On Wed, Feb 08, 2023 at 12:22:09PM +0100, Santiago Vila wrote: > El 8/2/23 a las 9:33, Andrius Merkys escribió: > > Hi Santiago, > > > > On 2023-02-06 19:36, Santiago Vila wrote: > > > This is fully reproducible at least on Hetzner virtual instances of type > > > CX21 and CX31. > > > > Thanks for giving this a look. Could you please provide the versions of > > python3-biopython and python3-numpy on your setup? ci.debian.net uses the > > following: > > > > python3-biopython (1.80+dfsg-4+b1) > > python3-numpy (1:1.24.1-2+b1) > > That's what I'm using as well. > > > I would like to rule out the possibility that version differences are to > > blame here. > > I think you can definitely rule out such possibility. > > See the two attached build logs. They were made the same day at nearly the > same time. One of them succeeded, and the other failed. > > The one that succeeded was made in a self-hosted virtual machine. > The one that failed was made in a Hetzner machine. > > So this seems to be machine-specific, maybe different optimizations for > Intel/AMD, > or failure to pass the tests when building in parallel (my self-hosted kvm > machine > has only one CPU),
The build succeeds on all little endian buildds[1] and in reproducible builds, therefore building in parallel is not the issue. > or any other similar thing which depends on the machine being used > (hence my offer to ssh into a Hetzner machine for you to test if you need it). The problem is that someone has to understand both chemistry and this software well enough, apparrently better than upstream so far: https://github.com/prody/ProDy/issues/1594#issuecomment-1419174201 > Thanks. cu Adrian [1] the tests never passed on big endian
