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, push_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         let mut m = m.to_string();
 126         m.push_str(": ");
 127         push_escaped_char(&mut m, c);
 128
 129         self.sess.span_diagnostic.struct_span_fatal(self.mk_sp(from_pos, to_pos), &m[..])
 130     }
 131
 132     /// Turns simple `rustc_lexer::TokenKind` enum into a rich
 133     /// `librustc_ast::TokenKind`. This turns strings into interned
 134     /// symbols and runs additional validation.
 135     fn cook_lexer_token(&self, token: rustc_lexer::TokenKind, start: BytePos) -> Option<TokenKind> {
 136         Some(match token {
 137             rustc_lexer::TokenKind::LineComment { doc_style } => {
 138                 // Skip non-doc comments
 139                 let doc_style = doc_style?;
 140
 141                 // Opening delimiter of the length 3 is not included into the symbol.
 142                 let content_start = start + BytePos(3);
 143                 let content = self.str_from(content_start);
 144                 self.cook_doc_comment(content_start, content, CommentKind::Line, doc_style)
 145             }
 146             rustc_lexer::TokenKind::BlockComment { doc_style, terminated } => {
 147                 if !terminated {
 148                     let msg = match doc_style {
 149                         Some(_) => "unterminated block doc-comment",
 150                         None => "unterminated block comment",
 151                     };
 152                     let last_bpos = self.pos;
 153                     self.sess
 154                         .span_diagnostic
 155                         .struct_span_fatal_with_code(
 156                             self.mk_sp(start, last_bpos),
 157                             msg,
 158                             error_code!(E0758),
 159                         )
 160                         .emit();
 161                     FatalError.raise();
 162                 }
 163
 164                 // Skip non-doc comments
 165                 let doc_style = doc_style?;
 166
 167                 // Opening delimiter of the length 3 and closing delimiter of the length 2
 168                 // are not included into the symbol.
 169                 let content_start = start + BytePos(3);
 170                 let content_end = self.pos - BytePos(if terminated { 2 } else { 0 });
 171                 let content = self.str_from_to(content_start, content_end);
 172                 self.cook_doc_comment(content_start, content, CommentKind::Block, doc_style)
 173             }
 174             rustc_lexer::TokenKind::Whitespace => return None,
 175             rustc_lexer::TokenKind::Ident | rustc_lexer::TokenKind::RawIdent => {
 176                 let is_raw_ident = token == rustc_lexer::TokenKind::RawIdent;
 177                 let mut ident_start = start;
 178                 if is_raw_ident {
 179                     ident_start = ident_start + BytePos(2);
 180                 }
 181                 let sym = nfc_normalize(self.str_from(ident_start));
 182                 let span = self.mk_sp(start, self.pos);
 183                 self.sess.symbol_gallery.insert(sym, span);
 184                 if is_raw_ident {
 185                     if !sym.can_be_raw() {
 186                         self.err_span(span, &format!("`{}` cannot be a raw identifier", sym));
 187                     }
 188                     self.sess.raw_identifier_spans.borrow_mut().push(span);
 189                 }
 190                 token::Ident(sym, is_raw_ident)
 191             }
 192             rustc_lexer::TokenKind::Literal { kind, suffix_start } => {
 193                 let suffix_start = start + BytePos(suffix_start as u32);
 194                 let (kind, symbol) = self.cook_lexer_literal(start, suffix_start, kind);
 195                 let suffix = if suffix_start < self.pos {
 196                     let string = self.str_from(suffix_start);
 197                     if string == "_" {
 198                         self.sess
 199                             .span_diagnostic
 200                             .struct_span_warn(
 201                                 self.mk_sp(suffix_start, self.pos),
 202                                 "underscore literal suffix is not allowed",
 203                             )
 204                             .warn(
 205                                 "this was previously accepted by the compiler but is \
 206                                    being phased out; it will become a hard error in \
 207                                    a future release!",
 208                             )
 209                             .note(
 210                                 "see issue #42326 \
 211                                  <https://github.com/rust-lang/rust/issues/42326> \
 212                                  for more information",
 213                             )
 214                             .emit();
 215                         None
 216                     } else {
 217                         Some(Symbol::intern(string))
 218                     }
 219                 } else {
 220                     None
 221                 };
 222                 token::Literal(token::Lit { kind, symbol, suffix })
 223             }
 224             rustc_lexer::TokenKind::Lifetime { starts_with_number } => {
 225                 // Include the leading `'` in the real identifier, for macro
 226                 // expansion purposes. See #12512 for the gory details of why
 227                 // this is necessary.
