Rollup merge of #76050 - matklad:pos, r=petrochenkov
Remove unused function
This commit is contained in:
commit
7d2460e5c4
@ -39,11 +39,6 @@
|
|||||||
mod parse {
|
mod parse {
|
||||||
#[cfg(test)]
|
#[cfg(test)]
|
||||||
mod tests;
|
mod tests;
|
||||||
#[cfg(test)]
|
|
||||||
mod lexer {
|
|
||||||
#[cfg(test)]
|
|
||||||
mod tests;
|
|
||||||
}
|
|
||||||
}
|
}
|
||||||
#[cfg(test)]
|
#[cfg(test)]
|
||||||
mod tokenstream {
|
mod tokenstream {
|
||||||
|
@ -1,252 +0,0 @@
|
|||||||
use rustc_ast::ast::AttrStyle;
|
|
||||||
use rustc_ast::token::{self, CommentKind, Token, TokenKind};
|
|
||||||
use rustc_data_structures::sync::Lrc;
|
|
||||||
use rustc_errors::{emitter::EmitterWriter, Handler};
|
|
||||||
use rustc_parse::lexer::StringReader;
|
|
||||||
use rustc_session::parse::ParseSess;
|
|
||||||
use rustc_span::source_map::{FilePathMapping, SourceMap};
|
|
||||||
use rustc_span::symbol::Symbol;
|
|
||||||
use rustc_span::with_default_session_globals;
|
|
||||||
use rustc_span::{BytePos, Span};
|
|
||||||
|
|
||||||
use std::io;
|
|
||||||
use std::path::PathBuf;
|
|
||||||
|
|
||||||
fn mk_sess(sm: Lrc<SourceMap>) -> ParseSess {
|
|
||||||
let emitter = EmitterWriter::new(
|
|
||||||
Box::new(io::sink()),
|
|
||||||
Some(sm.clone()),
|
|
||||||
false,
|
|
||||||
false,
|
|
||||||
false,
|
|
||||||
None,
|
|
||||||
false,
|
|
||||||
);
|
|
||||||
ParseSess::with_span_handler(Handler::with_emitter(true, None, Box::new(emitter)), sm)
|
|
||||||
}
|
|
||||||
|
|
||||||
// Creates a string reader for the given string.
|
|
||||||
fn setup<'a>(sm: &SourceMap, sess: &'a ParseSess, teststr: String) -> StringReader<'a> {
|
|
||||||
let sf = sm.new_source_file(PathBuf::from(teststr.clone()).into(), teststr);
|
|
||||||
StringReader::new(sess, sf, None)
|
|
||||||
}
|
|
||||||
|
|
||||||
#[test]
|
|
||||||
fn t1() {
|
|
||||||
with_default_session_globals(|| {
|
|
||||||
let sm = Lrc::new(SourceMap::new(FilePathMapping::empty()));
|
|
||||||
let sh = mk_sess(sm.clone());
|
|
||||||
let mut string_reader = setup(
|
|
||||||
&sm,
|
|
||||||
&sh,
|
|
||||||
"/* my source file */ fn main() { println!(\"zebra\"); }\n".to_string(),
|
|
||||||
);
|
|
||||||
assert_eq!(string_reader.next_token(), token::Comment);
|
|
||||||
assert_eq!(string_reader.next_token(), token::Whitespace);
|
|
||||||
let tok1 = string_reader.next_token();
|
|
||||||
let tok2 = Token::new(mk_ident("fn"), Span::with_root_ctxt(BytePos(21), BytePos(23)));
|
|
||||||
assert_eq!(tok1.kind, tok2.kind);
|
|
||||||
assert_eq!(tok1.span, tok2.span);
|
|
||||||
assert_eq!(string_reader.next_token(), token::Whitespace);
|
|
||||||
// Read another token.
|
|
||||||
let tok3 = string_reader.next_token();
|
|
||||||
assert_eq!(string_reader.pos(), BytePos(28));
|
|
||||||
let tok4 = Token::new(mk_ident("main"), Span::with_root_ctxt(BytePos(24), BytePos(28)));
|
|
||||||
assert_eq!(tok3.kind, tok4.kind);
|
|
||||||
assert_eq!(tok3.span, tok4.span);
|
|
||||||
|
|
||||||
assert_eq!(string_reader.next_token(), token::OpenDelim(token::Paren));
|
|
||||||
assert_eq!(string_reader.pos(), BytePos(29))
|
|
||||||
})
|
|
||||||
}
|
|
||||||
|
|
||||||
// Checks that the given reader produces the desired stream
|
|
||||||
// of tokens (stop checking after exhausting `expected`).
