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
;
39 /// It doesn't contain information about data that has been parsed,
40 /// only the type of the token and its size.
48 fn new(kind
: TokenKind
, len
: u32) -> Token
{
53 /// Enum representing common lexeme types.
54 #[derive(Clone, Copy, Debug, PartialEq, Eq)]
58 LineComment { doc_style: Option<DocStyle> }
,
60 /// `/* block comment */`
62 /// Block comments can be recursive, so a sequence like `/* /* */`
63 /// will not be considered terminated and will result in a parsing error.
64 BlockComment { doc_style: Option<DocStyle>, terminated: bool }
,
66 /// Any whitespace character sequence.
69 /// "ident" or "continue"
71 /// At this step, keywords are also considered identifiers.
74 /// Like the above, but containing invalid unicode codepoints.
80 /// An unknown prefix, like `foo#`, `foo'`, `foo"`.
82 /// Note that only the
83 /// prefix (`foo`) is included in the token, not the separator (which is
84 /// lexed as its own distinct token). In Rust 2021 and later, reserved
85 /// prefixes are reported as errors; in earlier editions, they result in a
86 /// (allowed by default) lint, and are treated as regular identifier
90 /// Examples: `12u8`, `1.0e-40`, `b"123"`. Note that `_` is an invalid
91 /// suffix, but may be present here on string and float literals. Users of
92 /// this type will need to check for and reject that case.
94 /// See [LiteralKind] for more details.
95 Literal { kind: LiteralKind, suffix_start: u32 }
,
98 Lifetime { starts_with_number: bool }
,
156 /// Unknown token, not expected by the lexer, e.g. "№"
163 #[derive(Clone, Copy, Debug, PartialEq, Eq)]
169 // Note that the suffix is *not* considered when deciding the `LiteralKind` in
170 // this type. This means that float literals like `1f32` are classified by this
171 // type as `Int`. (Compare against `rustc_ast::token::LitKind` and
172 // `rustc_ast::ast::LitKind.)
173 #[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord)]
174 pub enum LiteralKind
{
175 /// "12_u8", "0o100", "0b120i99", "1f32".
176 Int { base: Base, empty_int: bool }
,
177 /// "12.34f32", "1e3", but not "1f32`.
178 Float { base: Base, empty_exponent: bool }
,
179 /// "'a'", "'\\'", "'''", "';"
180 Char { terminated: bool }
,
181 /// "b'a'", "b'\\'", "b'''", "b';"
182 Byte { terminated: bool }
,
184 Str { terminated: bool }
,
185 /// "b"abc"", "b"abc"
186 ByteStr { terminated: bool }
,
187 /// "r"abc"", "r#"abc"#", "r####"ab"###"c"####", "r#"a". `None` indicates
188 /// an invalid literal.
189 RawStr { n_hashes: Option<u8> }
,
190 /// "br"abc"", "br#"abc"#", "br####"ab"###"c"####", "br#"a". `None`
191 /// indicates an invalid literal.
192 RawByteStr { n_hashes: Option<u8> }
,
195 #[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord)]
196 pub enum RawStrError
{
197 /// Non `#` characters exist between `r` and `"`, e.g. `r##~"abcde"##`
198 InvalidStarter { bad_char: char }
,
199 /// The string was not terminated, e.g. `r###"abcde"##`.
200 /// `possible_terminator_offset` is the number of characters after `r` or
201 /// `br` where they may have intended to terminate it.
202 NoTerminator { expected: u32, found: u32, possible_terminator_offset: Option<u32> }
,
203 /// More than 255 `#`s exist.
204 TooManyDelimiters { found: u32 }
,
207 /// Base of numeric literal encoding according to its prefix.
208 #[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord)]
210 /// Literal starts with "0b".
212 /// Literal starts with "0o".
214 /// Literal doesn't contain a prefix.
216 /// Literal starts with "0x".
220 /// `rustc` allows files to have a shebang, e.g. "#!/usr/bin/rustrun",
221 /// but shebang isn't a part of rust syntax.
222 pub fn strip_shebang(input
: &str) -> Option
<usize> {
223 // Shebang must start with `#!` literally, without any preceding whitespace.
224 // For simplicity we consider any line starting with `#!` a shebang,
225 // regardless of restrictions put on shebangs by specific platforms.