 228                 let lifetime_name = self.str_from(start);
 229                 if starts_with_number {
 230                     self.err_span_(start, self.pos, "lifetimes cannot start with a number");
 231                 }
 232                 let ident = Symbol::intern(lifetime_name);
 233                 token::Lifetime(ident)
 234             }
 235             rustc_lexer::TokenKind::Semi => token::Semi,
 236             rustc_lexer::TokenKind::Comma => token::Comma,
 237             rustc_lexer::TokenKind::Dot => token::Dot,
 238             rustc_lexer::TokenKind::OpenParen => token::OpenDelim(token::Paren),
 239             rustc_lexer::TokenKind::CloseParen => token::CloseDelim(token::Paren),
 240             rustc_lexer::TokenKind::OpenBrace => token::OpenDelim(token::Brace),
 241             rustc_lexer::TokenKind::CloseBrace => token::CloseDelim(token::Brace),
 242             rustc_lexer::TokenKind::OpenBracket => token::OpenDelim(token::Bracket),
 243             rustc_lexer::TokenKind::CloseBracket => token::CloseDelim(token::Bracket),
 244             rustc_lexer::TokenKind::At => token::At,
 245             rustc_lexer::TokenKind::Pound => token::Pound,
 246             rustc_lexer::TokenKind::Tilde => token::Tilde,
 247             rustc_lexer::TokenKind::Question => token::Question,
 248             rustc_lexer::TokenKind::Colon => token::Colon,
 249             rustc_lexer::TokenKind::Dollar => token::Dollar,
 250             rustc_lexer::TokenKind::Eq => token::Eq,
 251             rustc_lexer::TokenKind::Bang => token::Not,
 252             rustc_lexer::TokenKind::Lt => token::Lt,
 253             rustc_lexer::TokenKind::Gt => token::Gt,
 254             rustc_lexer::TokenKind::Minus => token::BinOp(token::Minus),
 255             rustc_lexer::TokenKind::And => token::BinOp(token::And),
 256             rustc_lexer::TokenKind::Or => token::BinOp(token::Or),
 257             rustc_lexer::TokenKind::Plus => token::BinOp(token::Plus),
 258             rustc_lexer::TokenKind::Star => token::BinOp(token::Star),
 259             rustc_lexer::TokenKind::Slash => token::BinOp(token::Slash),
 260             rustc_lexer::TokenKind::Caret => token::BinOp(token::Caret),
 261             rustc_lexer::TokenKind::Percent => token::BinOp(token::Percent),
 262
 263             rustc_lexer::TokenKind::Unknown => {
 264                 let c = self.str_from(start).chars().next().unwrap();
 265                 let mut err =
 266                     self.struct_fatal_span_char(start, self.pos, "unknown start of token", c);
 267                 // FIXME: the lexer could be used to turn the ASCII version of unicode homoglyphs,
 268                 // instead of keeping a table in `check_for_substitution`into the token. Ideally,
 269                 // this should be inside `rustc_lexer`. However, we should first remove compound
 270                 // tokens like `<<` from `rustc_lexer`, and then add fancier error recovery to it,
 271                 // as there will be less overall work to do this way.
 272                 let token = unicode_chars::check_for_substitution(self, start, c, &mut err);
 273                 err.emit();
 274                 token?
