Repository navigation
fix(base): preserve negative zero in F32.round - #1613
Merged
Lorenzobattistela merged 1 commit intoOct 16, 2026
Merged
Conversation
Lorenzobattistela
approved these changes
Oct 16, 2026
Merged
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes #1611.
When rounding produces zero, this restores the input sign instead of returning the positive zero created by the final addition. Other results and the existing ties-toward-positive-infinity rule stay unchanged.
The regression covers
-0, a negative fraction through its reciprocal, the negative half, and the adjacent value that must still round to-1. Before:0, False, 0, True. After:2147483648, True, 2147483648, Trueon both Bun and native C. The existing number kit also passes on both.Linux x86-64, Bun 1.4.3, clang 22.1.8. Current main has an unrelated compiler token-cap failure; this changes only Base and its regression, not the compiler or caps. The mini-cluster/Metal/CUDA gates were not run here. AI-assisted patch.