Re: failing CI builds

[Jonas Hahnfeld]

Am Samstag, den 27.06.2020, 18:34 +0200 schrieb Han-Wen Nienhuys:
> On Sat, Jun 27, 2020 at 6:29 PM Han-Wen Nienhuys <hanwenn@gmail.com> wrote:
> > On Fri, Jun 26, 2020 at 11:37 PM Jonas Hahnfeld <hahnjo@hahnjo.de> wrote:
> > > Hi all,
> > > 
> > > at least the following two MRs experience strange failures in CI:
> > > https://gitlab.com/lilypond/lilypond/-/merge_requests/191
> > > https://gitlab.com/lilypond/lilypond/-/merge_requests/201
> > > 
> > > The latter one is mine and I can reproduce this on my own runner, which
> > > means I can get the log files easily and even connect from the host
> > > with gdb and get a bracktrace (see attached files). It's failing
> > > somewhere deep inside libguile.so - any ideas?
> > > 
> > > When comparing the build artifacts, I noticed that the output files in
> > > mf/ are not reproducible. I'm not sure if that is related,
> > 
> > Unlikely. For unknown reasons metapost is dumping the current time in
> > its log output
> 
> PFA flies have creation timestamps. I suppose PFB and OTF have them too.

Sure, timestamps are one source of non-determinism. However, I'm also
seeing differences in the generated .svg files. I find this very
suspicious because my naive interpretation is that some glyphs differ
between multiple runs.

signature.asc
Description: This is a digitally signed message part