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;
7 error_code, Applicability, DiagnosticBuilder, ErrorGuaranteed, PResult, StashKey,
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,
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};
21 mod unescape_error_reporting;
24 use unescape_error_reporting::{emit_unescape_error, escaped_char};
26 // This type is used a lot. Make sure it doesn't unintentionally get bigger.
28 // This assertion is in this crate, rather than in `rustc_lexer`, because that
29 // crate cannot depend on `rustc_data_structures`.
30 #[cfg(all(target_arch = "x86_64", target_pointer_width = "64"))]
31 rustc_data_structures::static_assert_size!(rustc_lexer::Token, 12);
33 #[derive(Clone, Debug)]
34 pub struct UnmatchedBrace {
35 pub expected_delim: Delimiter,
36 pub found_delim: Option<Delimiter>,
38 pub unclosed_span: Option<Span>,
39 pub candidate_span: Option<Span>,
42 pub(crate) fn parse_token_trees<'a>(
45 mut start_pos: BytePos,
46 override_span: Option<Span>,
47 ) -> (PResult<'a, TokenStream>, Vec<UnmatchedBrace>) {
48 // Skip `#!`, if present.
49 if let Some(shebang_len) = rustc_lexer::strip_shebang(src) {
50 src = &src[shebang_len..];
51 start_pos = start_pos + BytePos::from_usize(shebang_len);
54 let cursor = Cursor::new(src);
56 StringReader { sess, start_pos, pos: start_pos, src, cursor, override_span };
57 tokentrees::TokenTreesReader::parse_all_token_trees(string_reader)
60 struct StringReader<'a> {
62 /// Initial position, read-only.
64 /// The absolute offset within the source_map of the current character.
66 /// Source text to tokenize.
68 /// Cursor for getting lexer tokens.
70 override_span: Option<Span>,
73 impl<'a> StringReader<'a> {
74 fn mk_sp(&self, lo: BytePos, hi: BytePos) -> Span {
75 self.override_span.unwrap_or_else(|| Span::with_root_ctxt(lo, hi))
78 /// Returns the next token, paired with a bool indicating if the token was
79 /// preceded by whitespace.
80 fn next_token(&mut self) -> (Token, bool) {
81 let mut preceded_by_whitespace = false;
82 let mut swallow_next_invalid = 0;
83 // Skip trivial (whitespace & comments) tokens
85 let token = self.cursor.advance_token();
87 self.pos = self.pos + BytePos(token.len);
89 debug!("next_token: {:?}({:?})", token.kind, self.str_from(start));
91 // Now "cook" the token, converting the simple `rustc_lexer::TokenKind` enum into a
92 // rich `rustc_ast::TokenKind`. This turns strings into interned symbols and runs
93 // additional validation.
94 let kind = match token.kind {
95 rustc_lexer::TokenKind::LineComment { doc_style } => {
96 // Skip non-doc comments
97 let Some(doc_style) = doc_style else {
98 self.lint_unicode_text_flow(start);
99 preceded_by_whitespace = true;
103 // Opening delimiter of the length 3 is not included into the symbol.
104 let content_start = start + BytePos(3);
105 let content = self.str_from(content_start);
106 self.cook_doc_comment(content_start, content, CommentKind::Line, doc_style)
108 rustc_lexer::TokenKind::BlockComment { doc_style, terminated } => {
110 self.report_unterminated_block_comment(start, doc_style);
113 // Skip non-doc comments
114 let Some(doc_style) = doc_style else {
115 self.lint_unicode_text_flow(start);
116 preceded_by_whitespace = true;
120 // Opening delimiter of the length 3 and closing delimiter of the length 2
121 // are not included into the symbol.
