On 2022-01-14 14:50, Sébastien Villemot wrote:
Le dimanche 02 janvier 2022 à 21:19 +0100, Paul Gevers a écrit :
With a recent upload of openblas the autopkgtest of hypre fails in
testing when that autopkgtest is run with the binary packages of
openblas from unstable on armhf due to a timeout after 2:47 h. It
passes
when run with only packages from testing in about 9-12 minutes. In
tabular form:
pass fail
openblas from testing 0.3.19+ds-1
hypre from testing 2.22.1-3
all others from testing from testing
...
First note that there has been no change to the ARM32-specific code of
openblas between 0.3.18+ds-1 and 0.3.19+ds-1.
Also, I tried to replicate the problem on a porterbox (abel), but the
autopkgtest succeeds (with hypre 2.22.1-3 and openblas 0.3.19+ds-2, in
an armhf chroot):
...
So at this point I don’t know how to tackle this issue.
Drew, any idea?
No clear ideas at this point. I ran a battery of tests on debci. One
armhf test passed on unstable, but 10 or so failed. So there's some
random irreproducibility of the problem.
The passing machine was ci-worker-armel-01, but it was also the host for
failing tests. So it's not a simple as some difference between
different CPU variants of armhf.
There's a new version waiting in NEW (actually there are 2 versions
waiting in NEW). There's not much more constructive testing I can do
until they get processed. Once they're cleared we'll be able to compare
the new version from experimental.
Drew