compiler/rustc_parse/src/lexer/mod.rs

   1 use crate::lexer::unicode_chars::UNICODE_ARRAY;
   2 use rustc_ast::ast::{self, AttrStyle};
   3 use rustc_ast::token::{self, CommentKind, Delimiter, Token, TokenKind};
   4 use rustc_ast::tokenstream::TokenStream;
   5 use rustc_ast::util::unicode::contains_text_flow_control_chars;
   6 use rustc_errors::{
   7     error_code, Applicability, DiagnosticBuilder, ErrorGuaranteed, PResult, StashKey,
   8 };
   9 use rustc_lexer::unescape::{self, Mode};
  10 use rustc_lexer::Cursor;
  11 use rustc_lexer::{Base, DocStyle, RawStrError};
  12 use rustc_session::lint::builtin::{
  13     RUST_2021_PREFIXES_INCOMPATIBLE_SYNTAX, TEXT_DIRECTION_CODEPOINT_IN_COMMENT,
  14 };
  15 use rustc_session::lint::BuiltinLintDiagnostics;
  16 use rustc_session::parse::ParseSess;
  17 use rustc_span::symbol::{sym, Symbol};
  18 use rustc_span::{edition::Edition, BytePos, Pos, Span};
  19
  20 mod diagnostics;
  21 mod tokentrees;
  22 mod unescape_error_reporting;
  23 mod unicode_chars;
  24
  25 use unescape_error_reporting::{emit_unescape_error, escaped_char};
  26
  27 // This type is used a lot. Make sure it doesn't unintentionally get bigger.
  28 //
  29 // This assertion is in this crate, rather than in `rustc_lexer`, because that
  30 // crate cannot depend on `rustc_data_structures`.
  31 #[cfg(all(target_arch = "x86_64", target_pointer_width = "64"))]
  32 rustc_data_structures::static_assert_size!(rustc_lexer::Token, 12);
  33
  34 #[derive(Clone, Debug)]
  35 pub struct UnmatchedBrace {
  36     pub expected_delim: Delimiter,
  37     pub found_delim: Option<Delimiter>,
  38     pub found_span: Span,
  39     pub unclosed_span: Option<Span>,
  40     pub candidate_span: Option<Span>,
  41 }
  42
  43 pub(crate) fn parse_token_trees<'a>(
  44     sess: &'a ParseSess,
  45     mut src: &'a str,
  46     mut start_pos: BytePos,
  47     override_span: Option<Span>,
  48 ) -> (PResult<'a, TokenStream>, Vec<UnmatchedBrace>) {
  49     // Skip `#!`, if present.
  50     if let Some(shebang_len) = rustc_lexer::strip_shebang(src) {
  51         src = &src[shebang_len..];
  52         start_pos = start_pos + BytePos::from_usize(shebang_len);
  53     }
  54
  55     let cursor = Cursor::new(src);
  56     let string_reader = StringReader {
  57         sess,
  58         start_pos,
  59         pos: start_pos,
  60         src,
  61         cursor,
  62         override_span,
  63         nbsp_is_whitespace: false,
  64     };
  65     tokentrees::TokenTreesReader::parse_all_token_trees(string_reader)
  66 }
  67
  68 struct StringReader<'a> {
  69     sess: &'a ParseSess,
  70     /// Initial position, read-only.
  71     start_pos: BytePos,
  72     /// The absolute offset within the source_map of the current character.
  73     pos: BytePos,
  74     /// Source text to tokenize.
  75     src: &'a str,
  76     /// Cursor for getting lexer tokens.
  77     cursor: Cursor<'a>,
  78     override_span: Option<Span>,
  79     /// When a "unknown start of token: \u{a0}" has already been emitted earlier
  80     /// in this file, it's safe to treat further occurrences of the non-breaking
  81     /// space character as whitespace.
  82     nbsp_is_whitespace: bool,
  83 }
  84
  85 impl<'a> StringReader<'a> {
  86     fn mk_sp(&self, lo: BytePos, hi: BytePos) -> Span {
  87         self.override_span.unwrap_or_else(|| Span::with_root_ctxt(lo, hi))
  88     }
  89
  90     /// Returns the next token, paired with a bool indicating if the token was
  91     /// preceded by whitespace.
