Bug#964973: Bug#954849: Bug#964780: libreoffice-writer crashes on startup after upgrade to 1:7.0.0~rc1-5
reassign 954849 mesa-opencl-icd
forcemerge 964973 954849
affects 964973 libreoffice,libreoffice-calc
thanks
Hi,
Am 17.07.20 um 01:19 schrieb Rogério Brito:
> On Thu, Jul 16, 2020 at 4:11 PM Rene Engelhard <rene@debian.org> wrote:
>> Am 15.07.20 um 17:42 schrieb Rene Engelhard:
>>> Am 15.07.20 um 17:34 schrieb Rene Engelhard:
>>>>> ,----
>>>>> | $ libreoffice
>>>>> | : CommandLine Error: Option 'polly' registered more than once!
>>>>> | LLVM ERROR: inconsistency in registered CommandLine options
>>>>> `----
>>>> I'd have a look what causes this. Wherever this "polly" "command-line"
>>>> option is set.
> In the context of LLVM polly is the integer polyhedral optimization
> "engine" (GCC also uses something similar called graphite, if I am not
> mistaken). What causes it to be included multiple times, I don't know
> (but I suspect, see below).
>
>> We should discuss there, probably. Different option but... I still
>> believe it's not a LO problem but a problem of something using LLVM or
>> someone configuring something (using LLVM) badly.
> I tested stuff a little bit more and, now, I can tell more about this
> issue. Are you ready for a journey? :-) I hope that you are.
[...]
> Then, remembering that Calc started using OpenCL (at one point or
> another), I installed calc. It *STILL* worked.
[...]
> Then, I installed beignet-opencl-icd. It *STILL* worked. (I have a
> notebook with an integrated intel chip and a discrete AMD card)
>
> Then, I installed mesa-opencl-icd. It pulled in a lot of packages,
> including clang-common-10-dev and other stuff. It *still* worked, to
> my amazement.
>
> Then, I decided, just for fun, to remove the directory
> ~/.config/libreoffice (actually, I simply renamed it to
> ~/.config/libreoffice.old). NOW, it didn't work and showed again the
> "polly" message.
>
> If I copy back the contents from ~/.config/libreoffice.old to
> ~/.config/libreoffice and invoke libreoffice, then it works again.
>
> OK, so, mesa-opencl-icd was the thing that caused some instabilities,
> right? I then, decided to remove it (and its dependencies), just
> keeping Intel's beignet and, bingo:
>
> I can fire up libreoffice with both a prepopulated
> ~/.config/libreoffice or not and it works without any problems.
>
> With mesa-opencl-icd/20.1.2-1 installed, I can only run libreoffice
> successfully if I have a preexisting ~/.config/libreoffice from a
> previous successful run (an empty directory prevents it from running).
Probably with a clean profiile it sees the stuff is there and enables
OpenCL while it doesn't if the ICD is not present?
Maybe we generally should disable OpenCL per default?
(as Fedora apparently does:
https://src.fedoraproject.org/rpms/libreoffice/blob/master/f/0001-Resolves-rhbz-1432468-disable-opencl-by-default.patch)
> So, part of the mystery is unveiled (but not solved)... The problem is
> definitely not with libreoffice, but with LLVM & Co.
>
> Just for the record, the packages that were installed when
> mesa-opencl-icd/20.1.2-1 were:
>
> * lib32gcc-s1/10.1.0-4
> * lib32stdc++6/10.1.0-4
> * libc6-i386/2.30-8
> * libclang-common-10-dev/1:10.0.1~+rc4-1
> * libclc-amdgcn/0.2.0+git20190
> * libclc-dev/0.2.0+git20190
> * libclc-r600/0.2.0+git20190
>
> Based on this, I guess that this bug should probably be reassigned to
> the LLVM people, perhaps (or to the mesa people)...
Yeah.
I just noticed there is already
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=964973
> More precise information, straces and logs are available if needed.
One could tell the mesa people :-)
Reassignig this bug :-)
> Thanks for taking care of libreoffice in Debian,
Thanks for trying and giving some light here.
And for the original one:
--- snip ---
in console I get message:
: CommandLine Error: Option 'limited-coverage-experimental' registered more
than once!
LLVM ERROR: inconsistency in registered CommandLine options
--- snip ---
there was https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=852746
(especially https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=852746#76
?), too...
Regards,
Rene
Reply to: