compiler/rustc_parse/src/lexer/mod.rs

   1 use rustc_ast::ast::AttrStyle;
   2 use rustc_ast::token::{self, CommentKind, Token, TokenKind};
   3 use rustc_ast::tokenstream::{Spacing, TokenStream};
   4 use rustc_errors::{error_code, Applicability, DiagnosticBuilder, FatalError, PResult};
   5 use rustc_lexer::unescape::{self, Mode};
   6 use rustc_lexer::{Base, DocStyle, RawStrError};
   7 use rustc_session::parse::ParseSess;
   8 use rustc_span::symbol::{sym, Symbol};
   9 use rustc_span::{BytePos, Pos, Span};
  10
  11 use tracing::debug;
  12
  13 mod tokentrees;
  14 mod unescape_error_reporting;
  15 mod unicode_chars;
  16
  17 use unescape_error_reporting::{emit_unescape_error, escaped_char};
  18
  19 #[derive(Clone, Debug)]
  20 pub struct UnmatchedBrace {
  21     pub expected_delim: token::DelimToken,
  22     pub found_delim: Option<token::DelimToken>,
  23     pub found_span: Span,
  24     pub unclosed_span: Option<Span>,
  25     pub candidate_span: Option<Span>,
  26 }
  27
  28 crate fn parse_token_trees<'a>(
  29     sess: &'a ParseSess,
  30     src: &'a str,
  31     start_pos: BytePos,
  32     override_span: Option<Span>,
  33 ) -> (PResult<'a, TokenStream>, Vec<UnmatchedBrace>) {
  34     StringReader { sess, start_pos, pos: start_pos, end_src_index: src.len(), src, override_span }
  35         .into_token_trees()
  36 }
  37
  38 struct StringReader<'a> {
  39     sess: &'a ParseSess,
  40     /// Initial position, read-only.
  41     start_pos: BytePos,
  42     /// The absolute offset within the source_map of the current character.
  43     pos: BytePos,
  44     /// Stop reading src at this index.
  45     end_src_index: usize,
  46     /// Source text to tokenize.
  47     src: &'a str,
  48     override_span: Option<Span>,
  49 }
  50
  51 impl<'a> StringReader<'a> {
  52     fn mk_sp(&self, lo: BytePos, hi: BytePos) -> Span {
  53         self.override_span.unwrap_or_else(|| Span::with_root_ctxt(lo, hi))
  54     }
  55
  56     /// Returns the next token, and info about preceding whitespace, if any.
  57     fn next_token(&mut self) -> (Spacing, Token) {
  58         let mut spacing = Spacing::Joint;
  59
  60         // Skip `#!` at the start of the file
  61         let start_src_index = self.src_index(self.pos);
  62         let text: &str = &self.src[start_src_index..self.end_src_index];
  63         let is_beginning_of_file = self.pos == self.start_pos;
  64         if is_beginning_of_file {
  65             if let Some(shebang_len) = rustc_lexer::strip_shebang(text) {
  66                 self.pos = self.pos + BytePos::from_usize(shebang_len);
  67                 spacing = Spacing::Alone;
  68             }
  69         }
  70
  71         // Skip trivial (whitespace & comments) tokens
  72         loop {
  73             let start_src_index = self.src_index(self.pos);
  74             let text: &str = &self.src[start_src_index..self.end_src_index];
  75
  76             if text.is_empty() {
  77                 let span = self.mk_sp(self.pos, self.pos);
  78                 return (spacing, Token::new(token::Eof, span));
  79             }
  80
  81             let token = rustc_lexer::first_token(text);
  82
  83             let start = self.pos;
  84             self.pos = self.pos + BytePos::from_usize(token.len);
  85
  86             debug!("next_token: {:?}({:?})", token.kind, self.str_from(start));
  87
  88             match self.cook_lexer_token(token.kind, start) {
  89                 Some(kind) => {
  90                     let span = self.mk_sp(start, self.pos);
  91                     return (spacing, Token::new(kind, span));
  92                 }
  93                 None => spacing = Spacing::Alone,
  94             }
  95         }
  96     }
  97
  98     /// Report a fatal lexical error with a given span.
