The problem turned out to be a real bug (null dereference) that only occurred occasionally due to poor testing procedures (test includes some randomization in the code path).
The bug has been patched and a separate RFS for fclib has been filed, #962684. regards, Steve On Mon, Jun 8, 2020 at 6:09 PM Stephen Sinclair <radars...@gmail.com> wrote: > > Hi Adam, > > Thanks for taking a look. > > On Mon, Jun 8, 2020 at 5:29 PM Adam Borowski <kilob...@angband.pl> wrote: > > > > On Mon, Jun 08, 2020 at 07:37:26AM +0200, Stephen Sinclair wrote: > > > * Package name : siconos > > > Version : 4.3.0+dfsg-1 > > > > But > > > > > Testing this package requires fclib 3.1.0+dfsg-1, which also needs > > > sponsorship. It can be found here: > > > > > > https://mentors.debian.net/package/fclib > > > > so thats actually two RFSes in one. And there are problems with fclib. > > I'm not sure what the right protocol is in this case, to be honest. > Should I have issued two RFSes or done this one differently? > > > > Changes since the last upload: > > > > > > * New upstream version. (Closes: #962219) (Closes: #961735) > > > * Add dependencies libboost-timer-dev, libboost-chrono-dev. > > > * Depend on openblas and lapacke instead of atlas. > > > * Require fclib 3.1.0 > > > * Update location of install paths. > > > * Enable WITH_GENERATION, now required for serialization. > > > * Install new siconos_export_raw_data tool. > > > * Fix cmake import targets to allow independent packages. > > > * Update patches for new upstream version. > > > * Add a flag for gfortran to avoid a regression in GCC-10. > > > (Closes: #957794) > > > * Remove unused build rule for swig3.0 symlink. > > > * Remove non-existent files from debian/copyright. > > > * Rewrite patch descriptions using gbp pq. > > > * Fix a Python warning about using 'is' with a literal. > > > > It looked ok on my box, and passed both all automated and manual review I've > > done. But, it fails on some of official buildds: at least on amd64 arm64 > > x32. > > > > I seem unable to reproduce the FTBFS locally -- in 15 tries on amd64, 1 on > > arm64, all passed. > > > > Thus, you'd need to investigate and fix that one in fclib first. > > I'm very confused about the error on buildd because I have indeed > built the package many times with a fresh debootstrap without such a > segfault. I will investigate and try to reproduce. > > I did my best to avoid problems but it seems I have missed something, > I'd like to understand the difference between buildd and my own > configuration to avoid this happening in the future. > > By the way Siconos due to its nature has been quite hard to get > working on any architecture other than amd64 unfortunately. Is there > a way to mark the package as amd64-only? > In the meantime I try to find solutions for other architectures but it > is slow going. > > > It'd be good if you filed a separate RFS for that. Let's leave this one > > for siconos 4.3.0+dfsg-1 > > Okay that sounds like a reasonable approach given this problem. > > regards, > Steve