122 let content_start = start + BytePos(3);
123 let content_end = self.pos - BytePos(if terminated { 2 } else { 0 });
124 let content = self.str_from_to(content_start, content_end);
125 self.cook_doc_comment(content_start, content, CommentKind::Block, doc_style)
127 rustc_lexer::TokenKind::Whitespace => {
128 preceded_by_whitespace = true;
131 rustc_lexer::TokenKind::Ident => {
132 let sym = nfc_normalize(self.str_from(start));
133 let span = self.mk_sp(start, self.pos);
134 self.sess.symbol_gallery.insert(sym, span);
135 token::Ident(sym, false)
137 rustc_lexer::TokenKind::RawIdent => {
138 let sym = nfc_normalize(self.str_from(start + BytePos(2)));
139 let span = self.mk_sp(start, self.pos);
140 self.sess.symbol_gallery.insert(sym, span);
141 if !sym.can_be_raw() {
142 self.err_span(span, &format!("`{}` cannot be a raw identifier", sym));
144 self.sess.raw_identifier_spans.borrow_mut().push(span);
145 token::Ident(sym, true)
147 rustc_lexer::TokenKind::UnknownPrefix => {
148 self.report_unknown_prefix(start);
149 let sym = nfc_normalize(self.str_from(start));
150 let span = self.mk_sp(start, self.pos);
151 self.sess.symbol_gallery.insert(sym, span);
152 token::Ident(sym, false)
154 rustc_lexer::TokenKind::InvalidIdent
155 // Do not recover an identifier with emoji if the codepoint is a confusable
156 // with a recoverable substitution token, like `➖`.
160 let sym = self.str_from(start);
161 sym.chars().count() == 1 && c == sym.chars().next().unwrap()
164 let sym = nfc_normalize(self.str_from(start));
165 let span = self.mk_sp(start, self.pos);
166 self.sess.bad_unicode_identifiers.borrow_mut().entry(sym).or_default()
168 token::Ident(sym, false)
170 rustc_lexer::TokenKind::Literal { kind, suffix_start } => {
171 let suffix_start = start + BytePos(suffix_start);
172 let (kind, symbol) = self.cook_lexer_literal(start, suffix_start, kind);
173 let suffix = if suffix_start < self.pos {
174 let string = self.str_from(suffix_start);
179 self.mk_sp(suffix_start, self.pos),
180 "underscore literal suffix is not allowed",
185 Some(Symbol::intern(string))
190 token::Literal(token::Lit { kind, symbol, suffix })
192 rustc_lexer::TokenKind::Lifetime { starts_with_number } => {
193 // Include the leading `'` in the real identifier, for macro
194 // expansion purposes. See #12512 for the gory details of why
195 // this is necessary.
196 let lifetime_name = self.str_from(start);
197 if starts_with_number {
198 let span = self.mk_sp(start, self.pos);
199 let mut diag = self.sess.struct_err("lifetimes cannot start with a number");
201 diag.stash(span, StashKey::LifetimeIsChar);
203 let ident = Symbol::intern(lifetime_name);
204 token::Lifetime(ident)
206 rustc_lexer::TokenKind::Semi => token::Semi,
207 rustc_lexer::TokenKind::Comma => token::Comma,
208 rustc_lexer::TokenKind::Dot => token::Dot,
209 rustc_lexer::TokenKind::OpenParen => token::OpenDelim(Delimiter::Parenthesis),
210 rustc_lexer::TokenKind::CloseParen => token::CloseDelim(Delimiter::Parenthesis),
211 rustc_lexer::TokenKind::OpenBrace => token::OpenDelim(Delimiter::Brace),
212 rustc_lexer::TokenKind::CloseBrace => token::CloseDelim(Delimiter::Brace),
213 rustc_lexer::TokenKind::OpenBracket => token::OpenDelim(Delimiter::Bracket),
214 rustc_lexer::TokenKind::CloseBracket => token::CloseDelim(Delimiter::Bracket),
215 rustc_lexer::TokenKind::At => token::At,
216 rustc_lexer::TokenKind::Pound => token::Pound,
217 rustc_lexer::TokenKind::Tilde => token::Tilde,