  99     fn fatal_span(&self, sp: Span, m: &str) -> FatalError {
 100         self.sess.span_diagnostic.span_fatal(sp, m)
 101     }
 102
 103     /// Report a lexical error with a given span.
 104     fn err_span(&self, sp: Span, m: &str) {
 105         self.sess.span_diagnostic.struct_span_err(sp, m).emit();
 106     }
 107
 108     /// Report a fatal error spanning [`from_pos`, `to_pos`).
 109     fn fatal_span_(&self, from_pos: BytePos, to_pos: BytePos, m: &str) -> FatalError {
 110         self.fatal_span(self.mk_sp(from_pos, to_pos), m)
 111     }
 112
 113     /// Report a lexical error spanning [`from_pos`, `to_pos`).
 114     fn err_span_(&self, from_pos: BytePos, to_pos: BytePos, m: &str) {
 115         self.err_span(self.mk_sp(from_pos, to_pos), m)
 116     }
 117
 118     fn struct_fatal_span_char(
 119         &self,
 120         from_pos: BytePos,
 121         to_pos: BytePos,
 122         m: &str,
 123         c: char,
 124     ) -> DiagnosticBuilder<'a> {
 125         self.sess
 126             .span_diagnostic
 127             .struct_span_fatal(self.mk_sp(from_pos, to_pos), &format!("{}: {}", m, escaped_char(c)))
 128     }
 129
 130     /// Turns simple `rustc_lexer::TokenKind` enum into a rich
 131     /// `librustc_ast::TokenKind`. This turns strings into interned
 132     /// symbols and runs additional validation.
 133     fn cook_lexer_token(&self, token: rustc_lexer::TokenKind, start: BytePos) -> Option<TokenKind> {
 134         Some(match token {
 135             rustc_lexer::TokenKind::LineComment { doc_style } => {
 136                 // Skip non-doc comments
 137                 let doc_style = doc_style?;
 138
 139                 // Opening delimiter of the length 3 is not included into the symbol.
 140                 let content_start = start + BytePos(3);
 141                 let content = self.str_from(content_start);
 142                 self.cook_doc_comment(content_start, content, CommentKind::Line, doc_style)
 143             }
 144             rustc_lexer::TokenKind::BlockComment { doc_style, terminated } => {
 145                 if !terminated {
 146                     let msg = match doc_style {
 147                         Some(_) => "unterminated block doc-comment",
 148                         None => "unterminated block comment",
 149                     };
 150                     let last_bpos = self.pos;
 151                     self.sess
 152                         .span_diagnostic
 153                         .struct_span_fatal_with_code(
 154                             self.mk_sp(start, last_bpos),
 155                             msg,
 156                             error_code!(E0758),
 157                         )
 158                         .emit();
 159                     FatalError.raise();
 160                 }
 161
 162                 // Skip non-doc comments
 163                 let doc_style = doc_style?;
 164
 165                 // Opening delimiter of the length 3 and closing delimiter of the length 2
 166                 // are not included into the symbol.
 167                 let content_start = start + BytePos(3);
 168                 let content_end = self.pos - BytePos(if terminated { 2 } else { 0 });
 169                 let content = self.str_from_to(content_start, content_end);
 170                 self.cook_doc_comment(content_start, content, CommentKind::Block, doc_style)
 171             }
 172             rustc_lexer::TokenKind::Whitespace => return None,
 173             rustc_lexer::TokenKind::Ident | rustc_lexer::TokenKind::RawIdent => {
 174                 let is_raw_ident = token == rustc_lexer::TokenKind::RawIdent;
 175                 let mut ident_start = start;
 176                 if is_raw_ident {
 177                     ident_start = ident_start + BytePos(2);
 178                 }
 179                 let sym = nfc_normalize(self.str_from(ident_start));
 180                 let span = self.mk_sp(start, self.pos);
 181                 self.sess.symbol_gallery.insert(sym, span);
 182                 if is_raw_ident {
 183                     if !sym.can_be_raw() {
 184                         self.err_span(span, &format!("`{}` cannot be a raw identifier", sym));
 185                     }
 186                     self.sess.raw_identifier_spans.borrow_mut().push(span);
 187                 }
 188                 token::Ident(sym, is_raw_ident)
 189             }
 190             rustc_lexer::TokenKind::Literal { kind, suffix_start } => {
 191                 let suffix_start = start + BytePos(suffix_start as u32);
 192                 let (kind, symbol) = self.cook_lexer_literal(start, suffix_start, kind);
 193                 let suffix = if suffix_start < self.pos {