  92     fn next_token(&mut self) -> (Token, bool) {
  93         let mut preceded_by_whitespace = false;
  94         let mut swallow_next_invalid = 0;
  95         // Skip trivial (whitespace & comments) tokens
  96         loop {
  97             let token = self.cursor.advance_token();
  98             let start = self.pos;
  99             self.pos = self.pos + BytePos(token.len);
 100
 101             debug!("next_token: {:?}({:?})", token.kind, self.str_from(start));
 102
 103             // Now "cook" the token, converting the simple `rustc_lexer::TokenKind` enum into a
 104             // rich `rustc_ast::TokenKind`. This turns strings into interned symbols and runs
 105             // additional validation.
 106             let kind = match token.kind {
 107                 rustc_lexer::TokenKind::LineComment { doc_style } => {
 108                     // Skip non-doc comments
 109                     let Some(doc_style) = doc_style else {
 110                         self.lint_unicode_text_flow(start);
 111                         preceded_by_whitespace = true;
 112                         continue;
 113                     };
 114
 115                     // Opening delimiter of the length 3 is not included into the symbol.
 116                     let content_start = start + BytePos(3);
 117                     let content = self.str_from(content_start);
 118                     self.cook_doc_comment(content_start, content, CommentKind::Line, doc_style)
 119                 }
 120                 rustc_lexer::TokenKind::BlockComment { doc_style, terminated } => {
 121                     if !terminated {
 122                         self.report_unterminated_block_comment(start, doc_style);
 123                     }
 124
 125                     // Skip non-doc comments
 126                     let Some(doc_style) = doc_style else {
 127                         self.lint_unicode_text_flow(start);
 128                         preceded_by_whitespace = true;
 129                         continue;
 130                     };
 131
 132                     // Opening delimiter of the length 3 and closing delimiter of the length 2
 133                     // are not included into the symbol.
 134                     let content_start = start + BytePos(3);
 135                     let content_end = self.pos - BytePos(if terminated { 2 } else { 0 });
 136                     let content = self.str_from_to(content_start, content_end);
 137                     self.cook_doc_comment(content_start, content, CommentKind::Block, doc_style)
 138                 }
 139                 rustc_lexer::TokenKind::Whitespace => {
 140                     preceded_by_whitespace = true;
 141                     continue;
 142                 }
 143                 rustc_lexer::TokenKind::Ident => {
 144                     let sym = nfc_normalize(self.str_from(start));
 145                     let span = self.mk_sp(start, self.pos);
 146                     self.sess.symbol_gallery.insert(sym, span);
 147                     token::Ident(sym, false)
 148                 }
 149                 rustc_lexer::TokenKind::RawIdent => {
 150                     let sym = nfc_normalize(self.str_from(start + BytePos(2)));
 151                     let span = self.mk_sp(start, self.pos);
 152                     self.sess.symbol_gallery.insert(sym, span);
 153                     if !sym.can_be_raw() {
 154                         self.err_span(span, &format!("`{}` cannot be a raw identifier", sym));
 155                     }
 156                     self.sess.raw_identifier_spans.borrow_mut().push(span);
 157                     token::Ident(sym, true)
 158                 }
 159                 rustc_lexer::TokenKind::UnknownPrefix => {
 160                     self.report_unknown_prefix(start);
 161                     let sym = nfc_normalize(self.str_from(start));
 162                     let span = self.mk_sp(start, self.pos);
 163                     self.sess.symbol_gallery.insert(sym, span);
 164                     token::Ident(sym, false)
 165                 }
 166                 rustc_lexer::TokenKind::InvalidIdent
 167                     // Do not recover an identifier with emoji if the codepoint is a confusable
 168                     // with a recoverable substitution token, like `➖`.