226 if let Some(input_tail
) = input
.strip_prefix("#!") {
227 // Ok, this is a shebang but if the next non-whitespace token is `[`,
228 // then it may be valid Rust code, so consider it Rust code.
229 let next_non_whitespace_token
= tokenize(input_tail
).map(|tok
| tok
.kind
).find(|tok
| {
232 TokenKind
::Whitespace
233 | TokenKind
::LineComment { doc_style: None }
234 | TokenKind
::BlockComment { doc_style: None, .. }
237 if next_non_whitespace_token
!= Some(TokenKind
::OpenBracket
) {
238 // No other choice than to consider this a shebang.
239 return Some(2 + input_tail
.lines().next().unwrap_or_default().len());
245 /// Validates a raw string literal. Used for getting more information about a
246 /// problem with a `RawStr`/`RawByteStr` with a `None` field.
248 pub fn validate_raw_str(input
: &str, prefix_len
: u32) -> Result
<(), RawStrError
> {
249 debug_assert
!(!input
.is_empty());
250 let mut cursor
= Cursor
::new(input
);
251 // Move past the leading `r` or `br`.
252 for _
in 0..prefix_len
{
253 cursor
.bump().unwrap();
255 cursor
.raw_double_quoted_string(prefix_len
).map(|_
| ())
258 /// Creates an iterator that produces tokens from the input string.
259 pub fn tokenize(input
: &str) -> impl Iterator
<Item
= Token
> + '_
{
260 let mut cursor
= Cursor
::new(input
);
261 std
::iter
::from_fn(move || {
262 let token
= cursor
.advance_token();
263 if token
.kind
!= TokenKind
::Eof { Some(token) }
else { None }
267 /// True if `c` is considered a whitespace according to Rust language definition.
268 /// See [Rust language reference](https://doc.rust-lang.org/reference/whitespace.html)
269 /// for definitions of these classes.
270 pub fn is_whitespace(c
: char) -> bool
{
271 // This is Pattern_White_Space.
273 // Note that this set is stable (ie, it doesn't change with different
274 // Unicode versions), so it's ok to just hard-code the values.
278 // Usual ASCII suspects
281 | '
\u{000B}'
// vertical tab
282 | '
\u{000C}'
// form feed
284 | '
\u{0020}'
// space
286 // NEXT LINE from latin1
290 | '
\u{200E}'
// LEFT-TO-RIGHT MARK
291 | '
\u{200F}'
// RIGHT-TO-LEFT MARK
293 // Dedicated whitespace characters from Unicode
294 | '
\u{2028}'
// LINE SEPARATOR
295 | '
\u{2029}'
// PARAGRAPH SEPARATOR
299 /// True if `c` is valid as a first character of an identifier.
300 /// See [Rust language reference](https://doc.rust-lang.org/reference/identifiers.html) for
301 /// a formal definition of valid identifier name.
302 pub fn is_id_start(c
: char) -> bool
{
303 // This is XID_Start OR '_' (which formally is not a XID_Start).
304 c
== '_'
|| unicode_xid
::UnicodeXID
::is_xid_start(c
)
307 /// True if `c` is valid as a non-first character of an identifier.
308 /// See [Rust language reference](https://doc.rust-lang.org/reference/identifiers.html) for
309 /// a formal definition of valid identifier name.
310 pub fn is_id_continue(c
: char) -> bool
{
311 unicode_xid
::UnicodeXID
::is_xid_continue(c
)
314 /// The passed string is lexically an identifier.
315 pub fn is_ident(string
: &str) -> bool
{
316 let mut chars
= string
.chars();
317 if let Some(start
) = chars
.next() {
318 is_id_start(start
) && chars
.all(is_id_continue
)
325 /// Parses a token from the input string.
326 pub fn advance_token(&mut self) -> Token
{
327 let first_char
= match self.bump() {
329 None
=> return Token
::new(TokenKind
::Eof
, 0),
331 let token_kind
= match first_char
{
332 // Slash, comment or block comment.
333 '
/'
=> match self.first() {
334 '
/'
=> self.line_comment(),
335 '
*'
=> self.block_comment(),
339 // Whitespace sequence.
340 c
if is_whitespace(c
) => self.whitespace(),
342 // Raw identifier, raw string literal or identifier.
