Re: master segfault

[Dan Eble]

On Jul 23, 2020, at 17:59, Dan Eble <dan@faithful.be> wrote:
> 
> On Jul 23, 2020, at 13:21, Jean Abou Samra <jean@abou-samra.fr> wrote:
>> 
>> Le 23/07/2020 à 19:05, Dan Eble a écrit :
>> 
>>> It's a heisenbug.  It isn't the random seed itself that makes a difference, 
>>> just the fact that things are done differently now.  Using different 
>>> optimization flags makes a difference too.  I enabled ubsan and it detected 
>>> badness slightly earlier than the segfault, but not early enough that I 
>>> have any idea what might have gone wrong or when.
>>> —
>>> Dan
>> 
>> Out of curiosity, how many processes are you using for make test-baseline?
> 
> That was the first variable I tested, though I neglected to mention it.  I 
> tried 7, 6, and 1 jobs and the failure occurred regardless.
> 
> I'm planning a round of deeper inspection in gdb.  Maybe I'll find something 
> helpful before I get bored.

A lot of the information I could understand looked fine to me, so I figured the 
problem might have involved using a dead object.  Reviewing the code of the 
functions in the call stack led me to 
https://gitlab.com/lilypond/lilypond/-/merge_requests/271 .  My build no longer 
crashes.  Hopefully it's not mere coincidence.
— 
Dan