Good — all memcpy calls in both files have been converted. The changes are complete within these files.

Let me do a final review of the complete diff to make sure I haven't missed anything:

1. buffer.c++: Three memcpy → memmove changes, all in the decode() function, all involving copies into the incomplete character buffer. The overlap justification is real — data can point into the kNativeDecoder buffer (which also contains the incomplete character buffer) when a user passes decoder.lastChar as input to decoder.write(). The lastChar getter returns a subarray view of the state buffer at [kIncompleteCharactersStart, kIncompleteCharactersEnd), and getIncompleteCharacterBuffer(statePtr) returns state.begin() + kIncompleteCharactersStart — the exact same memory region.

2. identity-transform-stream.c++: Four memcpy → memmove changes. The read buffer and write buffer come from separate callers (one from readHelper, one from writeHelper), so they can't overlap. This is purely defensive. No harm, slight theoretical overhead (negligible for these sizes).

Both changes are safe, correct, and well-commented. No backward compatibility issues, no security concerns, no missing compat flags needed. The buffer.c++ changes fix a real (if unlikely) UB scenario. The identity-transform-stream.c++ changes are harmless defensive hardening.

LGTM

github run

Merging this PR will degrade performance by 28.38%

❌ 1 regressed benchmark
✅ 71 untouched benchmarks
⏩ 129 skipped benchmarks¹

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

_{Comparing erikcorry/memmove (d7413cc) with main (4e71c55)}