 194                     let string = self.str_from(suffix_start);
 195                     if string == "_" {
 196                         self.sess
 197                             .span_diagnostic
 198                             .struct_span_warn(
 199                                 self.mk_sp(suffix_start, self.pos),
 200                                 "underscore literal suffix is not allowed",
 201                             )
 202                             .warn(
 203                                 "this was previously accepted by the compiler but is \
 204                                    being phased out; it will become a hard error in \
 205                                    a future release!",
 206                             )
 207                             .note(
 208                                 "see issue #42326 \
 209                                  <https://github.com/rust-lang/rust/issues/42326> \
 210                                  for more information",
 211                             )
 212                             .emit();
 213                         None
 214                     } else {
 215                         Some(Symbol::intern(string))
 216                     }
 217                 } else {
 218                     None
 219                 };
 220                 token::Literal(token::Lit { kind, symbol, suffix })
 221             }
 222             rustc_lexer::TokenKind::Lifetime { starts_with_number } => {
 223                 // Include the leading `'` in the real identifier, for macro
 224                 // expansion purposes. See #12512 for the gory details of why
 225                 // this is necessary.
 226                 let lifetime_name = self.str_from(start);
 227                 if starts_with_number {
 228                     self.err_span_(start, self.pos, "lifetimes cannot start with a number");
 229                 }
 230                 let ident = Symbol::intern(lifetime_name);
 231                 token::Lifetime(ident)
 232             }
 233             rustc_lexer::TokenKind::Semi => token::Semi,
 234             rustc_lexer::TokenKind::Comma => token::Comma,
 235             rustc_lexer::TokenKind::Dot => token::Dot,
 236             rustc_lexer::TokenKind::OpenParen => token::OpenDelim(token::Paren),
 237             rustc_lexer::TokenKind::CloseParen => token::CloseDelim(token::Paren),
 238             rustc_lexer::TokenKind::OpenBrace => token::OpenDelim(token::Brace),
 239             rustc_lexer::TokenKind::CloseBrace => token::CloseDelim(token::Brace),
 240             rustc_lexer::TokenKind::OpenBracket => token::OpenDelim(token::Bracket),
 241             rustc_lexer::TokenKind::CloseBracket => token::CloseDelim(token::Bracket),
 242             rustc_lexer::TokenKind::At => token::At,
 243             rustc_lexer::TokenKind::Pound => token::Pound,
 244             rustc_lexer::TokenKind::Tilde => token::Tilde,
 245             rustc_lexer::TokenKind::Question => token::Question,
 246             rustc_lexer::TokenKind::Colon => token::Colon,
 247             rustc_lexer::TokenKind::Dollar => token::Dollar,
 248             rustc_lexer::TokenKind::Eq => token::Eq,
 249             rustc_lexer::TokenKind::Bang => token::Not,
 250             rustc_lexer::TokenKind::Lt => token::Lt,
 251             rustc_lexer::TokenKind::Gt => token::Gt,
 252             rustc_lexer::TokenKind::Minus => token::BinOp(token::Minus),
 253             rustc_lexer::TokenKind::And => token::BinOp(token::And),
 254             rustc_lexer::TokenKind::Or => token::BinOp(token::Or),
 255             rustc_lexer::TokenKind::Plus => token::BinOp(token::Plus),
 256             rustc_lexer::TokenKind::Star => token::BinOp(token::Star),
 257             rustc_lexer::TokenKind::Slash => token::BinOp(token::Slash),
 258             rustc_lexer::TokenKind::Caret => token::BinOp(token::Caret),
 259             rustc_lexer::TokenKind::Percent => token::BinOp(token::Percent),
 260
 261             rustc_lexer::TokenKind::Unknown => {
 262                 let c = self.str_from(start).chars().next().unwrap();
 263                 let mut err =
 264                     self.struct_fatal_span_char(start, self.pos, "unknown start of token", c);
 265                 // FIXME: the lexer could be used to turn the ASCII version of unicode homoglyphs,
 266                 // instead of keeping a table in `check_for_substitution`into the token. Ideally,
 267                 // this should be inside `rustc_lexer`. However, we should first remove compound
 268                 // tokens like `<<` from `rustc_lexer`, and then add fancier error recovery to it,
 269                 // as there will be less overall work to do this way.