 169                     if !UNICODE_ARRAY
 170                         .iter()
 171                         .any(|&(c, _, _)| {
 172                             let sym = self.str_from(start);
 173                             sym.chars().count() == 1 && c == sym.chars().next().unwrap()
 174                         }) =>
 175                 {
 176                     let sym = nfc_normalize(self.str_from(start));
 177                     let span = self.mk_sp(start, self.pos);
 178                     self.sess.bad_unicode_identifiers.borrow_mut().entry(sym).or_default()
 179                         .push(span);
 180                     token::Ident(sym, false)
 181                 }
 182                 rustc_lexer::TokenKind::Literal { kind, suffix_start } => {
 183                     let suffix_start = start + BytePos(suffix_start);
 184                     let (kind, symbol) = self.cook_lexer_literal(start, suffix_start, kind);
 185                     let suffix = if suffix_start < self.pos {
 186                         let string = self.str_from(suffix_start);
 187                         if string == "_" {
 188                             self.sess
 189                                 .span_diagnostic
 190                                 .struct_span_err(
 191                                     self.mk_sp(suffix_start, self.pos),
 192                                     "underscore literal suffix is not allowed",
 193                                 )
 194                                 .emit();
 195                             None
 196                         } else {
 197                             Some(Symbol::intern(string))
 198                         }
 199                     } else {
 200                         None
 201                     };
 202                     token::Literal(token::Lit { kind, symbol, suffix })
 203                 }
 204                 rustc_lexer::TokenKind::Lifetime { starts_with_number } => {
 205                     // Include the leading `'` in the real identifier, for macro
 206                     // expansion purposes. See #12512 for the gory details of why
 207                     // this is necessary.
 208                     let lifetime_name = self.str_from(start);
 209                     if starts_with_number {
 210                         let span = self.mk_sp(start, self.pos);
 211                         let mut diag = self.sess.struct_err("lifetimes cannot start with a number");
 212                         diag.set_span(span);
 213                         diag.stash(span, StashKey::LifetimeIsChar);
 214                     }
 215                     let ident = Symbol::intern(lifetime_name);
 216                     token::Lifetime(ident)
 217                 }
 218                 rustc_lexer::TokenKind::Semi => token::Semi,
 219                 rustc_lexer::TokenKind::Comma => token::Comma,
 220                 rustc_lexer::TokenKind::Dot => token::Dot,
 221                 rustc_lexer::TokenKind::OpenParen => token::OpenDelim(Delimiter::Parenthesis),
 222                 rustc_lexer::TokenKind::CloseParen => token::CloseDelim(Delimiter::Parenthesis),
 223                 rustc_lexer::TokenKind::OpenBrace => token::OpenDelim(Delimiter::Brace),
 224                 rustc_lexer::TokenKind::CloseBrace => token::CloseDelim(Delimiter::Brace),
 225                 rustc_lexer::TokenKind::OpenBracket => token::OpenDelim(Delimiter::Bracket),
 226                 rustc_lexer::TokenKind::CloseBracket => token::CloseDelim(Delimiter::Bracket),
 227                 rustc_lexer::TokenKind::At => token::At,
 228                 rustc_lexer::TokenKind::Pound => token::Pound,
 229                 rustc_lexer::TokenKind::Tilde => token::Tilde,
 230                 rustc_lexer::TokenKind::Question => token::Question,
 231                 rustc_lexer::TokenKind::Colon => token::Colon,
 232                 rustc_lexer::TokenKind::Dollar => token::Dollar,
 233                 rustc_lexer::TokenKind::Eq => token::Eq,
 234                 rustc_lexer::TokenKind::Bang => token::Not,
 235                 rustc_lexer::TokenKind::Lt => token::Lt,
 236                 rustc_lexer::TokenKind::Gt => token::Gt,
 237                 rustc_lexer::TokenKind::Minus => token::BinOp(token::Minus),
 238                 rustc_lexer::TokenKind::And => token::BinOp(token::And),
 239                 rustc_lexer::TokenKind::Or => token::BinOp(token::Or),
 240                 rustc_lexer::TokenKind::Plus => token::BinOp(token::Plus),
 241                 rustc_lexer::TokenKind::Star => token::BinOp(token::Star),
 242                 rustc_lexer::TokenKind::Slash => token::BinOp(token::Slash),
 243                 rustc_lexer::TokenKind::Caret => token::BinOp(token::Caret),
 244                 rustc_lexer::TokenKind::Percent => token::BinOp(token::Percent),
 245
 246                 rustc_lexer::TokenKind::Unknown | rustc_lexer::TokenKind::InvalidIdent => {
 247                     // Don't emit diagnostics for sequences of the same invalid token
 248                     if swallow_next_invalid > 0 {
 249                         swallow_next_invalid -= 1;
 250                         continue;
 251                     }
 252                     let mut it = self.str_from_to_end(start).chars();
 253                     let c = it.next().unwrap();
 254                     if c == '\u{00a0}' {
 255                         // If an error has already been reported on non-breaking
 256                         // space characters earlier in the file, treat all
 257                         // subsequent occurrences as whitespace.
