1 //! The main parser interface.
3 #![feature(bool_to_option)]
4 #![feature(crate_visibility_modifier)]
7 use rustc_ast::token::{self, Nonterminal};
8 use rustc_ast::tokenstream::{self, TokenStream, TokenTree};
9 use rustc_ast_pretty::pprust;
10 use rustc_data_structures::sync::Lrc;
11 use rustc_errors::{Diagnostic, FatalError, Level, PResult};
12 use rustc_session::parse::ParseSess;
13 use rustc_span::{FileName, SourceFile, Span};
15 use std::path::{Path, PathBuf};
20 pub const MACRO_ARGUMENTS: Option<&'static str> = Some("macro arguments");
24 use parser::{emit_unclosed_delims, make_unclosed_delims_error, Parser};
26 pub mod validate_attr;
31 pub struct Directory {
33 pub ownership: DirectoryOwnership,
36 #[derive(Copy, Clone)]
37 pub enum DirectoryOwnership {
39 // None if `mod.rs`, `Some("foo")` if we're in `foo.rs`.
40 relative: Option<ast::Ident>,
46 // A bunch of utility functions of the form `parse_<thing>_from_<source>`
47 // where <thing> includes crate, expr, item, stmt, tts, and one that
48 // uses a HOF to parse anything, and <source> includes file and
51 /// A variant of 'panictry!' that works on a Vec<Diagnostic> instead of a single DiagnosticBuilder.
52 macro_rules! panictry_buffer {
53 ($handler:expr, $e:expr) => {{
54 use rustc_errors::FatalError;
55 use std::result::Result::{Err, Ok};
60 $handler.emit_diagnostic(&e);
68 pub fn parse_crate_from_file<'a>(input: &Path, sess: &'a ParseSess) -> PResult<'a, ast::Crate> {
69 let mut parser = new_parser_from_file(sess, input);
70 parser.parse_crate_mod()
73 pub fn parse_crate_attrs_from_file<'a>(
76 ) -> PResult<'a, Vec<ast::Attribute>> {
77 let mut parser = new_parser_from_file(sess, input);
78 parser.parse_inner_attributes()
81 pub fn parse_crate_from_source_str(
85 ) -> PResult<'_, ast::Crate> {
86 new_parser_from_source_str(sess, name, source).parse_crate_mod()
89 pub fn parse_crate_attrs_from_source_str(
93 ) -> PResult<'_, Vec<ast::Attribute>> {
94 new_parser_from_source_str(sess, name, source).parse_inner_attributes()
97 pub fn parse_stream_from_source_str(
101 override_span: Option<Span>,
103 let (stream, mut errors) =
104 source_file_to_stream(sess, sess.source_map().new_source_file(name, source), override_span);
105 emit_unclosed_delims(&mut errors, &sess);
109 /// Creates a new parser from a source string.
110 pub fn new_parser_from_source_str(sess: &ParseSess, name: FileName, source: String) -> Parser<'_> {
111 panictry_buffer!(&sess.span_diagnostic, maybe_new_parser_from_source_str(sess, name, source))
114 /// Creates a new parser from a source string. Returns any buffered errors from lexing the initial
116 pub fn maybe_new_parser_from_source_str(
120 ) -> Result<Parser<'_>, Vec<Diagnostic>> {
122 maybe_source_file_to_parser(sess, sess.source_map().new_source_file(name, source))?;
123 parser.recurse_into_file_modules = false;
127 /// Creates a new parser, handling errors as appropriate if the file doesn't exist.
128 pub fn new_parser_from_file<'a>(sess: &'a ParseSess, path: &Path) -> Parser<'a> {
129 source_file_to_parser(sess, file_to_source_file(sess, path, None))
132 /// Creates a new parser, returning buffered diagnostics if the file doesn't exist,
133 /// or from lexing the initial token stream.
134 pub fn maybe_new_parser_from_file<'a>(
137 ) -> Result<Parser<'a>, Vec<Diagnostic>> {
138 let file = try_file_to_source_file(sess, path, None).map_err(|db| vec![db])?;
139 maybe_source_file_to_parser(sess, file)
142 /// Given a session, a crate config, a path, and a span, add
143 /// the file at the given path to the `source_map`, and returns a parser.
144 /// On an error, uses the given span as the source of the problem.
145 pub fn new_sub_parser_from_file<'a>(
148 directory_ownership: DirectoryOwnership,
149 module_name: Option<String>,
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, maybe_source_file_to_parser(sess, source_file))
163 /// Given a `source_file` and config, return a parser. Returns any buffered errors from lexing the
164 /// initial token stream.
