1231243 - CSS transform on element scaled larger than the viewport does not repaint when the transform is animated and the element is scaled down

Status: RESOLVED FIXED

(Core :: Layout, defect)

Description

[Harley Mellifont]

User Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.11; rv:45.0) Gecko/20100101 Firefox/45.0
Build ID: 20151208030212

Steps to reproduce:

Scale an element larger than the viewport down to fit within the viewport.

Here's a video of the problem:
https://vid.me/mAiha

And a codepen:
http://codepen.io/metaeaux/pen/vLEMBa

The problem occurs when layers.offmainthreadcomposition.async-animations set to true.

Some other notes:
- Moving the mouse seems to trigger a paint of the element and the white borders disappear (as seen in the video)
- When the viewport is around 1920px wide (or larger) the problem doesn't seem to occur


Actual results:

The edges that exceeded the viewport size are painted white rather than painted with the content of the element


Expected results:

The content of the element should be painted.

Comment 1

[Harley Mellifont]

It seems to be related to this issue:
https://bugzilla.mozilla.org/show_bug.cgi?id=929362

It seems like a regression since the issue status is fixed.

Comment 2

[Mardeg]

Attachment: zoom out and transform testcase

Confirmed this behaviour in the latest nightly build

Comment 3

[David Baron :dbaron: (⌚️UTC-5, no longer working on Mozilla)]

Interestingly, I see this on Linux but don't see this on Windows.

Comment 4

[Markus Stange [:mstange]]

This is a regression from the nsDisplayBackgroundImage::GetBoundsInternal changes from bug 1166301.

Comment 5

[Botond Ballo [:botond]]

I will investigate.

Comment 6

[Botond Ballo [:botond]]

(In reply to Markus Stange [:mstange] from comment #4)
> This is a regression from the nsDisplayBackgroundImage::GetBoundsInternal
> changes from bug 1166301.

Confirmed, backing out the change to nsDisplayBackgroundImage::GetBoundsInternal() fixes the problem. Nice call!

Comment 7

[Botond Ballo [:botond]]

Note that, while this is a regression from bug 1166301, the issue only occurs with e10s and APZ enabled, so Firefox 45 and below are unaffected.

Comment 8

[Botond Ballo [:botond]]

The problem here is that bug 1166301 changed nsDisplayBackgroundImage::GetBoundsInternal() to return the bounds of the viewport rather than the bounds of the image if APZ is enabled. The intention was to return a size *no smaller* than the viewport size; we also want to return a size no smaller than the bounds of the image, so to handle cases like this where the image is larger than the viewport, we should be taking the union.

Comment 9

[Botond Ballo [:botond]]

Attachment: MozReview Request: Bug 1231243 - In nsDisplayBackgroundImage::GetBoundsInternal(), take the union of the image bounds and the viewport bounds if APZ is enabled. r=mstange

Review commit: https://reviewboard.mozilla.org/r/29481/diff/#index_header
See other reviews: https://reviewboard.mozilla.org/r/29481/

Comment 10

[Markus Stange [:mstange]]

Comment on attachment 8703832 [details]
MozReview Request: Bug 1231243 - In nsDisplayBackgroundImage::GetBoundsInternal(), take the union of the image bounds and the viewport bounds if APZ is enabled. r=mstange

https://reviewboard.mozilla.org/r/29481/#review26673

Comment 11

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/d15b9b50fbc4

Comment 12

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/d15b9b50fbc4

Comment 13

[Liz Henry (:lizzard) (privacy/fingerprinting team)]

Tracking for 46 since this is a regression - so if it reopens we'll notice it.