 258                         if self.nbsp_is_whitespace {
 259                             preceded_by_whitespace = true;
 260                             continue;
 261                         }
 262                         self.nbsp_is_whitespace = true;
 263                     }
 264                     let repeats = it.take_while(|c1| *c1 == c).count();
 265                     let mut err =
 266                         self.struct_err_span_char(start, self.pos + Pos::from_usize(repeats * c.len_utf8()), "unknown start of token", c);
 267                     // FIXME: the lexer could be used to turn the ASCII version of unicode
 268                     // homoglyphs, instead of keeping a table in `check_for_substitution`into the
 269                     // token. Ideally, this should be inside `rustc_lexer`. However, we should
 270                     // first remove compound tokens like `<<` from `rustc_lexer`, and then add
 271                     // fancier error recovery to it, as there will be less overall work to do this
 272                     // way.
 273                     let token = unicode_chars::check_for_substitution(self, start, c, &mut err, repeats+1);
 274                     if c == '\x00' {
 275                         err.help("source files must contain UTF-8 encoded text, unexpected null bytes might occur when a different encoding is used");
 276                     }
 277                     if repeats > 0 {
 278                         if repeats == 1 {
 279                             err.note(format!("character appears once more"));
 280                         } else {
 281                             err.note(format!("character appears {repeats} more times"));
 282                         }
 283                         swallow_next_invalid = repeats;
 284                     }
 285                     err.emit();
 286                     if let Some(token) = token {
 287                         token
 288                     } else {
 289                         preceded_by_whitespace = true;
 290                         continue;
 291                     }
 292                 }
 293                 rustc_lexer::TokenKind::Eof => token::Eof,
 294             };
 295             let span = self.mk_sp(start, self.pos);
 296             return (Token::new(kind, span), preceded_by_whitespace);
 297         }
 298     }
 299
 300     /// Report a fatal lexical error with a given span.
 301     fn fatal_span(&self, sp: Span, m: &str) -> ! {
 302         self.sess.span_diagnostic.span_fatal(sp, m)
 303     }
 304
 305     /// Report a lexical error with a given span.
 306     fn err_span(&self, sp: Span, m: &str) {
 307         self.sess.span_diagnostic.struct_span_err(sp, m).emit();
 308     }
 309
 310     /// Report a fatal error spanning [`from_pos`, `to_pos`).
 311     fn fatal_span_(&self, from_pos: BytePos, to_pos: BytePos, m: &str) -> ! {
 312         self.fatal_span(self.mk_sp(from_pos, to_pos), m)
 313     }
 314
 315     /// Report a lexical error spanning [`from_pos`, `to_pos`).
 316     fn err_span_(&self, from_pos: BytePos, to_pos: BytePos, m: &str) {
 317         self.err_span(self.mk_sp(from_pos, to_pos), m)
 318     }
 319
 320     fn struct_fatal_span_char(
 321         &self,
 322         from_pos: BytePos,
 323         to_pos: BytePos,
 324         m: &str,
 325         c: char,
 326     ) -> DiagnosticBuilder<'a, !> {
 327         self.sess
 328             .span_diagnostic
 329             .struct_span_fatal(self.mk_sp(from_pos, to_pos), &format!("{}: {}", m, escaped_char(c)))
 330     }
 331
 332     fn struct_err_span_char(
 333         &self,
 334         from_pos: BytePos,
 335         to_pos: BytePos,
 336         m: &str,
 337         c: char,
 338     ) -> DiagnosticBuilder<'a, ErrorGuaranteed> {
 339         self.sess
 340             .span_diagnostic
 341             .struct_span_err(self.mk_sp(from_pos, to_pos), &format!("{}: {}", m, escaped_char(c)))
 342     }
 343
 344     /// Detect usages of Unicode codepoints changing the direction of the text on screen and loudly
 345     /// complain about it.
