more persp fixes and gm

[bsalomon, 2011-09-12 18:11:43]

Looks like the raster code could use some perspective TLC, too. There's a weird
banding of a gradient radius.

[bsalomon, 2011-09-13 13:54:51]

On 2011/09/12 18:11:43, bsalomon wrote:
> Looks like the raster code could use some perspective TLC, too. There's a
weird
> banding of a gradient radius.

ping

[reed1, 2011-09-13 14:05:04]

Someday lets formalize some silent-error callback trick, so we can send the
client (if they want them) errors and warnings like "can't invert matrix"

With lazy nit: LGTM

http://codereview.appspot.com/4994048/diff/8001/gpu/src/GrContext.cpp
File gpu/src/GrContext.cpp (right):

http://codereview.appspot.com/4994048/diff/8001/gpu/src/GrContext.cpp#newcode620
gpu/src/GrContext.cpp:620: if (p != &tmpPaint) {
there exists a slightly fast way to do this lazy paint pattern.

SkTLazy<SkPaint> tmpPaint;  // we do no work at all at this stage
if (...) {
    p = tmpPaint.set(paint);  now p points to tmpPaint, initialized by paint
}

[bsalomon, 2011-09-13 20:05:25]

Checked in with lazy init at r2256,7.