1 //! The main parser interface.
3 #![feature(crate_visibility_modifier)]
6 use syntax::print::pprust;
7 use syntax::sess::ParseSess;
8 use syntax::token::{self, Nonterminal};
9 use syntax::tokenstream::{self, TokenStream, TokenTree};
11 use errors::{PResult, FatalError, Level, Diagnostic};
12 use rustc_data_structures::sync::Lrc;
13 use syntax_pos::{Span, SourceFile, FileName};
21 pub const MACRO_ARGUMENTS: Option<&'static str> = Some("macro arguments");
25 use parser::{Parser, emit_unclosed_delims, make_unclosed_delims_error};
27 pub mod validate_attr;
32 pub struct Directory<'a> {
33 pub path: Cow<'a, Path>,
34 pub ownership: DirectoryOwnership,
37 #[derive(Copy, Clone)]
38 pub enum DirectoryOwnership {
40 // None if `mod.rs`, `Some("foo")` if we're in `foo.rs`.
41 relative: Option<ast::Ident>,
47 // A bunch of utility functions of the form `parse_<thing>_from_<source>`
48 // where <thing> includes crate, expr, item, stmt, tts, and one that
49 // uses a HOF to parse anything, and <source> includes file and
52 /// A variant of 'panictry!' that works on a Vec<Diagnostic> instead of a single DiagnosticBuilder.
53 macro_rules! panictry_buffer {
54 ($handler:expr, $e:expr) => ({
55 use std::result::Result::{Ok, Err};
56 use errors::FatalError;
61 $handler.emit_diagnostic(&e);
69 pub fn parse_crate_from_file<'a>(input: &Path, sess: &'a ParseSess) -> PResult<'a, ast::Crate> {
70 let mut parser = new_parser_from_file(sess, input);
71 parser.parse_crate_mod()
74 pub fn parse_crate_attrs_from_file<'a>(input: &Path, sess: &'a ParseSess)
75 -> PResult<'a, Vec<ast::Attribute>> {
76 let mut parser = new_parser_from_file(sess, input);
77 parser.parse_inner_attributes()
80 pub fn parse_crate_from_source_str(name: FileName, source: String, sess: &ParseSess)
81 -> PResult<'_, ast::Crate> {
82 new_parser_from_source_str(sess, name, source).parse_crate_mod()
85 pub fn parse_crate_attrs_from_source_str(name: FileName, source: String, sess: &ParseSess)
86 -> PResult<'_, Vec<ast::Attribute>> {
87 new_parser_from_source_str(sess, name, source).parse_inner_attributes()
90 pub fn parse_stream_from_source_str(
94 override_span: Option<Span>,
96 let (stream, mut errors) = source_file_to_stream(
98 sess.source_map().new_source_file(name, source),
101 emit_unclosed_delims(&mut errors, &sess);
105 /// Creates a new parser from a source string.
106 pub fn new_parser_from_source_str(sess: &ParseSess, name: FileName, source: String) -> Parser<'_> {
107 panictry_buffer!(&sess.span_diagnostic, maybe_new_parser_from_source_str(sess, name, source))
110 /// Creates a new parser from a source string. Returns any buffered errors from lexing the initial
112 pub fn maybe_new_parser_from_source_str(sess: &ParseSess, name: FileName, source: String)
113 -> Result<Parser<'_>, Vec<Diagnostic>>
115 let mut parser = maybe_source_file_to_parser(sess,
116 sess.source_map().new_source_file(name, source))?;
117 parser.recurse_into_file_modules = false;
121 /// Creates a new parser, handling errors as appropriate if the file doesn't exist.
122 pub fn new_parser_from_file<'a>(sess: &'a ParseSess, path: &Path) -> Parser<'a> {
123 source_file_to_parser(sess, file_to_source_file(sess, path, None))
126 /// Creates a new parser, returning buffered diagnostics if the file doesn't exist,
127 /// or from lexing the initial token stream.
128 pub fn maybe_new_parser_from_file<'a>(sess: &'a ParseSess, path: &Path)
129 -> Result<Parser<'a>, Vec<Diagnostic>> {
130 let file = try_file_to_source_file(sess, path, None).map_err(|db| vec![db])?;
131 maybe_source_file_to_parser(sess, file)
134 /// Given a session, a crate config, a path, and a span, add
135 /// the file at the given path to the `source_map`, and returns a parser.
136 /// On an error, uses the given span as the source of the problem.
137 pub fn new_sub_parser_from_file<'a>(sess: &'a ParseSess,
139 directory_ownership: DirectoryOwnership,
140 module_name: Option<String>,
141 sp: Span) -> Parser<'a> {
142 let mut p = source_file_to_parser(sess, file_to_source_file(sess, path, Some(sp)));
143 p.directory.ownership = directory_ownership;
144 p.root_module_name = module_name;
148 /// Given a `source_file` and config, returns a parser.
149 fn source_file_to_parser(sess: &ParseSess, source_file: Lrc<SourceFile>) -> Parser<'_> {
150 panictry_buffer!(&sess.span_diagnostic,
151 maybe_source_file_to_parser(sess, source_file))
154 /// Given a `source_file` and config, return a parser. Returns any buffered errors from lexing the
155 /// initial token stream.
