Comment 272 for bug 217908

kenjo wrote:
> On Wed, 2009-11-18 at 01:57 +0000, Tom Jaeger wrote:
>>> First could someone explain exactly what feature in the graphics driver
>>> people think is wrong.
>> Many graphics drivers didn't use to accelerate composite operations with
>> repeat type PAD (and REFLECT) correctly. This was a year ago; i believe
>> these issues have long been fixed.
>
> not here I still have really bad scale problems with firefox.

This is not a driver problem, the driver renders exactly as firefox
instructs it to.

>>> Could a small test program be written to show the error ?? has maybe
>>> someone written one already.
>> There is a very simple test case attached to this very bug report. Its
>> source surface is only 2x2 (IIRC), which might not hit an accelerated
>> path on every driver;
>
> hmm I have attached the scale_error.tar.gz and that is simply a example
> html page with a image scale. what I wanted as example was C code that
> only uses X to provoke the problem.

comment #33

>
>> there's a test case with a larger source surface
>> attached to the -ati driver bug report.
>
> where is this ?? I cant find it.
At the top of the page:

https://bugs.freedesktop.org/show_bug.cgi?id=19712

>>> This needs to get solved one way or the other having a fix outside the
>>> normal distribution is not acceptable. but working around the problem in
>>> only firefox obviously is not the correct fix.
>> No, firefox is the correct place to fix this issue. Firefox asks for
>> nearest-neighbor interpolation because on unpatched cairo 1.8.6,
>> bilinear filtering forces a slow client-side fallback to insure correctness.
>>
> but if it's a driver bug there would be nothing to fix in firefox.
> if it is a firefox bug on the other hand the yes obviously

As explained many times before, firefox disables bilinear filtering due
to hitting a slow path in cairo. The underlying issues are fixed now,
but firefox continues to use the same workaround that was needed before.