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;
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 let c = self.str_from(start).chars().next().unwrap();
237 self.struct_err_span_char(start, self.pos, "unknown start of token", c);
238 // FIXME: the lexer could be used to turn the ASCII version of unicode
239 // homoglyphs, instead of keeping a table in `check_for_substitution`into the
240 // token. Ideally, this should be inside `rustc_lexer`. However, we should
241 // first remove compound tokens like `<<` from `rustc_lexer`, and then add
242 // fancier error recovery to it, as there will be less overall work to do this
244 let token = unicode_chars::check_for_substitution(self, start, c, &mut err);
246 err.help("source files must contain UTF-8 encoded text, unexpected null bytes might occur when a different encoding is used");
249 if let Some(token) = token {
252 preceded_by_whitespace = true;
256 rustc_lexer::TokenKind::Eof => token::Eof,
258 let span = self.mk_sp(start, self.pos);
259 return (Token::new(kind, span), preceded_by_whitespace);
263 /// Report a fatal lexical error with a given span.
264 fn fatal_span(&self, sp: Span, m: &str) -> ! {
265 self.sess.span_diagnostic.span_fatal(sp, m)
268 /// Report a lexical error with a given span.
269 fn err_span(&self, sp: Span, m: &str) {
270 self.sess.span_diagnostic.struct_span_err(sp, m).emit();
273 /// Report a fatal error spanning [`from_pos`, `to_pos`).
274 fn fatal_span_(&self, from_pos: BytePos, to_pos: BytePos, m: &str) -> ! {
275 self.fatal_span(self.mk_sp(from_pos, to_pos), m)
278 /// Report a lexical error spanning [`from_pos`, `to_pos`).
279 fn err_span_(&self, from_pos: BytePos, to_pos: BytePos, m: &str) {
280 self.err_span(self.mk_sp(from_pos, to_pos), m)
283 fn struct_fatal_span_char(
289 ) -> DiagnosticBuilder<'a, !> {
292 .struct_span_fatal(self.mk_sp(from_pos, to_pos), &format!("{}: {}", m, escaped_char(c)))
295 fn struct_err_span_char(
301 ) -> DiagnosticBuilder<'a, ErrorGuaranteed> {
304 .struct_span_err(self.mk_sp(from_pos, to_pos), &format!("{}: {}", m, escaped_char(c)))
307 /// Detect usages of Unicode codepoints changing the direction of the text on screen and loudly
308 /// complain about it.
309 fn lint_unicode_text_flow(&self, start: BytePos) {
310 // Opening delimiter of the length 2 is not included into the comment text.
311 let content_start = start + BytePos(2);
312 let content = self.str_from(content_start);
313 if contains_text_flow_control_chars(content) {
314 let span = self.mk_sp(start, self.pos);
315 self.sess.buffer_lint_with_diagnostic(
316 &TEXT_DIRECTION_CODEPOINT_IN_COMMENT,
319 "unicode codepoint changing visible direction of text present in comment",
320 BuiltinLintDiagnostics::UnicodeTextFlow(span, content.to_string()),
327 content_start: BytePos,
329 comment_kind: CommentKind,
332 if content.contains('\r') {
333 for (idx, _) in content.char_indices().filter(|&(_, c)| c == '\r') {
335 content_start + BytePos(idx as u32),
336 content_start + BytePos(idx as u32 + 1),
338 CommentKind::Line => "bare CR not allowed in doc-comment",
339 CommentKind::Block => "bare CR not allowed in block doc-comment",
345 let attr_style = match doc_style {
346 DocStyle::Outer => AttrStyle::Outer,
347 DocStyle::Inner => AttrStyle::Inner,
350 token::DocComment(comment_kind, attr_style, Symbol::intern(content))
353 fn cook_lexer_literal(
357 kind: rustc_lexer::LiteralKind,
358 ) -> (token::LitKind, Symbol) {
360 rustc_lexer::LiteralKind::Char { terminated } => {
362 self.sess.span_diagnostic.span_fatal_with_code(
363 self.mk_sp(start, end),
364 "unterminated character literal",
368 self.cook_quoted(token::Char, Mode::Char, start, end, 1, 1) // ' '
370 rustc_lexer::LiteralKind::Byte { terminated } => {
372 self.sess.span_diagnostic.span_fatal_with_code(
373 self.mk_sp(start + BytePos(1), end),
374 "unterminated byte constant",
378 self.cook_quoted(token::Byte, Mode::Byte, start, end, 2, 1) // b' '
380 rustc_lexer::LiteralKind::Str { terminated } => {
382 self.sess.span_diagnostic.span_fatal_with_code(
383 self.mk_sp(start, end),
384 "unterminated double quote string",
388 self.cook_quoted(token::Str, Mode::Str, start, end, 1, 1) // " "
