Bug#1082747: cod-tools: FTBFS (only) on riscv64

[Niko Tyni]

On Fri, Sep 27, 2024 at 06:29:12AM +0200, Aurelien Jarno wrote:
> On 2024-09-25 17:30, gregor herrmann wrote:
> > Source: cod-tools
> > Version: 3.10.0+dfsg-1.1
> > Severity: serious
> > Tags: help ftbfs
> > Justification: fails to build from source (but built successfully in the past)
> > X-Debbugs-Cc: debian-perl at lists.debian.org, debian-riscv at lists.debian.org
> > User: debian-perl at lists.debian.org
> > Usertags: perl-5.40-transition
> > 
> > I noticed that cod-tools 3.10.0+dfsg-1.1 fails to build but only on
> > riscv64:
> > 
> > https://buildd.debian.org/status/package.php?p=cod-tools
> > https://buildd.debian.org/status/logs.php?pkg=cod-tools
> > 
> > I care about this issue because 1) I uploaded the -1.1 NMU, and 2)
> > this will get in the way of the Perl 5.40 transition (#1082293),
> > which was the reason for the NMU in the first place, as it blocks the
> > migration of cod-tools.
> > 
> > Observations and thoughts:
> > * -1 built everywhere, including riscv64
> > * -1.1 still builds everywhere, except for riscv64
> > * I don't think the changes from the NMU (Niko's Perl exporter()
> >   patch) are related
> > * The build on riscv64 failed 4 times, so not a
> >   parallelization/timing issue probably (and d/rules uses
> >   --no-parallel)
> 
> I confirm that the build reproducibly failed 4 times on 4 different
> build daemons. That said on my side I am not able to reproduce the
> problem, neither on two different boards at home, nor on the porterbox
> ricci.d.o (which runs the same hardware as the buildds).
> 
> If someone is able to reproduce the issue, even if not able to debug it,
> that would be very helpful, so do not hesitate to point that in the bug
> log (that's why I kept the Cc: to debian-riscv at l.d.o).

It seems to be a readdir() order thing. I was able to reproduce it on amd64
under 'disorderfs --sort-dirents=yes --reverse-dirents=no'.

Didn't have time to debug it further yet.

Hope this helps,
-- 
Niko