feat: RegexCapability - #196
Conversation
01e092e to
fa8c15a
Compare
80d8df4 to
81d4a77
Compare
Overall package sizeSelf size: 30.29 MB Dependency sizes| name | version | self size | total size | |------|---------|-----------|------------|🤖 This report was automatically generated by heaviest-objects-in-the-universe |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 81d4a77fc2
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
| // `g` is required to iterate all matches with exec; `d` is required for the | ||
| // per-group indices used by `capturesAll`. | ||
| module.exports.compile = function (pattern) { | ||
| return new RegExp(pattern, 'gd') |
There was a problem hiding this comment.
When callers use a pattern that can consume one UTF-16 code unit, such as ., against non-BMP input like "😀", this regex is not in Unicode mode, so exec matches only the high surrogate and findFirst reports [0, 2]. Those values are returned to Rust as UTF-8 byte offsets, but byte 2 is inside the four-byte character; consumers that slice strings or expect Rust regex-style match boundaries will get incorrect results or panic. Compile/iterate with Unicode-aware semantics, or otherwise prevent matches from splitting surrogate pairs.
Useful? React with 👍 / 👎.
| for (let g = 0; g < groupCount; g++) { | ||
| const idx = m.indices[g] | ||
| if (idx === undefined) out.push(-1, -1) | ||
| else out.push(local[idx[0] - matchStart], local[idx[1] - matchStart]) |
There was a problem hiding this comment.
For JS patterns with captures in lookahead/lookbehind, a capture group's indices can fall outside the overall match span; for example (?=(a)) on "a" has match [0, 0] but group [0, 1]. Because the local byte-offset map only covers matchStart..matchEnd, local[idx[1] - matchStart] is undefined and is later converted to 0 in the returned Int32Array, so Rust receives incorrect capture ranges. Build the offset map over the full min/max range of all present group indices, or fall back to walking from the haystack start for out-of-span groups.
Useful? React with 👍 / 👎.
| } | ||
|
|
||
| fn captures(handle: &Self::Handle, haystack: &str) -> Option<Captures> { | ||
| decode_captures(js_captures_all(&handle.re, haystack)) |
There was a problem hiding this comment.
When callers ask for only the first capture set, this still invokes capturesAll, which runs the global regexp over the entire haystack and allocates/decodes every match before discarding all but the first. On large trace payloads or broad patterns this makes captures() scale with total match count instead of the first match; add a JS helper that performs a single exec and decodes only that result.
Useful? React with 👍 / 👎.
| // `g` is required to iterate all matches with exec; `d` is required for the | ||
| // per-group indices used by `capturesAll`. | ||
| module.exports.compile = function (pattern) { | ||
| return new RegExp(pattern, 'gd') |
There was a problem hiding this comment.
When the agent's /info filters include a pathological pattern such as (a+)+$, this compiles it into V8's backtracking RegExp; evaluating it against a non-matching resource like a long run of as plus ! can block the Node event loop for seconds, whereas the native capability uses Rust's linear-time regex engine. Since these filters are applied while processing trace payloads, a configured filter can stall trace flushing; use the Rust/wasm regex engine or otherwise reject patterns that can trigger catastrophic backtracking.
Useful? React with 👍 / 👎.
1 participant
Wasm still gets compiled and is in the dependancy tree through libdd-common, but wasm-ld trims it out for it is not used