Fixed GrContext.cpp Windows compiler complaints

In all cases I attempted to just make explicit what was already happening rather than changing the behavior (e.g., using SkScalarCeilToInt on the width & heights). None of the new asserts fired during the execution of gm, sampleapp and bench.

[bsalomon, 2012-05-08 14:22:55]

On 2012/05/08 13:48:47, robertphillips wrote:

LGTM

[Stephen White, 2012-05-09 15:32:37]

Peanut gallery:  These warnings feel like overkill to me.  Also, if they're only
enabled on one platform, people committing from the other platforms are likely
to introduce them inadvertently.  Out of curiosity, have they caught actual
bugs?

http://codereview.appspot.com/6198051/diff/1/src/gpu/GrContext.cpp
File src/gpu/GrContext.cpp (right):

http://codereview.appspot.com/6198051/diff/1/src/gpu/GrContext.cpp#newcode2096
src/gpu/GrContext.cpp:2096: GrAssert(SkScalarIsInt(srcRect.fLeft));
I think it would be better to create an SkIRect from the scaled-down srcRect
here (call it srcIRect?), and use that in all the places that expect an int
value.  Then we wouldn't need to repeatedly call SkScalarFloorToInt() on it.

[robertphillips, 2012-05-09 22:45:01]

W.r.t. the warnings, I think we have invested a lot of effort getting the Linux
port to compile cleanly but have neglected the Windows port a bit. I think
cleaning them up has clarified the developer's intent in many situations but it
is unlikely doing so has fixed any existing bugs. I don't think we just want to
turn off the warning (i.e., warning C4244: 'argument' : conversion from <X> to
<Y>, possible loss of data) since it could be indicative of a real problem in
future code.

http://codereview.appspot.com/6198051/diff/1/src/gpu/GrContext.cpp
File src/gpu/GrContext.cpp (right):

http://codereview.appspot.com/6198051/diff/1/src/gpu/GrContext.cpp#newcode2096
src/gpu/GrContext.cpp:2096: GrAssert(SkScalarIsInt(srcRect.fLeft));
On 2012/05/09 15:32:37, Stephen White wrote:
> I think it would be better to create an SkIRect from the scaled-down srcRect
> here (call it srcIRect?), and use that in all the places that expect an int
> value.  Then we wouldn't need to repeatedly call SkScalarFloorToInt() on it.

Done.

[Stephen White, 2012-05-10 14:00:25]

http://codereview.appspot.com/6198051/diff/6001/src/gpu/GrContext.cpp
File src/gpu/GrContext.cpp (right):

http://codereview.appspot.com/6198051/diff/6001/src/gpu/GrContext.cpp#newcode...
src/gpu/GrContext.cpp:2096: SkIRect srcIRect = SkIRect::MakeLTRB(
This could be:

SkIRect srcIRect;
srcRect.roundOut(&srcIRect);

(or round(), equivalent in this case).

[robertphillips, 2012-05-10 14:48:31]

http://codereview.appspot.com/6198051/diff/6001/src/gpu/GrContext.cpp
File src/gpu/GrContext.cpp (right):

http://codereview.appspot.com/6198051/diff/6001/src/gpu/GrContext.cpp#newcode...
src/gpu/GrContext.cpp:2096: SkIRect srcIRect = SkIRect::MakeLTRB(
Done - much cleaner.

[Stephen White, 2012-05-10 14:52:56]

LGTM

[robertphillips, 2012-05-10 15:11:52]

committed as r3895