 346     fn lint_unicode_text_flow(&self, start: BytePos) {
 347         // Opening delimiter of the length 2 is not included into the comment text.
 348         let content_start = start + BytePos(2);
 349         let content = self.str_from(content_start);
 350         if contains_text_flow_control_chars(content) {
 351             let span = self.mk_sp(start, self.pos);
 352             self.sess.buffer_lint_with_diagnostic(
 353                 &TEXT_DIRECTION_CODEPOINT_IN_COMMENT,
 354                 span,
 355                 ast::CRATE_NODE_ID,
 356                 "unicode codepoint changing visible direction of text present in comment",
 357                 BuiltinLintDiagnostics::UnicodeTextFlow(span, content.to_string()),
 358             );
 359         }
 360     }
 361
 362     fn cook_doc_comment(
 363         &self,
 364         content_start: BytePos,
 365         content: &str,
 366         comment_kind: CommentKind,
 367         doc_style: DocStyle,
 368     ) -> TokenKind {
 369         if content.contains('\r') {
 370             for (idx, _) in content.char_indices().filter(|&(_, c)| c == '\r') {
 371                 self.err_span_(
 372                     content_start + BytePos(idx as u32),
 373                     content_start + BytePos(idx as u32 + 1),
 374                     match comment_kind {
 375                         CommentKind::Line => "bare CR not allowed in doc-comment",
 376                         CommentKind::Block => "bare CR not allowed in block doc-comment",
 377                     },
 378                 );
 379             }
 380         }
 381
 382         let attr_style = match doc_style {
 383             DocStyle::Outer => AttrStyle::Outer,
 384             DocStyle::Inner => AttrStyle::Inner,
 385         };
 386
 387         token::DocComment(comment_kind, attr_style, Symbol::intern(content))
 388     }
 389
 390     fn cook_lexer_literal(
 391         &self,
 392         start: BytePos,
 393         end: BytePos,
 394         kind: rustc_lexer::LiteralKind,
 395     ) -> (token::LitKind, Symbol) {
 396         match kind {
 397             rustc_lexer::LiteralKind::Char { terminated } => {
 398                 if !terminated {
 399                     self.sess.span_diagnostic.span_fatal_with_code(
 400                         self.mk_sp(start, end),
 401                         "unterminated character literal",
 402                         error_code!(E0762),
 403                     )
 404                 }
 405                 self.cook_quoted(token::Char, Mode::Char, start, end, 1, 1) // ' '
 406             }
 407             rustc_lexer::LiteralKind::Byte { terminated } => {
 408                 if !terminated {
 409                     self.sess.span_diagnostic.span_fatal_with_code(
 410                         self.mk_sp(start + BytePos(1), end),
 411                         "unterminated byte constant",
 412                         error_code!(E0763),
 413                     )
 414                 }
 415                 self.cook_quoted(token::Byte, Mode::Byte, start, end, 2, 1) // b' '
 416             }
 417             rustc_lexer::LiteralKind::Str { terminated } => {
 418                 if !terminated {
 419                     self.sess.span_diagnostic.span_fatal_with_code(
 420                         self.mk_sp(start, end),
 421                         "unterminated double quote string",
 422                         error_code!(E0765),
 423                     )
 424                 }
 425                 self.cook_quoted(token::Str, Mode::Str, start, end, 1, 1) // " "
 426             }
 427             rustc_lexer::LiteralKind::ByteStr { terminated } => {
 428                 if !terminated {
 429                     self.sess.span_diagnostic.span_fatal_with_code(
 430                         self.mk_sp(start + BytePos(1), end),
 431                         "unterminated double quote byte string",
 432                         error_code!(E0766),
 433                     )
 434                 }
 435                 self.cook_quoted(token::ByteStr, Mode::ByteStr, start, end, 2, 1) // b" "
 436             }
 437             rustc_lexer::LiteralKind::RawStr { n_hashes } => {
 438                 if let Some(n_hashes) = n_hashes {
 439                     let n = u32::from(n_hashes);