343 'r'
=> match (self.first(), self.second()) {
344 ('
#', c1) if is_id_start(c1) => self.raw_ident(),
345 ('
#', _) | ('"', _) => {
346 let res = self.raw_double_quoted_string(1);
347 let suffix_start = self.pos_within_token();
349 self.eat_literal_suffix();
351 let kind = RawStr { n_hashes: res.ok() };
352 Literal { kind, suffix_start }
354 _ => self.ident_or_unknown_prefix(),
357 // Byte literal, byte string literal, raw byte string literal or identifier.
358 'b' => match (self.first(), self.second()) {
361 let terminated = self.single_quoted_string();
362 let suffix_start = self.pos_within_token();
364 self.eat_literal_suffix();
366 let kind = Byte { terminated };
367 Literal { kind, suffix_start }
371 let terminated
= self.double_quoted_string();
372 let suffix_start
= self.pos_within_token();
374 self.eat_literal_suffix();
376 let kind
= ByteStr { terminated }
;
377 Literal { kind, suffix_start }
379 ('r'
, '
"') | ('r', '#') => {
381 let res = self.raw_double_quoted_string(2);
382 let suffix_start = self.pos_within_token();
384 self.eat_literal_suffix();
386 let kind = RawByteStr { n_hashes: res.ok() };
387 Literal { kind, suffix_start }
389 _ => self.ident_or_unknown_prefix(),
392 // Identifier (this should be checked after other variant that can
393 // start as identifier).
394 c if is_id_start(c) => self.ident_or_unknown_prefix(),
398 let literal_kind = self.number(c);
399 let suffix_start = self.pos_within_token();
400 self.eat_literal_suffix();
401 TokenKind::Literal { kind: literal_kind, suffix_start }
404 // One-symbol tokens.
432 // Lifetime or character literal.
433 '\'' => self.lifetime_or_char(),
437 let terminated
= self.double_quoted_string();
438 let suffix_start
= self.pos_within_token();
440 self.eat_literal_suffix();
442 let kind
= Str { terminated }
;
443 Literal { kind, suffix_start }
445 // Identifier starting with an emoji. Only lexed for graceful error recovery.
446 c
if !c
.is_ascii() && unic_emoji_char
::is_emoji(c
) => {
447 self.fake_ident_or_unknown_prefix()
451 let res
= Token
::new(token_kind
, self.pos_within_token());
452 self.reset_pos_within_token();
456 fn line_comment(&mut self) -> TokenKind
{
457 debug_assert
!(self.prev() == '
/'
&& self.first() == '
/'
);
460 let doc_style
= match self.first() {
461 // `//!` is an inner line doc comment.
462 '
!'
=> Some(DocStyle
::Inner
),
463 // `////` (more than 3 slashes) is not considered a doc comment.
464 '
/'
if self.second() != '
/'
=> Some(DocStyle
::Outer
),
468 self.eat_while(|c
| c
!= '
\n'
);
469 LineComment { doc_style }
472 fn block_comment(&mut self) -> TokenKind
{
473 debug_assert
!(self.prev() == '
/'
&& self.first() == '
*'
);
476 let doc_style
= match self.first() {
477 // `/*!` is an inner block doc comment.
478 '
!'
=> Some(DocStyle
::Inner
),
479 // `/***` (more than 2 stars) is not considered a doc comment.
480 // `/**/` is not considered a doc comment.
481 '
*'
if !matches
!(self.second(), '
*'
| '
/'
) => Some(DocStyle
::Outer
),
485 let mut depth
= 1usize
;
486 while let Some(c
) = self.bump() {
488 '
/'
if self.first() == '
*'
=> {
492 '
*'
if self.first() == '
/'
=> {
496 // This block comment is closed, so for a construction like "/* */ */"
497 // there will be a successfully parsed block comment "/* */"
498 // and " */" will be processed separately.
506 BlockComment { doc_style, terminated: depth == 0 }
509 fn whitespace(&mut self) -> TokenKind
{
510 debug_assert
!(is_whitespace(self.prev()));
511 self.eat_while(is_whitespace
);
515 fn raw_ident(&mut self) -> TokenKind
{
516 debug_assert
!(self.prev() == 'r'
&& self.first() == '
#' && is_id_start(self.second()));
519 // Eat the identifier part of RawIdent.
