[kixelated]

Problem

rust2-ts refused retain, retain_mut, dedup, dedup_by, dedup_by_key, insert, remove, and swap_remove on a Vec<u8>, and last, reverse, swap, and the six sorts on a [u8]: the mappings in library.rs skipped u8 elements, which are an rt.ByteVec or rt.ByteSlice rather than an array, and nothing mapped them instead. Found by #204. No moq consumer: rust2 aims to support all of std.

Approach

• Every !is_u8 skip in library.rs now has a byte counterpart. Those were all of them; the other Vec and slice mappings already handled bytes.
• Vec<u8>: new rt.byteVecInsert, Remove, SwapRemove, Retain, RetainMut, Dedup, DedupByMut, and DedupByKeyMut, which edit the Uint8Array in place (copyWithin for the shifts). The representation is unchanged.
• [u8]: rt.byteSliceLast and rt.byteSliceSwap, beside byteSliceFirst. reverse and the sorts without a closure call the typed array's own reverse and sort on byteSliceView. A typed array sorts by value, and equal bytes can't be told apart, so stability doesn't matter. The closure sorts run rt.sliceSort* over the view, which now take any indexable.
• retain_mut, dedup_by, and dedup_by_key give their closure a &mut, and a byte is a value JS copies, so it has to be passed as an rt.Place. place_call now lists those three for Vec, and the mapping picks a *Mut variant that passes an AtIndex whenever the element needs_place. That fixes Vec<u16> and other copied elements too, which were refused for the same reason. One problem, one mechanism, so it's in this PR.
• vecRetain, vecDedupBy, and the byte versions share two index-based helpers, retainAt and dedupAt, so the logic exists once.
• ("slice", "reverse") is now elements(s).reverse() for every element type, and rt.sliceReverse is deleted.

Tests: a new edits::bytes case in the lang fixture checks against native wasm32 (50k random cases; 495,916 in total with the exhaustive short ones):

• each method on every short prefix of the input (empty, one byte, two bytes) and on the whole input;
• a subslice that starts past offset 0 of its storage (v[1..]'s sort, reverse, swap, and last);
• out-of-bounds panics from insert, remove, swap_remove, swap, and a subslice range;
• stable sort_by_key with tied keys, and unstable sorts only by keys or comparators with no ties, so std's order is the only valid one;
• closures that log their arguments and change bytes through their places;
• u16 versions of retain_mut, dedup_by, and dedup_by_key.

As a mutation check, breaking byteVecDedupByMut's argument order and byteVecSwapRemove made the differential report 263 differences. There are no unwind tests.

reject used <[u8]>::sort as its example of an unsupported call. It now uses rotate_left, which is still unsupported, so line numbers don't shift.

Census

moq-net: 3089 → 3087 refusals (the two <[u8]>::last calls in zlib-rs inflate.rs). model and kio are unchanged.

Impact

• Output: the methods above now translate on Vec<u8> and [u8], and retain_mut, dedup_by, and dedup_by_key translate on a Vec of any element JS copies.
• Runtime additions: rt.byteSliceLast, rt.byteSliceSwap, rt.byteVecInsert, rt.byteVecRemove, rt.byteVecSwapRemove, rt.byteVecRetain, rt.byteVecRetainMut, rt.byteVecDedup, rt.byteVecDedupByMut, rt.byteVecDedupByKeyMut, rt.vecRetainMut, rt.vecDedupByMut, and rt.vecDedupByKeyMut. rt.sliceSort, rt.sliceSortBy, and rt.sliceSortByKey accept any indexable.
• Breaking: rt.sliceReverse is removed. There's a MIGRATE.md entry.
• The codec's output is unchanged (its snapshot passes), so no bench run.

Alternatives

• Running every byte method over a byteSliceView: for last and swap that builds a view on each call, which costs several times a ByteSlice. So those got dedicated functions like first; only the O(n) methods use the view.
• Lending places only for bytes: Vec<u16>::retain_mut would still be refused for the same reason.

Follow-ups

None new. u8::rotate_left and [u8]::rotate_left are still unsupported; the reject fixture records the second.

🤖 Generated with Claude Code

(Written by Claude Opus 5.5)

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-29T09:31:17.464364Z	c8a468b	New commits
🔒 Security Review	✅ Completed 2026-10-29T09:33:16.329324Z	c8a468b	New commits

ℹ️ About Codex in GitHub

Your team has set up Codex to 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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[kixelated]

Automated review by review (OpenAI)

Reviewed commit: 28d7467

No actionable bugs found in the changed mappings and runtime paths. Reusing index-based retain/dedup helpers while lending copied elements through AtIndex is a reasonable shared fix for bytes and other scalar elements. The byte-specific paths preserve slice offsets and vector length/capacity boundaries; direct numeric typed-array sorting avoids unnecessary comparator machinery. The added differential cases cover mutation callbacks, dedup argument order, empty inputs, subslices and bounds failures.

Verification: static diff, runtime helpers, lowering/place-call handling and generated fixture inspection only. No tests, differential or mutation runs, builds or benchmarks performed; reported results were not independently reproduced. Unwind behavior was not validated.

(Written by OpenAI)