390 rustc_lexer::LiteralKind::ByteStr { terminated } => {
392 self.sess.span_diagnostic.span_fatal_with_code(
393 self.mk_sp(start + BytePos(1), end),
394 "unterminated double quote byte string",
398 self.cook_quoted(token::ByteStr, Mode::ByteStr, start, end, 2, 1) // b" "
400 rustc_lexer::LiteralKind::RawStr { n_hashes } => {
401 if let Some(n_hashes) = n_hashes {
402 let n = u32::from(n_hashes);
403 let kind = token::StrRaw(n_hashes);
404 self.cook_quoted(kind, Mode::RawStr, start, end, 2 + n, 1 + n) // r##" "##
406 self.report_raw_str_error(start, 1);
409 rustc_lexer::LiteralKind::RawByteStr { n_hashes } => {
410 if let Some(n_hashes) = n_hashes {
411 let n = u32::from(n_hashes);
412 let kind = token::ByteStrRaw(n_hashes);
413 self.cook_quoted(kind, Mode::RawByteStr, start, end, 3 + n, 1 + n) // br##" "##
415 self.report_raw_str_error(start, 2);
418 rustc_lexer::LiteralKind::Int { base, empty_int } => {
422 .struct_span_err_with_code(
423 self.mk_sp(start, end),
424 "no valid digits found for number",
428 (token::Integer, sym::integer(0))
430 if matches!(base, Base::Binary | Base::Octal) {
431 let base = base as u32;
432 let s = self.str_from_to(start + BytePos(2), end);
433 for (idx, c) in s.char_indices() {
434 if c != '_' && c.to_digit(base).is_none() {
436 start + BytePos::from_usize(2 + idx),
437 start + BytePos::from_usize(2 + idx + c.len_utf8()),
438 &format!("invalid digit for a base {} literal", base),
443 (token::Integer, self.symbol_from_to(start, end))
446 rustc_lexer::LiteralKind::Float { base, empty_exponent } => {
448 self.err_span_(start, self.pos, "expected at least one digit in exponent");
451 Base::Hexadecimal => {
452 self.err_span_(start, end, "hexadecimal float literal is not supported")
455 self.err_span_(start, end, "octal float literal is not supported")
458 self.err_span_(start, end, "binary float literal is not supported")
462 (token::Float, self.symbol_from_to(start, end))
468 fn src_index(&self, pos: BytePos) -> usize {
469 (pos - self.start_pos).to_usize()
472 /// Slice of the source text from `start` up to but excluding `self.pos`,
473 /// meaning the slice does not include the character `self.ch`.
474 fn str_from(&self, start: BytePos) -> &str {
475 self.str_from_to(start, self.pos)
478 /// As symbol_from, with an explicit endpoint.
479 fn symbol_from_to(&self, start: BytePos, end: BytePos) -> Symbol {
480 debug!("taking an ident from {:?} to {:?}", start, end);
481 Symbol::intern(self.str_from_to(start, end))
484 /// Slice of the source text spanning from `start` up to but excluding `end`.
485 fn str_from_to(&self, start: BytePos, end: BytePos) -> &str {
486 &self.src[self.src_index(start)..self.src_index(end)]
489 fn report_raw_str_error(&self, start: BytePos, prefix_len: u32) -> ! {
490 match rustc_lexer::validate_raw_str(self.str_from(start), prefix_len) {
491 Err(RawStrError::InvalidStarter { bad_char }) => {
492 self.report_non_started_raw_string(start, bad_char)
494 Err(RawStrError::NoTerminator { expected, found, possible_terminator_offset }) => self
495 .report_unterminated_raw_string(start, expected, possible_terminator_offset, found),
496 Err(RawStrError::TooManyDelimiters { found }) => {
497 self.report_too_many_hashes(start, found)
499 Ok(()) => panic!("no error found for supposedly invalid raw string literal"),
503 fn report_non_started_raw_string(&self, start: BytePos, bad_char: char) -> ! {
504 self.struct_fatal_span_char(
507 "found invalid character; only `#` is allowed in raw string delimitation",
513 fn report_unterminated_raw_string(
517 possible_offset: Option<u32>,
518 found_terminators: u32,
520 let mut err = self.sess.span_diagnostic.struct_span_fatal_with_code(
521 self.mk_sp(start, start),
522 "unterminated raw string",
526 err.span_label(self.mk_sp(start, start), "unterminated raw string");
530 "this raw string should be terminated with `\"{}`",
531 "#".repeat(n_hashes as usize)
535 if let Some(possible_offset) = possible_offset {
536 let lo = start + BytePos(possible_offset as u32);
537 let hi = lo + BytePos(found_terminators as u32);
538 let span = self.mk_sp(lo, hi);
541 "consider terminating the string here",
542 "#".repeat(n_hashes as usize),
543 Applicability::MaybeIncorrect,