156 fn maybe_source_file_to_parser(
158 source_file: Lrc<SourceFile>,
159 ) -> Result<Parser<'_>, Vec<Diagnostic>> {
160 let end_pos = source_file.end_pos;
161 let (stream, unclosed_delims) = maybe_file_to_stream(sess, source_file, None)?;
162 let mut parser = stream_to_parser(sess, stream, None);
163 parser.unclosed_delims = unclosed_delims;
164 if parser.token == token::Eof && parser.token.span.is_dummy() {
165 parser.token.span = Span::new(end_pos, end_pos, parser.token.span.ctxt());
171 // Must preserve old name for now, because `quote!` from the *existing*
172 // compiler expands into it.
173 pub fn new_parser_from_tts(sess: &ParseSess, tts: Vec<TokenTree>) -> Parser<'_> {
174 stream_to_parser(sess, tts.into_iter().collect(), crate::MACRO_ARGUMENTS)
180 /// Given a session and a path and an optional span (for error reporting),
181 /// add the path to the session's source_map and return the new source_file or
182 /// error when a file can't be read.
183 fn try_file_to_source_file(sess: &ParseSess, path: &Path, spanopt: Option<Span>)
184 -> Result<Lrc<SourceFile>, Diagnostic> {
185 sess.source_map().load_file(path)
187 let msg = format!("couldn't read {}: {}", path.display(), e);
188 let mut diag = Diagnostic::new(Level::Fatal, &msg);
189 if let Some(sp) = spanopt {
196 /// Given a session and a path and an optional span (for error reporting),
197 /// adds the path to the session's `source_map` and returns the new `source_file`.
198 fn file_to_source_file(sess: &ParseSess, path: &Path, spanopt: Option<Span>)
200 match try_file_to_source_file(sess, path, spanopt) {
201 Ok(source_file) => source_file,
203 sess.span_diagnostic.emit_diagnostic(&d);
209 /// Given a `source_file`, produces a sequence of token trees.
210 pub fn source_file_to_stream(
212 source_file: Lrc<SourceFile>,
213 override_span: Option<Span>,
214 ) -> (TokenStream, Vec<lexer::UnmatchedBrace>) {
215 panictry_buffer!(&sess.span_diagnostic, maybe_file_to_stream(sess, source_file, override_span))
218 /// Given a source file, produces a sequence of token trees. Returns any buffered errors from
219 /// parsing the token stream.
220 pub fn maybe_file_to_stream(
222 source_file: Lrc<SourceFile>,
223 override_span: Option<Span>,
224 ) -> Result<(TokenStream, Vec<lexer::UnmatchedBrace>), Vec<Diagnostic>> {
225 let srdr = lexer::StringReader::new(sess, source_file, override_span);
226 let (token_trees, unmatched_braces) = srdr.into_token_trees();
229 Ok(stream) => Ok((stream, unmatched_braces)),
231 let mut buffer = Vec::with_capacity(1);
232 err.buffer(&mut buffer);
233 // Not using `emit_unclosed_delims` to use `db.buffer`
234 for unmatched in unmatched_braces {
235 if let Some(err) = make_unclosed_delims_error(unmatched, &sess) {
236 err.buffer(&mut buffer);
244 /// Given a stream and the `ParseSess`, produces a parser.
245 pub fn stream_to_parser<'a>(
248 subparser_name: Option<&'static str>,
250 Parser::new(sess, stream, None, true, false, subparser_name)
253 /// Given a stream, the `ParseSess` and the base directory, produces a parser.
255 /// Use this function when you are creating a parser from the token stream
256 /// and also care about the current working directory of the parser (e.g.,
257 /// you are trying to resolve modules defined inside a macro invocation).
261 /// The main usage of this function is outside of rustc, for those who uses
262 /// libsyntax as a library. Please do not remove this function while refactoring
263 /// just because it is not used in rustc codebase!
264 pub fn stream_to_parser_with_base_dir<'a>(
267 base_dir: Directory<'a>,
269 Parser::new(sess, stream, Some(base_dir), true, false, None)
272 /// Runs the given subparser `f` on the tokens of the given `attr`'s item.
273 pub fn parse_in_attr<'a, T>(
275 attr: &ast::Attribute,
276 mut f: impl FnMut(&mut Parser<'a>) -> PResult<'a, T>,
277 ) -> PResult<'a, T> {
278 let mut parser = Parser::new(
280 // FIXME(#66940, Centril | petrochenkov): refactor this function so it doesn't
281 // require reconstructing and immediately re-parsing delimiters.
282 attr.get_normal_item().args.outer_tokens(),
288 let result = f(&mut parser)?;
289 if parser.token != token::Eof {
290 parser.unexpected()?;
295 // NOTE(Centril): The following probably shouldn't be here but it acknowledges the
296 // fact that architecturally, we are using parsing (read on below to understand why).
298 pub fn nt_to_tokenstream(nt: &Nonterminal, sess: &ParseSess, span: Span) -> TokenStream {
299 // A `Nonterminal` is often a parsed AST item. At this point we now
300 // need to convert the parsed AST to an actual token stream, e.g.
