1 //! The main parser interface.
3 #![feature(bool_to_option)]
4 #![feature(crate_visibility_modifier)]
5 #![feature(slice_patterns)]
8 use syntax::print::pprust;
9 use syntax::sess::ParseSess;
10 use syntax::token::{self, Nonterminal};
11 use syntax::tokenstream::{self, TokenStream, TokenTree};
13 use rustc_errors::{PResult, FatalError, Level, Diagnostic};
14 use rustc_data_structures::sync::Lrc;
15 use syntax_pos::{Span, SourceFile, FileName};
23 pub const MACRO_ARGUMENTS: Option<&'static str> = Some("macro arguments");
27 use parser::{Parser, emit_unclosed_delims, make_unclosed_delims_error};
29 pub mod validate_attr;
34 pub struct Directory<'a> {
35 pub path: Cow<'a, Path>,
36 pub ownership: DirectoryOwnership,
39 #[derive(Copy, Clone)]
40 pub enum DirectoryOwnership {
42 // None if `mod.rs`, `Some("foo")` if we're in `foo.rs`.
43 relative: Option<ast::Ident>,
49 // A bunch of utility functions of the form `parse_<thing>_from_<source>`
50 // where <thing> includes crate, expr, item, stmt, tts, and one that
51 // uses a HOF to parse anything, and <source> includes file and
54 /// A variant of 'panictry!' that works on a Vec<Diagnostic> instead of a single DiagnosticBuilder.
55 macro_rules! panictry_buffer {
56 ($handler:expr, $e:expr) => ({
57 use std::result::Result::{Ok, Err};
58 use rustc_errors::FatalError;
63 $handler.emit_diagnostic(&e);
71 pub fn parse_crate_from_file<'a>(input: &Path, sess: &'a ParseSess) -> PResult<'a, ast::Crate> {
72 let mut parser = new_parser_from_file(sess, input);
73 parser.parse_crate_mod()
76 pub fn parse_crate_attrs_from_file<'a>(input: &Path, sess: &'a ParseSess)
77 -> PResult<'a, Vec<ast::Attribute>> {
78 let mut parser = new_parser_from_file(sess, input);
79 parser.parse_inner_attributes()
82 pub fn parse_crate_from_source_str(name: FileName, source: String, sess: &ParseSess)
83 -> PResult<'_, ast::Crate> {
84 new_parser_from_source_str(sess, name, source).parse_crate_mod()
87 pub fn parse_crate_attrs_from_source_str(name: FileName, source: String, sess: &ParseSess)
88 -> PResult<'_, Vec<ast::Attribute>> {
89 new_parser_from_source_str(sess, name, source).parse_inner_attributes()
92 pub fn parse_stream_from_source_str(
96 override_span: Option<Span>,
98 let (stream, mut errors) = source_file_to_stream(
100 sess.source_map().new_source_file(name, source),
103 emit_unclosed_delims(&mut errors, &sess);
107 /// Creates a new parser from a source string.
108 pub fn new_parser_from_source_str(sess: &ParseSess, name: FileName, source: String) -> Parser<'_> {
109 panictry_buffer!(&sess.span_diagnostic, maybe_new_parser_from_source_str(sess, name, source))
112 /// Creates a new parser from a source string. Returns any buffered errors from lexing the initial
114 pub fn maybe_new_parser_from_source_str(sess: &ParseSess, name: FileName, source: String)
115 -> Result<Parser<'_>, Vec<Diagnostic>>
117 let mut parser = maybe_source_file_to_parser(sess,
118 sess.source_map().new_source_file(name, source))?;
119 parser.recurse_into_file_modules = false;
123 /// Creates a new parser, handling errors as appropriate if the file doesn't exist.
124 pub fn new_parser_from_file<'a>(sess: &'a ParseSess, path: &Path) -> Parser<'a> {
125 source_file_to_parser(sess, file_to_source_file(sess, path, None))
128 /// Creates a new parser, returning buffered diagnostics if the file doesn't exist,
129 /// or from lexing the initial token stream.
130 pub fn maybe_new_parser_from_file<'a>(sess: &'a ParseSess, path: &Path)
131 -> Result<Parser<'a>, Vec<Diagnostic>> {
132 let file = try_file_to_source_file(sess, path, None).map_err(|db| vec![db])?;
133 maybe_source_file_to_parser(sess, file)
136 /// Given a session, a crate config, a path, and a span, add
137 /// the file at the given path to the `source_map`, and returns a parser.
138 /// On an error, uses the given span as the source of the problem.
139 pub fn new_sub_parser_from_file<'a>(sess: &'a ParseSess,
141 directory_ownership: DirectoryOwnership,
142 module_name: Option<String>,
143 sp: Span) -> Parser<'a> {
144 let mut p = source_file_to_parser(sess, file_to_source_file(sess, path, Some(sp)));
145 p.directory.ownership = directory_ownership;
146 p.root_module_name = module_name;
150 /// Given a `source_file` and config, returns a parser.
