moonbitlang / core / 6264

91%

push

github

fix(base64): keep the vector fast paths free of wrapping bounds

The block loops tested `index + 16 <= end`, which wraps to a true
comparison for a view sitting at the very end of a large backing buffer,
and `encode_v128` sized its UTF-16 buffer as `char_count * 2` without
checking that the product fits in an `Int`. Neither is reachable through
any realistic caller -- both need an allocation close to two gigabytes --
but the vector loads and stores they guard are not bounds checked, so a
wrapped index reads and writes outside the buffer rather than failing.

Both loops now bound themselves by subtraction, and the encoder hands an
input whose buffer size would wrap to the scalar encoder, which builds
the string without ever materializing that buffer.

Found by Codex CLI while reviewing the equivalent code in #4122, which
fixes the same two defects in `encoding/hex`.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

1 of 2 new or added lines in 2 files covered. (50.0%)

16592 of 18275 relevant lines covered (90.79%)

306064.36 hits per line