moonbitlang / core / 6879

89%

push

github

fix(float): correct Float::round for signed zero, the half boundary, and large integers

On native, wasm and wasm-gc `Float::round` was `floor(x + 0.5)`, the same
naive formula #3939 replaced for `Double::round`:

- odd integers of magnitude in [2^23, 2^24) moved by one, because
  `x + 0.5` is a tie that rounds to even (8388609 -> 8388610);
- nextDown(0.5) rounded to 1, because `x + 0.5` rounds up to 1.0;
- inputs in [-0.5, 0), and -0 itself, rounded to +0 instead of -0.

Port the guarded logic of `Double::round` so every backend agrees with js
(`Math.round`): return the input for zero, NaN and |x| >= 2^23, return +0
for 0 < x < 0.5 and -0 for -0.5 <= x < 0, and use `floor(x + 0.5)`
otherwise. The regression test now runs on js as well.

Signed-off-by: Codex CLI <codex@openai.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

8 of 8 new or added lines in 1 file covered. (100.0%)

16657 of 18626 relevant lines covered (89.43%)

332898.82 hits per line