218 rustc_lexer::TokenKind::Question => token::Question,
219 rustc_lexer::TokenKind::Colon => token::Colon,
220 rustc_lexer::TokenKind::Dollar => token::Dollar,
221 rustc_lexer::TokenKind::Eq => token::Eq,
222 rustc_lexer::TokenKind::Bang => token::Not,
223 rustc_lexer::TokenKind::Lt => token::Lt,
224 rustc_lexer::TokenKind::Gt => token::Gt,
225 rustc_lexer::TokenKind::Minus => token::BinOp(token::Minus),
226 rustc_lexer::TokenKind::And => token::BinOp(token::And),
227 rustc_lexer::TokenKind::Or => token::BinOp(token::Or),
228 rustc_lexer::TokenKind::Plus => token::BinOp(token::Plus),
229 rustc_lexer::TokenKind::Star => token::BinOp(token::Star),
230 rustc_lexer::TokenKind::Slash => token::BinOp(token::Slash),
231 rustc_lexer::TokenKind::Caret => token::BinOp(token::Caret),
232 rustc_lexer::TokenKind::Percent => token::BinOp(token::Percent),
234 rustc_lexer::TokenKind::Unknown | rustc_lexer::TokenKind::InvalidIdent => {
235 // Don't emit diagnostics for sequences of the same invalid token
236 if swallow_next_invalid > 0 {
237 swallow_next_invalid -= 1;
240 let mut it = self.str_from_to_end(start).chars();
241 let c = it.next().unwrap();
242 let repeats = it.take_while(|c1| *c1 == c).count();
244 self.struct_err_span_char(start, self.pos + Pos::from_usize(repeats * c.len_utf8()), "unknown start of token", c);
245 // FIXME: the lexer could be used to turn the ASCII version of unicode
246 // homoglyphs, instead of keeping a table in `check_for_substitution`into the
247 // token. Ideally, this should be inside `rustc_lexer`. However, we should
248 // first remove compound tokens like `<<` from `rustc_lexer`, and then add
249 // fancier error recovery to it, as there will be less overall work to do this
251 let token = unicode_chars::check_for_substitution(self, start, c, &mut err, repeats+1);
253 err.help("source files must contain UTF-8 encoded text, unexpected null bytes might occur when a different encoding is used");
257 err.note(format!("character appears once more"));
259 err.note(format!("character appears {repeats} more times"));
261 swallow_next_invalid = repeats;
264 if let Some(token) = token {
267 preceded_by_whitespace = true;
271 rustc_lexer::TokenKind::Eof => token::Eof,
273 let span = self.mk_sp(start, self.pos);
274 return (Token::new(kind, span), preceded_by_whitespace);
278 /// Report a fatal lexical error with a given span.
279 fn fatal_span(&self, sp: Span, m: &str) -> ! {
280 self.sess.span_diagnostic.span_fatal(sp, m)
283 /// Report a lexical error with a given span.
284 fn err_span(&self, sp: Span, m: &str) {
285 self.sess.span_diagnostic.struct_span_err(sp, m).emit();
288 /// Report a fatal error spanning [`from_pos`, `to_pos`).
289 fn fatal_span_(&self, from_pos: BytePos, to_pos: BytePos, m: &str) -> ! {
290 self.fatal_span(self.mk_sp(from_pos, to_pos), m)
293 /// Report a lexical error spanning [`from_pos`, `to_pos`).
294 fn err_span_(&self, from_pos: BytePos, to_pos: BytePos, m: &str) {
295 self.err_span(self.mk_sp(from_pos, to_pos), m)
298 fn struct_fatal_span_char(
304 ) -> DiagnosticBuilder<'a, !> {
307 .struct_span_fatal(self.mk_sp(from_pos, to_pos), &format!("{}: {}", m, escaped_char(c)))
310 fn struct_err_span_char(
316 ) -> DiagnosticBuilder<'a, ErrorGuaranteed> {
319 .struct_span_err(self.mk_sp(from_pos, to_pos), &format!("{}: {}", m, escaped_char(c)))
322 /// Detect usages of Unicode codepoints changing the direction of the text on screen and loudly
323 /// complain about it.
