1 //! Low-level Rust lexer.
3 //! The idea with `rustc_lexer` is to make a reusable library,
4 //! by separating out pure lexing and rustc-specific concerns, like spans,
5 //! error reporting, and interning. So, rustc_lexer operates directly on `&str`,
6 //! produces simple tokens which are a pair of type-tag and a bit of original text,
7 //! and does not report errors, instead storing them as flags on the token.
9 //! Tokens produced by this lexer are not yet ready for parsing the Rust syntax.
10 //! For that see [`rustc_parse::lexer`], which converts this basic token stream
11 //! into wide tokens used by actual parser.
13 //! The purpose of this crate is to convert raw sources into a labeled sequence
14 //! of well-known token types, so building an actual Rust token stream will
17 //! The main entity of this crate is the [`TokenKind`] enum which represents common
20 //! [`rustc_parse::lexer`]: ../rustc_parse/lexer/index.html
21 #![deny(rustc::untranslatable_diagnostic)]
22 #![deny(rustc::diagnostic_outside_of_impl)]
23 // We want to be able to build this crate with a stable compiler, so no
24 // `#![feature]` attributes should be added.
32 pub use crate::cursor::Cursor;
34 use self::LiteralKind::*;
35 use self::TokenKind::*;
36 use crate::cursor::EOF_CHAR;
37 use std::convert::TryFrom;
40 /// It doesn't contain information about data that has been parsed,
41 /// only the type of the token and its size.
49 fn new(kind: TokenKind, len: u32) -> Token {
54 /// Enum representing common lexeme types.
55 #[derive(Clone, Copy, Debug, PartialEq, Eq)]
59 LineComment { doc_style: Option<DocStyle> },
61 /// `/* block comment */`
63 /// Block comments can be recursive, so a sequence like `/* /* */`
64 /// will not be considered terminated and will result in a parsing error.
65 BlockComment { doc_style: Option<DocStyle>, terminated: bool },
67 /// Any whitespace character sequence.
70 /// "ident" or "continue"
72 /// At this step, keywords are also considered identifiers.
75 /// Like the above, but containing invalid unicode codepoints.
81 /// An unknown prefix, like `foo#`, `foo'`, `foo"`.
83 /// Note that only the
84 /// prefix (`foo`) is included in the token, not the separator (which is
85 /// lexed as its own distinct token). In Rust 2021 and later, reserved
86 /// prefixes are reported as errors; in earlier editions, they result in a
87 /// (allowed by default) lint, and are treated as regular identifier
91 /// Examples: `12u8`, `1.0e-40`, `b"123"`. Note that `_` is an invalid
92 /// suffix, but may be present here on string and float literals. Users of
93 /// this type will need to check for and reject that case.
95 /// See [LiteralKind] for more details.
96 Literal { kind: LiteralKind, suffix_start: u32 },
99 Lifetime { starts_with_number: bool },
157 /// Unknown token, not expected by the lexer, e.g. "№"
164 #[derive(Clone, Copy, Debug, PartialEq, Eq)]
170 #[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord)]
171 pub enum LiteralKind {
172 /// "12_u8", "0o100", "0b120i99"
173 Int { base: Base, empty_int: bool },
174 /// "12.34f32", "0b100.100"
175 Float { base: Base, empty_exponent: bool },
176 /// "'a'", "'\\'", "'''", "';"
177 Char { terminated: bool },
178 /// "b'a'", "b'\\'", "b'''", "b';"
179 Byte { terminated: bool },
181 Str { terminated: bool },
182 /// "b"abc"", "b"abc"
183 ByteStr { terminated: bool },
184 /// "r"abc"", "r#"abc"#", "r####"ab"###"c"####", "r#"a". `None` indicates
185 /// an invalid literal.
186 RawStr { n_hashes: Option<u8> },
187 /// "br"abc"", "br#"abc"#", "br####"ab"###"c"####", "br#"a". `None`
188 /// indicates an invalid literal.
189 RawByteStr { n_hashes: Option<u8> },
192 #[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord)]
193 pub enum RawStrError {
194 /// Non `#` characters exist between `r` and `"`, e.g. `r##~"abcde"##`
195 InvalidStarter { bad_char: char },
196 /// The string was not terminated, e.g. `r###"abcde"##`.
