1661869 - Firefox Screenshots Fail to Capture Canvas Elements

Status: VERIFIED FIXED

(Core :: Graphics, defect, P2)

Description

[artemx100]

User Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:82.0) Gecko/20100101 Firefox/82.0

Steps to reproduce:

1. Visit https://teemo.gg/model-viewer?skinid=nami-15&model-type=champions
2. Scroll down and try to take a screenshot of the canvas element

Actual results:

The canvas element is invisible in the screenshot

Expected results:

The canvas element should appear in the screenshot.

Comment 1

[artemx100]

For what it matters, I tested and it works properly in Firefox 66.0.5, however it does not work in Firefox 79.0, meaning the regression happened somewhere between those two versions.

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

[Alice0775 White]

I can reproduce the problem on Nightly82.0a1 Win10.

Regression window:
https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=4879dea5cdd64080b5ae0c6c877bd87d76d06fda&tochange=7ba2f2cf6bdedd2cafd4e8717fa548e5b54dd184

Comment 4

[Julien Cristau [:jcristau] (back April 22)]

Jeff can you take a look?

Comment 5

[Jens Stutte [:jstutte]]

Hi Kris, can you set the severity for this bug, please?

Comment 6

[Mike Taylor [:miketaylr]]

ni? emma to put on your radar.

Comment 7

[Emma Malysz]

Thanks for the regression window, Alice!

Comment 8

[Jim Mathies [:jimm]]

Wondering if this should be higher than P3.

Comment 9

[Gian-Carlo Pascutto [:gcp]]

Any particular reason? The regressor identified dates back to July 9, and this wasn't found until 2 months later, so it doesn't look like this is frequently hit. Maybe we can make an argument that it should go before trying to fix the perf issues, perhaps.

Comment 10

[Emma Malysz]

Hey :artemx100, thanks for reporting this. Are you still seeing this issue?

I believe this may have been fixed by Bug 1660651, as I can no longer reproduce

Comment 11

[artemx100]

(In reply to Emma Malysz from comment #10)

Hey :artemx100, thanks for reporting this. Are you still seeing this issue?

I believe this may have been fixed by Bug 1660651, as I can no longer reproduce

Yes, I still can. The issue is still unsolved in the most recent nightly build.
Tested with both my current and a fresh profile.
Version 84.0a1
Build ID 20201103095421

Comment 12

[Emma Malysz]

Thanks! It looks like this issue does not affect Mac, but I was able to reproduce on Windows

Comment 13

[Ryan VanderMeulen [:RyanVM]]

This feels like a regression that should be getting on someone's radar rather than slipping release after release. Any chance we can revisit the prioritization here?

Comment 14

[Ryan VanderMeulen [:RyanVM]]

Attachment: screenshot

This still trivially reproduces for me on any WebGL content on Windows 10, even simple content like the MDN page below:
https://developer.mozilla.org/en-US/docs/Web/API/WebGL_API/By_example/Clearing_with_colors

Jim, does a simpler testcase make this easier to prioritize?

Comment 15

[Jim Mathies [:jimm]]

WebGL related, gcp and JeffG know about this.

Comment 16

[Jim Mathies [:jimm]]

I just noticed this bug impacts taking screenshots of Google Maps. Seems kinda bad. Do we have any idea what's broken here and how to fix it? I might be able to find someone to work on it if JeffG is too busy.

Comment 17

[Gian-Carlo Pascutto [:gcp]]

Jim, I think Jeff tried to debug this, and then ran into the bug this depends on, i.e. that the feature is completely broken in debug mode, which complicated things. I think he was going to look again when that gets fixed...but we can't seem to get traction in that team.

If you feel it's urgent and have availability, feel free to take it.

Comment 18

[Gian-Carlo Pascutto [:gcp]]

Hmm, there's at least a WIP patch in the other bug now.

Comment 19

[Emma Malysz]

I can revisit the patch in Bug 1685388.

Comment 20

[Gian-Carlo Pascutto [:gcp]]

Thank you very much Emma! That will probably help whoever ends up tackling that bug.

Comment 21

[Jim Mathies [:jimm]]

I ran into this today taking a screenshot of google maps. I'd like to try and figure out what's going wrong here and get it fixed.

Comment 22

[Evgeniy Karyakin]

Reproducible in 91.0.1 x64. Check Google Maps or, for example, Johns Hopkins University's COVID-19 statistics page.
https://www.arcgis.com/apps/dashboards/bda7594740fd40299423467b48e9ecf6
Unfortunately, cannot tell exactly when it started but a few releases ago (~ v85.x) it worked well.

