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> },
60 /// `/* block comment */`
62 /// Block comments can be recursive, so the sequence like `/* /* */`
63 /// will not be considered terminated and will result in a parsing error.
64 BlockComment { doc_style: Option<DocStyle>, terminated: bool },
65 /// Any whitespace characters sequence.
67 /// "ident" or "continue"
68 /// At this step keywords are also considered identifiers.
70 /// Like the above, but containing invalid unicode codepoints.
74 /// An unknown prefix like `foo#`, `foo'`, `foo"`. Note that only the
75 /// prefix (`foo`) is included in the token, not the separator (which is
76 /// lexed as its own distinct token). In Rust 2021 and later, reserved
77 /// prefixes are reported as errors; in earlier editions, they result in a
78 /// (allowed by default) lint, and are treated as regular identifier
81 /// "12_u8", "1.0e-40", "b"123"". See `LiteralKind` for more details.
82 Literal { kind: LiteralKind, suffix_start: u32 },
84 Lifetime { starts_with_number: bool },
142 /// Unknown token, not expected by the lexer, e.g. "№"
149 #[derive(Clone, Copy, Debug, PartialEq, Eq)]
155 #[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord)]
156 pub enum LiteralKind {
157 /// "12_u8", "0o100", "0b120i99"
158 Int { base: Base, empty_int: bool },
159 /// "12.34f32", "0b100.100"
160 Float { base: Base, empty_exponent: bool },
161 /// "'a'", "'\\'", "'''", "';"
162 Char { terminated: bool },
163 /// "b'a'", "b'\\'", "b'''", "b';"
164 Byte { terminated: bool },
166 Str { terminated: bool },
167 /// "b"abc"", "b"abc"
168 ByteStr { terminated: bool },
169 /// "r"abc"", "r#"abc"#", "r####"ab"###"c"####", "r#"a". `None` indicates
170 /// an invalid literal.
171 RawStr { n_hashes: Option<u8> },
172 /// "br"abc"", "br#"abc"#", "br####"ab"###"c"####", "br#"a". `None`
173 /// indicates an invalid literal.
174 RawByteStr { n_hashes: Option<u8> },
177 #[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord)]
178 pub enum RawStrError {
179 /// Non `#` characters exist between `r` and `"`, e.g. `r##~"abcde"##`
180 InvalidStarter { bad_char: char },
181 /// The string was not terminated, e.g. `r###"abcde"##`.
182 /// `possible_terminator_offset` is the number of characters after `r` or
183 /// `br` where they may have intended to terminate it.
184 NoTerminator { expected: u32, found: u32, possible_terminator_offset: Option<u32> },
185 /// More than 255 `#`s exist.
186 TooManyDelimiters { found: u32 },
189 /// Base of numeric literal encoding according to its prefix.
190 #[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord)]
192 /// Literal starts with "0b".
194 /// Literal starts with "0o".
196 /// Literal starts with "0x".
198 /// Literal doesn't contain a prefix.
202 /// `rustc` allows files to have a shebang, e.g. "#!/usr/bin/rustrun",
203 /// but shebang isn't a part of rust syntax.
204 pub fn strip_shebang(input: &str) -> Option<usize> {
205 // Shebang must start with `#!` literally, without any preceding whitespace.
206 // For simplicity we consider any line starting with `#!` a shebang,
207 // regardless of restrictions put on shebangs by specific platforms.
208 if let Some(input_tail) = input.strip_prefix("#!") {
209 // Ok, this is a shebang but if the next non-whitespace token is `[`,
210 // then it may be valid Rust code, so consider it Rust code.
211 let next_non_whitespace_token = tokenize(input_tail).map(|tok| tok.kind).find(|tok| {
214 TokenKind::Whitespace
215 | TokenKind::LineComment { doc_style: None }
216 | TokenKind::BlockComment { doc_style: None, .. }
219 if next_non_whitespace_token != Some(TokenKind::OpenBracket) {
220 // No other choice than to consider this a shebang.
