Flate image and r1083565 (Was: Re: Walkers mass-test - help needed)

[Cyrille Berger]

On Tuesday 02 February 2010, Boudewijn Rempt wrote:
> On Monday 01 February 2010, Dmitry Kazakov wrote:
> > http://dimula73.narod.ru/walkers_singlepatch_v5.diff<http://dimula73.naro
> >d. ru/walkers_singlepatch_v4.diff>
> >
> > o Fixed undo after Flatten Image.
> > o Fixed refresh after creation of the layer
> >
> > o Reverted Cyrille's commit r1083565 to eliminate crash
> 
> For this, I think you should open a bug: "memory leak on document
>  destruction" with the history of this patch. We should fix the memory
>  leak, but that might be hard.
> 
Yes and I am really sceptical that that change has anything to do. The 
destructor is only deleted by the document. To make sure of that I added a 
qFatal in the destructor, and did:
* open krita
* close krita
* qFatal triggered

And then I did
* open krita
* add layer
* flaten image
* undo (twice, not sure why flatten image is not grouped anymore)
* assert somewhere else

So in my case the destructor of KisShapeController (which is the only thing 
changed by r1083565) is only called on document destruction at krita exit, and 
not during flaten image. So can someone that can reproduce the flatten image bug 
with change r1083565 and not r1083565, add a qFatal("something");, in 
KisShapeController::~KisShapeController, and send me the backtrace that goes 
to the destructor ? (alternatively, you can use a break point in gdb, but I do 
tend to not find them trustworthy).

-- 
Cyrille Berger