|
|
||||||
fn check_tokenization(mut string_reader: StringReader<'_>, expected: Vec<TokenKind>) {
|
|
||||||
for expected_tok in &expected {
|
|
||||||
assert_eq!(&string_reader.next_token(), expected_tok);
|
|
||||||
}
|
|
||||||
}
|
|
||||||
|
|
||||||
// Makes the identifier by looking up the string in the interner.
|
|
||||||
fn mk_ident(id: &str) -> TokenKind {
|
|
||||||
token::Ident(Symbol::intern(id), false)
|
|
||||||
}
|
|
||||||
|
|
||||||
fn mk_lit(kind: token::LitKind, symbol: &str, suffix: Option<&str>) -> TokenKind {
|
|
||||||
TokenKind::lit(kind, Symbol::intern(symbol), suffix.map(Symbol::intern))
|
|
||||||
}
|
|
||||||
|
|
||||||
#[test]
|
|
||||||
fn doublecolon_parsing() {
|
|
||||||
with_default_session_globals(|| {
|
|
||||||
let sm = Lrc::new(SourceMap::new(FilePathMapping::empty()));
|
|
||||||
let sh = mk_sess(sm.clone());
|
|
||||||
check_tokenization(
|
|
||||||
setup(&sm, &sh, "a b".to_string()),
|
|
||||||
vec![mk_ident("a"), token::Whitespace, mk_ident("b")],
|
|
||||||
);
|
|
||||||
})
|
|
||||||
}
|
|
||||||
|
|
||||||
#[test]
|
|
||||||
fn doublecolon_parsing_2() {
|
|
||||||
with_default_session_globals(|| {
|
|
||||||
let sm = Lrc::new(SourceMap::new(FilePathMapping::empty()));
|
|
||||||
let sh = mk_sess(sm.clone());
|
|
||||||
check_tokenization(
|
|
||||||
setup(&sm, &sh, "a::b".to_string()),
|
|
||||||
vec![mk_ident("a"), token::Colon, token::Colon, mk_ident("b")],
|
|
||||||
);
|
|
||||||
})
|
|
||||||
}
|
|
||||||
|
|
||||||
#[test]
|
|
||||||
fn doublecolon_parsing_3() {
|
|
||||||
with_default_session_globals(|| {
|
|
||||||
let sm = Lrc::new(SourceMap::new(FilePathMapping::empty()));
|
|
||||||
let sh = mk_sess(sm.clone());
|
|
||||||
check_tokenization(
|
|
||||||
setup(&sm, &sh, "a ::b".to_string()),
|
|
||||||
vec![mk_ident("a"), token::Whitespace, token::Colon, token::Colon, mk_ident("b")],
|
|
||||||
);
|
|
||||||
})
|
|
||||||
}
|
|
||||||
|
|
||||||
#[test]
|
|
||||||
fn doublecolon_parsing_4() {
|
|
||||||
with_default_session_globals(|| {
|
|
||||||
let sm = Lrc::new(SourceMap::new(FilePathMapping::empty()));
|
|
||||||
let sh = mk_sess(sm.clone());
|
|
||||||
check_tokenization(
|
|
||||||
setup(&sm, &sh, "a:: b".to_string()),
|
|
||||||
vec![mk_ident("a"), token::Colon, token::Colon, token::Whitespace, mk_ident("b")],
|
|
||||||
);
|
|
||||||
})
|
|
||||||
}
|
|
||||||
|
|
||||||
#[test]
|
|
||||||
fn character_a() {
|
|
||||||
with_default_session_globals(|| {
|
|
||||||
let sm = Lrc::new(SourceMap::new(FilePathMapping::empty()));
|
|
||||||
let sh = mk_sess(sm.clone());
|
|
||||||
assert_eq!(setup(&sm, &sh, "'a'".to_string()).next_token(), mk_lit(token::Char, "a", None),);
|
|
||||||
})
|
|
||||||
}
|
|
||||||
|
|
||||||
#[test]
|
|
||||||
fn character_space() {
|
|
||||||
with_default_session_globals(|| {
|
|
||||||
let sm = Lrc::new(SourceMap::new(FilePathMapping::empty()));
|
|
||||||
let sh = mk_sess(sm.clone());
|
|
||||||
assert_eq!(setup(&sm, &sh, "' '".to_string()).next_token(), mk_lit(token::Char, " ", None),);
|
|
||||||