221 return Some(2 + input_tail.lines().next().unwrap_or_default().len());
227 /// Validates a raw string literal. Used for getting more information about a
228 /// problem with a `RawStr`/`RawByteStr` with a `None` field.
230 pub fn validate_raw_str(input: &str, prefix_len: u32) -> Result<(), RawStrError> {
231 debug_assert!(!input.is_empty());
232 let mut cursor = Cursor::new(input);
233 // Move past the leading `r` or `br`.
234 for _ in 0..prefix_len {
235 cursor.bump().unwrap();
237 cursor.raw_double_quoted_string(prefix_len).map(|_| ())
240 /// Creates an iterator that produces tokens from the input string.
241 pub fn tokenize(input: &str) -> impl Iterator<Item = Token> + '_ {
242 let mut cursor = Cursor::new(input);
243 std::iter::from_fn(move || {
244 let token = cursor.advance_token();
245 if token.kind != TokenKind::Eof { Some(token) } else { None }
249 /// True if `c` is considered a whitespace according to Rust language definition.
250 /// See [Rust language reference](https://doc.rust-lang.org/reference/whitespace.html)
251 /// for definitions of these classes.
252 pub fn is_whitespace(c: char) -> bool {
253 // This is Pattern_White_Space.
255 // Note that this set is stable (ie, it doesn't change with different
256 // Unicode versions), so it's ok to just hard-code the values.
260 // Usual ASCII suspects
263 | '\u{000B}' // vertical tab
264 | '\u{000C}' // form feed
266 | '\u{0020}' // space
268 // NEXT LINE from latin1
272 | '\u{200E}' // LEFT-TO-RIGHT MARK
273 | '\u{200F}' // RIGHT-TO-LEFT MARK
275 // Dedicated whitespace characters from Unicode
276 | '\u{2028}' // LINE SEPARATOR
277 | '\u{2029}' // PARAGRAPH SEPARATOR
281 /// True if `c` is valid as a first character of an identifier.
282 /// See [Rust language reference](https://doc.rust-lang.org/reference/identifiers.html) for
283 /// a formal definition of valid identifier name.
284 pub fn is_id_start(c: char) -> bool {
285 // This is XID_Start OR '_' (which formally is not a XID_Start).
286 c == '_' || unicode_xid::UnicodeXID::is_xid_start(c)
289 /// True if `c` is valid as a non-first character of an identifier.
290 /// See [Rust language reference](https://doc.rust-lang.org/reference/identifiers.html) for
291 /// a formal definition of valid identifier name.
292 pub fn is_id_continue(c: char) -> bool {
293 unicode_xid::UnicodeXID::is_xid_continue(c)
296 /// The passed string is lexically an identifier.
297 pub fn is_ident(string: &str) -> bool {
298 let mut chars = string.chars();
299 if let Some(start) = chars.next() {
300 is_id_start(start) && chars.all(is_id_continue)
307 /// Parses a token from the input string.
308 pub fn advance_token(&mut self) -> Token {
309 let first_char = match self.bump() {
311 None => return Token::new(TokenKind::Eof, 0),
313 let token_kind = match first_char {
314 // Slash, comment or block comment.
315 '/' => match self.first() {
316 '/' => self.line_comment(),
317 '*' => self.block_comment(),
321 // Whitespace sequence.
322 c if is_whitespace(c) => self.whitespace(),
324 // Raw identifier, raw string literal or identifier.
325 'r' => match (self.first(), self.second()) {
326 ('#', c1) if is_id_start(c1) => self.raw_ident(),
327 ('#', _) | ('"', _) => {
328 let res = self.raw_double_quoted_string(1);
329 let suffix_start = self.pos_within_token();
331 self.eat_literal_suffix();
333 let kind = RawStr { n_hashes: res.ok() };
334 Literal { kind, suffix_start }
336 _ => self.ident_or_unknown_prefix(),
339 // Byte literal, byte string literal, raw byte string literal or identifier.