324 fn lint_unicode_text_flow(&self, start: BytePos) {
325 // Opening delimiter of the length 2 is not included into the comment text.
326 let content_start = start + BytePos(2);
327 let content = self.str_from(content_start);
328 if contains_text_flow_control_chars(content) {
329 let span = self.mk_sp(start, self.pos);
330 self.sess.buffer_lint_with_diagnostic(
331 &TEXT_DIRECTION_CODEPOINT_IN_COMMENT,
334 "unicode codepoint changing visible direction of text present in comment",
335 BuiltinLintDiagnostics::UnicodeTextFlow(span, content.to_string()),
342 content_start: BytePos,
344 comment_kind: CommentKind,
347 if content.contains('\r') {
348 for (idx, _) in content.char_indices().filter(|&(_, c)| c == '\r') {
350 content_start + BytePos(idx as u32),
351 content_start + BytePos(idx as u32 + 1),
353 CommentKind::Line => "bare CR not allowed in doc-comment",
354 CommentKind::Block => "bare CR not allowed in block doc-comment",
360 let attr_style = match doc_style {
361 DocStyle::Outer => AttrStyle::Outer,
362 DocStyle::Inner => AttrStyle::Inner,
365 token::DocComment(comment_kind, attr_style, Symbol::intern(content))
368 fn cook_lexer_literal(
372 kind: rustc_lexer::LiteralKind,
373 ) -> (token::LitKind, Symbol) {
375 rustc_lexer::LiteralKind::Char { terminated } => {
377 self.sess.span_diagnostic.span_fatal_with_code(
378 self.mk_sp(start, end),
379 "unterminated character literal",
383 self.cook_quoted(token::Char, Mode::Char, start, end, 1, 1) // ' '
385 rustc_lexer::LiteralKind::Byte { terminated } => {
387 self.sess.span_diagnostic.span_fatal_with_code(
388 self.mk_sp(start + BytePos(1), end),
389 "unterminated byte constant",
393 self.cook_quoted(token::Byte, Mode::Byte, start, end, 2, 1) // b' '
395 rustc_lexer::LiteralKind::Str { terminated } => {
397 self.sess.span_diagnostic.span_fatal_with_code(
398 self.mk_sp(start, end),
399 "unterminated double quote string",
403 self.cook_quoted(token::Str, Mode::Str, start, end, 1, 1) // " "
405 rustc_lexer::LiteralKind::ByteStr { terminated } => {
407 self.sess.span_diagnostic.span_fatal_with_code(
408 self.mk_sp(start + BytePos(1), end),
409 "unterminated double quote byte string",
413 self.cook_quoted(token::ByteStr, Mode::ByteStr, start, end, 2, 1) // b" "
415 rustc_lexer::LiteralKind::RawStr { n_hashes } => {
416 if let Some(n_hashes) = n_hashes {
417 let n = u32::from(n_hashes);
418 let kind = token::StrRaw(n_hashes);
419 self.cook_quoted(kind, Mode::RawStr, start, end, 2 + n, 1 + n) // r##" "##
421 self.report_raw_str_error(start, 1);
424 rustc_lexer::LiteralKind::RawByteStr { n_hashes } => {
425 if let Some(n_hashes) = n_hashes {
426 let n = u32::from(n_hashes);
427 let kind = token::ByteStrRaw(n_hashes);
428 self.cook_quoted(kind, Mode::RawByteStr, start, end, 3 + n, 1 + n) // br##" "##
430 self.report_raw_str_error(start, 2);
433 rustc_lexer::LiteralKind::Int { base, empty_int } => {
437 .struct_span_err_with_code(
438 self.mk_sp(start, end),
439 "no valid digits found for number",
443 (token::Integer, sym::integer(0))
445 if matches!(base, Base::Binary | Base::Octal) {
446 let base = base as u32;
447 let s = self.str_from_to(start + BytePos(2), end);
448 for (idx, c) in s.char_indices() {
449 if c != '_' && c.to_digit(base).is_none() {
451 start + BytePos::from_usize(2 + idx),
452 start + BytePos::from_usize(2 + idx + c.len_utf8()),
453 &format!("invalid digit for a base {} literal", base),
458 (token::Integer, self.symbol_from_to(start, end))
461 rustc_lexer::LiteralKind::Float { base, empty_exponent } => {
463 self.err_span_(start, self.pos, "expected at least one digit in exponent");
466 Base::Hexadecimal => {
467 self.err_span_(start, end, "hexadecimal float literal is not supported")
470 self.err_span_(start, end, "octal float literal is not supported")
473 self.err_span_(start, end, "binary float literal is not supported")
477 (token::Float, self.symbol_from_to(start, end))
483 fn src_index(&self, pos: BytePos) -> usize {
484 (pos - self.start_pos).to_usize()
487 /// Slice of the source text from `start` up to but excluding `self.pos`,
488 /// meaning the slice does not include the character `self.ch`.
