[webkit-qt] Inconsistent acceleration of linear gradients and atlas maps

[Jonathan Morton]

>>> There are several cases where the painter falls back to the raster
>>> engine.
>>
>> Is anyone at all looking into this yet?
>
> Nobody in Oslo is looking at it yet.
>
> You should start by adding a reduction to https://bugs.webkit.org/show_bug.cgi?id=41682 in order to get people interested.

AFAIK, a reduced test case is already attached to that bug.  We tried
to reduce it further, but found that such efforts made the
reproduction unreliable - the problem didn't go away entirely, but did
not always appear.

> You also mentioned you modified Qt, but you do not contribute the code back to Qt. So people have now way reproduce the problem you mention, and some contributors probably feel you take code but don't contribute back.

I believe that the atlas-maps case (which is attached to the bug) is
reproducible with stock Qt.  All of the cases that we're concerned
about are also reproducible when hacking an AllFeatures support
declaration into the X11 PaintEngine - which would normally be
expected to produce rendering artefacts but zero fallbacks, and
therefore maximum performance.

I had, however, forgotten that accurately reproducing gradient-based
test cases would require our patch to Qt.  Let me see what I can sort
out today.  We were already in favour of contributing this code via
our local Nokia contacts, on the grounds that it is incomplete - it
has known problems but shows valid techniques and is useful.

 - Jonathan Morton