197 /// `possible_terminator_offset` is the number of characters after `r` or
198 /// `br` where they may have intended to terminate it.
199 NoTerminator { expected: u32, found: u32, possible_terminator_offset: Option<u32> },
200 /// More than 255 `#`s exist.
201 TooManyDelimiters { found: u32 },
204 /// Base of numeric literal encoding according to its prefix.
205 #[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord)]
207 /// Literal starts with "0b".
209 /// Literal starts with "0o".
211 /// Literal doesn't contain a prefix.
213 /// Literal starts with "0x".
217 /// `rustc` allows files to have a shebang, e.g. "#!/usr/bin/rustrun",
218 /// but shebang isn't a part of rust syntax.
219 pub fn strip_shebang(input: &str) -> Option<usize> {
220 // Shebang must start with `#!` literally, without any preceding whitespace.
221 // For simplicity we consider any line starting with `#!` a shebang,
222 // regardless of restrictions put on shebangs by specific platforms.
223 if let Some(input_tail) = input.strip_prefix("#!") {
224 // Ok, this is a shebang but if the next non-whitespace token is `[`,
225 // then it may be valid Rust code, so consider it Rust code.
226 let next_non_whitespace_token = tokenize(input_tail).map(|tok| tok.kind).find(|tok| {
229 TokenKind::Whitespace
230 | TokenKind::LineComment { doc_style: None }
231 | TokenKind::BlockComment { doc_style: None, .. }
234 if next_non_whitespace_token != Some(TokenKind::OpenBracket) {
235 // No other choice than to consider this a shebang.
236 return Some(2 + input_tail.lines().next().unwrap_or_default().len());
242 /// Validates a raw string literal. Used for getting more information about a
243 /// problem with a `RawStr`/`RawByteStr` with a `None` field.
245 pub fn validate_raw_str(input: &str, prefix_len: u32) -> Result<(), RawStrError> {
246 debug_assert!(!input.is_empty());
247 let mut cursor = Cursor::new(input);
248 // Move past the leading `r` or `br`.
249 for _ in 0..prefix_len {
250 cursor.bump().unwrap();
252 cursor.raw_double_quoted_string(prefix_len).map(|_| ())
255 /// Creates an iterator that produces tokens from the input string.
256 pub fn tokenize(input: &str) -> impl Iterator<Item = Token> + '_ {
257 let mut cursor = Cursor::new(input);
258 std::iter::from_fn(move || {
259 let token = cursor.advance_token();
260 if token.kind != TokenKind::Eof { Some(token) } else { None }
264 /// True if `c` is considered a whitespace according to Rust language definition.
265 /// See [Rust language reference](https://doc.rust-lang.org/reference/whitespace.html)
266 /// for definitions of these classes.
267 pub fn is_whitespace(c: char) -> bool {
268 // This is Pattern_White_Space.
270 // Note that this set is stable (ie, it doesn't change with different
271 // Unicode versions), so it's ok to just hard-code the values.
275 // Usual ASCII suspects
278 | '\u{000B}' // vertical tab
279 | '\u{000C}' // form feed
281 | '\u{0020}' // space
283 // NEXT LINE from latin1
287 | '\u{200E}' // LEFT-TO-RIGHT MARK
288 | '\u{200F}' // RIGHT-TO-LEFT MARK
290 // Dedicated whitespace characters from Unicode
291 | '\u{2028}' // LINE SEPARATOR
292 | '\u{2029}' // PARAGRAPH SEPARATOR
296 /// True if `c` is valid as a first character of an identifier.
297 /// See [Rust language reference](https://doc.rust-lang.org/reference/identifiers.html) for
298 /// a formal definition of valid identifier name.
299 pub fn is_id_start(c: char) -> bool {
300 // This is XID_Start OR '_' (which formally is not a XID_Start).
301 c == '_' || unicode_xid::UnicodeXID::is_xid_start(c)
304 /// True if `c` is valid as a non-first character of an identifier.
305 /// See [Rust language reference](https://doc.rust-lang.org/reference/identifiers.html) for
306 /// a formal definition of valid identifier name.
307 pub fn is_id_continue(c: char) -> bool {
308 unicode_xid::UnicodeXID::is_xid_continue(c)
311 /// The passed string is lexically an identifier.
