[cairo] broken pixman bilinear filtering?

[Bill Spitzak]

Vladimir Vukicevic wrote:
> On 2/13/06, Bertram Felgenhauer <bertram.felgenhauer at googlemail.com> wrote:
> 
>>Vladimir Vukicevic wrote:
>>
>>>The attached test produces a png like the attached output; the first
>>>square is the 3x3 image scaled up 20x with bilinear filtering, and the
>>>second is the image scaled up with nearest neighbour.  It looks like
>>>for bilinear the center of the coverage region for the pixel at source
>>>0,0 is placed at 0,0 in the destination, when instead it should be
>>>offset by half the size of the destination area covered by a single
>>>pixel.
>>>
>>>I don't dare delve into pixman myself to look into this, I don't think :)
>>
>>the patch at https://bugs.freedesktop.org/show_bug.cgi?id=2488 should
>>fix that. (the attached picture was created using it)
> 
> 
> Ah, great!  I didn't think to check bugzilla first, I should have.  I
> notice that this bug exists in the latest version of fbcompose in the
> X server as well; this is unfortunate, as the patch was posted well
> before xorg 6.9 shipped =/  Keith (or davidr?  I'm not sure who's
> maintaining pixman/fb at the moemnt) should probably take a look at
> the patch and integrate it..

Is this going to change XRender? That may be a bad idea. I have worked 
around this bug, and I assumme many other people have, in XRender, by 
adjusting the transformation before doing any scaling of images. If this 
is fixed in XRender it will actually break these fixed programs. Another 
alternative would be a test to tell if the implementation is the old or 
new one.

Fixing Cairo or any other API that is not in common use yet is entirely 
appropriate, however.