 440                     let kind = token::StrRaw(n_hashes);
 441                     self.cook_quoted(kind, Mode::RawStr, start, end, 2 + n, 1 + n) // r##" "##
 442                 } else {
 443                     self.report_raw_str_error(start, 1);
 444                 }
 445             }
 446             rustc_lexer::LiteralKind::RawByteStr { n_hashes } => {
 447                 if let Some(n_hashes) = n_hashes {
 448                     let n = u32::from(n_hashes);
 449                     let kind = token::ByteStrRaw(n_hashes);
 450                     self.cook_quoted(kind, Mode::RawByteStr, start, end, 3 + n, 1 + n) // br##" "##
 451                 } else {
 452                     self.report_raw_str_error(start, 2);
 453                 }
 454             }
 455             rustc_lexer::LiteralKind::Int { base, empty_int } => {
 456                 if empty_int {
 457                     self.sess
 458                         .span_diagnostic
 459                         .struct_span_err_with_code(
 460                             self.mk_sp(start, end),
 461                             "no valid digits found for number",
 462                             error_code!(E0768),
 463                         )
 464                         .emit();
 465                     (token::Integer, sym::integer(0))
 466                 } else {
 467                     if matches!(base, Base::Binary | Base::Octal) {
 468                         let base = base as u32;
 469                         let s = self.str_from_to(start + BytePos(2), end);
 470                         for (idx, c) in s.char_indices() {
 471                             if c != '_' && c.to_digit(base).is_none() {
 472                                 self.err_span_(
 473                                     start + BytePos::from_usize(2 + idx),
 474                                     start + BytePos::from_usize(2 + idx + c.len_utf8()),
 475                                     &format!("invalid digit for a base {} literal", base),
 476                                 );
 477                             }
 478                         }
 479                     }
 480                     (token::Integer, self.symbol_from_to(start, end))
 481                 }
 482             }
 483             rustc_lexer::LiteralKind::Float { base, empty_exponent } => {
 484                 if empty_exponent {
 485                     self.err_span_(start, self.pos, "expected at least one digit in exponent");
 486                 }
 487                 match base {
 488                     Base::Hexadecimal => {
 489                         self.err_span_(start, end, "hexadecimal float literal is not supported")
 490                     }
 491                     Base::Octal => {
 492                         self.err_span_(start, end, "octal float literal is not supported")
 493                     }
 494                     Base::Binary => {
 495                         self.err_span_(start, end, "binary float literal is not supported")
 496                     }
 497                     _ => {}
 498                 }
 499                 (token::Float, self.symbol_from_to(start, end))
 500             }
 501         }
 502     }
 503
 504     #[inline]
 505     fn src_index(&self, pos: BytePos) -> usize {
 506         (pos - self.start_pos).to_usize()
 507     }
 508
 509     /// Slice of the source text from `start` up to but excluding `self.pos`,
 510     /// meaning the slice does not include the character `self.ch`.
 511     fn str_from(&self, start: BytePos) -> &'a str {
 512         self.str_from_to(start, self.pos)
 513     }
 514
 515     /// As symbol_from, with an explicit endpoint.
 516     fn symbol_from_to(&self, start: BytePos, end: BytePos) -> Symbol {
 517         debug!("taking an ident from {:?} to {:?}", start, end);
 518         Symbol::intern(self.str_from_to(start, end))
 519     }
 520
 521     /// Slice of the source text spanning from `start` up to but excluding `end`.
 522     fn str_from_to(&self, start: BytePos, end: BytePos) -> &'a str {
 523         &self.src[self.src_index(start)..self.src_index(end)]
 524     }
 525
 526     /// Slice of the source text spanning from `start` until the end
 527     fn str_from_to_end(&self, start: BytePos) -> &'a str {
 528         &self.src[self.src_index(start)..]