165 fn maybe_source_file_to_parser(
167 source_file: Lrc<SourceFile>,
168 ) -> Result<Parser<'_>, Vec<Diagnostic>> {
169 let end_pos = source_file.end_pos;
170 let (stream, unclosed_delims) = maybe_file_to_stream(sess, source_file, None)?;
171 let mut parser = stream_to_parser(sess, stream, None);
172 parser.unclosed_delims = unclosed_delims;
173 if parser.token == token::Eof {
174 parser.token.span = Span::new(end_pos, end_pos, parser.token.span.ctxt());
180 // Must preserve old name for now, because `quote!` from the *existing*
181 // compiler expands into it.
182 pub fn new_parser_from_tts(sess: &ParseSess, tts: Vec<TokenTree>) -> Parser<'_> {
183 stream_to_parser(sess, tts.into_iter().collect(), crate::MACRO_ARGUMENTS)
188 /// Given a session and a path and an optional span (for error reporting),
189 /// add the path to the session's source_map and return the new source_file or
190 /// error when a file can't be read.
191 fn try_file_to_source_file(
194 spanopt: Option<Span>,
195 ) -> Result<Lrc<SourceFile>, Diagnostic> {
196 sess.source_map().load_file(path).map_err(|e| {
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>) -> Lrc<SourceFile> {
209 match try_file_to_source_file(sess, path, spanopt) {
210 Ok(source_file) => source_file,
212 sess.span_diagnostic.emit_diagnostic(&d);
218 /// Given a `source_file`, produces a sequence of token trees.
219 pub fn source_file_to_stream(
221 source_file: Lrc<SourceFile>,
222 override_span: Option<Span>,
223 ) -> (TokenStream, Vec<lexer::UnmatchedBrace>) {
224 panictry_buffer!(&sess.span_diagnostic, maybe_file_to_stream(sess, source_file, override_span))
227 /// Given a source file, produces a sequence of token trees. Returns any buffered errors from
228 /// parsing the token stream.
229 pub fn maybe_file_to_stream(
231 source_file: Lrc<SourceFile>,
232 override_span: Option<Span>,
233 ) -> Result<(TokenStream, Vec<lexer::UnmatchedBrace>), Vec<Diagnostic>> {
234 let srdr = lexer::StringReader::new(sess, source_file, override_span);
235 let (token_trees, unmatched_braces) = srdr.into_token_trees();
238 Ok(stream) => Ok((stream, unmatched_braces)),
240 let mut buffer = Vec::with_capacity(1);
241 err.buffer(&mut buffer);
242 // Not using `emit_unclosed_delims` to use `db.buffer`
243 for unmatched in unmatched_braces {
244 if let Some(err) = make_unclosed_delims_error(unmatched, &sess) {
245 err.buffer(&mut buffer);
253 /// Given a stream and the `ParseSess`, produces a parser.
254 pub fn stream_to_parser<'a>(
257 subparser_name: Option<&'static str>,
259 Parser::new(sess, stream, None, true, false, subparser_name)
262 /// Given a stream, the `ParseSess` and the base directory, produces a parser.
264 /// Use this function when you are creating a parser from the token stream
265 /// and also care about the current working directory of the parser (e.g.,
266 /// you are trying to resolve modules defined inside a macro invocation).
270 /// The main usage of this function is outside of rustc, for those who uses
271 /// librustc_ast as a library. Please do not remove this function while refactoring
272 /// just because it is not used in rustc codebase!
273 pub fn stream_to_parser_with_base_dir<'a>(
278 Parser::new(sess, stream, Some(base_dir), true, false, None)
281 /// Runs the given subparser `f` on the tokens of the given `attr`'s item.
282 pub fn parse_in<'a, T>(
286 mut f: impl FnMut(&mut Parser<'a>) -> PResult<'a, T>,
287 ) -> PResult<'a, T> {
288 let mut parser = Parser::new(sess, tts, None, false, false, Some(name));
289 let result = f(&mut parser)?;
290 if parser.token != token::Eof {
291 parser.unexpected()?;
296 // NOTE(Centril): The following probably shouldn't be here but it acknowledges the
297 // fact that architecturally, we are using parsing (read on below to understand why).
299 pub fn nt_to_tokenstream(nt: &Nonterminal, sess: &ParseSess, span: Span) -> TokenStream {
300 // A `Nonterminal` is often a parsed AST item. At this point we now
301 // need to convert the parsed AST to an actual token stream, e.g.
302 // un-parse it basically.
