[Libreoffice-bugs] [Bug 124124] Writer crashes in KDE5 after document reconstruction / at start

[bugzilla-daemon at bugs.documentfoundation.org]

https://bugs.documentfoundation.org/show_bug.cgi?id=124124

--- Comment #21 from Michael Weghorn <m.weghorn at posteo.de> ---
(In reply to Sandra Weddig from comment #20)
> I did add a new backtrace, one with a different window style and one of the
> debug masterbuild (this seems not to run, do I need a different JDK?).

Thanks.
The easiest workaround is probably to just disable Java by temporarily starting
with the gtk3 VCL plugin, i.e. set the environment variable
SAL_USE_VCLPLUGIN=gtk3 before starting LibreOffice and then in "Tools" ->
"Options" -> "Advanced" disable the checkbox for "Use a Java runtime
environment".

I've set up an openSUSE Tumbleweed VM for testing and could reproduce the JDK
issue, but unfortunately not the crash, which makes this hard to analyse.

Looking at the backtrace from comment 18 and the involved code, I'd guess that
there's a division by zero (causing the "SIGFPE, Arithmetic exception"),
possibly because all the variables in the divisor (i.e. "range" and
"scrollbar->pageStep") are 0.

It would be interesting to have that verified, so it'd be of help if you could
start the debug version (with the same widget style as for the backtrace in
comment 19) with

   gdb --args <SOME_PATH>/instdir/program/soffice.bin --writer

at which point a '(gdb)' prompt appears.
The command 'r' will start the program. If the same crash appears, it'd be good
to print the values of the variables involved and the backtrace with these
commands then:

 thread apply all bt
 p range
 p scrollbar->pageStep
 p scrollbar->maximum
 p scrollbar->minimum


and attach the result here. Since this is not really straightforward, please
feel free to just leave out those steps and just attach the backtrace.

In the meantime, I had the same issue a few times, but couldn't reliable
reproduce. Any idea what might help to reproduce this?


AFAICS, the relevant values for the 'QStyleOptionSlider option' variable being
passed are set at [1] based on values passed to that method by the
ImplControlValue param.


[1]
https://gerrit.libreoffice.org/plugins/gitiles/core/+/master/vcl/qt5/Qt5Graphics_Controls.cxx#490

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/libreoffice-bugs/attachments/20190403/c61b6bf2/attachment.html>