On Sun, Jan 15, 2023 at 04:36:39PM +0100, Santiago Vila wrote: > El 15/1/23 a las 15:45, Nilesh Patra escribió: > > On Sun, Jan 15, 2023 at 02:02:15PM +0100, Santiago Vila wrote: > > > I can reproduce this build failure on Hetzner virtual machines with 2 > > > CPUs, > > > where it fails randomly 50% of the time. > > > > > > Therefore, to reproduce the randomness, you have to try many times. > > > > I built it 50 times on barriere.debian.org (4 CPUs) and did not hit > > the failure even once. I don't think I have the time/energy for further > > debug this any further; I'd have tried if it was properly reproducible. > > The randomness is 100% reproducible on *certain* machines. Unfortunately, > as you have just checked, barriere was not one of them, but I can offer > ssh access to such a machine. Please contact me privately for details. > > Packages which FTBFS randomly are the worst, that's why I always offer ssh > access to a machine where they happen when I report one of those bugs.
Thank you Santiago for providing me access to your instance and helping
me emulate the failure there. I am able to repro it on 5-6 attempt at
re-building.
The test code that fails in question does not make a lot of sense to me,
as some of it seems redundant and there's probably a bug there. I tried
to apply a small patch which I consider what upstream was trying to do,
and I built the package ~25 times with this script (in lots of 15 + 10)
#!/bin/bash
for i in {1..10}; do
cd /home/nilesh/macsyfinder && sbuild -d bookworm --no-clean-source
2>&1 > /tmp/log
fgrep -q "cannot join current thread"
/home/nilesh/macsyfinder_2.0-2_amd64.build
x=$?
if [ "$x" != 1 ]; then exit $x; fi
echo -n "Loop complete: "
echo $i
done
And I could see it being successful (I also checked the last 50 lines of
each *.build file and the status was successful). I had also sent the patch in
comments to upstream in the issue linked with this bug report
https://github.com/gem-pasteur/macsyfinder/issues/58
I've now added a comment saying that the patch fixes these random
issues. I am tempted to merge this patch and upload, but I am not 100%
confident if it is correct. And hence, a review would be nice (Etienne,
if you are reading this if you could check) or I'll wait for
upstream action. They seem kind of active.
Again, thank you Santiago!
--
Best,
Nilesh
signature.asc
Description: PGP signature
