[PySide] random crashes in pyside during signals

[Joel B. Mohler]

On 4/26/2013 1:13 PM, Luc-Eric Rousseau wrote:
> On Wed, Apr 24, 2013 at 7:34 PM, John Ehresman <jpe at wingware.com> wrote:
>> Do you have a test case?  Do you have a subclass of QCheckBox?  The
>> immediate cause is that mo in SignalManager::retriveMetaObject ends up
>> being NULL, but the bug probably occurred earlier.
>>
>> Thanks,
> I've been trying to isolate in a smaller test case and I haven't been
> able to yet.
> There was mention of a checkbox in that callstack, however in other
> times it's elsewhere (for example as below). I do not have a subclass
> of QCheckBox, it might have been  the checkbox part of QT's
> QListWidget if there was one.  The way I'm reproducing the crash 100%
> of the time is to click a push button a few times fast that changes
> the selection in a QListBox, which then causes another pane to refresh
> with new controls. However, if you don't do that, every once in a
> while you can click another control elsewhere and *poof*, you'll crash
> with the same call stack. It's just harder to reproduce an other way.
> So I've come to the conclusion that there isn't anything wrong
> specially with that QListWidget UI, it's just much easier to reproduce
> it there.

I'm not sure how helpful this is, but I add an anecdote.  I reported 
both https://bugreports.qt-project.org/browse/PYSIDE-160 and 
https://bugreports.qt-project.org/browse/PYSIDE-144 which are memory 
issues of some sort.  Both resulted in crashes which seem similar to 
yours.  The only way I found them was by eliminating code paths bit by 
bit until I found the crucial element.  Both manifested most repeatedly 
after calls to QLayout::takeAt to replace widgets in views related to a 
list.  I'm assuming this is because the shiboken pointer lists become 
corrupt and takeAt and subsequent garbage collection aggravated that 
corruption.

Joel