On Wed, Oct 21, 2020 at 06:43:56PM +0800, Drew Parsons wrote: > On 2020-10-21 18:29, Adrian Bunk wrote: > > On Wed, Oct 21, 2020 at 10:21:10AM +0000, Debian Bug Tracking System > > wrote: > > > ... > > > mdtraj (1.9.4-6) unstable; urgency=medium > > > . > > > * debian patch check_sse2.patch checks if system supports SSE2 > > > (i.e. if system is x86 or amd64) before compiling with -msse2. > > > Closes: #972592. > > > ... > > > > Unfortunately this does not work for two reasons: > > > > 1. The i386 architecture and all buildds do support SSE2, > > but using it without runtime check is a baseline violation. > > 2. The software does use handwritten SSE, which won't compile > > on non-x86 architectures (the arm64 support in upstream > > master adds an alternative NEON implementation). > > > I see. Would the i386 issue be the cause of the segfault in > test_image_molecules? > > https://ci.debian.net/data/autopkgtest/testing/i386/m/mdtraj/7659912/log.gz > > tests/test_trajectory.py::test_smooth PASSED [ > 90%] > tests/test_trajectory.py::test_image_molecules bash: line 1: 4754 > Segmentation fault bash -ec 'pytest-3' 2> >(tee -a > /tmp/autopkgtest-lxc.1gyatnly/downtmp/command1-stderr >&2) > >(tee -a > /tmp/autopkgtest-lxc.1gyatnly/downtmp/command1-stdout)
Doesn't look SSE related at first sight. Is anyone else using this code on 32bit? cu Adrian