 275             }
 276         })
 277     }
 278
 279     fn cook_doc_comment(
 280         &self,
 281         content_start: BytePos,
 282         content: &str,
 283         comment_kind: CommentKind,
 284         doc_style: DocStyle,
 285     ) -> TokenKind {
 286         if content.contains('\r') {
 287             for (idx, _) in content.char_indices().filter(|&(_, c)| c == '\r') {
 288                 self.err_span_(
 289                     content_start + BytePos(idx as u32),
 290                     content_start + BytePos(idx as u32 + 1),
 291                     match comment_kind {
 292                         CommentKind::Line => "bare CR not allowed in doc-comment",
 293                         CommentKind::Block => "bare CR not allowed in block doc-comment",
 294                     },
 295                 );
 296             }
 297         }
 298
 299         let attr_style = match doc_style {
 300             DocStyle::Outer => AttrStyle::Outer,
 301             DocStyle::Inner => AttrStyle::Inner,
 302         };
 303
 304         token::DocComment(comment_kind, attr_style, Symbol::intern(content))
 305     }
 306
 307     fn cook_lexer_literal(
 308         &self,
 309         start: BytePos,
 310         suffix_start: BytePos,
 311         kind: rustc_lexer::LiteralKind,
 312     ) -> (token::LitKind, Symbol) {
 313         // prefix means `"` or `br"` or `r###"`, ...
 314         let (lit_kind, mode, prefix_len, postfix_len) = match kind {
 315             rustc_lexer::LiteralKind::Char { terminated } => {
 316                 if !terminated {
 317                     self.sess
 318                         .span_diagnostic
 319                         .struct_span_fatal_with_code(
 320                             self.mk_sp(start, suffix_start),
 321                             "unterminated character literal",
 322                             error_code!(E0762),
 323                         )
 324                         .emit();
 325                     FatalError.raise();
 326                 }
 327                 (token::Char, Mode::Char, 1, 1) // ' '
 328             }
 329             rustc_lexer::LiteralKind::Byte { terminated } => {
 330                 if !terminated {
 331                     self.sess
 332                         .span_diagnostic
 333                         .struct_span_fatal_with_code(
 334                             self.mk_sp(start + BytePos(1), suffix_start),
 335                             "unterminated byte constant",
 336                             error_code!(E0763),
 337                         )
 338                         .emit();
 339                     FatalError.raise();
 340                 }
 341                 (token::Byte, Mode::Byte, 2, 1) // b' '
 342             }
 343             rustc_lexer::LiteralKind::Str { terminated } => {
 344                 if !terminated {
 345                     self.sess
 346                         .span_diagnostic
 347                         .struct_span_fatal_with_code(
 348                             self.mk_sp(start, suffix_start),
 349                             "unterminated double quote string",
 350                             error_code!(E0765),
 351                         )
 352                         .emit();
 353                     FatalError.raise();
 354                 }
 355                 (token::Str, Mode::Str, 1, 1) // " "
 356             }
 357             rustc_lexer::LiteralKind::ByteStr { terminated } => {
 358                 if !terminated {
 359                     self.sess
 360                         .span_diagnostic
 361                         .struct_span_fatal_with_code(
 362                             self.mk_sp(start + BytePos(1), suffix_start),
 363                             "unterminated double quote byte string",
