Re: fixing gdk-pixbuf image loaders

[Larry Ewing <lewing ximian com>]

On 22 Jan 2001 19:32:57 -0500, John Ellis wrote:
> Havoc Pennington wrote:
> > 
> > Hi,
> > 
> > For GTK 2.0 we want the pixbuf image loaders to be robust against
> > corrupt image data. At the moment, some of them are completely broken
> > ("gee, I have a number from the network - let's dereference this array
> > with it!").
> 
> While on the topic of gdk-pixbuf, some of the scaling functions seem to have
> problems rendering when the offset is non-zero and the resulting render area of
> the source aligns with the right-side of the image.
> 

Yeah I reported this almost a full year ago with a very simple test
case.  No one has taken the time to really track it down yet.  Federico
and I spent a fair amount of time looking for the problem but never got
anywhere.

> (by problems I mean random pixels along the right side of the resulting image,
> and at worst, crashing)
> 
> With CVS gdk-pixbuf stable, I found this while working to make gdk-pixbuf sane
> with GQview:
> 
> gdk_pixbuf_scale: Broken when GdkInterpType is anything but GDK_INTERP_NEAREST
> gdk_pixbuf_composite: same ?
>

NEAREST used to crash with nonzero offsets, someone finally tracked that
one down. 

> gdk_pixbuf_composite_with_color: Only GDK_INTERP_TILES is broken, I suspect
> because the other render types are used in eog, they were fixed (eog does not
> have the TILES option).
> 
> Would be happy to help, but got lost trying to read the pixops.c file :/
> 

hopefully we can add this to the list of things to take care of in this
cleanup.

--Larry