340 'b' => match (self.first(), self.second()) {
343 let terminated = self.single_quoted_string();
344 let suffix_start = self.pos_within_token();
346 self.eat_literal_suffix();
348 let kind = Byte { terminated };
349 Literal { kind, suffix_start }
353 let terminated = self.double_quoted_string();
354 let suffix_start = self.pos_within_token();
356 self.eat_literal_suffix();
358 let kind = ByteStr { terminated };
359 Literal { kind, suffix_start }
361 ('r', '"') | ('r', '#') => {
363 let res = self.raw_double_quoted_string(2);
364 let suffix_start = self.pos_within_token();
366 self.eat_literal_suffix();
368 let kind = RawByteStr { n_hashes: res.ok() };
369 Literal { kind, suffix_start }
371 _ => self.ident_or_unknown_prefix(),
374 // Identifier (this should be checked after other variant that can
375 // start as identifier).
376 c if is_id_start(c) => self.ident_or_unknown_prefix(),
380 let literal_kind = self.number(c);
381 let suffix_start = self.pos_within_token();
382 self.eat_literal_suffix();
383 TokenKind::Literal { kind: literal_kind, suffix_start }
386 // One-symbol tokens.
414 // Lifetime or character literal.
415 '\'' => self.lifetime_or_char(),
419 let terminated = self.double_quoted_string();
420 let suffix_start = self.pos_within_token();
422 self.eat_literal_suffix();
424 let kind = Str { terminated };
425 Literal { kind, suffix_start }
427 // Identifier starting with an emoji. Only lexed for graceful error recovery.
428 c if !c.is_ascii() && unic_emoji_char::is_emoji(c) => {
429 self.fake_ident_or_unknown_prefix()
433 let res = Token::new(token_kind, self.pos_within_token());
434 self.reset_pos_within_token();
438 fn line_comment(&mut self) -> TokenKind {
439 debug_assert!(self.prev() == '/' && self.first() == '/');
442 let doc_style = match self.first() {
443 // `//!` is an inner line doc comment.
444 '!' => Some(DocStyle::Inner),
445 // `////` (more than 3 slashes) is not considered a doc comment.
446 '/' if self.second() != '/' => Some(DocStyle::Outer),
450 self.eat_while(|c| c != '\n');
451 LineComment { doc_style }
454 fn block_comment(&mut self) -> TokenKind {
455 debug_assert!(self.prev() == '/' && self.first() == '*');
458 let doc_style = match self.first() {
459 // `/*!` is an inner block doc comment.
460 '!' => Some(DocStyle::Inner),
461 // `/***` (more than 2 stars) is not considered a doc comment.
462 // `/**/` is not considered a doc comment.
463 '*' if !matches!(self.second(), '*' | '/') => Some(DocStyle::Outer),
467 let mut depth = 1usize;
468 while let Some(c) = self.bump() {
470 '/' if self.first() == '*' => {
474 '*' if self.first() == '/' => {
478 // This block comment is closed, so for a construction like "/* */ */"
479 // there will be a successfully parsed block comment "/* */"
480 // and " */" will be processed separately.
488 BlockComment { doc_style, terminated: depth == 0 }
491 fn whitespace(&mut self) -> TokenKind {
492 debug_assert!(is_whitespace(self.prev()));
493 self.eat_while(is_whitespace);
497 fn raw_ident(&mut self) -> TokenKind {
498 debug_assert!(self.prev() == 'r' && self.first() == '#' && is_id_start(self.second()));
501 // Eat the identifier part of RawIdent.
502 self.eat_identifier();
506 fn ident_or_unknown_prefix(&mut self) -> TokenKind {
507 debug_assert!(is_id_start(self.prev()));
508 // Start is already eaten, eat the rest of identifier.