})
|
|
||||||
}
|
|
||||||
|
|
||||||
#[test]
|
|
||||||
fn character_escaped() {
|
|
||||||
with_default_session_globals(|| {
|
|
||||||
let sm = Lrc::new(SourceMap::new(FilePathMapping::empty()));
|
|
||||||
let sh = mk_sess(sm.clone());
|
|
||||||
assert_eq!(
|
|
||||||
setup(&sm, &sh, "'\\n'".to_string()).next_token(),
|
|
||||||
mk_lit(token::Char, "\\n", None),
|
|
||||||
);
|
|
||||||
})
|
|
||||||
}
|
|
||||||
|
|
||||||
#[test]
|
|
||||||
fn lifetime_name() {
|
|
||||||
with_default_session_globals(|| {
|
|
||||||
let sm = Lrc::new(SourceMap::new(FilePathMapping::empty()));
|
|
||||||
let sh = mk_sess(sm.clone());
|
|
||||||
assert_eq!(
|
|
||||||
setup(&sm, &sh, "'abc".to_string()).next_token(),
|
|
||||||
token::Lifetime(Symbol::intern("'abc")),
|
|
||||||
);
|
|
||||||
})
|
|
||||||
}
|
|
||||||
|
|
||||||
#[test]
|
|
||||||
fn raw_string() {
|
|
||||||
with_default_session_globals(|| {
|
|
||||||
let sm = Lrc::new(SourceMap::new(FilePathMapping::empty()));
|
|
||||||
let sh = mk_sess(sm.clone());
|
|
||||||
assert_eq!(
|
|
||||||
setup(&sm, &sh, "r###\"\"#a\\b\x00c\"\"###".to_string()).next_token(),
|
|
||||||
mk_lit(token::StrRaw(3), "\"#a\\b\x00c\"", None),
|
|
||||||
);
|
|
||||||
})
|
|
||||||
}
|
|
||||||
|
|
||||||
#[test]
|
|
||||||
fn literal_suffixes() {
|
|
||||||
with_default_session_globals(|| {
|
|
||||||
let sm = Lrc::new(SourceMap::new(FilePathMapping::empty()));
|
|
||||||
let sh = mk_sess(sm.clone());
|
|
||||||
macro_rules! test {
|
|
||||||
($input: expr, $tok_type: ident, $tok_contents: expr) => {{
|
|
||||||
assert_eq!(
|
|
||||||
setup(&sm, &sh, format!("{}suffix", $input)).next_token(),
|
|
||||||
mk_lit(token::$tok_type, $tok_contents, Some("suffix")),
|
|
||||||
);
|
|
||||||
// with a whitespace separator
|
|
||||||
assert_eq!(
|
|
||||||
setup(&sm, &sh, format!("{} suffix", $input)).next_token(),
|
|
||||||
mk_lit(token::$tok_type, $tok_contents, None),
|
|
||||||
);
|
|
||||||
}};
|
|
||||||
}
|
|
||||||
|
|
||||||
test!("'a'", Char, "a");
|
|
||||||
test!("b'a'", Byte, "a");
|
|
||||||
test!("\"a\"", Str, "a");
|
|
||||||
test!("b\"a\"", ByteStr, "a");
|
|
||||||
test!("1234", Integer, "1234");
|
|
||||||
test!("0b101", Integer, "0b101");
|
|
||||||
test!("0xABC", Integer, "0xABC");
|
|
||||||
test!("1.0", Float, "1.0");
|
|
||||||
test!("1.0e10", Float, "1.0e10");
|
|
||||||
|
|
||||||
assert_eq!(
|
|
||||||
setup(&sm, &sh, "2us".to_string()).next_token(),
|
|
||||||
mk_lit(token::Integer, "2", Some("us")),
|
|
||||||
);
|
|
||||||
assert_eq!(
|
|
||||||
setup(&sm, &sh, "r###\"raw\"###suffix".to_string()).next_token(),
|
|
||||||
mk_lit(token::StrRaw(3), "raw", Some("suffix")),
|
|
||||||
);
|
|
||||||
assert_eq!(
|
|
||||||
setup(&sm, &sh, "br###\"raw\"###suffix".to_string()).next_token(),
|
|
||||||
mk_lit(token::ByteStrRaw(3), "raw", Some("suffix")),
|
|
||||||
);
|
|
||||||
})
|
|
||||||
}
|
|
||||||
|
|
||||||
#[test]
|
|
||||||
fn nested_block_comments() {
|
|
||||||
with_default_session_globals(|| {
|
|
||||||
