remove bogus nearlyzero checks

remove bogus nearlyzero checks
Committed: https://code.google.com/p/skia/source/detail?r=4014

[reed1, 2012-05-21 16:32:24]

2 svg layouttest failures related to this (on linux)

svg/transforms/svg-css-transforms.xhtml
transforms/svg-vs-css.xhtml

Both cases show single pixel diff, attributable to this CL. These should be
rebaselined.

[epoger, 2012-05-23 15:32:23]

Why no reviewers on this CL that affects visible behavior?

[reed1, 2012-05-23 17:04:36]

On 2012/05/23 15:32:23, epoger wrote:
> Why no reviewers on this CL that affects visible behavior?

It did not affect visible behavior. The fix affected a new test case, the very
very long dash, hence the additional drawing added to this GM. This was in
response to a chrome bug with the same conditions. I saw no GM changes as a
result of this change, and only the 4 layouttests (out of many that exercise
dashing).

I am very interested in understanding what the chrome unittests are actually
producing, and will follow up on that when I'm back tomorrow (if vandebo doesn't
figure it out first).

[epoger, 2012-05-23 17:15:31]

On 2012/05/23 17:04:36, reed1 wrote:
> On 2012/05/23 15:32:23, epoger wrote:
> > Why no reviewers on this CL that affects visible behavior?
> 
> It did not affect visible behavior.

Well, the layout test differences seem to indicate that this *did* change
visible behavior.

But I shouldn't have muddied the water with the visibility aspect... why not get
changes reviewed *in general*?  Isn't code review a healthy thing in most
circumstances?

> I am very interested in understanding what the chrome unittests are actually
> producing, and will follow up on that when I'm back tomorrow (if vandebo
doesn't
> figure it out first).

If you are referring to http://crbug.com/129331 ('VectorCanvasTest.PathEffects
fails on WebKit Win (deps)'), I'm working on it. (Although so far, I have been
blocked on rather pedestrian issues around updating my Chrome tree.)