On 07.11.2011 12:25, Pedro Giffuni wrote:
I agree with everyone :).
Right now it doesnt make sense to spend time on this,
however if Erics patch avoids the crash for now it
would be an acceptable solution.
I would like a bugzilla issue that we can keep open
so that we dont forget about the underlying issues,
maybe 118576 serves that purpose already.
I agree.
What is the status of Erics patch? Does it work and can it be applied
under the Apache license?
Regards, Andre
Cheers,
Pedro.
--- On Mon, 11/7/11, Andre Fischer<[email protected]> wrote:
Waiting, what about adopt a compromise, using the
patch I wrote (or a
better one) instead ? See :
http://ftp.educoo.org/home/ericb/patches/apache_ooo/configmgr_windows/configmgr_fixed.diff
From what I can see at a first glance, this fixes only the
symptoms but
not the root cause. That, however, is not a bad
thing, since the root
cause exists since before issue 112786 was fixed.
The config items still exist when the config manager is
destroyed. They
should have been removed before that. Their life time
control does not
work. Boost shared_ptrs might help here. The
owners of the config
items (in the two cases I investigated so far) use their
own, hand-made
reference counting, sometimes two layers deep.
Replacing this with
shared_ptrs or scoped_ptrs might fix this. But once
you start cleaning
up this old code you won't be able to stop.
Another question coming to my mind is now: can we
integrate the patch,
or must we write another solution ? (I tried to
discuss with Caolan on
IRC, but no answer yet)
Thanks in advance for any suggestion :-)
If your patch fixes the crash(es) then that is as good as
removing the
offending patch of issue 112786.
When we have a little time at our hands then we still
should fix the
life time control of the config items.
Regards,
Andre
Regards,
Eric
Note : I manualy added the changes, since the patch
Caolan provided uses
deep changes in cppuhelper, not directly compatible
with our code imho.
The orignal link of the fix is :
https://bugs.freedesktop.org/show_bug.cgi?id=31494