312 pub fn is_ident(string: &str) -> bool {
313 let mut chars = string.chars();
314 if let Some(start) = chars.next() {
315 is_id_start(start) && chars.all(is_id_continue)
322 /// Parses a token from the input string.
323 pub fn advance_token(&mut self) -> Token {
324 let first_char = match self.bump() {
326 None => return Token::new(TokenKind::Eof, 0),
328 let token_kind = match first_char {
329 // Slash, comment or block comment.
330 '/' => match self.first() {
331 '/' => self.line_comment(),
332 '*' => self.block_comment(),
336 // Whitespace sequence.
337 c if is_whitespace(c) => self.whitespace(),
339 // Raw identifier, raw string literal or identifier.
340 'r' => match (self.first(), self.second()) {
341 ('#', c1) if is_id_start(c1) => self.raw_ident(),
342 ('#', _) | ('"', _) => {
343 let res = self.raw_double_quoted_string(1);
344 let suffix_start = self.pos_within_token();
346 self.eat_literal_suffix();
348 let kind = RawStr { n_hashes: res.ok() };
349 Literal { kind, suffix_start }
351 _ => self.ident_or_unknown_prefix(),
354 // Byte literal, byte string literal, raw byte string literal or identifier.
355 'b' => match (self.first(), self.second()) {
358 let terminated = self.single_quoted_string();
359 let suffix_start = self.pos_within_token();
361 self.eat_literal_suffix();
363 let kind = Byte { terminated };
364 Literal { kind, suffix_start }
368 let terminated = self.double_quoted_string();
369 let suffix_start = self.pos_within_token();
371 self.eat_literal_suffix();
373 let kind = ByteStr { terminated };
374 Literal { kind, suffix_start }
376 ('r', '"') | ('r', '#') => {
378 let res = self.raw_double_quoted_string(2);
379 let suffix_start = self.pos_within_token();
381 self.eat_literal_suffix();
383 let kind = RawByteStr { n_hashes: res.ok() };
384 Literal { kind, suffix_start }
386 _ => self.ident_or_unknown_prefix(),
389 // Identifier (this should be checked after other variant that can
390 // start as identifier).
391 c if is_id_start(c) => self.ident_or_unknown_prefix(),
395 let literal_kind = self.number(c);
396 let suffix_start = self.pos_within_token();
397 self.eat_literal_suffix();
398 TokenKind::Literal { kind: literal_kind, suffix_start }
401 // One-symbol tokens.
429 // Lifetime or character literal.
430 '\'' => self.lifetime_or_char(),
434 let terminated = self.double_quoted_string();
435 let suffix_start = self.pos_within_token();
437 self.eat_literal_suffix();
439 let kind = Str { terminated };
440 Literal { kind, suffix_start }
442 // Identifier starting with an emoji. Only lexed for graceful error recovery.
443 c if !c.is_ascii() && unic_emoji_char::is_emoji(c) => {
444 self.fake_ident_or_unknown_prefix()
448 let res = Token::new(token_kind, self.pos_within_token());
449 self.reset_pos_within_token();
453 fn line_comment(&mut self) -> TokenKind {
454 debug_assert!(self.prev() == '/' && self.first() == '/');
457 let doc_style = match self.first() {
458 // `//!` is an inner line doc comment.
459 '!' => Some(DocStyle::Inner),
460 // `////` (more than 3 slashes) is not considered a doc comment.
461 '/' if self.second() != '/' => Some(DocStyle::Outer),
465 self.eat_while(|c| c != '\n');
466 LineComment { doc_style }
469 fn block_comment(&mut self) -> TokenKind {
470 debug_assert!(self.prev() == '/' && self.first() == '*');
473 let doc_style = match self.first() {
474 // `/*!` is an inner block doc comment.
475 '!' => Some(DocStyle::Inner),
476 // `/***` (more than 2 stars) is not considered a doc comment.
477 // `/**/` is not considered a doc comment.
478 '*' if !matches!(self.second(), '*' | '/') => Some(DocStyle::Outer),
482 let mut depth = 1usize;
483 while let Some(c) = self.bump() {
485 '/' if self.first() == '*' => {
489 '*' if self.first() == '/' => {
493 // This block comment is closed, so for a construction like "/* */ */"
494 // there will be a successfully parsed block comment "/* */"
495 // and " */" will be processed separately.
