Change tilegrid test to test it directly, rather than through SkPicture

Change tilegrid test to test it directly, rather than through SkPicture
This is necessary because it makes assumptions that picture will draw all the rects that match the grids, which may not hold if picture decides to improve the accuracy of the results.

BUG=skia:2125
Committed: http://code.google.com/p/skia/source/detail?r=13827

[mtklein, 2014-03-16 17:27:20]

lgtm

[tomhudson, 2014-03-17 09:48:46]

LGTM, but before I CQ it, can't we now delete class MockCanvas?

At which point:

Nit: commit message isn't quite clear - what do you mean "if picture decides to
improve the accuracy of the results"?

[ian_cottrell, 2014-03-17 09:52:49]

On 2014/03/17 09:48:46, tomhudson wrote:
> LGTM, but before I CQ it, can't we now delete class MockCanvas?
> 
> At which point:
> 
> Nit: commit message isn't quite clear - what do you mean "if picture decides
to
> improve the accuracy of the results"?

The test is checking that exactly the rects that fall into tiles that intersect
the canvas end up being rendered, even when the rects themselves fall outside
the rendering bounds. This is the behaviour of the tilegrid, but the picture
layer in between could easily decide to cull the rects before render, it has no
contract that implies it would not, and in fact, that is exactly the change I
made that made me notice this test was flawed. I kind of think that is a flawed
test anyway, as it's checking an implementation detail not a contract, but I did
not want to change that as well.

[mtklein, 2014-03-17 09:58:33]

>I kind of think that is a flawed
> test anyway, as it's checking an implementation detail not a contract, but I
did
> not want to change that as well.

Just FYI, this is working-as-intended under tests/.  Generally, tests/ tends to
be a mix of both white box, black box, and regression tests.  The downside is
that tests sometimes become irrelevant or wrong, but usually is a way that's
obvious to whoever's changing the code.  Upside is, delete or change tests as
they decay!

[iancottrell, 2014-03-17 10:30:39]

On 2014/03/17 09:58:33, mtklein wrote:
> >I kind of think that is a flawed
> > test anyway, as it's checking an implementation detail not a contract, but I
> did
> > not want to change that as well.
> 
> Just FYI, this is working-as-intended under tests/.  Generally, tests/ tends
to
> be a mix of both white box, black box, and regression tests.  The downside is
> that tests sometimes become irrelevant or wrong, but usually is a way that's
> obvious to whoever's changing the code.  Upside is, delete or change tests as
> they decay!

And MockCanvas is gone.

[tomhudson, 2014-03-17 10:31:43]

lgtm

[commit-bot: I haz the power, 2014-03-17 10:31:45]

CQ is trying da patch. Follow status at
https://skia-tree-status.appspot.com/cq/iancottrell@google.com/199083004/20001

[commit-bot: I haz the power, 2014-03-17 10:51:46]

Change committed as 13827