509 self.eat_while(is_id_continue);
510 // Known prefixes must have been handled earlier. So if
511 // we see a prefix here, it is definitely an unknown prefix.
513 '#' | '"' | '\'' => UnknownPrefix,
514 c if !c.is_ascii() && unic_emoji_char::is_emoji(c) => {
515 self.fake_ident_or_unknown_prefix()
521 fn fake_ident_or_unknown_prefix(&mut self) -> TokenKind {
522 // Start is already eaten, eat the rest of identifier.
524 unicode_xid::UnicodeXID::is_xid_continue(c)
525 || (!c.is_ascii() && unic_emoji_char::is_emoji(c))
528 // Known prefixes must have been handled earlier. So if
529 // we see a prefix here, it is definitely an unknown prefix.
531 '#' | '"' | '\'' => UnknownPrefix,
536 fn number(&mut self, first_digit: char) -> LiteralKind {
537 debug_assert!('0' <= self.prev() && self.prev() <= '9');
538 let mut base = Base::Decimal;
539 if first_digit == '0' {
540 // Attempt to parse encoding base.
541 let has_digits = match self.first() {
545 self.eat_decimal_digits()
550 self.eat_decimal_digits()
553 base = Base::Hexadecimal;
555 self.eat_hexadecimal_digits()
557 // Not a base prefix.
558 '0'..='9' | '_' | '.' | 'e' | 'E' => {
559 self.eat_decimal_digits();
563 _ => return Int { base, empty_int: false },
565 // Base prefix was provided, but there were no digits
566 // after it, e.g. "0x".
568 return Int { base, empty_int: true };
571 // No base prefix, parse number in the usual way.
572 self.eat_decimal_digits();
576 // Don't be greedy if this is actually an
577 // integer literal followed by field/method access or a range pattern
578 // (`0..2` and `12.foo()`)
579 '.' if self.second() != '.' && !is_id_start(self.second()) => {
580 // might have stuff after the ., and if it does, it needs to start
583 let mut empty_exponent = false;
584 if self.first().is_digit(10) {
585 self.eat_decimal_digits();
589 empty_exponent = !self.eat_float_exponent();
594 Float { base, empty_exponent }
598 let empty_exponent = !self.eat_float_exponent();
599 Float { base, empty_exponent }
601 _ => Int { base, empty_int: false },
605 fn lifetime_or_char(&mut self) -> TokenKind {
606 debug_assert!(self.prev() == '\'');
608 let can_be_a_lifetime = if self.second() == '\'' {
609 // It's surely not a lifetime.
612 // If the first symbol is valid for identifier, it can be a lifetime.
613 // Also check if it's a number for a better error reporting (so '0 will
614 // be reported as invalid lifetime and not as unterminated char literal).
615 is_id_start(self.first()) || self.first().is_digit(10)
618 if !can_be_a_lifetime {
619 let terminated = self.single_quoted_string();
620 let suffix_start = self.pos_within_token();
622 self.eat_literal_suffix();
624 let kind = Char { terminated };
625 return Literal { kind, suffix_start };
628 // Either a lifetime or a character literal with
629 // length greater than 1.
631 let starts_with_number = self.first().is_digit(10);
633 // Skip the literal contents.
634 // First symbol can be a number (which isn't a valid identifier start),
635 // so skip it without any checks.
637 self.eat_while(is_id_continue);
639 // Check if after skipping literal contents we've met a closing
640 // single quote (which means that user attempted to create a
641 // string with single quotes).
642 if self.first() == '\'' {
644 let kind = Char { terminated: true };
645 Literal { kind, suffix_start: self.pos_within_token() }
647 Lifetime { starts_with_number }
651 fn single_quoted_string(&mut self) -> bool {
652 debug_assert!(self.prev() == '\'');
653 // Check if it's a one-symbol literal.
654 if self.second() == '\'' && self.first() != '\\' {
660 // Literal has more than one symbol.
662 // Parse until either quotes are terminated or error is detected.