151 fn source_file_to_parser(sess: &ParseSess, source_file: Lrc<SourceFile>) -> Parser<'_> {
152 panictry_buffer!(&sess.span_diagnostic,
153 maybe_source_file_to_parser(sess, source_file))
156 /// Given a `source_file` and config, return a parser. Returns any buffered errors from lexing the
157 /// initial token stream.
158 fn maybe_source_file_to_parser(
160 source_file: Lrc<SourceFile>,
161 ) -> Result<Parser<'_>, Vec<Diagnostic>> {
162 let end_pos = source_file.end_pos;
163 let (stream, unclosed_delims) = maybe_file_to_stream(sess, source_file, None)?;
164 let mut parser = stream_to_parser(sess, stream, None);
165 parser.unclosed_delims = unclosed_delims;
166 if parser.token == token::Eof && parser.token.span.is_dummy() {
167 parser.token.span = Span::new(end_pos, end_pos, parser.token.span.ctxt());
173 // Must preserve old name for now, because `quote!` from the *existing*
174 // compiler expands into it.
175 pub fn new_parser_from_tts(sess: &ParseSess, tts: Vec<TokenTree>) -> Parser<'_> {
176 stream_to_parser(sess, tts.into_iter().collect(), crate::MACRO_ARGUMENTS)
182 /// Given a session and a path and an optional span (for error reporting),
183 /// add the path to the session's source_map and return the new source_file or
184 /// error when a file can't be read.
185 fn try_file_to_source_file(sess: &ParseSess, path: &Path, spanopt: Option<Span>)
186 -> Result<Lrc<SourceFile>, Diagnostic> {
187 sess.source_map().load_file(path)
189 let msg = format!("couldn't read {}: {}", path.display(), e);
190 let mut diag = Diagnostic::new(Level::Fatal, &msg);
191 if let Some(sp) = spanopt {
198 /// Given a session and a path and an optional span (for error reporting),
199 /// adds the path to the session's `source_map` and returns the new `source_file`.
200 fn file_to_source_file(sess: &ParseSess, path: &Path, spanopt: Option<Span>)
202 match try_file_to_source_file(sess, path, spanopt) {
203 Ok(source_file) => source_file,
205 sess.span_diagnostic.emit_diagnostic(&d);
211 /// Given a `source_file`, produces a sequence of token trees.
212 pub fn source_file_to_stream(
214 source_file: Lrc<SourceFile>,
215 override_span: Option<Span>,
216 ) -> (TokenStream, Vec<lexer::UnmatchedBrace>) {
217 panictry_buffer!(&sess.span_diagnostic, maybe_file_to_stream(sess, source_file, override_span))
220 /// Given a source file, produces a sequence of token trees. Returns any buffered errors from
221 /// parsing the token stream.
222 pub fn maybe_file_to_stream(
224 source_file: Lrc<SourceFile>,
225 override_span: Option<Span>,
226 ) -> Result<(TokenStream, Vec<lexer::UnmatchedBrace>), Vec<Diagnostic>> {
227 let srdr = lexer::StringReader::new(sess, source_file, override_span);
228 let (token_trees, unmatched_braces) = srdr.into_token_trees();
231 Ok(stream) => Ok((stream, unmatched_braces)),
233 let mut buffer = Vec::with_capacity(1);
234 err.buffer(&mut buffer);
235 // Not using `emit_unclosed_delims` to use `db.buffer`
236 for unmatched in unmatched_braces {
237 if let Some(err) = make_unclosed_delims_error(unmatched, &sess) {
238 err.buffer(&mut buffer);
246 /// Given a stream and the `ParseSess`, produces a parser.
247 pub fn stream_to_parser<'a>(
250 subparser_name: Option<&'static str>,
252 Parser::new(sess, stream, None, true, false, subparser_name)
255 /// Given a stream, the `ParseSess` and the base directory, produces a parser.
257 /// Use this function when you are creating a parser from the token stream
258 /// and also care about the current working directory of the parser (e.g.,
259 /// you are trying to resolve modules defined inside a macro invocation).
263 /// The main usage of this function is outside of rustc, for those who uses
264 /// libsyntax as a library. Please do not remove this function while refactoring
265 /// just because it is not used in rustc codebase!
266 pub fn stream_to_parser_with_base_dir<'a>(
269 base_dir: Directory<'a>,
271 Parser::new(sess, stream, Some(base_dir), true, false, None)
274 /// Runs the given subparser `f` on the tokens of the given `attr`'s item.
275 pub fn parse_in<'a, T>(
279 mut f: impl FnMut(&mut Parser<'a>) -> PResult<'a, T>,
280 ) -> PResult<'a, T> {
281 let mut parser = Parser::new(sess, tts, None, false, false, Some(name));
282 let result = f(&mut parser)?;
283 if parser.token != token::Eof {
284 parser.unexpected()?;
289 // NOTE(Centril): The following probably shouldn't be here but it acknowledges the
290 // fact that architecturally, we are using parsing (read on below to understand why).
