[webkit-dev] "No red pixels below" automation

[Konstantin Tokarev]

Yet another example: fast/layers/scroll-with-transform-composited-layer.html

"All the boxes should only contain green content."

Such tests would be harder to grep.

16.07.2016, 19:30, "Konstantin Tokarev" <annulen at yandex.ru>:
> 16.07.2016, 18:56, "Darin Adler" <darin at apple.com>:
>>>   On Jul 16, 2016, at 8:45 AM, Konstantin Tokarev <annulen at yandex.ru> wrote:
>>>
>>>   For example, I'm looking at fast/canvas/shadow-offset-* now. They are pixel tests now.
>>
>>  The simplest efficient way to turn these into reference tests that check that no red is visible would be by having the test function in shadow-offset.js file take an argument to tell it whether it’s the reference or the test. For the reference, it would skip the step where we fill with red (skip the fillRect that happens just after setting fillStyle to red). Then the test would fail if any red shows.
>>
>>  The downside of switching to this is that we’d no longer have pixel tests, and so no longer be testing anything else other than that the red square is obscured. If our initial premise is that we’re not making real use of pixel test results, then we shouldn’t worry about preserving these tests that we are not using.
>>
>>  With a bit more work we could make better reference tests that would truly fully replace the pixel tests; we would have to write alternate code to generate each of the reference images without using shadows. I think that would likely be relatively straightforward, but it would require a bit more thought about drawing multiple rectangles. Should be easy for the two tests with platform independent results (1 and 2), but harder for the ones that involve blurs, because we’d have to find some other way to render the blurs.
>
> In attachment there is a list (maybe incomplete) of layout tests containing text "no red", which are not ref tests.
>
> --
> Regards,
> Konstantin
> ,
>
> _______________________________________________
> webkit-dev mailing list
> webkit-dev at lists.webkit.org
> https://lists.webkit.org/mailman/listinfo/webkit-dev

-- 
Regards,
Konstantin