503 BlockComment { doc_style, terminated: depth == 0 }
506 fn whitespace(&mut self) -> TokenKind {
507 debug_assert!(is_whitespace(self.prev()));
508 self.eat_while(is_whitespace);
512 fn raw_ident(&mut self) -> TokenKind {
513 debug_assert!(self.prev() == 'r' && self.first() == '#' && is_id_start(self.second()));
516 // Eat the identifier part of RawIdent.
517 self.eat_identifier();
521 fn ident_or_unknown_prefix(&mut self) -> TokenKind {
522 debug_assert!(is_id_start(self.prev()));
523 // Start is already eaten, eat the rest of identifier.
524 self.eat_while(is_id_continue);
525 // Known prefixes must have been handled earlier. So if
526 // we see a prefix here, it is definitely an unknown prefix.
528 '#' | '"' | '\'' => UnknownPrefix,
529 c if !c.is_ascii() && unic_emoji_char::is_emoji(c) => {
530 self.fake_ident_or_unknown_prefix()
536 fn fake_ident_or_unknown_prefix(&mut self) -> TokenKind {
537 // Start is already eaten, eat the rest of identifier.
539 unicode_xid::UnicodeXID::is_xid_continue(c)
540 || (!c.is_ascii() && unic_emoji_char::is_emoji(c))
543 // Known prefixes must have been handled earlier. So if
544 // we see a prefix here, it is definitely an unknown prefix.
546 '#' | '"' | '\'' => UnknownPrefix,
551 fn number(&mut self, first_digit: char) -> LiteralKind {
552 debug_assert!('0' <= self.prev() && self.prev() <= '9');
553 let mut base = Base::Decimal;
554 if first_digit == '0' {
555 // Attempt to parse encoding base.
556 let has_digits = match self.first() {
560 self.eat_decimal_digits()
565 self.eat_decimal_digits()
568 base = Base::Hexadecimal;
570 self.eat_hexadecimal_digits()
572 // Not a base prefix.
573 '0'..='9' | '_' | '.' | 'e' | 'E' => {
574 self.eat_decimal_digits();
578 _ => return Int { base, empty_int: false },
580 // Base prefix was provided, but there were no digits
581 // after it, e.g. "0x".
583 return Int { base, empty_int: true };
586 // No base prefix, parse number in the usual way.
587 self.eat_decimal_digits();
591 // Don't be greedy if this is actually an
592 // integer literal followed by field/method access or a range pattern
593 // (`0..2` and `12.foo()`)
594 '.' if self.second() != '.' && !is_id_start(self.second()) => {
595 // might have stuff after the ., and if it does, it needs to start
598 let mut empty_exponent = false;
599 if self.first().is_digit(10) {
600 self.eat_decimal_digits();
604 empty_exponent = !self.eat_float_exponent();
609 Float { base, empty_exponent }
613 let empty_exponent = !self.eat_float_exponent();
614 Float { base, empty_exponent }
616 _ => Int { base, empty_int: false },
620 fn lifetime_or_char(&mut self) -> TokenKind {
621 debug_assert!(self.prev() == '\'');
623 let can_be_a_lifetime = if self.second() == '\'' {
624 // It's surely not a lifetime.
627 // If the first symbol is valid for identifier, it can be a lifetime.
628 // Also check if it's a number for a better error reporting (so '0 will
629 // be reported as invalid lifetime and not as unterminated char literal).
630 is_id_start(self.first()) || self.first().is_digit(10)
633 if !can_be_a_lifetime {
634 let terminated = self.single_quoted_string();
635 let suffix_start = self.pos_within_token();
637 self.eat_literal_suffix();
639 let kind = Char { terminated };
640 return Literal { kind, suffix_start };
643 // Either a lifetime or a character literal with
644 // length greater than 1.
646 let starts_with_number = self.first().is_digit(10);
648 // Skip the literal contents.
649 // First symbol can be a number (which isn't a valid identifier start),
650 // so skip it without any checks.
652 self.eat_while(is_id_continue);
654 // Check if after skipping literal contents we've met a closing
655 // single quote (which means that user attempted to create a
656 // string with single quotes).
