[webkit-dev] How to interpret repaint test results?

[Xianzhu (Drew) Wang 王显著]

On Thu, Feb 9, 2012 at 10:33 AM, Simon Fraser <simon.fraser at apple.com>wrote:

> On Feb 9, 2012, at 7:03 PM, Xianzhu (Drew) Wang 王显著 wrote:
>
> I searched code and found that the repaint rects are only enabled and used
> on Mac. Chromium just paint the gray mask in layoutTestController.display()
> and let the later repaints to clear the mask automatically. Haven't check
> how other platforms do, but I think the different implementations cause
> different pixel expectations and extra efforts of maintaining them.
>
> How about just dumping the repaint rects as text?
>
>
> I'd suggest adding layoutTestController.repaintRectsAsText() or something,
> so that tests can do whatever they like with the data.
>
> I still think that repaint rects should show in pixel results, because
> that makes it much easier to spot correct and incorrect behavior.
>
>
Agreed. Then I'd suggest using fast/repaint/resources/repaint.js to draw
the repaint rects based on the data, which can avoid different
implementations of platforms, many rebased pixel results and interference
of scrolling.

> Moreover, based on this we could also dump the repaint rects of composited
> layers so make some tests feasible (for example,
> https://bugs.webkit.org/show_bug.cgi?id=75638).
>
>
> Yes, that would help.
>
> Simon
>
>
> On Wed, Feb 8, 2012 at 11:55 PM, Simon Fraser <simon.fraser at apple.com>wrote:
>
>> On Feb 9, 2012, at 12:16 AM, Xianzhu (Drew) Wang 王显著 wrote:
>>
>> > I'm confused with the expected results of some repaint tests, for
>> example, fast/repaint/fixed-scroll-simple.html. The expected pixel result
>> (platform/mac/fast/repaint/fixed-scroll-simple-expected.png) is all masked
>> by dark gray. Does this mean that no part of the page is repainted?
>> However, the test seems to expect that some part of the page is repainted
>> so that the red box is fully covered by the green box and is not left on
>> the page. What did I miss?
>>
>> Repaint tests now make use of code in
>> FrameView::repaintContentRectangle(), which records which rectangles were
>> repainted, but only via calls to this method.
>>
>> I'm guessing that scrolling does invalidate, but not via
>> FrameView::repaintContentRectangle().
>>
>> Simon
>>
>>
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.webkit.org/pipermail/webkit-dev/attachments/20120209/4a024fbd/attachment.html>