292 pub fn nt_to_tokenstream(nt: &Nonterminal, sess: &ParseSess, span: Span) -> TokenStream {
293 // A `Nonterminal` is often a parsed AST item. At this point we now
294 // need to convert the parsed AST to an actual token stream, e.g.
295 // un-parse it basically.
297 // Unfortunately there's not really a great way to do that in a
298 // guaranteed lossless fashion right now. The fallback here is to just
299 // stringify the AST node and reparse it, but this loses all span
302 // As a result, some AST nodes are annotated with the token stream they
303 // came from. Here we attempt to extract these lossless token streams
304 // before we fall back to the stringification.
305 let tokens = match *nt {
306 Nonterminal::NtItem(ref item) => {
307 prepend_attrs(sess, &item.attrs, item.tokens.as_ref(), span)
309 Nonterminal::NtTraitItem(ref item) | Nonterminal::NtImplItem(ref item) => {
310 prepend_attrs(sess, &item.attrs, item.tokens.as_ref(), span)
312 Nonterminal::NtIdent(ident, is_raw) => {
313 Some(tokenstream::TokenTree::token(token::Ident(ident.name, is_raw), ident.span).into())
315 Nonterminal::NtLifetime(ident) => {
316 Some(tokenstream::TokenTree::token(token::Lifetime(ident.name), ident.span).into())
318 Nonterminal::NtTT(ref tt) => {
319 Some(tt.clone().into())
324 // FIXME(#43081): Avoid this pretty-print + reparse hack
325 let source = pprust::nonterminal_to_string(nt);
326 let filename = FileName::macro_expansion_source_code(&source);
327 let tokens_for_real = parse_stream_from_source_str(filename, source, sess, Some(span));
329 // During early phases of the compiler the AST could get modified
330 // directly (e.g., attributes added or removed) and the internal cache
331 // of tokens my not be invalidated or updated. Consequently if the
332 // "lossless" token stream disagrees with our actual stringification
333 // (which has historically been much more battle-tested) then we go
334 // with the lossy stream anyway (losing span information).
336 // Note that the comparison isn't `==` here to avoid comparing spans,
337 // but it *also* is a "probable" equality which is a pretty weird
338 // definition. We mostly want to catch actual changes to the AST
339 // like a `#[cfg]` being processed or some weird `macro_rules!`
342 // What we *don't* want to catch is the fact that a user-defined
343 // literal like `0xf` is stringified as `15`, causing the cached token
344 // stream to not be literal `==` token-wise (ignoring spans) to the
345 // token stream we got from stringification.
347 // Instead the "probably equal" check here is "does each token
348 // recursively have the same discriminant?" We basically don't look at
349 // the token values here and assume that such fine grained token stream
350 // modifications, including adding/removing typically non-semantic
351 // tokens such as extra braces and commas, don't happen.
352 if let Some(tokens) = tokens {
353 if tokens.probably_equal_for_proc_macro(&tokens_for_real) {
356 info!("cached tokens found, but they're not \"probably equal\", \
357 going with stringified version");
359 return tokens_for_real
364 attrs: &[ast::Attribute],
365 tokens: Option<&tokenstream::TokenStream>,
366 span: syntax_pos::Span
367 ) -> Option<tokenstream::TokenStream> {
368 let tokens = tokens?;
369 if attrs.len() == 0 {
370 return Some(tokens.clone())
372 let mut builder = tokenstream::TokenStreamBuilder::new();
374 assert_eq!(attr.style, ast::AttrStyle::Outer,
375 "inner attributes should prevent cached tokens from existing");
377 let source = pprust::attribute_to_string(attr);
378 let macro_filename = FileName::macro_expansion_source_code(&source);
380 let item = match attr.kind {
381 ast::AttrKind::Normal(ref item) => item,
382 ast::AttrKind::DocComment(_) => {
383 let stream = parse_stream_from_source_str(macro_filename, source, sess, Some(span));
384 builder.push(stream);
389 // synthesize # [ $path $tokens ] manually here
390 let mut brackets = tokenstream::TokenStreamBuilder::new();
392 // For simple paths, push the identifier directly
393 if item.path.segments.len() == 1 && item.path.segments[0].args.is_none() {
394 let ident = item.path.segments[0].ident;
395 let token = token::Ident(ident.name, ident.as_str().starts_with("r#"));
396 brackets.push(tokenstream::TokenTree::token(token, ident.span));
398 // ... and for more complicated paths, fall back to a reparse hack that
399 // should eventually be removed.
401 let stream = parse_stream_from_source_str(macro_filename, source, sess, Some(span));
402 brackets.push(stream);
405 brackets.push(item.args.outer_tokens());
407 // The span we list here for `#` and for `[ ... ]` are both wrong in
408 // that it encompasses more than each token, but it hopefully is "good
409 // enough" for now at least.
410 builder.push(tokenstream::TokenTree::token(token::Pound, attr.span));
411 let delim_span = tokenstream::DelimSpan::from_single(attr.span);
412 builder.push(tokenstream::TokenTree::Delimited(
413 delim_span, token::DelimToken::Bracket, brackets.build().into()));
415 builder.push(tokens.clone());
416 Some(builder.build())