[Mageia-dev] urpmi segfaults on the build system (was Re: Our BS is about to eat packages?)

[Olivier Blin]

nicolas vigier <boklm at mars-attacks.org> writes:

> On Mon, 29 Aug 2011, Michael Scherer wrote:

[...]

>> > After looking at the problem, I suspect it could be linked to texmf.
>> > Each time iurt install it, the process stop.  
>> 
>> Boklm found the problem, it was urpmi segfaulting.
>> 
>> We remove the chroot, and thing seems to be ok now again.
>
> But we probably still have the problem, as segfaults seems to be
> happenning not all the time.
>
> After succesfully reproducing the error using the old chroot and running
> urpmi several times, the backtrace :
>
> #0  0xf7405b29 in _int_malloc () from /lib/i686/libc.so.6
> #1  0xf7408177 in malloc () from /lib/i686/libc.so.6
> #2  0xf6eadfa8 in __os_malloc (env=0xb442db8, size=4149199776, storep=0xffbbf5cc) at ../os/os_alloc.c:253
> #3  0xf6eac47c in __memp_sync_int (env=0xb442db8, dbmfp=0xb893130, trickle_max=0, flags=8, wrote_totalp=0x0, interruptedp=0x0)
>     at ../mp/mp_sync.c:294
> #4  0xf6eacfaf in __memp_fsync (dbmfp=0xb893130) at ../mp/mp_sync.c:202
> #5  0xf6e3cdf9 in __db_sync (dbp=0xb8963e8) at ../db/db_am.c:706
> #6  0xf6e3aa68 in __db_refresh (dbp=0xb8963e8, txn=0x0, flags=0, deferred_closep=0xffbbf6fc, reuse=0) at ../db/db.c:819
> #7  0xf6e3adc5 in __db_close (dbp=0xb8963e8, txn=0x0, flags=0) at ../db/db.c:695
> #8  0xf6e52a38 in __db_close_pp (dbp=0xb8963e8, flags=0) at ../db/db_iface.c:253
> #9  0xf72fec45 in db3close (dbi=0xb893030, flags=0) at backend/db3.c:494
> #10 0xf73071c6 in dbiClose (db=0xb2b6950) at ../lib/rpmdb_internal.h:453
> #11 rpmdbClose (db=0xb2b6950) at rpmdb.c:852
> #12 0xf73379d6 in rpmtsCloseDB (ts=0xb231b20) at rpmts.c:64
> #13 0xf7337a48 in rpmtsFree (ts=0xb231b20) at rpmts.c:567
> #14 0xf736c7a1 in XS_URPM__DB_DESTROY (my_perl=0x8d42008, cv=0x8e2d420) at URPM.xs:2794

[...]

Hello,

After hitting the issue yesterday with the "git" package, we had a look
at the issue together with pterjan.
I had to merge some patches in valgrind to support the new compressed
debug info first, but then valgrind spotted some interesting issues.

pterjan fixed some of them, fixes are now in perl-URPM 4.12.
It should help preventing "build loops" in the BS.

Attached is the valgrind log with remaining issues, maybe insightful for
Titi.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: valgrind.log.bz2
Type: application/x-bzip2
Size: 7174 bytes
Desc: not available
URL: </pipermail/mageia-dev/attachments/20120811/3ce980ee/attachment.bz2>
-------------- next part --------------

-- 
Olivier Blin - blino