[PyKDE] Crash with unnotified event

[Phil Thompson]

On Wednesday 17 January 2007 3:52 pm, Giovanni Bajo wrote:
> On 1/17/2007 2:07 PM, Phil Thompson wrote:
> >> this script causes a segfault for me with latest SIP snapshot and PyQt
> >> 3.17:
> >>
> >> ================================================
> >> from qt import *
> >> app = QApplication([])
> >> k = QWidget(None)
> >> qApp.postEvent(k, QCustomEvent(QEvent.User))
> >> del app
> >> ================================================
> >
> > Is this a recent problem with current snapshots?
>
> This specific testcase crashes also with older versions of SIP and PyQt
> (4.4 and 3.16).
>
> I'm getting a serie of crashes, all related to QApplication destruction
> and event delivery, which never triggered before but now trigger with
> recent SIP snapshots. This is the first testcase reduction I made (they
> are pretty difficult to reduce!). I have probably at least another to
> report that I'm still reducing.
>
> > Is there a real use case (you'd never actually code the above)? Otherwise
> > just remove the "del".
>
> Yes, the real use case is an automatic testsuite. For each test, I need
> to create and destroy a new QApplication to start afresh. It kind of
> works in C++, but I have historically always had troubles with PyQt. A
> generic workaround for me right now is to put a qApp.processEvents()
> before app destruction, but I would prefer if the bugs were fixed (and I
> thought you preferred it too...).

I'm not convinced it's a bug given that you have no control on the order in 
which things get torn down from Python. Even if it is a bug, I'm not sure it 
is possible to fix it (for the same reason).

If you can't avoid the "del app" then preceding it with a "del k" also avoids 
the crash - which is just another way of imposing an order on the normal 
chaos that happens when the interpreter exits. That would suggest that if, in 
C++, you destroyed the QApplication before the QWidget you would also get the 
crash.

Phil