Bug 51119 – REGRESSION: Element with rotation does not respond to mouse events after r73385

• Component: Layout and Rendering
• Status: RESOLVED FIXED

[Mihai Parparita]

Test case: http://persistent.info/webkit/test-cases/crbug-67004.html

The third div (which has -webkit-transform: rotate(180deg)) does not turn green when hovered over. 

This happened between r73372 and r73408, r73385 is the most likely cause. Reported on the Chromium side as http://crbug.com/67004

[Simon Fraser (smfr)]

Created attachment 76667 [details]
Testcase

[Simon Fraser (smfr)]

<rdar://problem/8772995>

[Dave Hyatt]

Created attachment 76674 [details]
Patch

[Mihai Parparita]

Seems like this could have a layout test (using eventSender).

[Simon Fraser (smfr)]

elementFromPoint() could be used to test, without any need for events.

I would like to understand the rounding issues, and why they are new.

[Dave Hyatt]

They aren't new.  The old code used to use the entire layer bounds as the hit test area.  This area was a rect of 0, 0, rightOverflow(), bottomOverflow().  Since that isn't right for RTL documents or for vertical writing modes that could have left and top overflow, I opted to try to tighten up the hit test area to just be the point.

Tightening up to a 1x1 rect though exposed rounding issues with the mapping into localHitTestRect in hitTestLayer.

The patch is basically going back to what the old code did, but also incorporating left and top overflow into the hit test area.

[Dave Hyatt]

Comment on attachment 76674 [details]
Patch

I'll add a test.

[Simon Fraser (smfr)]

I think the problem is that

            localHitTestRect = newTransformState->mappedQuad().enclosingBoundingBox();

maps the point outside the 1x1 hitTestRect when there's a flip.

[Simon Fraser (smfr)]

Specifically, say I hover over the lower right corner of a 200x200 box. The local point is 195, 195, and the hitTestRest is a  195,195 1x1 rectangle.

When that rectangle is mapped through the 180deg rotation, it becomes a 5, 5 -1x-1 rectangle, which gets bounded as 4,4 1x1. So the point at 5,5 is no longer inside.

[Simon Fraser (smfr)]

I agree with the attached patch; mapping a point inside a 1x1 rectangle through transforms is not going to work.

[Erik Arvidsson]

Created attachment 77561 [details]
Test case with mouse event

[Beth Dakin]

(In reply to comment #2)
> <rdar://problem/8772995>

See also: <rdar://problem/8783620>

[Dave Hyatt]

Created attachment 78453 [details]
LayoutTests portion.

[Simon Fraser (smfr)]

Comment on attachment 78453 [details]
LayoutTests portion.

View in context: https://bugs.webkit.org/attachment.cgi?id=78453&action=review

> LayoutTests/fast/transforms/transform-hit-test-flipped.html:4
> +<meta charset="utf-8" />

Whu?

> LayoutTests/fast/transforms/transform-hit-test-flipped.html:5
> +<title>Untitled</title>

Double whu?

> LayoutTests/fast/transforms/transform-hit-test-flipped.html:23
> +   var elt = document.elementFromPoint(50, 50);

I'm pretty sure document.elementFromPoint() will update layout.

> LayoutTests/fast/transforms/transform-hit-test-flipped.html:37
> +</html>
>  \ No newline at end of file

Boo!

[Beth Dakin]

I committed Hyatt's patch with: http://trac.webkit.org/changeset/75472

[WebKit Review Bot]

http://trac.webkit.org/changeset/75472 might have broken GTK Linux 64-bit Debug
The following tests are not passing:
media/adopt-node-crash.html

[Andy Estes]

*** Bug 51808 has been marked as a duplicate of this bug. ***