IRC channel logs


back to list of logs

<rekado>found a reproducibility problem
<rekado>c-blosc is built with AVX2 instructions.
<rekado>this fails at runtime on the MDC cluster nodes
<rekado>python-numcodecs bundles c-blosc and has the same problem.
<rekado>I’ma fix these
<civodul>great that you were able to pinpoint the issue
<rekado>maybe we could have a build phase or a lint thing that checks the sources for AVX2, SSE4, march=native, etc
<civodul>a build phase maybe
<civodul>or a compiler wrapper... but that's evil!
<efraim>Can we try to run programs with using limited CPU instructions without using qemu?
<civodul>efraim: i think qemu or specific hardware is the only way to test that
<civodul>there are bioconductor-related failures in the disarchive jobset:
<civodul>does that ring a bell?
<civodul>comes from the latest r-* upgrade in the a0f1476..56dcaff range:
<rekado>it often happens that right after a bioconductor upgrade people notice problems and then quickly upgrade some packages
<rekado>usually before we have fetched their sources
<rekado>so I guess I’ll have to run the upgrade once more to catch these upgraded tarballs
<civodul>heh ok
<civodul>horror stories