 270                 let token = unicode_chars::check_for_substitution(self, start, c, &mut err);
 271                 if c == '\x00' {
 272                     err.help("source files must contain UTF-8 encoded text, unexpected null bytes might occur when a different encoding is used");
 273                 }
 274                 err.emit();
 275                 token?
 276             }
 277         })
 278     }
 279
 280     fn cook_doc_comment(
 281         &self,
 282         content_start: BytePos,
 283         content: &str,
 284         comment_kind: CommentKind,
 285         doc_style: DocStyle,
 286     ) -> TokenKind {
 287         if content.contains('\r') {
 288             for (idx, _) in content.char_indices().filter(|&(_, c)| c == '\r') {
 289                 self.err_span_(
 290                     content_start + BytePos(idx as u32),
 291                     content_start + BytePos(idx as u32 + 1),
 292                     match comment_kind {
 293                         CommentKind::Line => "bare CR not allowed in doc-comment",
 294                         CommentKind::Block => "bare CR not allowed in block doc-comment",
 295                     },
 296                 );
 297             }
 298         }
 299
 300         let attr_style = match doc_style {
 301             DocStyle::Outer => AttrStyle::Outer,
 302             DocStyle::Inner => AttrStyle::Inner,
 303         };
 304
 305         token::DocComment(comment_kind, attr_style, Symbol::intern(content))
 306     }
 307
 308     fn cook_lexer_literal(
 309         &self,
 310         start: BytePos,
 311         suffix_start: BytePos,
 312         kind: rustc_lexer::LiteralKind,
 313     ) -> (token::LitKind, Symbol) {
 314         // prefix means `"` or `br"` or `r###"`, ...
 315         let (lit_kind, mode, prefix_len, postfix_len) = match kind {
 316             rustc_lexer::LiteralKind::Char { terminated } => {
 317                 if !terminated {
 318                     self.sess
 319                         .span_diagnostic
 320                         .struct_span_fatal_with_code(
 321                             self.mk_sp(start, suffix_start),
 322                             "unterminated character literal",
 323                             error_code!(E0762),
 324                         )
 325                         .emit();
 326                     FatalError.raise();
 327                 }
 328                 (token::Char, Mode::Char, 1, 1) // ' '
 329             }
 330             rustc_lexer::LiteralKind::Byte { terminated } => {
 331                 if !terminated {
 332                     self.sess
 333                         .span_diagnostic
 334                         .struct_span_fatal_with_code(
 335                             self.mk_sp(start + BytePos(1), suffix_start),
 336                             "unterminated byte constant",
 337                             error_code!(E0763),
 338                         )
 339                         .emit();
 340                     FatalError.raise();
 341                 }
 342                 (token::Byte, Mode::Byte, 2, 1) // b' '
 343             }
 344             rustc_lexer::LiteralKind::Str { terminated } => {
 345                 if !terminated {
 346                     self.sess
 347                         .span_diagnostic
 348                         .struct_span_fatal_with_code(
 349                             self.mk_sp(start, suffix_start),
 350                             "unterminated double quote string",
 351                             error_code!(E0765),
 352                         )
 353                         .emit();
 354                     FatalError.raise();
 355                 }
 356                 (token::Str, Mode::Str, 1, 1) // " "
 357             }
 358             rustc_lexer::LiteralKind::ByteStr { terminated } => {
 359                 if !terminated {
 360                     self.sess
 361                         .span_diagnostic
 362                         .struct_span_fatal_with_code(
