn maybe_source_file_to_parser(
168 source_file: Lrc<SourceFile>,
169 ) -> Result<Parser<'_>, Vec<Diagnostic>> {
170 let end_pos = source_file.end_pos;
171 let (stream, unclosed_delims) = maybe_file_to_stream(sess, source_file, None)?;
172 let mut parser = stream_to_parser(sess, stream, None);
173 parser.unclosed_delims = unclosed_delims;
174 if parser.token == token::Eof && parser.token.span.is_dummy() {
175 parser.token.span = Span::new(end_pos, end_pos, parser.token.span.ctxt());
181 // Must preserve old name for now, because `quote!` from the *existing*
182 // compiler expands into it.
183 pub fn new_parser_from_tts(sess: &ParseSess, tts: Vec<TokenTree>) -> Parser<'_> {
184 stream_to_parser(sess, tts.into_iter().collect(), crate::MACRO_ARGUMENTS)
190 /// Given a session and a path and an optional span (for error reporting),
191 /// add the path to the session's source_map and return the new source_file or
192 /// error when a file can't be read.
193 fn try_file_to_source_file(sess: &ParseSess, path: &Path, spanopt: Option<Span>)
194 -> Result<Lrc<SourceFile>, Diagnostic> {
195 sess.source_map().load_file(path)
197 let msg = format!("couldn't read {}: {}", path.display(), e);
198 let mut diag = Diagnostic::new(Level::Fatal, &msg);
199 if let Some(sp) = spanopt {
206 /// Given a session and a path and an optional span (for error reporting),
207 /// adds the path to the session's `source_map` and returns the new `source_file`.
208 fn file_to_source_file(sess: &ParseSess, path: &Path, spanopt: Option<Span>)
210 match try_file_to_source_file(sess, path, spanopt) {
211 Ok(source_file) => source_file,
213 sess.span_diagnostic.emit_diagnostic(&d);
219 /// Given a `source_file`, produces a sequence of token trees.
220 pub fn source_file_to_stream(
222 source_file: Lrc<SourceFile>,
223 override_span: Option<Span>,
224 ) -> (TokenStream, Vec<lexer::UnmatchedBrace>) {
225 panictry_buffer!(&sess.span_diagnostic, maybe_file_to_stream(sess, source_file, override_span))
228 /// Given a source file, produces a sequence of token trees. Returns any buffered errors from
229 /// parsing the token stream.
230 pub fn maybe_file_to_stream(
232 source_file: Lrc<SourceFile>,
233 override_span: Option<Span>,
234 ) -> Result<(TokenStream, Vec<lexer::UnmatchedBrace>), Vec<Diagnostic>> {
235 let srdr = lexer::StringReader::new(sess, source_file, override_span);
236 let (token_trees, unmatched_braces) = srdr.into_token_trees();
239 Ok(stream) => Ok((stream, unmatched_braces)),
241 let mut buffer = Vec::with_capacity(1);
242 err.buffer(&mut buffer);
243 // Not using `emit_unclosed_delims` to use `db.buffer`
244 for unmatched in unmatched_braces {
245 if let Some(err) = make_unclosed_delims_error(unmatched, &sess) {
246 err.buffer(&mut buffer);
254 /// Given a stream and the `ParseSess`, produces a parser.
255 pub fn stream_to_parser<'a>(
258 subparser_name: Option<&'static str>,
260 Parser::new(sess, stream, None, true, false, subparser_name)
263 /// Given a stream, the `ParseSess` and the base directory, produces a parser.
265 /// Use this function when you are creating a parser from the token stream
266 /// and also care about the current working directory of the parser (e.g.,
267 /// you are trying to resolve modules defined inside a macro invocation).
271 /// The main usage of this function is outside of rustc, for those who uses
272 /// libsyntax as a library. Please do not remove this function while refactoring
273 /// just because it is not used in rustc codebase!
274 pub fn stream_to_parser_with_base_dir<'a>(
277 base_dir: Directory<'a>,
279 Parser::new(sess, stream, Some(base_dir), true, false, None)
282 /// Runs the given subparser `f` on the tokens of the given `attr`'s item.
283 pub fn parse_in_attr<'a, T>(
285 attr: &ast::Attribute,
286 mut f: impl FnMut(&mut Parser<'a>) -> PResult<'a, T>,
287 ) -> PResult<'a, T> {
288 let mut parser = Parser::new(
296 let result = f(&mut parser)?;
297 if parser.token != token::Eof {
298 parser.unexpected()?;
303 // NOTE(Centril): The following probably shouldn't be here but it acknowledges the
304 // fact that architecturally, we are using parsing (read on below to understand why).
306 pub fn nt_to_tokenstream(nt: &Nonterminal, sess: &ParseSess, span: Span) -> TokenStream {
307 // A `Nonterminal` is often a parsed AST item. At this point we now
308 // need to convert the parsed AST to an actual token stream, e.g.
