[PyQt] Bug report: multiple QApplication instances cause a segfault

[Algis Kabaila]

On Thu, 23 Jun 2011 03:23:36 PM David Townshend wrote:
> On Thu, Jun 23, 2011 at 6:14 AM, Algis Kabaila <akabaila at pcug.org.au> wrote:
> > On Thu, 23 Jun 2011 08:17:04 AM Baz Walter wrote:
> > > On 22/06/11 20:49, David Townshend wrote:
> > > > On Wed, Jun 22, 2011 at 2:34 PM, Baz Walter<bazwal at ftml.net>  wrote:
> > > >> according to the qt docs, only one application object should be
> > 
> > created
> > 
> > > >> at a time, and there is a global qApp pointer which refers to the
> > > >> current instance.
> > > >> 
> > > >> this suggests that, in pyqt, something like the following should
> > > >> work:
> > > >> 
> > > >> from PyQt4 import QtGui
> > > >> QtGui.qApp = app = QtGui.QApplication([])
> > > >> app.quit()
> > > >> QtGui.qApp = app = None
> > > 
> > > [snip]
> > > 
> > > > I've investigated further and found that its caused by the gtk style
> > > > which I am using, since I am running gnome. Simply changing the qt
> > 
> > style
> > 
> > > > avoids the problem. I recall reading about similar issues before with
> > > > the gtk-qt style, so I think that this is a know problem.  However, I
> > > > noticed that after changing the style, python still segfaults on
> > > > exit. gdb indicates that it is something in the QApplication
> > > > destructor, and it seems that deleting the
> > > > 
> > > > instance after quitting solves this, e.g.:
> > > >>>> from PyQt4.QtGui import QApplication
> > > > 
> > > >>>> for i in range(10):
> > > > ...     app = QApplication([])
> > > > ...     app.quit()
> > > > ...     del app
> > > > ...
> > > > 
> > > >>>> exit()
> > > > 
> > > > I experimented a bit with setting qApp = None, but it didn't make a
> > > > difference.
> > > 
> > > my point was simply that you need to make sure you remove all
> > > references to the current qpplication instance before creating a new
> > > one.
> > > 
> > > in general, setting qApp to the instance of qapplication isn't
> > > necessary. but it's worth noting that qApp won't refer to the same
> > > *pyqt* object unless it's explicitly over-written. this may become
> > > relevant when working with subclasses of qapplication. of course, if
> > > you don't use qApp, it won't make any difference, as you say ;-)
> > 
> > Baz,
> > 
> > This does not explain why the IDLE quits so early in the test, even
> > before CR
> > is pressed... Actually, in python3.2 if one programs in IDLE, one finds
> > that
> > IDLE is useless with PyQt as it quits (just vanishes from the screen...)
> > if one tries to run a script after modifying it.  Basically, IDLE works
> > only if
> > it is invoked from scratch each time one wants to run a program.
> > 
> > I suggest that because of such misbehaviour it is not useful to look for
> > faults in the test case - there is a fault, either in Python3 IDLE,
> > Python 3,
> > PyQt, Qt itself.
> > 
> > As for the PyQt bug tracker - I was not able to find it, though I saw bug
> > reports in several places, though I said earlier that I did see it
> > somewhere.
> > 
> > I would write it down to my aging memory....
> > 
> > OldAl.
> > _______________________________________________
> > PyQt mailing list    PyQt at riverbankcomputing.com
> > http://www.riverbankcomputing.com/mailman/listinfo/pyqt
> 
> Baz - Sorry, I think I misunderstood you slightly, but I reached the same
> conclusion - to delete all references to the instance.  I don't think that
> omitting this should cause a segfault though, so it does look like a bug,
> but at least its easy to work around.
> 
> OldAl - I don't get that behaviour.  Is this is linux or windows, have you
> tried to get a backtrace to see where the problem is?  I have to admit I
> never use IDLE, I much prefer the console interface, so I haven't had much
> opportunity to see if this happens to me.
> 
> David
No backtrace possible - it's vanishing act of IDLE...  It only happens with 
IDLE for python 3 (3.2).  My os is kubuntu 11.04 "natty".  Sip and PyQt are 
compiled with python 3.2, but Qt is from debian (ubuntu) binaries.

As I could not make use of IDLE, I compiled eric5 from source.  It is a really 
nice IDE!  [press "(" gives (), press "," - adds a space after it - and the 
code completion is only the beginning!] - free sales pitch.  It is free as in 
beer and OSS.

Here are my system details:
ak at supremo:~$ lsb_release -a
No LSB modules are available.
Distributor ID: Ubuntu
Description:    Ubuntu 11.04
Release:        11.04
Codename:       natty
ak at supremo:~$ 
ak at supremo:~$ python3
Python 3.2 (r32:88445, Mar 25 2011, 19:56:22) 
[GCC 4.5.2] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> from PyQt4 import QtCore
>>> QtCore.PYQT_VERSION_STR
'4.8.4'
>>> QtCore.QT_VERSION_STR
'4.7.2'

The key point is that IDLE is for Python3.2 and sip, PyQt are compiled with 
Python3.2.

I have observed this kind of errant behaviour with PySide too.  (PySide does 
not yet have a Python3 version).

OldAl.