fix(rust) reuse the string rules inside attributes - #4461
Conversation
The `meta` mode for Rust attributes declared its own minimal string rule, so only plain `"..."` literals were recognised inside `#[...]`. Raw strings and byte strings were mis-highlighted: in `#[doc = r#"a "quoted" word"#]` the literal was cut at the first inner quote, leaving the rest of the attribute wrongly scoped. Attributes hold arbitrary token trees, so any string literal that is valid elsewhere is valid there too. Extract the existing string modes into named constants and reuse them in `meta` instead of duplicating a weaker version. The character-literal mode is deliberately left out of `meta` so that lifetimes in attribute token trees (`#[foo(Bar<'a>)]`) are not treated as an unterminated char literal. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
| "assert_ne!", | ||
| "debug_assert_ne!" | ||
| ]; | ||
| const QUOTE_STRING = hljs.inherit(hljs.QUOTE_STRING_MODE, { |
There was a problem hiding this comment.
What are we still getting from QUOTE_STRING_MODE? escapes?
There was a problem hiding this comment.
Yes — the end: /"/ and the BACKSLASH_ESCAPE handling. illegal: null clears QUOTE_STRING_MODE's illegal: '\n' so multiline strings still highlight. (This rule is unchanged from before — the PR just extracts it into a named constant so the attribute rule can reuse it.)
| }); | ||
| const RAW_STRING = { | ||
| scope: 'string', | ||
| begin: /b?r(#*)"(.|\n)*?"\1(?!#)/ |
There was a problem hiding this comment.
Can the run of # really be infinite?
There was a problem hiding this comment.
If so lets add a ################ sample to the specs, just cause. If not lets add the max size (if it's reasonable) and tighten this rule.
There was a problem hiding this comment.
No — rustc caps raw string delimiters at 255 # symbols ("too many # symbols: raw strings may be delimited by up to 255 # symbols"). Tightened the rule to #{0,255} and added a many-hash sample to the specs. Verified the boundary: 255 hashes highlights as one raw string, while 256 hashes (invalid Rust) no longer matches the raw-string rule.
rustc rejects raw string literals delimited by more than 255 # symbols,
so tighten the rule from #* to #{0,255} and add a many-hash spec sample.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
3 participants
Resolves #3817
Changes
The
metamode for Rust attributes declared its own minimal string rule ("…"+BACKSLASH_ESCAPE), so only plain double-quoted literals were recognised inside#[...]. Raw strings and byte strings were mis-highlighted:#[doc = r#"a "quoted" word"#]was cut at the first inner quote, so
quotedfell out of the string and the rest of the attribute was wrongly scoped.Attributes hold arbitrary token trees, so any string literal that is valid elsewhere is valid there too — which answers the question raised in the issue. This extracts the existing string modes into named constants (
QUOTE_STRING,RAW_STRING,CHARACTER) and reusesQUOTE_STRING/RAW_STRINGinsidemetainstead of duplicating a weaker version. The top-levelcontainsorder is unchanged.The character-literal mode is deliberately not added to
meta, so lifetimes in attribute token trees (#[foo(Bar<'a>)]) are not treated as an unterminated char literal.Note on the original report: the exact snippet in the issue (
#[error("\" appears in a string")]) has been highlighted correctly since 11.10.0, but the underlying cause it pointed at —metanot reusing the string rules — was still present. Before/after for the remaining cases:#[doc = r#"a "quoted" word"#]r#outside the string, literal split in twor#"…"#is one string#[doc = b"bytes"]boutside the stringb"bytes"is one stringr##"…"##in an attributeHow to test
Or in the demo/playground, paste the Rust snippets above.
Checklist
test/markup/rust/strings.txt/.expect.txt— raw, byte, multi-line raw strings in attributes, plus a lifetime-in-attribute regression case)CHANGES.md🤖 Generated with Claude Code