 529     }
 530
 531     fn report_raw_str_error(&self, start: BytePos, prefix_len: u32) -> ! {
 532         match rustc_lexer::validate_raw_str(self.str_from(start), prefix_len) {
 533             Err(RawStrError::InvalidStarter { bad_char }) => {
 534                 self.report_non_started_raw_string(start, bad_char)
 535             }
 536             Err(RawStrError::NoTerminator { expected, found, possible_terminator_offset }) => self
 537                 .report_unterminated_raw_string(start, expected, possible_terminator_offset, found),
 538             Err(RawStrError::TooManyDelimiters { found }) => {
 539                 self.report_too_many_hashes(start, found)
 540             }
 541             Ok(()) => panic!("no error found for supposedly invalid raw string literal"),
 542         }
 543     }
 544
 545     fn report_non_started_raw_string(&self, start: BytePos, bad_char: char) -> ! {
 546         self.struct_fatal_span_char(
 547             start,
 548             self.pos,
 549             "found invalid character; only `#` is allowed in raw string delimitation",
 550             bad_char,
 551         )
 552         .emit()
 553     }
 554
 555     fn report_unterminated_raw_string(
 556         &self,
 557         start: BytePos,
 558         n_hashes: u32,
 559         possible_offset: Option<u32>,
 560         found_terminators: u32,
 561     ) -> ! {
 562         let mut err = self.sess.span_diagnostic.struct_span_fatal_with_code(
 563             self.mk_sp(start, start),
 564             "unterminated raw string",
 565             error_code!(E0748),
 566         );
 567
 568         err.span_label(self.mk_sp(start, start), "unterminated raw string");
 569
 570         if n_hashes > 0 {
 571             err.note(&format!(
 572                 "this raw string should be terminated with `\"{}`",
 573                 "#".repeat(n_hashes as usize)
 574             ));
 575         }
 576
 577         if let Some(possible_offset) = possible_offset {
 578             let lo = start + BytePos(possible_offset as u32);
 579             let hi = lo + BytePos(found_terminators as u32);
 580             let span = self.mk_sp(lo, hi);
 581             err.span_suggestion(
 582                 span,
 583                 "consider terminating the string here",
 584                 "#".repeat(n_hashes as usize),
 585                 Applicability::MaybeIncorrect,
 586             );
 587         }
 588
 589         err.emit()
 590     }
 591
 592     fn report_unterminated_block_comment(&self, start: BytePos, doc_style: Option<DocStyle>) {
 593         let msg = match doc_style {
 594             Some(_) => "unterminated block doc-comment",
 595             None => "unterminated block comment",
 596         };
 597         let last_bpos = self.pos;
 598         let mut err = self.sess.span_diagnostic.struct_span_fatal_with_code(
 599             self.mk_sp(start, last_bpos),
 600             msg,
 601             error_code!(E0758),
 602         );
 603         let mut nested_block_comment_open_idxs = vec![];
 604         let mut last_nested_block_comment_idxs = None;
 605         let mut content_chars = self.str_from(start).char_indices().peekable();
 606
 607         while let Some((idx, current_char)) = content_chars.next() {
 608             match content_chars.peek() {
 609                 Some((_, '*')) if current_char == '/' => {
 610                     nested_block_comment_open_idxs.push(idx);
 611                 }
 612                 Some((_, '/')) if current_char == '*' => {
 613                     last_nested_block_comment_idxs =
 614                         nested_block_comment_open_idxs.pop().map(|open_idx| (open_idx, idx));
 615                 }
 616                 _ => {}
 617             };
 618         }
 619
 620         if let Some((nested_open_idx, nested_close_idx)) = last_nested_block_comment_idxs {
 621             err.span_label(self.mk_sp(start, start + BytePos(2)), msg)
 622                 .span_label(
 623                     self.mk_sp(
 624                         start + BytePos(nested_open_idx as u32),
 625                         start + BytePos(nested_open_idx as u32 + 2),
 626                     ),
 627                     "...as last nested comment starts here, maybe you want to close this instead?",
 628                 )
 629                 .span_label(
 630                     self.mk_sp(
 631                         start + BytePos(nested_close_idx as u32),
 632                         start + BytePos(nested_close_idx as u32 + 2),
 633                     ),
 634                     "...and last nested comment terminates here.",
 635                 );
 636         }
 637
 638         err.emit();
 639     }
 640
 641     // RFC 3101 introduced the idea of (reserved) prefixes. As of Rust 2021,
 642     // using a (unknown) prefix is an error. In earlier editions, however, they
 643     // only result in a (allowed by default) lint, and are treated as regular
 644     // identifier tokens.