309 // un-parse it basically.
311 // Unfortunately there's not really a great way to do that in a
312 // guaranteed lossless fashion right now. The fallback here is to just
313 // stringify the AST node and reparse it, but this loses all span
316 // As a result, some AST nodes are annotated with the token stream they
317 // came from. Here we attempt to extract these lossless token streams
318 // before we fall back to the stringification.
319 let tokens = match *nt {
320 Nonterminal::NtItem(ref item) => {
321 prepend_attrs(sess, &item.attrs, item.tokens.as_ref(), span)
323 Nonterminal::NtTraitItem(ref item) => {
324 prepend_attrs(sess, &item.attrs, item.tokens.as_ref(), span)
326 Nonterminal::NtImplItem(ref item) => {
327 prepend_attrs(sess, &item.attrs, item.tokens.as_ref(), span)
329 Nonterminal::NtIdent(ident, is_raw) => {
330 Some(tokenstream::TokenTree::token(token::Ident(ident.name, is_raw), ident.span).into())
332 Nonterminal::NtLifetime(ident) => {
333 Some(tokenstream::TokenTree::token(token::Lifetime(ident.name), ident.span).into())
335 Nonterminal::NtTT(ref tt) => {
336 Some(tt.clone().into())
341 // FIXME(#43081): Avoid this pretty-print + reparse hack
342 let source = pprust::nonterminal_to_string(nt);
343 let filename = FileName::macro_expansion_source_code(&source);
344 let tokens_for_real = parse_stream_from_source_str(filename, source, sess, Some(span));
346 // During early phases of the compiler the AST could get modified
347 // directly (e.g., attributes added or removed) and the internal cache
348 // of tokens my not be invalidated or updated. Consequently if the
349 // "lossless" token stream disagrees with our actual stringification
350 // (which has historically been much more battle-tested) then we go
351 // with the lossy stream anyway (losing span information).
353 // Note that the comparison isn't `==` here to avoid comparing spans,
354 // but it *also* is a "probable" equality which is a pretty weird
355 // definition. We mostly want to catch actual changes to the AST
356 // like a `#[cfg]` being processed or some weird `macro_rules!`
359 // What we *don't* want to catch is the fact that a user-defined
360 // literal like `0xf` is stringified as `15`, causing the cached token
361 // stream to not be literal `==` token-wise (ignoring spans) to the
362 // token stream we got from stringification.
364 // Instead the "probably equal" check here is "does each token
365 // recursively have the same discriminant?" We basically don't look at
366 // the token values here and assume that such fine grained token stream
367 // modifications, including adding/removing typically non-semantic
368 // tokens such as extra braces and commas, don't happen.
369 if let Some(tokens) = tokens {
370 if tokens.probably_equal_for_proc_macro(&tokens_for_real) {
373 info!("cached tokens found, but they're not \"probably equal\", \
374 going with stringified version");
376 return tokens_for_real
381 attrs: &[ast::Attribute],
382 tokens: Option<&tokenstream::TokenStream>,
383 span: syntax_pos::Span
384 ) -> Option<tokenstream::TokenStream> {
385 let tokens = tokens?;
386 if attrs.len() == 0 {
387 return Some(tokens.clone())
389 let mut builder = tokenstream::TokenStreamBuilder::new();
391 assert_eq!(attr.style, ast::AttrStyle::Outer,
392 "inner attributes should prevent cached tokens from existing");
394 let source = pprust::attribute_to_string(attr);
395 let macro_filename = FileName::macro_expansion_source_code(&source);
396 if attr.is_sugared_doc {
397 let stream = parse_stream_from_source_str(macro_filename, source, sess, Some(span));
398 builder.push(stream);
402 // synthesize # [ $path $tokens ] manually here
403 let mut brackets = tokenstream::TokenStreamBuilder::new();
405 // For simple paths, push the identifier directly
406 if attr.path.segments.len() == 1 && attr.path.segments[0].args.is_none() {
407 let ident = attr.path.segments[0].ident;
408 let token = token::Ident(ident.name, ident.as_str().starts_with("r#"));
409 brackets.push(tokenstream::TokenTree::token(token, ident.span));
411 // ... and for more complicated paths, fall back to a reparse hack that
412 // should eventually be removed.
414 let stream = parse_stream_from_source_str(macro_filename, source, sess, Some(span));
415 brackets.push(stream);
418 brackets.push(attr.tokens.clone());
420 // The span we list here for `#` and for `[ ... ]` are both wrong in
421 // that it encompasses more than each token, but it hopefully is "good
422 // enough" for now at least.
423 builder.push(tokenstream::TokenTree::token(token::Pound, attr.span));
424 let delim_span = tokenstream::DelimSpan::from_single(attr.span);
425 builder.push(tokenstream::TokenTree::Delimited(
426 delim_span, token::DelimToken::Bracket, brackets.build().into()));
428 builder.push(tokens.clone());
429 Some(builder.build())