let sm = Lrc::new(SourceMap::new(FilePathMapping::empty()));
|
|
||||||
let sh = mk_sess(sm.clone());
|
|
||||||
let mut lexer = setup(&sm, &sh, "/* /* */ */'a'".to_string());
|
|
||||||
assert_eq!(lexer.next_token(), token::Comment);
|
|
||||||
assert_eq!(lexer.next_token(), mk_lit(token::Char, "a", None));
|
|
||||||
})
|
|
||||||
}
|
|
||||||
|
|
||||||
#[test]
|
|
||||||
fn crlf_comments() {
|
|
||||||
with_default_session_globals(|| {
|
|
||||||
let sm = Lrc::new(SourceMap::new(FilePathMapping::empty()));
|
|
||||||
let sh = mk_sess(sm.clone());
|
|
||||||
let mut lexer = setup(&sm, &sh, "// test\r\n/// test\r\n".to_string());
|
|
||||||
let comment = lexer.next_token();
|
|
||||||
assert_eq!(comment.kind, token::Comment);
|
|
||||||
assert_eq!((comment.span.lo(), comment.span.hi()), (BytePos(0), BytePos(7)));
|
|
||||||
assert_eq!(lexer.next_token(), token::Whitespace);
|
|
||||||
assert_eq!(
|
|
||||||
lexer.next_token(),
|
|
||||||
token::DocComment(CommentKind::Line, AttrStyle::Outer, Symbol::intern(" test"))
|
|
||||||
);
|
|
||||||
})
|
|
||||||
}
|
|
@ -128,6 +128,34 @@ fn check_lexing(src: &str, expect: Expect) {
|
|||||||
expect.assert_eq(&actual)
|
expect.assert_eq(&actual)
|
||||||
}
|
}
|
||||||
|
|
||||||
|
#[test]
|
||||||
|
fn smoke_test() {
|
||||||
|
check_lexing(
|
||||||
|
"/* my source file */ fn main() { println!(\"zebra\"); }\n",
|
||||||
|
expect![[r#"
|
||||||
|
Token { kind: BlockComment { doc_style: None, terminated: true }, len: 20 }
|
||||||
|
Token { kind: Whitespace, len: 1 }
|
||||||
|
Token { kind: Ident, len: 2 }
|
||||||
|
Token { kind: Whitespace, len: 1 }
|
||||||
|
Token { kind: Ident, len: 4 }
|
||||||
|
Token { kind: OpenParen, len: 1 }
|
||||||
|
Token { kind: CloseParen, len: 1 }
|
||||||
|
Token { kind: Whitespace, len: 1 }
|
||||||
|
Token { kind: OpenBrace, len: 1 }
|
||||||
|
Token { kind: Whitespace, len: 1 }
|
||||||
|
Token { kind: Ident, len: 7 }
|
||||||
|
Token { kind: Bang, len: 1 }
|
||||||
|
Token { kind: OpenParen, len: 1 }
|
||||||
|
Token { kind: Literal { kind: Str { terminated: true }, suffix_start: 7 }, len: 7 }
|
||||||
|
Token { kind: CloseParen, len: 1 }
|
||||||
|
Token { kind: Semi, len: 1 }
|
||||||
|
Token { kind: Whitespace, len: 1 }
|
||||||
|
Token { kind: CloseBrace, len: 1 }
|
||||||
|
Token { kind: Whitespace, len: 1 }
|
||||||
|
"#]],
|
||||||
|
)
|
||||||
|
}
|
||||||
|
|
||||||
#[test]
|
#[test]
|
||||||
fn comment_flavors() {
|
fn comment_flavors() {
|
||||||
check_lexing(
|
check_lexing(
|
||||||
@ -143,25 +171,117 @@ fn comment_flavors() {
|
|||||||
/*! inner doc block */
|
/*! inner doc block */
|
||||||
",
|
",
|
||||||
expect![[r#"
|
expect![[r#"
|
||||||
Token { kind: Whitespace, len: 1 }
|
Token { kind: Whitespace, len: 1 }
|
||||||
Token { kind: LineComment { doc_style: None }, len: 7 }
|
Token { kind: LineComment { doc_style: None }, len: 7 }
|
||||||
Token { kind: Whitespace, len: 1 }
|
Token { kind: Whitespace, len: 1 }
|
||||||
Token { kind: LineComment { doc_style: None }, len: 17 }
|
Token { kind: LineComment { doc_style: None }, len: 17 }
