Closed BrieucF closed 5 months ago
This looks oddly similar to what we see in the k4MarlinWrapper CI at the moment: https://github.com/key4hep/k4MarlinWrapper/actions/runs/7785263794/job/21264263282#step:3:6809
Do the Alma9 and CentOS7 nightlies have the same date at the moment, or is one of them lagging behind due to other issues? Can you easily check that?
They are both from today (2024-02-06)
OK. Thanks for checking. I will have a look at the k4MarlinWrapper one, and I think that should also solve this, since in the end also CLD reco is using that.
Quick update on the little things I have figured out so far using CLIC reco (but I would assume CLD reco is the same / similar). Playing around with commenting out a few algorithms, I think I have narrowed it down to LCFIPlus being the offender here. That would also make sense, as that uses some ROOT things internally and the segmentation violation is inside ROOT. I haven't been able to debug it much further and I also don't know why it only happens on CentOS7 only yet.
edit: As a side note: LCFIPlus hasn't changed in 2-ish years.
It also seems to have been present for at least three weeks now, or at least I can find CI runs that show the issue from 3 weeks ago, e.g. https://github.com/key4hep/k4MarlinWrapper/actions/runs/7519772256/job/20473078519
Some hints by comparing to a working version, it doesn't work when:
In case it's LCFIPlus by checking the dates, it's not due to https://github.com/iLCSoft/LCFIVertex/pull/9 and the other dependencies remain unchanged
Since there aren't going to be any more releases nor nightlies with CentOS 7 and it's unlikely anyone will have a look and fix it I'm going to close this.
The CLD reconstruction seems to be broken with CentOS 7 but works with the Alma9 flavor.
Reproducer:
Stack trace: stack_trace_CLD_reco_COS7.txt