304 // Unfortunately there's not really a great way to do that in a
305 // guaranteed lossless fashion right now. The fallback here is to just
306 // stringify the AST node and reparse it, but this loses all span
309 // As a result, some AST nodes are annotated with the token stream they
310 // came from. Here we attempt to extract these lossless token streams
311 // before we fall back to the stringification.
312 let tokens = match *nt {
313 Nonterminal::NtItem(ref item) => {
314 prepend_attrs(sess, &item.attrs, item.tokens.as_ref(), span)
316 Nonterminal::NtIdent(ident, is_raw) => {
317 Some(tokenstream::TokenTree::token(token::Ident(ident.name, is_raw), ident.span).into())
319 Nonterminal::NtLifetime(ident) => {
320 Some(tokenstream::TokenTree::token(token::Lifetime(ident.name), ident.span).into())
322 Nonterminal::NtTT(ref tt) => Some(tt.clone().into()),
326 // FIXME(#43081): Avoid this pretty-print + reparse hack
327 let source = pprust::nonterminal_to_string(nt);
328 let filename = FileName::macro_expansion_source_code(&source);
329 let tokens_for_real = parse_stream_from_source_str(filename, source, sess, Some(span));
331 // During early phases of the compiler the AST could get modified
332 // directly (e.g., attributes added or removed) and the internal cache
333 // of tokens my not be invalidated or updated. Consequently if the
334 // "lossless" token stream disagrees with our actual stringification
335 // (which has historically been much more battle-tested) then we go
336 // with the lossy stream anyway (losing span information).
338 // Note that the comparison isn't `==` here to avoid comparing spans,
339 // but it *also* is a "probable" equality which is a pretty weird
340 // definition. We mostly want to catch actual changes to the AST
341 // like a `#[cfg]` being processed or some weird `macro_rules!`
344 // What we *don't* want to catch is the fact that a user-defined
345 // literal like `0xf` is stringified as `15`, causing the cached token
346 // stream to not be literal `==` token-wise (ignoring spans) to the
347 // token stream we got from stringification.
349 // Instead the "probably equal" check here is "does each token
350 // recursively have the same discriminant?" We basically don't look at
351 // the token values here and assume that such fine grained token stream
352 // modifications, including adding/removing typically non-semantic
353 // tokens such as extra braces and commas, don't happen.
354 if let Some(tokens) = tokens {
355 if tokens.probably_equal_for_proc_macro(&tokens_for_real) {
359 "cached tokens found, but they're not \"probably equal\", \
360 going with stringified version"
363 return tokens_for_real;
368 attrs: &[ast::Attribute],
369 tokens: Option<&tokenstream::TokenStream>,
370 span: rustc_span::Span,
371 ) -> Option<tokenstream::TokenStream> {
372 let tokens = tokens?;
373 if attrs.is_empty() {
374 return Some(tokens.clone());
376 let mut builder = tokenstream::TokenStreamBuilder::new();
380 ast::AttrStyle::Outer,
381 "inner attributes should prevent cached tokens from existing"
384 let source = pprust::attribute_to_string(attr);
385 let macro_filename = FileName::macro_expansion_source_code(&source);
387 let item = match attr.kind {
388 ast::AttrKind::Normal(ref item) => item,
389 ast::AttrKind::DocComment(_) => {
390 let stream = parse_stream_from_source_str(macro_filename, source, sess, Some(span));
391 builder.push(stream);
396 // synthesize # [ $path $tokens ] manually here
397 let mut brackets = tokenstream::TokenStreamBuilder::new();
399 // For simple paths, push the identifier directly
400 if item.path.segments.len() == 1 && item.path.segments[0].args.is_none() {
401 let ident = item.path.segments[0].ident;
402 let token = token::Ident(ident.name, ident.as_str().starts_with("r#"));
403 brackets.push(tokenstream::TokenTree::token(token, ident.span));
405 // ... and for more complicated paths, fall back to a reparse hack that
406 // should eventually be removed.
408 let stream = parse_stream_from_source_str(macro_filename, source, sess, Some(span));
409 brackets.push(stream);
412 brackets.push(item.args.outer_tokens());
414 // The span we list here for `#` and for `[ ... ]` are both wrong in
415 // that it encompasses more than each token, but it hopefully is "good
416 // enough" for now at least.
417 builder.push(tokenstream::TokenTree::token(token::Pound, attr.span));
418 let delim_span = tokenstream::DelimSpan::from_single(attr.span);
419 builder.push(tokenstream::TokenTree::Delimited(
421 token::DelimToken::Bracket,
425 builder.push(tokens.clone());
426 Some(builder.build())