 363                             self.mk_sp(start + BytePos(1), suffix_start),
 364                             "unterminated double quote byte string",
 365                             error_code!(E0766),
 366                         )
 367                         .emit();
 368                     FatalError.raise();
 369                 }
 370                 (token::ByteStr, Mode::ByteStr, 2, 1) // b" "
 371             }
 372             rustc_lexer::LiteralKind::RawStr { n_hashes, err } => {
 373                 self.report_raw_str_error(start, err);
 374                 let n = u32::from(n_hashes);
 375                 (token::StrRaw(n_hashes), Mode::RawStr, 2 + n, 1 + n) // r##" "##
 376             }
 377             rustc_lexer::LiteralKind::RawByteStr { n_hashes, err } => {
 378                 self.report_raw_str_error(start, err);
 379                 let n = u32::from(n_hashes);
 380                 (token::ByteStrRaw(n_hashes), Mode::RawByteStr, 3 + n, 1 + n) // br##" "##
 381             }
 382             rustc_lexer::LiteralKind::Int { base, empty_int } => {
 383                 return if empty_int {
 384                     self.sess
 385                         .span_diagnostic
 386                         .struct_span_err_with_code(
 387                             self.mk_sp(start, suffix_start),
 388                             "no valid digits found for number",
 389                             error_code!(E0768),
 390                         )
 391                         .emit();
 392                     (token::Integer, sym::integer(0))
 393                 } else {
 394                     self.validate_int_literal(base, start, suffix_start);
 395                     (token::Integer, self.symbol_from_to(start, suffix_start))
 396                 };
 397             }
 398             rustc_lexer::LiteralKind::Float { base, empty_exponent } => {
 399                 if empty_exponent {
 400                     self.err_span_(start, self.pos, "expected at least one digit in exponent");
 401                 }
 402
 403                 match base {
 404                     Base::Hexadecimal => self.err_span_(
 405                         start,
 406                         suffix_start,
 407                         "hexadecimal float literal is not supported",
 408                     ),
 409                     Base::Octal => {
 410                         self.err_span_(start, suffix_start, "octal float literal is not supported")
 411                     }
 412                     Base::Binary => {
 413                         self.err_span_(start, suffix_start, "binary float literal is not supported")
 414                     }
 415                     _ => (),
 416                 }
 417
 418                 let id = self.symbol_from_to(start, suffix_start);
 419                 return (token::Float, id);
 420             }
 421         };
 422         let content_start = start + BytePos(prefix_len);
 423         let content_end = suffix_start - BytePos(postfix_len);
 424         let id = self.symbol_from_to(content_start, content_end);
 425         self.validate_literal_escape(mode, content_start, content_end, prefix_len, postfix_len);
 426         (lit_kind, id)
 427     }
 428
 429     #[inline]
 430     fn src_index(&self, pos: BytePos) -> usize {
 431         (pos - self.start_pos).to_usize()
 432     }
 433
 434     /// Slice of the source text from `start` up to but excluding `self.pos`,
 435     /// meaning the slice does not include the character `self.ch`.
 436     fn str_from(&self, start: BytePos) -> &str {
 437         self.str_from_to(start, self.pos)
 438     }
 439
 440     /// As symbol_from, with an explicit endpoint.
 441     fn symbol_from_to(&self, start: BytePos, end: BytePos) -> Symbol {
 442         debug!("taking an ident from {:?} to {:?}", start, end);
 443         Symbol::intern(self.str_from_to(start, end))
 444     }
 445
 446     /// Slice of the source text spanning from `start` up to but excluding `end`.