 364                             error_code!(E0766),
 365                         )
 366                         .emit();
 367                     FatalError.raise();
 368                 }
 369                 (token::ByteStr, Mode::ByteStr, 2, 1) // b" "
 370             }
 371             rustc_lexer::LiteralKind::RawStr { n_hashes, err } => {
 372                 self.report_raw_str_error(start, err);
 373                 let n = u32::from(n_hashes);
 374                 (token::StrRaw(n_hashes), Mode::RawStr, 2 + n, 1 + n) // r##" "##
 375             }
 376             rustc_lexer::LiteralKind::RawByteStr { n_hashes, err } => {
 377                 self.report_raw_str_error(start, err);
 378                 let n = u32::from(n_hashes);
 379                 (token::ByteStrRaw(n_hashes), Mode::RawByteStr, 3 + n, 1 + n) // br##" "##
 380             }
 381             rustc_lexer::LiteralKind::Int { base, empty_int } => {
 382                 return if empty_int {
 383                     self.sess
 384                         .span_diagnostic
 385                         .struct_span_err_with_code(
 386                             self.mk_sp(start, suffix_start),
 387                             "no valid digits found for number",
 388                             error_code!(E0768),
 389                         )
 390                         .emit();
 391                     (token::Integer, sym::integer(0))
 392                 } else {
 393                     self.validate_int_literal(base, start, suffix_start);
 394                     (token::Integer, self.symbol_from_to(start, suffix_start))
 395                 };
 396             }
 397             rustc_lexer::LiteralKind::Float { base, empty_exponent } => {
 398                 if empty_exponent {
 399                     self.err_span_(start, self.pos, "expected at least one digit in exponent");
 400                 }
 401
 402                 match base {
 403                     Base::Hexadecimal => self.err_span_(
 404                         start,
 405                         suffix_start,
 406                         "hexadecimal float literal is not supported",
 407                     ),
 408                     Base::Octal => {
 409                         self.err_span_(start, suffix_start, "octal float literal is not supported")
 410                     }
 411                     Base::Binary => {
 412                         self.err_span_(start, suffix_start, "binary float literal is not supported")
 413                     }
 414                     _ => (),
 415                 }
 416
 417                 let id = self.symbol_from_to(start, suffix_start);
 418                 return (token::Float, id);
 419             }
 420         };
 421         let content_start = start + BytePos(prefix_len);
 422         let content_end = suffix_start - BytePos(postfix_len);
 423         let id = self.symbol_from_to(content_start, content_end);
 424         self.validate_literal_escape(mode, content_start, content_end);
 425         (lit_kind, id)
 426     }
 427
 428     #[inline]
 429     fn src_index(&self, pos: BytePos) -> usize {
 430         (pos - self.start_pos).to_usize()
 431     }
 432
 433     /// Slice of the source text from `start` up to but excluding `self.pos`,
 434     /// meaning the slice does not include the character `self.ch`.
 435     fn str_from(&self, start: BytePos) -> &str {
 436         self.str_from_to(start, self.pos)
 437     }
 438
 439     /// As symbol_from, with an explicit endpoint.
 440     fn symbol_from_to(&self, start: BytePos, end: BytePos) -> Symbol {
 441         debug!("taking an ident from {:?} to {:?}", start, end);
 442         Symbol::intern(self.str_from_to(start, end))
 443     }
 444
 445     /// Slice of the source text spanning from `start` up to but excluding `end`.