550 fn report_unterminated_block_comment(&self, start: BytePos, doc_style: Option<DocStyle>) {
551 let msg = match doc_style {
552 Some(_) => "unterminated block doc-comment",
553 None => "unterminated block comment",
555 let last_bpos = self.pos;
556 let mut err = self.sess.span_diagnostic.struct_span_fatal_with_code(
557 self.mk_sp(start, last_bpos),
561 let mut nested_block_comment_open_idxs = vec![];
562 let mut last_nested_block_comment_idxs = None;
563 let mut content_chars = self.str_from(start).char_indices().peekable();
565 while let Some((idx, current_char)) = content_chars.next() {
566 match content_chars.peek() {
567 Some((_, '*')) if current_char == '/' => {
568 nested_block_comment_open_idxs.push(idx);
570 Some((_, '/')) if current_char == '*' => {
571 last_nested_block_comment_idxs =
572 nested_block_comment_open_idxs.pop().map(|open_idx| (open_idx, idx));
578 if let Some((nested_open_idx, nested_close_idx)) = last_nested_block_comment_idxs {
579 err.span_label(self.mk_sp(start, start + BytePos(2)), msg)
582 start + BytePos(nested_open_idx as u32),
583 start + BytePos(nested_open_idx as u32 + 2),
585 "...as last nested comment starts here, maybe you want to close this instead?",
589 start + BytePos(nested_close_idx as u32),
590 start + BytePos(nested_close_idx as u32 + 2),
592 "...and last nested comment terminates here.",
599 // RFC 3101 introduced the idea of (reserved) prefixes. As of Rust 2021,
600 // using a (unknown) prefix is an error. In earlier editions, however, they
601 // only result in a (allowed by default) lint, and are treated as regular
602 // identifier tokens.
603 fn report_unknown_prefix(&self, start: BytePos) {
604 let prefix_span = self.mk_sp(start, self.pos);
605 let prefix_str = self.str_from_to(start, self.pos);
606 let msg = format!("prefix `{}` is unknown", prefix_str);
608 let expn_data = prefix_span.ctxt().outer_expn_data();
610 if expn_data.edition >= Edition::Edition2021 {
611 // In Rust 2021, this is a hard error.
612 let mut err = self.sess.span_diagnostic.struct_span_err(prefix_span, &msg);
613 err.span_label(prefix_span, "unknown prefix");
614 if prefix_str == "rb" {
615 err.span_suggestion_verbose(
617 "use `br` for a raw byte string",
619 Applicability::MaybeIncorrect,
621 } else if expn_data.is_root() {
622 err.span_suggestion_verbose(
623 prefix_span.shrink_to_hi(),
624 "consider inserting whitespace here",
626 Applicability::MaybeIncorrect,
629 err.note("prefixed identifiers and literals are reserved since Rust 2021");
632 // Before Rust 2021, only emit a lint for migration.
633 self.sess.buffer_lint_with_diagnostic(
634 &RUST_2021_PREFIXES_INCOMPATIBLE_SYNTAX,
638 BuiltinLintDiagnostics::ReservedPrefix(prefix_span),
643 fn report_too_many_hashes(&self, start: BytePos, found: u32) -> ! {
648 "too many `#` symbols: raw strings may be delimited \
649 by up to 255 `#` symbols, but found {}",
657 kind: token::LitKind,
663 ) -> (token::LitKind, Symbol) {
664 let mut has_fatal_err = false;
665 let content_start = start + BytePos(prefix_len);
666 let content_end = end - BytePos(postfix_len);
667 let lit_content = self.str_from_to(content_start, content_end);
668 unescape::unescape_literal(lit_content, mode, &mut |range, result| {
669 // Here we only check for errors. The actual unescaping is done later.
670 if let Err(err) = result {
671 let span_with_quotes = self.mk_sp(start, end);
672 let (start, end) = (range.start as u32, range.end as u32);
673 let lo = content_start + BytePos(start);
674 let hi = lo + BytePos(end - start);
675 let span = self.mk_sp(lo, hi);
677 has_fatal_err = true;
680 &self.sess.span_diagnostic,
691 // We normally exclude the quotes for the symbol, but for errors we
692 // include it because it results in clearer error messages.
694 (kind, Symbol::intern(lit_content))
696 (token::Err, self.symbol_from_to(start, end))
701 pub fn nfc_normalize(string: &str) -> Symbol {
702 use unicode_normalization::{is_nfc_quick, IsNormalized, UnicodeNormalization};
703 match is_nfc_quick(string.chars()) {
704 IsNormalized::Yes => Symbol::intern(string),
706 let normalized_str: String = string.chars().nfc().collect();
707 Symbol::intern(&normalized_str)