|
||||||
Token { kind: Whitespace, len: 1 }
|
Token { kind: Whitespace, len: 1 }
|
||||||
Token { kind: LineComment { doc_style: Some(Outer) }, len: 18 }
|
Token { kind: LineComment { doc_style: Some(Outer) }, len: 18 }
|
||||||
Token { kind: Whitespace, len: 1 }
|
Token { kind: Whitespace, len: 1 }
|
||||||
Token { kind: LineComment { doc_style: Some(Inner) }, len: 18 }
|
Token { kind: LineComment { doc_style: Some(Inner) }, len: 18 }
|
||||||
Token { kind: Whitespace, len: 1 }
|
Token { kind: Whitespace, len: 1 }
|
||||||
Token { kind: BlockComment { doc_style: None, terminated: true }, len: 11 }
|
Token { kind: BlockComment { doc_style: None, terminated: true }, len: 11 }
|
||||||
Token { kind: Whitespace, len: 1 }
|
Token { kind: Whitespace, len: 1 }
|
||||||
Token { kind: BlockComment { doc_style: None, terminated: true }, len: 4 }
|
Token { kind: BlockComment { doc_style: None, terminated: true }, len: 4 }
|
||||||
Token { kind: Whitespace, len: 1 }
|
Token { kind: Whitespace, len: 1 }
|
||||||
Token { kind: BlockComment { doc_style: None, terminated: true }, len: 18 }
|
Token { kind: BlockComment { doc_style: None, terminated: true }, len: 18 }
|
||||||
Token { kind: Whitespace, len: 1 }
|
Token { kind: Whitespace, len: 1 }
|
||||||
Token { kind: BlockComment { doc_style: Some(Outer), terminated: true }, len: 22 }
|
Token { kind: BlockComment { doc_style: Some(Outer), terminated: true }, len: 22 }
|
||||||
Token { kind: Whitespace, len: 1 }
|
Token { kind: Whitespace, len: 1 }
|
||||||
Token { kind: BlockComment { doc_style: Some(Inner), terminated: true }, len: 22 }
|
Token { kind: BlockComment { doc_style: Some(Inner), terminated: true }, len: 22 }
|
||||||
Token { kind: Whitespace, len: 1 }
|
Token { kind: Whitespace, len: 1 }
|
||||||
"#]],
|
"#]],
|
||||||
|
)
|
||||||
|
}
|
||||||
|
|
||||||
|
#[test]
|
||||||
|
fn nested_block_comments() {
|
||||||
|
check_lexing(
|
||||||
|
"/* /* */ */'a'",
|
||||||
|
expect![[r#"
|
||||||
|
Token { kind: BlockComment { doc_style: None, terminated: true }, len: 11 }
|
||||||
|
Token { kind: Literal { kind: Char { terminated: true }, suffix_start: 3 }, len: 3 }
|
||||||
|
"#]],
|
||||||
|
)
|
||||||
|
}
|
||||||
|
|
||||||
|
#[test]
|
||||||
|
fn characters() {
|
||||||
|
check_lexing(
|
||||||
|
"'a' ' ' '\\n'",
|
||||||
|
expect![[r#"
|
||||||
|
Token { kind: Literal { kind: Char { terminated: true }, suffix_start: 3 }, len: 3 }
|
||||||
|
Token { kind: Whitespace, len: 1 }
|
||||||
|
Token { kind: Literal { kind: Char { terminated: true }, suffix_start: 3 }, len: 3 }
|
||||||
|
Token { kind: Whitespace, len: 1 }
|
||||||
|
Token { kind: Literal { kind: Char { terminated: true }, suffix_start: 4 }, len: 4 }
|
||||||
|
"#]],
|
||||||
|
);
|
||||||
|
}
|
||||||
|
|
||||||
|
#[test]
|
||||||
|
fn lifetime() {
|
||||||
|
check_lexing(
|
||||||
|
"'abc",
|
||||||
|
expect![[r#"
|
||||||
|
Token { kind: Lifetime { starts_with_number: false }, len: 4 }
|
||||||
|
"#]],
|
||||||
|
);
|
||||||
|
}
|
||||||
|
|
||||||
|