665 // Quotes are terminated, finish parsing.
670 // Probably beginning of the comment, which we don't want to include
671 // to the error report.
673 // Newline without following '\'' means unclosed quote, stop parsing.
674 '\n' if self.second() != '\'' => break,
675 // End of file, stop parsing.
676 EOF_CHAR if self.is_eof() => break,
677 // Escaped slash is considered one character, so bump twice.
682 // Skip the character.
688 // String was not terminated.
692 /// Eats double-quoted string and returns true
693 /// if string is terminated.
694 fn double_quoted_string(&mut self) -> bool {
695 debug_assert!(self.prev() == '"');
696 while let Some(c) = self.bump() {
701 '\\' if self.first() == '\\' || self.first() == '"' => {
702 // Bump again to skip escaped character.
708 // End of file reached.
712 /// Eats the double-quoted string and returns `n_hashes` and an error if encountered.
713 fn raw_double_quoted_string(&mut self, prefix_len: u32) -> Result<u8, RawStrError> {
714 // Wrap the actual function to handle the error with too many hashes.
715 // This way, it eats the whole raw string.
716 let n_hashes = self.raw_string_unvalidated(prefix_len)?;
717 // Only up to 255 `#`s are allowed in raw strings
718 match u8::try_from(n_hashes) {
720 Err(_) => Err(RawStrError::TooManyDelimiters { found: n_hashes }),
724 fn raw_string_unvalidated(&mut self, prefix_len: u32) -> Result<u32, RawStrError> {
725 debug_assert!(self.prev() == 'r');
726 let start_pos = self.pos_within_token();
727 let mut possible_terminator_offset = None;
728 let mut max_hashes = 0;
730 // Count opening '#' symbols.
732 while self.first() == '#' {
736 let n_start_hashes = eaten;
738 // Check that string is started.
742 let c = c.unwrap_or(EOF_CHAR);
743 return Err(RawStrError::InvalidStarter { bad_char: c });
747 // Skip the string contents and on each '#' character met, check if this is
748 // a raw string termination.
750 self.eat_while(|c| c != '"');
753 return Err(RawStrError::NoTerminator {
754 expected: n_start_hashes,
756 possible_terminator_offset,
760 // Eat closing double quote.
763 // Check that amount of closing '#' symbols
764 // is equal to the amount of opening ones.
765 // Note that this will not consume extra trailing `#` characters:
766 // `r###"abcde"####` is lexed as a `RawStr { n_hashes: 3 }`
767 // followed by a `#` token.
768 let mut n_end_hashes = 0;
769 while self.first() == '#' && n_end_hashes < n_start_hashes {
774 if n_end_hashes == n_start_hashes {
775 return Ok(n_start_hashes);
776 } else if n_end_hashes > max_hashes {
777 // Keep track of possible terminators to give a hint about
778 // where there might be a missing terminator
779 possible_terminator_offset =
780 Some(self.pos_within_token() - start_pos - n_end_hashes + prefix_len);
781 max_hashes = n_end_hashes;
786 fn eat_decimal_digits(&mut self) -> bool {
787 let mut has_digits = false;
803 fn eat_hexadecimal_digits(&mut self) -> bool {
804 let mut has_digits = false;
810 '0'..='9' | 'a'..='f' | 'A'..='F' => {
820 /// Eats the float exponent. Returns true if at least one digit was met,
821 /// and returns false otherwise.
822 fn eat_float_exponent(&mut self) -> bool {
823 debug_assert!(self.prev() == 'e' || self.prev() == 'E');
824 if self.first() == '-' || self.first() == '+' {
827 self.eat_decimal_digits()
830 // Eats the suffix of the literal, e.g. "_u8".
831 fn eat_literal_suffix(&mut self) {
832 self.eat_identifier();
835 // Eats the identifier.
836 fn eat_identifier(&mut self) {
837 if !is_id_start(self.first()) {
842 self.eat_while(is_id_continue);