520 self.eat_identifier();
524 fn ident_or_unknown_prefix(&mut self) -> TokenKind
{
525 debug_assert
!(is_id_start(self.prev()));
526 // Start is already eaten, eat the rest of identifier.
527 self.eat_while(is_id_continue
);
528 // Known prefixes must have been handled earlier. So if
529 // we see a prefix here, it is definitely an unknown prefix.
531 '
#' | '"' | '\'' => UnknownPrefix,
532 c if !c.is_ascii() && unic_emoji_char::is_emoji(c) => {
533 self.fake_ident_or_unknown_prefix()
539 fn fake_ident_or_unknown_prefix(&mut self) -> TokenKind {
540 // Start is already eaten, eat the rest of identifier.
542 unicode_xid::UnicodeXID::is_xid_continue(c)
543 || (!c.is_ascii() && unic_emoji_char::is_emoji(c))
546 // Known prefixes must have been handled earlier. So if
547 // we see a prefix here, it is definitely an unknown prefix.
549 '#' | '"' | '\'' => UnknownPrefix,
554 fn number(&mut self, first_digit
: char) -> LiteralKind
{
555 debug_assert
!('
0'
<= self.prev() && self.prev() <= '
9'
);
556 let mut base
= Base
::Decimal
;
557 if first_digit
== '
0'
{
558 // Attempt to parse encoding base.
559 let has_digits
= match self.first() {
563 self.eat_decimal_digits()
568 self.eat_decimal_digits()
571 base
= Base
::Hexadecimal
;
573 self.eat_hexadecimal_digits()
575 // Not a base prefix.
576 '
0'
..='
9'
| '_'
| '
.'
| 'e'
| 'E'
=> {
577 self.eat_decimal_digits();
581 _
=> return Int { base, empty_int: false }
,
583 // Base prefix was provided, but there were no digits
584 // after it, e.g. "0x".
586 return Int { base, empty_int: true }
;
589 // No base prefix, parse number in the usual way.
590 self.eat_decimal_digits();
594 // Don't be greedy if this is actually an
595 // integer literal followed by field/method access or a range pattern
596 // (`0..2` and `12.foo()`)
597 '
.'
if self.second() != '
.'
&& !is_id_start(self.second()) => {
598 // might have stuff after the ., and if it does, it needs to start
601 let mut empty_exponent
= false;
602 if self.first().is_digit(10) {
603 self.eat_decimal_digits();
607 empty_exponent
= !self.eat_float_exponent();
612 Float { base, empty_exponent }
616 let empty_exponent
= !self.eat_float_exponent();
617 Float { base, empty_exponent }
619 _
=> Int { base, empty_int: false }
,
623 fn lifetime_or_char(&mut self) -> TokenKind
{
624 debug_assert
!(self.prev() == '
\''
);
626 let can_be_a_lifetime
= if self.second() == '
\''
{
627 // It's surely not a lifetime.
630 // If the first symbol is valid for identifier, it can be a lifetime.
631 // Also check if it's a number for a better error reporting (so '0 will
632 // be reported as invalid lifetime and not as unterminated char literal).
633 is_id_start(self.first()) || self.first().is_digit(10)
636 if !can_be_a_lifetime
{
637 let terminated
= self.single_quoted_string();
638 let suffix_start
= self.pos_within_token();
640 self.eat_literal_suffix();
642 let kind
= Char { terminated }
;
643 return Literal { kind, suffix_start }
;
646 // Either a lifetime or a character literal with
647 // length greater than 1.
649 let starts_with_number
= self.first().is_digit(10);
651 // Skip the literal contents.
652 // First symbol can be a number (which isn't a valid identifier start),
653 // so skip it without any checks.
655 self.eat_while(is_id_continue
);
657 // Check if after skipping literal contents we've met a closing
658 // single quote (which means that user attempted to create a
659 // string with single quotes).
660 if self.first() == '
\''
{
662 let kind
= Char { terminated: true }
;
663 Literal { kind, suffix_start: self.pos_within_token() }
665 Lifetime { starts_with_number }
669 fn single_quoted_string(&mut self) -> bool
{
670 debug_assert
!(self.prev() == '
\''
);
671 // Check if it's a one-symbol literal.