#[test]
|
||||||
|
fn raw_string() {
|
||||||
|
check_lexing(
|
||||||
|
"r###\"\"#a\\b\x00c\"\"###",
|
||||||
|
expect![[r#"
|
||||||
|
Token { kind: Literal { kind: RawStr { n_hashes: 3, err: None }, suffix_start: 17 }, len: 17 }
|
||||||
|
"#]],
|
||||||
|
)
|
||||||
|
}
|
||||||
|
|
||||||
|
#[test]
|
||||||
|
fn literal_suffixes() {
|
||||||
|
check_lexing(
|
||||||
|
r####"
|
||||||
|
'a'
|
||||||
|
b'a'
|
||||||
|
"a"
|
||||||
|
b"a"
|
||||||
|
1234
|
||||||
|
0b101
|
||||||
|
0xABC
|
||||||
|
1.0
|
||||||
|
1.0e10
|
||||||
|
2us
|
||||||
|
r###"raw"###suffix
|
||||||
|
br###"raw"###suffix
|
||||||
|
"####,
|
||||||
|
expect![[r#"
|
||||||
|
Token { kind: Whitespace, len: 1 }
|
||||||
|
Token { kind: Literal { kind: Char { terminated: true }, suffix_start: 3 }, len: 3 }
|
||||||
|
Token { kind: Whitespace, len: 1 }
|
||||||
|
Token { kind: Literal { kind: Byte { terminated: true }, suffix_start: 4 }, len: 4 }
|
||||||
|
Token { kind: Whitespace, len: 1 }
|
||||||
|
Token { kind: Literal { kind: Str { terminated: true }, suffix_start: 3 }, len: 3 }
|
||||||
|
Token { kind: Whitespace, len: 1 }
|
||||||
|
Token { kind: Literal { kind: ByteStr { terminated: true }, suffix_start: 4 }, len: 4 }
|
||||||
|
Token { kind: Whitespace, len: 1 }
|
||||||
|
Token { kind: Literal { kind: Int { base: Decimal, empty_int: false }, suffix_start: 4 }, len: 4 }
|
||||||
|
Token { kind: Whitespace, len: 1 }
|
||||||
|
Token { kind: Literal { kind: Int { base: Binary, empty_int: false }, suffix_start: 5 }, len: 5 }
|
||||||
|
Token { kind: Whitespace, len: 1 }
|
||||||
|
Token { kind: Literal { kind: Int { base: Hexadecimal, empty_int: false }, suffix_start: 5 }, len: 5 }
|
||||||
|
Token { kind: Whitespace, len: 1 }
|
||||||
|
Token { kind: Literal { kind: Float { base: Decimal, empty_exponent: false }, suffix_start: 3 }, len: 3 }
|
||||||
|
Token { kind: Whitespace, len: 1 }
|
||||||
|
Token { kind: Literal { kind: Float { base: Decimal, empty_exponent: false }, suffix_start: 6 }, len: 6 }
|
||||||
|
Token { kind: Whitespace, len: 1 }
|
||||||
|
Token { kind: Literal { kind: Int { base: Decimal, empty_int: false }, suffix_start: 1 }, len: 3 }
|
||||||
|
Token { kind: Whitespace, len: 1 }
|
||||||
|
Token { kind: Literal { kind: RawStr { n_hashes: 3, err: None }, suffix_start: 12 }, len: 18 }
|
||||||
|
Token { kind: Whitespace, len: 1 }
|
||||||
|
Token { kind: Literal { kind: RawByteStr { n_hashes: 3, err: None }, suffix_start: 13 }, len: 19 }
|
||||||
|
Token { kind: Whitespace, len: 1 }
|
||||||
|
"#]],
|
||||||
)
|
)
|
||||||
}
|
}
|
||||||
|
@ -439,10 +439,6 @@ fn cook_lexer_literal(
|
|||||||
(lit_kind, id)
|
(lit_kind, id)
|
||||||
}
|
}
|
||||||
|
|
||||||
pub fn pos(&self) -> BytePos {
|
|
||||||
self.pos
|
|
||||||
}
|
|
||||||
|
|
||||||
#[inline]
|
#[inline]
|
||||||
fn src_index(&self, pos: BytePos) -> usize {
|
fn src_index(&self, pos: BytePos) -> usize {
|
||||||
(pos - self.start_pos).to_usize()
|
(pos - self.start_pos).to_usize()
|
||||||
|
Loading…
Reference in New Issue
Block a user