Re: Fixing regressions and serious issues

[Jean Abou Samra]

Le 18/09/2022 à 14:46, Jonas Hahnfeld a écrit :
> On Fri, 2022-09-16 at 13:36 +0200, Jonas Hahnfeld via Discussions on
> LilyPond development wrote:
> > On Wed, 2022-09-14 at 22:35 +0200, Jean Abou Samra wrote:
> > > Le 14/09/2022 à 22:16, Jonas Hahnfeld a écrit :
> > > > On Wed, 2022-07-20 at 11:39 +0200, Jonas Hahnfeld via Discussions on
> > > > LilyPond development wrote:
> > > > What do we do about this one? Over the past couple of weeks, I tried
> > > > quite a number of ideas, with no success so far.
> > > Thanks a lot for working on this even if it didn't succeed so far.
> > >
> > > Just in case for others: Jonas shared some details about what he tried
> > > in https://github.com/ivmai/bdwgc/issues/454#issuecomment-1244375504
> > I've tried some more and probably developed an understanding of what's
> > happening - I will post on the upstream issue later. For our use case,
> > however, we can "cheat" a bit because we statically link both bdwgc and
> > libguile; https://gitlab.com/lilypond/lilypond/-/merge_requests/1627
> > should work around the crashes (at least they do for me in Wine).
> Jean thankfully tested the binaries I produced with that change, and
> they don't seem to completely break while fixing the crashes with the
> originally reported scores.




I'm sorry for not having been clear on this point: yes, I
haven't observed crashes on the original reports; on the other
hand, I tried 2.23.10 (supposed to be buggy), and didn't get
crashes either, so I dunno :(  This is the glory of Heisenbugs.

The MR at least fixed the crashes you observed with Wine, so
hopefully it should at least reduce the frequency of crashes.
At this point, there seems to be a good chance that we are dealing
with a bug in BDWGC itself, so I think we can call it good enough
if it doesn't pop up for real-world use cases on our end, and
allow the 2.24 in that case, even if the underlying issue in
BDWGC is not fully understood.

The conclusion is that I am happy this change was included in
2.23.13, as we will be able to ask Windows users if these binaries
seem to fix the issue for them.

By the way, if you happen to still have a binary of what is now
2.23.13 with debugging symbols, that would save me time. I'd like
to investigate the crash when loading the edition-engraver, which
is 100% reproducible for me, but at a glance doesn't seem related
to this GC Heisenbug.