Note

• This issue is reproducible on Windows 10 (x64) with ATI Radeon HD 3000 graphics.
• It is not reproducible on system with Intel HD 630.

Affected versions

• Firefox 66.0.2
• Firefox Beta 67.0b8
• Firefox Nightly 68.0a1

Affected platforms

• Windows 10 (x64)

Steps to reproduce

1. Start Firefox with a clean profile
2. Open this link: http://www.alpacaprojects.com/inferno/en/
3. Click on start and select one of the selectable characters to zoom in

Expected result

• There is no rendering issues after zooming in

Actual result

• There is a flash after zooming in, and seeking through the characters with the arrows results in a flashing (see the attachment)

Regression range

• First bad: (20171112100435) aabfc14671b55983e1c3053989a4c3b7c5691aaa
• Last good: (20171113100232) fc194660762d1b92e1679d860a8bf41116d0f54f
• Pushlog: https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=aabfc14671b55983e1c3053989a4c3b7c5691aaa&tochange=fc194660762d1b92e1679d860a8bf41116d0f54f
• Potential regressor:
• This is a manual regression as mozregression would skip every build between 2017-11-12 and 2017-11-13.

Additional notes

• If the pref layers.acceleration.disabled is set to true, and hardware acceleration is disabled, the issue is no longer reproducible but while seeking through the characters, a thin grid will appear and disappear after instead of the flashing.

Thank for the report Catalin! Are you sure about that regression range, however? I am testing on Linux rather than windows, but for me https://hg.mozilla.org/integration/mozilla-inbound/rev/0901f80a81db is when things turned bad.

I'm seeing the following warnings a lot:
[GFX3-]: Surface size too large (exceeds allocation limit)!
[GFX3-]: Surface size too large (exceeds extent limit)!

So my guess is when we move perhaps the transform being animated causes us to make a massive active layer. This attempts to allocate a massive surface, which fails, so there's a blank flash. Then when we become steady again we have a normal sized layer and paint correctly.

Hi Jamie, I tried finding the exact regression under Windows 10, and in the Nightly from 2017-11-12 the issue is not reproducing, but on the 13th, the issue was there. I couldn't do a regression between those two respective days, as mozregression would skip all the builds between and would result in a error.

During testings, I have seen the issue only on Windows 10, on the same system.

Bulk change of P3 carryover bugs to wontfix for 68.

Set release status flags based on info from the regressing bug 828240

Set release status flags based on info from the regressing bug 828240