489 fn str_from(&self, start: BytePos) -> &str {
490 self.str_from_to(start, self.pos)
493 /// As symbol_from, with an explicit endpoint.
494 fn symbol_from_to(&self, start: BytePos, end: BytePos) -> Symbol {
495 debug!("taking an ident from {:?} to {:?}", start, end);
496 Symbol::intern(self.str_from_to(start, end))
499 /// Slice of the source text spanning from `start` up to but excluding `end`.
500 fn str_from_to(&self, start: BytePos, end: BytePos) -> &str {
501 &self.src[self.src_index(start)..self.src_index(end)]
504 /// Slice of the source text spanning from `start` until the end
505 fn str_from_to_end(&self, start: BytePos) -> &str {
506 &self.src[self.src_index(start)..]
509 fn report_raw_str_error(&self, start: BytePos, prefix_len: u32) -> ! {
510 match rustc_lexer::validate_raw_str(self.str_from(start), prefix_len) {
511 Err(RawStrError::InvalidStarter { bad_char }) => {
512 self.report_non_started_raw_string(start, bad_char)
514 Err(RawStrError::NoTerminator { expected, found, possible_terminator_offset }) => self
515 .report_unterminated_raw_string(start, expected, possible_terminator_offset, found),
516 Err(RawStrError::TooManyDelimiters { found }) => {
517 self.report_too_many_hashes(start, found)
519 Ok(()) => panic!("no error found for supposedly invalid raw string literal"),
523 fn report_non_started_raw_string(&self, start: BytePos, bad_char: char) -> ! {
524 self.struct_fatal_span_char(
527 "found invalid character; only `#` is allowed in raw string delimitation",
533 fn report_unterminated_raw_string(
537 possible_offset: Option<u32>,
538 found_terminators: u32,
540 let mut err = self.sess.span_diagnostic.struct_span_fatal_with_code(
541 self.mk_sp(start, start),
542 "unterminated raw string",
546 err.span_label(self.mk_sp(start, start), "unterminated raw string");
550 "this raw string should be terminated with `\"{}`",
551 "#".repeat(n_hashes as usize)
555 if let Some(possible_offset) = possible_offset {
556 let lo = start + BytePos(possible_offset as u32);
557 let hi = lo + BytePos(found_terminators as u32);
558 let span = self.mk_sp(lo, hi);
561 "consider terminating the string here",
562 "#".repeat(n_hashes as usize),
563 Applicability::MaybeIncorrect,
570 fn report_unterminated_block_comment(&self, start: BytePos, doc_style: Option<DocStyle>) {
571 let msg = match doc_style {
572 Some(_) => "unterminated block doc-comment",
573 None => "unterminated block comment",
575 let last_bpos = self.pos;
576 let mut err = self.sess.span_diagnostic.struct_span_fatal_with_code(
577 self.mk_sp(start, last_bpos),
581 let mut nested_block_comment_open_idxs = vec![];
582 let mut last_nested_block_comment_idxs = None;
583 let mut content_chars = self.str_from(start).char_indices().peekable();
585 while let Some((idx, current_char)) = content_chars.next() {
586 match content_chars.peek() {
587 Some((_, '*')) if current_char == '/' => {
588 nested_block_comment_open_idxs.push(idx);
590 Some((_, '/')) if current_char == '*' => {
591 last_nested_block_comment_idxs =
592 nested_block_comment_open_idxs.pop().map(|open_idx| (open_idx, idx));
598 if let Some((nested_open_idx, nested_close_idx)) = last_nested_block_comment_idxs {
599 err.span_label(self.mk_sp(start, start + BytePos(2)), msg)
602 start + BytePos(nested_open_idx as u32),
603 start + BytePos(nested_open_idx as u32 + 2),
605 "...as last nested comment starts here, maybe you want to close this instead?",
609 start + BytePos(nested_close_idx as u32),
610 start + BytePos(nested_close_idx as u32 + 2),
612 "...and last nested comment terminates here.",
619 // RFC 3101 introduced the idea of (reserved) prefixes. As of Rust 2021,
620 // using a (unknown) prefix is an error. In earlier editions, however, they
621 // only result in a (allowed by default) lint, and are treated as regular
622 // identifier tokens.