Comment 26

[Ryan VanderMeulen [:RyanVM]]

This is not an S3 bug if it affects major sites like Google Maps. We need to bump this up in priority and figure out a path forward.

Comment 27

[Emma Malysz]

Since I believe this is an issue with browser.tabs.captureTab, I want to point out that the API is also used to report a site issue (https://searchfox.org/mozilla-central/rev/4d90ff4537330d6b17cb956c0fadf759086d9bb7/browser/extensions/report-site-issue/background.js#106-109).

Maybe triggering from that flow will unblock this bug from Bug 1685388.

Regardless, I'm going to be prioritizing the screenshots work. Waiting on another debug build, hopefully this time I'll be able to reproduce

Comment 28

[Emma Malysz]

We weren't able to reproduce the issue in Bug 1685388. Provided no one is having the issue with screenshots in a debug build, I think this bug is considered unblocked and can be worked on

Comment 29

[Kelsey Gilbert [:jgilbert]]

Here's an instant-loading example that screenshotting fails on: https://jdashg.github.io/misc/webgl/ping-pong.html

Comment 30

[Kelsey Gilbert [:jgilbert]]

Attachment: Bug 1661869 - Fix screenshotting of WebGL canvases.

For WebGL canvases with preserveDrawingBuffer:false, frontbuffer
snapshotting was not working properly.
If the frontbuffer had no user-fb associated with it,
was inexplicably calling BindDefaultFBForRead, which binds
webgl's default FB, not gl's default FB 0.
This caused frontbuffer snappshotting to actually snapshot the (cleared)
backbuffer instead of the visible frontbuffer contents.

Comment 31

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/6b9c61565d44
Fix screenshotting of WebGL canvases. r=gfx-reviewers,lsalzman

Comment 32

[Atila Butkovits]

https://hg.mozilla.org/mozilla-central/rev/6b9c61565d44

Comment 33

[Ryan VanderMeulen [:RyanVM]]

Is there any way we can land a test for this? Also, please nominate for Beta and ESR91 approval when you get a chance.

Comment 34

[Kelsey Gilbert [:jgilbert]]

Yeah, in two stages I think:

1. Test the related functionality via canvas.toDataURL()
2. Add actual Screenshots test here: https://searchfox.org/mozilla-central/source/browser/extensions/screenshots/test/browser/browser.ini

Comment 35

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

The patch landed in nightly and beta is affected.
:jgilbert, is this bug important enough to require an uplift?
If not please set status_beta to wontfix.

For more information, please visit auto_nag documentation.

Comment 36

[Pascal Chevrel:pascalc]

Jeff, are you planning an uplift into our last beta or should we mark it wontfix for 93? Thanks

Comment 37

[Kelsey Gilbert [:jgilbert]]

[Tracking Requested - why for this release]:
It sucks for users when screenshotting inexplicably doesn't work for some content, which users may not even realize is webgl.

Comment 38

[Ryan VanderMeulen [:RyanVM]]

Per discussion with Jeff, better to let this bake more given where we are in the cycle. Let's aim to get this into 91.3esr next cycle.

Comment 42

[Kelsey Gilbert [:jgilbert]]

Comment on attachment 9241879 [details]
Bug 1661869 - Fix screenshotting of WebGL canvases.

ESR Uplift Approval Request

• If this is not a sec:{high,crit} bug, please state case for ESR consideration: Fix screenshotting of webgl canvases, such as Google Maps
• User impact if declined: Screenshotting of some canvases will continue to be broken.
• Fix Landed on Version: 94
• Risk to taking this patch: Low
• Why is the change risky/not risky? (and alternatives if risky): It has baked in 94 now, and few changes have happened between 91 and 94 in this area.
• String or UUID changes made by this patch: none

Comment 43

[Alin Ilea, Desktop QA]

Reproduced the issue in Release 92, on google maps, using Win10.
Verified - Fixed in Beta 94.0b2 (build id: 20211005185813) and latest Nightly 95.0a1 (build id: 20211005094529). The screenshots are displayed accordingly, using google maps.

Comment 44

[Ryan VanderMeulen [:RyanVM]]

Comment on attachment 9241879 [details]
Bug 1661869 - Fix screenshotting of WebGL canvases.

Approved for 91.4esr.

Comment 45

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/releases/mozilla-esr91/rev/2d2658a85cef

Comment 46

[Alin Ilea, Desktop QA]

Verified - Fixed in 91.3.0esr (build id: 20211007190139) (https://hg.mozilla.org/releases/mozilla-esr91/rev/2d2658a85cef) on Win10. The screenshots are displayed correctly, using google maps.