1601898 - SVG breaks on Windows 10 with Hardware Accelaration

Status: RESOLVED WONTFIX

(Core :: Graphics, defect, P3)

Description

[Gerd Riesselmann]

Attachment: Shows the wrongly rendered page

User Agent: Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/78.0.3904.108 Safari/537.36

Steps to reproduce:

1.) On Windows 10, enable hardware accelaration
2.) Open https://audisto.com => SVG is broken

3.) Disable hardware acceleation
4.) Open https://audisto.com => SVG shows correctly

Actual results:

The SVG above the fold shows a grid that ends suddenly

Expected results:

The SVG's grid should blend into the backgroud.

It generally should show the same behavior as with FF 70 on Linux, MacOS and Windows with Hardware Acceleration disabled.

It should show the same result than other browsers (Chrome, Edge) on the same machine.

Comment 1

[Gerd Riesselmann]

Attachment: Shows the correctly rendered site (FF 70 on Linux)

Comment 2

[BugBot [:suhaib / :marco/ :calixte]]

Bugbug thinks this bug should belong to this component, but please revert this change in case of error.

Comment 3

[Jonathan Kew [:jfkthame]]

Given that it's specific to hardware-accelerated rendering on Windows, this seems likely to be more of a gfx issue -- moving to Graphics for investigation.

Comment 4

[Jeff Muizelaar [:jrmuizel]]

Gerd, is this a regression?

Comment 5

[Gerd Riesselmann]

Possibly, yes. We never had any complains nor issues when we implemented and tested this layout around two years ago. I must confess, though, this may well be due to not not properly testing on Windows...

Comment 6

[BugBot [:suhaib / :marco/ :calixte]]

Bugbug thinks this bug is a regression, but please revert this change in case of error.

Comment 7

[BugBot [:suhaib / :marco/ :calixte]]

The priority flag is not set for this bug.
:jbonisteel, could you have a look please?

For more information, please visit auto_nag documentation.

Comment 8

[Jessie [:jbonisteel] pls NI]

Will update priority if we can get a regression window and figure out what's up

Comment 9

[Gerd Riesselmann]

Since two weeks have passed since I filed this bug: It is possible, that someone will try to reproduce the steps to at least confirm the issue?

What's going to happen next seems rather vague to me at the moment. Who will do the outstanding progression testing? And who will try to figure out what's up? When?

Since hardware acceleration is default under Windows, this affects quite a lot of our users. We therefore won't leave it like this forever, but will try to work around this in the near future. Which at some point will make further testing impossible.

Wading through the open bugs related to SVG and/or hardware acceleration, I also think this bug is quite prototypical for a whole class of bugs.

Comment 10

[Daniel Bodea [:danibodea]]

I am attempting to confirm the issue, but I am not sure what issue I am looking for. I have rendered this webpage with and without hardware acceleration on Firefox Release v70.0, Firefox Release v71.0 and Nightly v74.0a1. No differences are being observed.
What do you mean by "The SVG above the fold shows a grid that ends suddenly."
Can you provide pictures with the actual and expected results for comparison?

Comment 11

[Gerd Riesselmann]

(In reply to Bodea Daniel [:danibodea] from comment #10)

I am attempting to confirm the issue, but I am not sure what issue I am looking for. I have rendered this webpage with and without hardware acceleration on Firefox Release v70.0, Firefox Release v71.0 and Nightly v74.0a1. No differences are being observed.
What do you mean by "The SVG above the fold shows a grid that ends suddenly."
Can you provide pictures with the actual and expected results for comparison?

I did already. A screenshot of the actual - wrongly rendered - result is already attached to the opening post: https://bugzilla.mozilla.org/attachment.cgi?id=9114103, a screenshot of the expected result is in comment #1: https://bugzilla.mozilla.org/attachment.cgi?id=9114105.

Comment 12

[Gerd Riesselmann]

[:danibodea]

I checked back with some of our users, who reported this to me (I do not have Windows myself).

• A user that is tied to version "68.3.0esr 64bit" (due to corporate policies) confirmed it
• Users with version 71 and higher reported the issue as not occurring any more

I therefore think, it has been (maybe accidentally) fixed.

Comment 13

[Daniel Bodea [:danibodea]]

Attachment: issue comparison.png

I finally understood what the issue was. I have made a screenshot comparing the affected build ESR v68.4.0esr ane the unaffected Firefox Release v72.0.

Furthermore, I can also confirm that the issue does not occur anymore on Beta v73.0b2 or Nightly v74.0a1.

Comment 14

[Daniel Bodea [:danibodea]]

The issue seems to be that the image used as a background in a few sections of the webpage is too visible on the affected build, when it should be vaguer / blend into the background.
I can also confirm the fact that it does not occur on the Ubuntu 18 or Mac OS 10.13.

Comment 15

[Jessie [:jbonisteel] pls NI]

Since we know what versions are broken and what are fixed, can you run Mozregression to get a regression range?

Comment 16

[Tobias Schwarz]

Nightly Builds:

2017-06-08 is good
2017-06-09 is bad

This fixed the Bug:

Bug 1524284. Enable WebRender by default on modern Intel desktop gpus. r=kats
This enables WebRender on a small subset of modern Intel gpus.
Differential Revision: https://phabricator.services.mozilla.com/D18242

Comment 17

[Daniel Bodea [:danibodea]]

I have regressed the fix and my result is the following:

2020-01-09T18:22:05: INFO : Narrowed inbound regression window from [1a3fc48f, 52cff0e3] (3 builds) to [8399ab60, 52cff0e3] (2 builds) (~1 steps left)
2020-01-09T18:22:05: DEBUG : Starting merge handling...
2020-01-09T18:22:05: DEBUG : Using url: https://hg.mozilla.org/integration/autoland/json-pushes?changeset=52cff0e398023169d4e13473d3de88e6da45fb92&full=1
2020-01-09T18:22:06: DEBUG : Found commit message:
Bug 1522422 - Fixing opacity issue on SVG USE elements in D2D. r=jrmuizel
Differential Revision: https://phabricator.services.mozilla.com/D44059
2020-01-09T18:22:06: DEBUG : Did not find a branch, checking all integration branches
2020-01-09T18:22:06: INFO : The bisection is done.
2020-01-09T18:22:06: INFO : Stopped

So it appears that this issue was fixed by bug 1522422.
The fix could be uplifted to firefox-esr68.

Furthermore, I attempted to regress the introduction of the issue and the result is the following:
Last known good build: Nightly v55.0a1
First known bad build: Nightly v60.0a1
The regression could not be finished because there are not enough builds to bisect.

Comment 18

[Daniel Bodea [:danibodea]]

It appears two different results have emerged. I hope the real regressor can be determined from the 2 cases. If the regression should be reattempted, please NI me.

Comment 19

[Jeff Muizelaar [:jrmuizel]]

I suspect this was originally regressed by bug 1359527 in FF 55.

Comment 20

[Jeff Muizelaar [:jrmuizel]]

We haven't had any issues with the fix in bug 1522422 so it may be worth uplifting to ESR