 645     fn report_unknown_prefix(&self, start: BytePos) {
 646         let prefix_span = self.mk_sp(start, self.pos);
 647         let prefix_str = self.str_from_to(start, self.pos);
 648         let msg = format!("prefix `{}` is unknown", prefix_str);
 649
 650         let expn_data = prefix_span.ctxt().outer_expn_data();
 651
 652         if expn_data.edition >= Edition::Edition2021 {
 653             // In Rust 2021, this is a hard error.
 654             let mut err = self.sess.span_diagnostic.struct_span_err(prefix_span, &msg);
 655             err.span_label(prefix_span, "unknown prefix");
 656             if prefix_str == "rb" {
 657                 err.span_suggestion_verbose(
 658                     prefix_span,
 659                     "use `br` for a raw byte string",
 660                     "br",
 661                     Applicability::MaybeIncorrect,
 662                 );
 663             } else if expn_data.is_root() {
 664                 err.span_suggestion_verbose(
 665                     prefix_span.shrink_to_hi(),
 666                     "consider inserting whitespace here",
 667                     " ",
 668                     Applicability::MaybeIncorrect,
 669                 );
 670             }
 671             err.note("prefixed identifiers and literals are reserved since Rust 2021");
 672             err.emit();
 673         } else {
 674             // Before Rust 2021, only emit a lint for migration.
 675             self.sess.buffer_lint_with_diagnostic(
 676                 &RUST_2021_PREFIXES_INCOMPATIBLE_SYNTAX,
 677                 prefix_span,
 678                 ast::CRATE_NODE_ID,
 679                 &msg,
 680                 BuiltinLintDiagnostics::ReservedPrefix(prefix_span),
 681             );
 682         }
 683     }
 684
 685     fn report_too_many_hashes(&self, start: BytePos, found: u32) -> ! {
 686         self.fatal_span_(
 687             start,
 688             self.pos,
 689             &format!(
 690                 "too many `#` symbols: raw strings may be delimited \
 691                 by up to 255 `#` symbols, but found {}",
 692                 found
 693             ),
 694         )
 695     }
 696
 697     fn cook_quoted(
 698         &self,
 699         kind: token::LitKind,
 700         mode: Mode,
 701         start: BytePos,
 702         end: BytePos,
 703         prefix_len: u32,
 704         postfix_len: u32,
 705     ) -> (token::LitKind, Symbol) {
 706         let mut has_fatal_err = false;
 707         let content_start = start + BytePos(prefix_len);
 708         let content_end = end - BytePos(postfix_len);
 709         let lit_content = self.str_from_to(content_start, content_end);
 710         unescape::unescape_literal(lit_content, mode, &mut |range, result| {
 711             // Here we only check for errors. The actual unescaping is done later.
 712             if let Err(err) = result {
 713                 let span_with_quotes = self.mk_sp(start, end);
 714                 let (start, end) = (range.start as u32, range.end as u32);
 715                 let lo = content_start + BytePos(start);
 716                 let hi = lo + BytePos(end - start);
 717                 let span = self.mk_sp(lo, hi);
 718                 if err.is_fatal() {
 719                     has_fatal_err = true;
 720                 }
 721                 emit_unescape_error(
 722                     &self.sess.span_diagnostic,
 723                     lit_content,
 724                     span_with_quotes,
 725                     span,
 726                     mode,
 727                     range,
 728                     err,
 729                 );
 730             }
 731         });
 732
 733         // We normally exclude the quotes for the symbol, but for errors we
 734         // include it because it results in clearer error messages.
 735         if !has_fatal_err {
 736             (kind, Symbol::intern(lit_content))
 737         } else {
 738             (token::Err, self.symbol_from_to(start, end))
 739         }
 740     }
 741 }
 742
 743 pub fn nfc_normalize(string: &str) -> Symbol {
 744     use unicode_normalization::{is_nfc_quick, IsNormalized, UnicodeNormalization};
 745     match is_nfc_quick(string.chars()) {
 746         IsNormalized::Yes => Symbol::intern(string),
 747         _ => {
 748             let normalized_str: String = string.chars().nfc().collect();
 749             Symbol::intern(&normalized_str)
 750         }
 751     }
 752 }