301 // un-parse it basically.
303 // Unfortunately there's not really a great way to do that in a
304 // guaranteed lossless fashion right now. The fallback here is to just
305 // stringify the AST node and reparse it, but this loses all span
308 // As a result, some AST nodes are annotated with the token stream they
309 // came from. Here we attempt to extract these lossless token streams
310 // before we fall back to the stringification.
311 let tokens = match *nt {
312 Nonterminal::NtItem(ref item) => {
313 prepend_attrs(sess, &item.attrs, item.tokens.as_ref(), span)
315 Nonterminal::NtTraitItem(ref item) => {
316 prepend_attrs(sess, &item.attrs, item.tokens.as_ref(), span)
318 Nonterminal::NtImplItem(ref item) => {
319 prepend_attrs(sess, &item.attrs, item.tokens.as_ref(), span)
321 Nonterminal::NtIdent(ident, is_raw) => {
322 Some(tokenstream::TokenTree::token(token::Ident(ident.name, is_raw), ident.span).into())
324 Nonterminal::NtLifetime(ident) => {
325 Some(tokenstream::TokenTree::token(token::Lifetime(ident.name), ident.span).into())
327 Nonterminal::NtTT(ref tt) => {
328 Some(tt.clone().into())
333 // FIXME(#43081): Avoid this pretty-print + reparse hack
334 let source = pprust::nonterminal_to_string(nt);
335 let filename = FileName::macro_expansion_source_code(&source);
336 let tokens_for_real = parse_stream_from_source_str(filename, source, sess, Some(span));
338 // During early phases of the compiler the AST could get modified
339 // directly (e.g., attributes added or removed) and the internal cache
340 // of tokens my not be invalidated or updated. Consequently if the
341 // "lossless" token stream disagrees with our actual stringification
342 // (which has historically been much more battle-tested) then we go
343 // with the lossy stream anyway (losing span information).
345 // Note that the comparison isn't `==` here to avoid comparing spans,
346 // but it *also* is a "probable" equality which is a pretty weird
347 // definition. We mostly want to catch actual changes to the AST
348 // like a `#[cfg]` being processed or some weird `macro_rules!`
351 // What we *don't* want to catch is the fact that a user-defined
352 // literal like `0xf` is stringified as `15`, causing the cached token
353 // stream to not be literal `==` token-wise (ignoring spans) to the
354 // token stream we got from stringification.
356 // Instead the "probably equal" check here is "does each token
357 // recursively have the same discriminant?" We basically don't look at
358 // the token values here and assume that such fine grained token stream
359 // modifications, including adding/removing typically non-semantic
360 // tokens such as extra braces and commas, don't happen.
361 if let Some(tokens) = tokens {
362 if tokens.probably_equal_for_proc_macro(&tokens_for_real) {
365 info!("cached tokens found, but they're not \"probably equal\", \
366 going with stringified version");
368 return tokens_for_real
373 attrs: &[ast::Attribute],
374 tokens: Option<&tokenstream::TokenStream>,
375 span: syntax_pos::Span
376 ) -> Option<tokenstream::TokenStream> {
377 let tokens = tokens?;
378 if attrs.len() == 0 {
379 return Some(tokens.clone())
381 let mut builder = tokenstream::TokenStreamBuilder::new();
383 assert_eq!(attr.style, ast::AttrStyle::Outer,
384 "inner attributes should prevent cached tokens from existing");
386 let source = pprust::attribute_to_string(attr);
387 let macro_filename = FileName::macro_expansion_source_code(&source);
389 let item = match attr.kind {
390 ast::AttrKind::Normal(ref item) => item,
391 ast::AttrKind::DocComment(_) => {
392 let stream = parse_stream_from_source_str(macro_filename, source, sess, Some(span));
393 builder.push(stream);
398 // synthesize # [ $path $tokens ] manually here
399 let mut brackets = tokenstream::TokenStreamBuilder::new();
401 // For simple paths, push the identifier directly
402 if item.path.segments.len() == 1 && item.path.segments[0].args.is_none() {
403 let ident = item.path.segments[0].ident;
404 let token = token::Ident(ident.name, ident.as_str().starts_with("r#"));
405 brackets.push(tokenstream::TokenTree::token(token, ident.span));
407 // ... and for more complicated paths, fall back to a reparse hack that
408 // should eventually be removed.
410 let stream = parse_stream_from_source_str(macro_filename, source, sess, Some(span));
411 brackets.push(stream);
414 brackets.push(item.args.outer_tokens());
416 // The span we list here for `#` and for `[ ... ]` are both wrong in
417 // that it encompasses more than each token, but it hopefully is "good
418 // enough" for now at least.
419 builder.push(tokenstream::TokenTree::token(token::Pound, attr.span));
420 let delim_span = tokenstream::DelimSpan::from_single(attr.span);
421 builder.push(tokenstream::TokenTree::Delimited(
422 delim_span, token::DelimToken::Bracket, brackets.build().into()));
424 builder.push(tokens.clone());
425 Some(builder.build())