False positive?

[FrancescAlted]

Hi. In Blosc/c-blosc2#567 we have been trying to cope with a possible issue raised by oss-fuzz. However, we cannot reproduce it in our Ubuntu boxes, and furthermore, we don't think we are doing anything controversial in the pointed function.

Is there a way to silence a potential false positive using oss-fuzz?

Thanks in advance.

[DavidKorczynski]

I tried reproducing and was also unable to.

Is there a way you can guard against the potential issue suggested here Blosc/c-blosc2#567 (comment) ? When OSS-Fuzz builds fuzzers there is a fuzzer-specific macro defined: FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION see here for more details -- would it be possible to use this macro? around the area that complains with some alternative logic?

[FrancescAlted]

Recently I did some commit that seems to fix the issues with OSS-Fuzz. Not sure why they are passing now, because I did not touch anything special affecting the Linux/Clang platform that uses OSS-Fuzz. Did you fixed something on the OSS-Fuzz side of the things? Anyway, happy this seems fixed now.

[DavidKorczynski]

Did you fixed something on the OSS-Fuzz side of the things?

No, I think it's the commit you did that caused the change.

It seems the root cause was missing support for AVX512 in the machines that run CIFuzz (not necessarily OSS-Fuzz as such).

Closing as it's fixed.

[FrancescAlted]

Happy to see the issue gone (but still scratching my head :-)