[Pixman] [PATCH 1/2] Remove the 8e extra safety margin in COVER_CLIP analysis

[Ben Avison]

On Fri, 11 Sep 2015 12:30:23 +0100, Pekka Paalanen <ppaalanen at gmail.com> wrote:

> As samplers are allowed to fetch the pixel at x2 unconditionally, we
> require
>     x1 >= 0
>     x2 < width

I may be getting picky, but that's circular logic - the samplers are only
allowed to fetch the pixel at x2 unconditionally because of the way
BILINEAR_COVER_CLIP is defined in this piece of code, so you can't use it
as its own justification. As I wrote the commit log originally, the
premise is that some samplers expect to be able to fetch the pixel at x2
unconditionally (on at least one axis) and the conclusion is that we need
to define BILINEAR_COVER_CLIP to allow that (for now).

How about:
Because samplers may fetch the pixel at x2 unconditionally, we require...

Other than that, the series looks good - good thinking about the impact
of projective transforms (though maybe someday that could do with a more
rigorous examination).

Also good to see Bill's way of phrasing the ultimate aim - that it'll be
defined to be safe to fetch all pixels that have a non-zero weight - made
it into patch 2. I think it's a clear and concise description.

I had to look up what bikeshedding meant since you've used it a couple of
times - very apt. I'll have to drop it into conversation sometime :)

Ben