 446     fn str_from_to(&self, start: BytePos, end: BytePos) -> &str {
 447         &self.src[self.src_index(start)..self.src_index(end)]
 448     }
 449
 450     fn report_raw_str_error(&self, start: BytePos, opt_err: Option<RawStrError>) {
 451         match opt_err {
 452             Some(RawStrError::InvalidStarter { bad_char }) => {
 453                 self.report_non_started_raw_string(start, bad_char)
 454             }
 455             Some(RawStrError::NoTerminator { expected, found, possible_terminator_offset }) => self
 456                 .report_unterminated_raw_string(start, expected, possible_terminator_offset, found),
 457             Some(RawStrError::TooManyDelimiters { found }) => {
 458                 self.report_too_many_hashes(start, found)
 459             }
 460             None => (),
 461         }
 462     }
 463
 464     fn report_non_started_raw_string(&self, start: BytePos, bad_char: char) -> ! {
 465         self.struct_fatal_span_char(
 466             start,
 467             self.pos,
 468             "found invalid character; only `#` is allowed in raw string delimitation",
 469             bad_char,
 470         )
 471         .emit();
 472         FatalError.raise()
 473     }
 474
 475     fn report_unterminated_raw_string(
 476         &self,
 477         start: BytePos,
 478         n_hashes: usize,
 479         possible_offset: Option<usize>,
 480         found_terminators: usize,
 481     ) -> ! {
 482         let mut err = self.sess.span_diagnostic.struct_span_fatal_with_code(
 483             self.mk_sp(start, start),
 484             "unterminated raw string",
 485             error_code!(E0748),
 486         );
 487
 488         err.span_label(self.mk_sp(start, start), "unterminated raw string");
 489
 490         if n_hashes > 0 {
 491             err.note(&format!(
 492                 "this raw string should be terminated with `\"{}`",
 493                 "#".repeat(n_hashes)
 494             ));
 495         }
 496
 497         if let Some(possible_offset) = possible_offset {
 498             let lo = start + BytePos(possible_offset as u32);
 499             let hi = lo + BytePos(found_terminators as u32);
 500             let span = self.mk_sp(lo, hi);
 501             err.span_suggestion(
 502                 span,
 503                 "consider terminating the string here",
 504                 "#".repeat(n_hashes),
 505                 Applicability::MaybeIncorrect,
 506             );
 507         }
 508
 509         err.emit();
 510         FatalError.raise()
 511     }
 512
 513     /// Note: It was decided to not add a test case, because it would be to big.
 514     /// https://github.com/rust-lang/rust/pull/50296#issuecomment-392135180
 515     fn report_too_many_hashes(&self, start: BytePos, found: usize) -> ! {
 516         self.fatal_span_(
 517             start,
 518             self.pos,
 519             &format!(
 520                 "too many `#` symbols: raw strings may be delimited \
 521                 by up to 65535 `#` symbols, but found {}",
 522                 found
 523             ),
 524         )
 525         .raise();
 526     }
 527
 528     fn validate_literal_escape(&self, mode: Mode, content_start: BytePos, content_end: BytePos) {
 529         let lit_content = self.str_from_to(content_start, content_end);
 530         unescape::unescape_literal(lit_content, mode, &mut |range, result| {
 531             // Here we only check for errors. The actual unescaping is done later.
 532             if let Err(err) = result {
 533                 let span_with_quotes =
 534                     self.mk_sp(content_start - BytePos(1), content_end + BytePos(1));
 535                 emit_unescape_error(
 536                     &self.sess.span_diagnostic,
 537                     lit_content,
 538                     span_with_quotes,
 539                     mode,
 540                     range,
 541                     err,
 542                 );
 543             }
 544         });
 545     }
 546
 547     fn validate_int_literal(&self, base: Base, content_start: BytePos, content_end: BytePos) {
 548         let base = match base {
 549             Base::Binary => 2,
 550             Base::Octal => 8,
 551             _ => return,
 552         };
 553         let s = self.str_from_to(content_start + BytePos(2), content_end);
 554         for (idx, c) in s.char_indices() {
 555             let idx = idx as u32;
 556             if c != '_' && c.to_digit(base).is_none() {
 557                 let lo = content_start + BytePos(2 + idx);
 558                 let hi = content_start + BytePos(2 + idx + c.len_utf8() as u32);
 559                 self.err_span_(lo, hi, &format!("invalid digit for a base {} literal", base));
 560             }
 561         }
 562     }
 563 }
 564
 565 pub fn nfc_normalize(string: &str) -> Symbol {
 566     use unicode_normalization::{is_nfc_quick, IsNormalized, UnicodeNormalization};
 567     match is_nfc_quick(string.chars()) {
 568         IsNormalized::Yes => Symbol::intern(string),
 569         _ => {
 570             let normalized_str: String = string.chars().nfc().collect();
 571             Symbol::intern(&normalized_str)
 572         }
 573     }
 574 }