672 if self.second() == '
\''
&& self.first() != '
\\'
{
678 // Literal has more than one symbol.
680 // Parse until either quotes are terminated or error is detected.
683 // Quotes are terminated, finish parsing.
688 // Probably beginning of the comment, which we don't want to include
689 // to the error report.
691 // Newline without following '\'' means unclosed quote, stop parsing.
692 '
\n'
if self.second() != '
\''
=> break,
693 // End of file, stop parsing.
694 EOF_CHAR
if self.is_eof() => break,
695 // Escaped slash is considered one character, so bump twice.
700 // Skip the character.
706 // String was not terminated.
710 /// Eats double-quoted string and returns true
711 /// if string is terminated.
712 fn double_quoted_string(&mut self) -> bool
{
713 debug_assert
!(self.prev() == '
"');
714 while let Some(c) = self.bump() {
719 '
\\'
if self.first() == '
\\'
|| self.first() == '
"' => {
720 // Bump again to skip escaped character.
726 // End of file reached.
730 /// Eats the double-quoted string and returns `n_hashes` and an error if encountered.
731 fn raw_double_quoted_string(&mut self, prefix_len: u32) -> Result<u8, RawStrError> {
732 // Wrap the actual function to handle the error with too many hashes.
733 // This way, it eats the whole raw string.
734 let n_hashes = self.raw_string_unvalidated(prefix_len)?;
735 // Only up to 255 `#`s are allowed in raw strings
736 match u8::try_from(n_hashes) {
738 Err(_) => Err(RawStrError::TooManyDelimiters { found: n_hashes }),
742 fn raw_string_unvalidated(&mut self, prefix_len: u32) -> Result<u32, RawStrError> {
743 debug_assert!(self.prev() == 'r');
744 let start_pos = self.pos_within_token();
745 let mut possible_terminator_offset = None;
746 let mut max_hashes = 0;
748 // Count opening '#' symbols.
750 while self.first() == '#' {
754 let n_start_hashes = eaten;
756 // Check that string is started.
760 let c
= c
.unwrap_or(EOF_CHAR
);
761 return Err(RawStrError
::InvalidStarter { bad_char: c }
);
765 // Skip the string contents and on each '#' character met, check if this is
766 // a raw string termination.
768 self.eat_while(|c
| c
!= '
"');
771 return Err(RawStrError::NoTerminator {
772 expected: n_start_hashes,
774 possible_terminator_offset,
778 // Eat closing double quote.
781 // Check that amount of closing '#' symbols
782 // is equal to the amount of opening ones.
783 // Note that this will not consume extra trailing `#` characters:
784 // `r###"abcde
"####` is lexed as a `RawStr { n_hashes: 3 }`
785 // followed by a `#` token.
786 let mut n_end_hashes = 0;
787 while self.first() == '#' && n_end_hashes < n_start_hashes {
792 if n_end_hashes == n_start_hashes {
793 return Ok(n_start_hashes);
794 } else if n_end_hashes > max_hashes {
795 // Keep track of possible terminators to give a hint about
796 // where there might be a missing terminator
797 possible_terminator_offset =
798 Some(self.pos_within_token() - start_pos - n_end_hashes + prefix_len);
799 max_hashes = n_end_hashes;
804 fn eat_decimal_digits(&mut self) -> bool {
805 let mut has_digits = false;
821 fn eat_hexadecimal_digits(&mut self) -> bool {
822 let mut has_digits = false;
828 '0'..='9' | 'a'..='f' | 'A'..='F' => {
838 /// Eats the float exponent. Returns true if at least one digit was met,
839 /// and returns false otherwise.
840 fn eat_float_exponent(&mut self) -> bool {
841 debug_assert!(self.prev() == 'e' || self.prev() == 'E');
842 if self.first() == '-' || self.first() == '+' {
845 self.eat_decimal_digits()
848 // Eats the suffix of the literal, e.g. "u8".
849 fn eat_literal_suffix(&mut self) {
850 self.eat_identifier();
853 // Eats the identifier. Note: succeeds on `_`, which isn't a valid
855 fn eat_identifier(&mut self) {
856 if !is_id_start(self.first()) {
861 self.eat_while(is_id_continue);