623 fn report_unknown_prefix(&self, start: BytePos) {
624 let prefix_span = self.mk_sp(start, self.pos);
625 let prefix_str = self.str_from_to(start, self.pos);
626 let msg = format!("prefix `{}` is unknown", prefix_str);
628 let expn_data = prefix_span.ctxt().outer_expn_data();
630 if expn_data.edition >= Edition::Edition2021 {
631 // In Rust 2021, this is a hard error.
632 let mut err = self.sess.span_diagnostic.struct_span_err(prefix_span, &msg);
633 err.span_label(prefix_span, "unknown prefix");
634 if prefix_str == "rb" {
635 err.span_suggestion_verbose(
637 "use `br` for a raw byte string",
639 Applicability::MaybeIncorrect,
641 } else if expn_data.is_root() {
642 err.span_suggestion_verbose(
643 prefix_span.shrink_to_hi(),
644 "consider inserting whitespace here",
646 Applicability::MaybeIncorrect,
649 err.note("prefixed identifiers and literals are reserved since Rust 2021");
652 // Before Rust 2021, only emit a lint for migration.
653 self.sess.buffer_lint_with_diagnostic(
654 &RUST_2021_PREFIXES_INCOMPATIBLE_SYNTAX,
658 BuiltinLintDiagnostics::ReservedPrefix(prefix_span),
663 fn report_too_many_hashes(&self, start: BytePos, found: u32) -> ! {
668 "too many `#` symbols: raw strings may be delimited \
669 by up to 255 `#` symbols, but found {}",
677 kind: token::LitKind,
683 ) -> (token::LitKind, Symbol) {
684 let mut has_fatal_err = false;
685 let content_start = start + BytePos(prefix_len);
686 let content_end = end - BytePos(postfix_len);
687 let lit_content = self.str_from_to(content_start, content_end);
688 unescape::unescape_literal(lit_content, mode, &mut |range, result| {
689 // Here we only check for errors. The actual unescaping is done later.
690 if let Err(err) = result {
691 let span_with_quotes = self.mk_sp(start, end);
692 let (start, end) = (range.start as u32, range.end as u32);
693 let lo = content_start + BytePos(start);
694 let hi = lo + BytePos(end - start);
695 let span = self.mk_sp(lo, hi);
697 has_fatal_err = true;
700 &self.sess.span_diagnostic,
711 // We normally exclude the quotes for the symbol, but for errors we
712 // include it because it results in clearer error messages.
714 (kind, Symbol::intern(lit_content))
716 (token::Err, self.symbol_from_to(start, end))
721 pub fn nfc_normalize(string: &str) -> Symbol {
722 use unicode_normalization::{is_nfc_quick, IsNormalized, UnicodeNormalization};
723 match is_nfc_quick(string.chars()) {
724 IsNormalized::Yes => Symbol::intern(string),
726 let normalized_str: String = string.chars().nfc().collect();
727 Symbol::intern(&normalized_str)