https://bugs.documentfoundation.org/show_bug.cgi?id=122643
--- Comment #9 from Michael Meeks <[email protected]> ---
Thanks Julien; as you say - all false positives from glibc internals assuming
knowledge of their own allocator (which is fine).
The DispatchUserEvents thing really looks like memory corruption though:
warn:vcl:11210:11210:vcl/source/app/salusereventlist.cxx:120: Uncaught
St12out_of_range multi_type_vector::position#1570: block position not found!
(logical pos=1048580, block size=4, logical size=1048576)
I would assume that in fact under valgrind we don't get a crash either ;-)
which is annoying - presumably it perterbs threading so the issue doesn't
happen.
What might help is getting another valgrind trace when run with:
--fair-sched=yes
It is also possible that we need to substantially reduce this eg:
coregrind/m_scheduler/scheduler.c:#define SCHEDULING_QUANTUM 1200 // 100000
To get a better approximation of fast context switching to catch the race. It
is somewhat odd that this isn't a cmd-line option for valgrind particularly
since there is no performance benefit of it not being configurable; possibly an
easy-hack for valgrind lurks there =)
--
You are receiving this mail because:
You are the assignee for the bug.
_______________________________________________
Libreoffice-bugs mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/libreoffice-bugs
