[Kde-pim] Kontact3 crash related to catalogues(bug #139652)

[Raúl Sánchez Siles]

David Faure wrote:

  Thanks a ton for your comforting mail. It's a relief to know that someone
cares about such a difficult problem as this could be.

> On Monday 17 September 2007, Raúl Sánchez Siles wrote:
>> Moreover I don't know what should I be looking for. Any suggestion?
> Use valgrind. Oh wait, you have already done that :)
 Yes and as you could imagine is _very_ painful. Moreover, since I'm
pretending to be a tough guy I've done for about 5 more times till you
answered this. I have got some more crashed which I reported convenientely
in the BTS, but I'm out of luck trying to repeat this very important one.

> Let's see... wow your ld suppressions don't work, many false hits there
> (and you can remove leak-check=yes when not hunting leaks).

  I think this is working now, still some Qt and libc errors visible.

> 
> But there's one real bug here (Frank?):
> ==9351== Conditional jump or move depends on uninitialised value(s)
> ==9351==    at 0xC5CFBD7: Akregator::Article::hash() const
> (article.cpp:379)
> ==9351==    by 0xC5D6F64: Akregator::Feed::appendArticles(RSS::Document
> const&) (feed.cpp:478)
> ==9351==    by 0xC5D72D3: Akregator::Feed::fetchCompleted(RSS::Loader*,
> RSS::Document, RSS::Status) (feed.cpp:648)
> ==9351==    by 0xC5D758C: Akregator::Feed::qt_invoke(int, QUObject*)
> (feed.moc:218)

  I think I have seen some more akregrator bugs like the one referring to
closeTab.

> 
> However this is "just" undefined behavior, not a reason for a crash. I
> can't see any catalogue-related crash... Can you try again with valgrind,
> using "valgrind kontact --nofork"? (Possibly after double-checking the
> suppression files or recompiling valgrind, which will then pick the right
> suppression files automatically, hopefully.)
> 

  As the good suggestion you did, I did it with the --nofork parameter. You
can see the latest run attached. This time kontact crashed because of a
problem is knode(KNNetAccess::slotThreadSignal) which is not related to the
catalogue unfortunately, every time I had a crash I prayed for it to be the
catalogue problem, I should try with different Gods next time... :P

  The valgrind log I attach includes file descriptors info in case you see
whatever valuable information. Maybe it's a little bit more thoroughful to
read, but I didn't want to avoid any valuable info.

 Taken this into account I think the starving hypothesis takes force. I'm
running kontact without valgrind now, I'm fed up of collapsing my system.
Another idea I have is that maybe watchdir related stuff could be causing
the problem. I had some crash related to this(supposed to be solved on
3.5.8). So unless other actions suggestted, what I'll do is:

  - Running it till crashes to check that problem is still present.
  - Avoid using parts from within kontact. I'll start with akregator,
running it standalone, then knode will follow. Maybe I could be able to
isolate the problem a little more. A painful valgrinded akregator should be
less painful than running a full kontact session again.

  The only way I have to hard debug this is attaching to kontact when the
apps crashes and it's the only thing I could do for the moment.

  Needless to say that whatever idea, as crazy as it could be could be
useful for me, so I'd ask you not to be shy and share it with me.

  Thanks.
  

-- 
Raúl Sánchez Siles
-------------- next part --------------
A non-text attachment was scrubbed...
Name: kontact.vg.29363.bz2
Type: application/x-bzip2
Size: 21853 bytes
Desc: not available
URL: <http://mail.kde.org/pipermail/kde-pim/attachments/20070920/168f5102/attachment.bin>
-------------- next part --------------
_______________________________________________
KDE PIM mailing list kde-pim at kde.org
https://mail.kde.org/mailman/listinfo/kde-pim
KDE PIM home page at http://pim.kde.org/