 447     fn str_from_to(&self, start: BytePos, end: BytePos) -> &str {
 448         &self.src[self.src_index(start)..self.src_index(end)]
 449     }
 450
 451     fn report_raw_str_error(&self, start: BytePos, opt_err: Option<RawStrError>) {
 452         match opt_err {
 453             Some(RawStrError::InvalidStarter { bad_char }) => {
 454                 self.report_non_started_raw_string(start, bad_char)
 455             }
 456             Some(RawStrError::NoTerminator { expected, found, possible_terminator_offset }) => self
 457                 .report_unterminated_raw_string(start, expected, possible_terminator_offset, found),
 458             Some(RawStrError::TooManyDelimiters { found }) => {
 459                 self.report_too_many_hashes(start, found)
 460             }
 461             None => (),
 462         }
 463     }
 464
 465     fn report_non_started_raw_string(&self, start: BytePos, bad_char: char) -> ! {
 466         self.struct_fatal_span_char(
 467             start,
 468             self.pos,
 469             "found invalid character; only `#` is allowed in raw string delimitation",
 470             bad_char,
 471         )
 472         .emit();
 473         FatalError.raise()
 474     }
 475
 476     fn report_unterminated_raw_string(
 477         &self,
 478         start: BytePos,
 479         n_hashes: usize,
 480         possible_offset: Option<usize>,
 481         found_terminators: usize,
 482     ) -> ! {
 483         let mut err = self.sess.span_diagnostic.struct_span_fatal_with_code(
 484             self.mk_sp(start, start),
 485             "unterminated raw string",
 486             error_code!(E0748),
 487         );
 488
 489         err.span_label(self.mk_sp(start, start), "unterminated raw string");
 490
 491         if n_hashes > 0 {
 492             err.note(&format!(
 493                 "this raw string should be terminated with `\"{}`",
 494                 "#".repeat(n_hashes)
 495             ));
 496         }
 497
 498         if let Some(possible_offset) = possible_offset {
 499             let lo = start + BytePos(possible_offset as u32);
 500             let hi = lo + BytePos(found_terminators as u32);
 501             let span = self.mk_sp(lo, hi);
 502             err.span_suggestion(
 503                 span,
 504                 "consider terminating the string here",
 505                 "#".repeat(n_hashes),
 506                 Applicability::MaybeIncorrect,
 507             );
 508         }
 509
 510         err.emit();
 511         FatalError.raise()
 512     }
 513
 514     /// Note: It was decided to not add a test case, because it would be too big.
 515     /// <https://github.com/rust-lang/rust/pull/50296#issuecomment-392135180>
 516     fn report_too_many_hashes(&self, start: BytePos, found: usize) -> ! {
 517         self.fatal_span_(
 518             start,
 519             self.pos,
 520             &format!(
 521                 "too many `#` symbols: raw strings may be delimited \
 522                 by up to 65535 `#` symbols, but found {}",
 523                 found
 524             ),
 525         )
 526         .raise();
 527     }
 528
 529     fn validate_literal_escape(
 530         &self,
 531         mode: Mode,
 532         content_start: BytePos,
 533         content_end: BytePos,
 534         prefix_len: u32,
 535         postfix_len: u32,
 536     ) {
 537         let lit_content = self.str_from_to(content_start, content_end);
 538         unescape::unescape_literal(lit_content, mode, &mut |range, result| {
 539             // Here we only check for errors. The actual unescaping is done later.
 540             if let Err(err) = result {
 541                 let span_with_quotes = self
 542                     .mk_sp(content_start - BytePos(prefix_len), content_end + BytePos(postfix_len));
 543                 let (start, end) = (range.start as u32, range.end as u32);
 544                 let lo = content_start + BytePos(start);
 545                 let hi = lo + BytePos(end - start);
 546                 let span = self.mk_sp(lo, hi);
 547                 emit_unescape_error(
 548                     &self.sess.span_diagnostic,
 549                     lit_content,
 550                     span_with_quotes,
 551                     span,
 552                     mode,
 553                     range,
 554                     err,
 555                 );
 556             }
 557         });
 558     }
 559
 560     fn validate_int_literal(&self, base: Base, content_start: BytePos, content_end: BytePos) {
 561         let base = match base {
 562             Base::Binary => 2,
 563             Base::Octal => 8,
 564             _ => return,
 565         };
 566         let s = self.str_from_to(content_start + BytePos(2), content_end);
 567         for (idx, c) in s.char_indices() {
 568             let idx = idx as u32;
 569             if c != '_' && c.to_digit(base).is_none() {
 570                 let lo = content_start + BytePos(2 + idx);
 571                 let hi = content_start + BytePos(2 + idx + c.len_utf8() as u32);
 572                 self.err_span_(lo, hi, &format!("invalid digit for a base {} literal", base));
 573             }
 574         }
 575     }
 576 }
 577
 578 pub fn nfc_normalize(string: &str) -> Symbol {
 579     use unicode_normalization::{is_nfc_quick, IsNormalized, UnicodeNormalization};
 580     match is_nfc_quick(string.chars()) {
 581         IsNormalized::Yes => Symbol::intern(string),
 582         _ => {
 583             let normalized_str: String = string.chars().nfc().collect();
 584             Symbol::intern(&normalized_str)
 585         }
 586     }
 587 }