[PyKDE] RE: Slow memory leak when paintCell is overridden in QListViewItem

[Tuvi, Selim]

I re-tested against sip-snapshot-20041121 (also rebuilt PyQt 3.13) and I still observe the slow memory leak. With the paintCell override VM size increased from 3000k to around 11000k in about 5 minutes. This is about the same rate that I had observed with SIP 4.1 and PyQt 3.13, so nothing seems to have changed between the two SIP versions.

I also tried the latest SIP snapshot by not overriding paintCell and I did observe a very small memory increase initially but I am not sure if I can call it a leak. It increased from 2980k to 3032k after 5 minutes of running and kept that value afterwards (I ran it for about half an hour).

-Selim

> -----Original Message-----
> From: Phil Thompson [mailto:phil at riverbankcomputing.co.uk]
> Sent: Sunday, November 21, 2004 5:12 AM
> To: Tuvi, Selim
> Cc: pykde at mats.imk.fraunhofer.de
> Subject: Re: Slow memory leak when paintCell is overridden in
> QListViewItem
> 
> On Saturday 20 November 2004 2:40 am, Tuvi, Selim wrote:
> > Hi, we have a module which displays some statistics in a QListView. We
> > wanted to change how the list view looked like so we subclassed the
> > QListViewItem and implemented our own paintCell method.
> >
> > When we override the paintCell method, we observe a slow memory leak. If
> we
> > do not override the paintCell method then the memory leak does not
> occur.
> >
> > I am attaching a standalone version of the module that you can test.
> >
> > We are running under Windows XP, PyQt 3.13, SIP 4.1, Qt 3.3.3 and Python
> > 2.3.4 all compiled with Visual Studio .NET 2003.
> >
> > The memory leak can be observed under the task manager, Processes tab,
> > under the VM Size column.
> >
> > I haven't verified yet if this is a problem with Qt or PyQt or maybe it
> is
> > my code.
> 
> With SIP v4.0.1 I saw a very slow leak with your "working" code and a not
> so
> very slow leak with your "broken" code.
> 
> With the current SIP snapshot I saw a very slow leak with both versions.
> Subjectively I think the leak was slower than with v4.0.1.
> 
> I'll continue looking, but if you can try the snapshot I'd be interested
> to
> know if you think it improved.
> 
> Phil