657 if self.first() == '\'' {
659 let kind = Char { terminated: true };
660 Literal { kind, suffix_start: self.pos_within_token() }
662 Lifetime { starts_with_number }
666 fn single_quoted_string(&mut self) -> bool {
667 debug_assert!(self.prev() == '\'');
668 // Check if it's a one-symbol literal.
669 if self.second() == '\'' && self.first() != '\\' {
675 // Literal has more than one symbol.
677 // Parse until either quotes are terminated or error is detected.
680 // Quotes are terminated, finish parsing.
685 // Probably beginning of the comment, which we don't want to include
686 // to the error report.
688 // Newline without following '\'' means unclosed quote, stop parsing.
689 '\n' if self.second() != '\'' => break,
690 // End of file, stop parsing.
691 EOF_CHAR if self.is_eof() => break,
692 // Escaped slash is considered one character, so bump twice.
697 // Skip the character.
703 // String was not terminated.
707 /// Eats double-quoted string and returns true
708 /// if string is terminated.
709 fn double_quoted_string(&mut self) -> bool {
710 debug_assert!(self.prev() == '"');
711 while let Some(c) = self.bump() {
716 '\\' if self.first() == '\\' || self.first() == '"' => {
717 // Bump again to skip escaped character.
723 // End of file reached.
727 /// Eats the double-quoted string and returns `n_hashes` and an error if encountered.
728 fn raw_double_quoted_string(&mut self, prefix_len: u32) -> Result<u8, RawStrError> {
729 // Wrap the actual function to handle the error with too many hashes.
730 // This way, it eats the whole raw string.
731 let n_hashes = self.raw_string_unvalidated(prefix_len)?;
732 // Only up to 255 `#`s are allowed in raw strings
733 match u8::try_from(n_hashes) {
735 Err(_) => Err(RawStrError::TooManyDelimiters { found: n_hashes }),
739 fn raw_string_unvalidated(&mut self, prefix_len: u32) -> Result<u32, RawStrError> {
740 debug_assert!(self.prev() == 'r');
741 let start_pos = self.pos_within_token();
742 let mut possible_terminator_offset = None;
743 let mut max_hashes = 0;
745 // Count opening '#' symbols.
747 while self.first() == '#' {
751 let n_start_hashes = eaten;
753 // Check that string is started.
757 let c = c.unwrap_or(EOF_CHAR);
758 return Err(RawStrError::InvalidStarter { bad_char: c });
762 // Skip the string contents and on each '#' character met, check if this is
763 // a raw string termination.
765 self.eat_while(|c| c != '"');
768 return Err(RawStrError::NoTerminator {
769 expected: n_start_hashes,
771 possible_terminator_offset,
775 // Eat closing double quote.
778 // Check that amount of closing '#' symbols
779 // is equal to the amount of opening ones.
780 // Note that this will not consume extra trailing `#` characters:
781 // `r###"abcde"####` is lexed as a `RawStr { n_hashes: 3 }`
782 // followed by a `#` token.
783 let mut n_end_hashes = 0;
784 while self.first() == '#' && n_end_hashes < n_start_hashes {
789 if n_end_hashes == n_start_hashes {
790 return Ok(n_start_hashes);
791 } else if n_end_hashes > max_hashes {
792 // Keep track of possible terminators to give a hint about
793 // where there might be a missing terminator
794 possible_terminator_offset =
795 Some(self.pos_within_token() - start_pos - n_end_hashes + prefix_len);
796 max_hashes = n_end_hashes;
801 fn eat_decimal_digits(&mut self) -> bool {
802 let mut has_digits = false;
818 fn eat_hexadecimal_digits(&mut self) -> bool {
819 let mut has_digits = false;
825 '0'..='9' | 'a'..='f' | 'A'..='F' => {
835 /// Eats the float exponent. Returns true if at least one digit was met,
836 /// and returns false otherwise.
837 fn eat_float_exponent(&mut self) -> bool {
838 debug_assert!(self.prev() == 'e' || self.prev() == 'E');
839 if self.first() == '-' || self.first() == '+' {
842 self.eat_decimal_digits()
845 // Eats the suffix of the literal, e.g. "u8".
846 fn eat_literal_suffix(&mut self) {
847 self.eat_identifier();
850 // Eats the identifier. Note: succeeds on `_`, which isn't a valid
852 fn eat_identifier(&mut self) {
853 if !is_id_start(self.first()) {
858 self.eat_while(is_id_continue);