Closed Bug 1687554 Opened 4 years ago Closed 4 years ago

[Bug] Pages are broken on 1/19 Firefox Nightly

Categories

(Core :: Graphics: WebRender, defect)

Unspecified
Android
defect

Tracking

()

RESOLVED FIXED
86 Branch
Tracking Status
firefox-esr78 --- unaffected
firefox84 --- unaffected
firefox85 --- unaffected
firefox86 + fixed

People

(Reporter: amoya, Assigned: jnicol)

References

(Regression)

Details

(Keywords: regression)

Attachments

(4 files)

From github: https://github.com/mozilla-mobile/fenix/issues/17524.

Steps to reproduce

  1. Open Firefox Nightly and open a few pages (I had wikipedia.org, and cnn.com).
  2. Observe the pages.

Expected behavior

No issues encountered. Pages are loaded correctly, the content is displayed without glitches.

Actual behavior

Pages are broken! The user cannot read any of the content displayed. Reloading the page, doesn't fix the problem.
Cleaning the profile, doesn't fix the problem.

Device information

  • Android device: HTC 10 (Android 8, Andreno 530), and Huawei MediaPad M2 (Android 5.1.1)
  • Fenix version: 1/19 Nightly
  • Not reproducible on Beta 85.0.0-beta.9, nor on RC 84.1.4

logcat.txt

<img src="https://user-images.githubusercontent.com/40694158/105018048-2f735000-5a4d-11eb-91d4-76009a817bec.png" width="300" height="550" />

Change performed by the Move to Bugzilla add-on.

Rough regression range between known-good and known-bad GV versions:
https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=8dd1cdbc4076ce53dfa259b1ece26e67189ea140&tochange=46713b21267a0769462b4a0cb6cbff4d3cbee901

Unfortunately we went a few days on Fenix Nightly with the same GV build due to some breaking changes in Android Components that needed adapting for, so this is the best we can do unless someone is able to bisect with GVE by way of mozregression.

I could repro this on Thunderbird somewhat consistently (though I can repro on Desktop too), and came up with the following range:

https://hg.mozilla.org/mozilla-central/pushloghtml?changeset=0f5e4a3c6f0a6ab0fcd77c46cedc819e2bd0b3f8

https://hg.mozilla.org/mozilla-central/rev/58ec75f99b23d99ad4e094de44c5ba32c0b4b57d looks the most suspicious according to our chat in #gfx. Android probably doesn't have sub-pixel glyphs, so that's why it probably shows up more easily than on desktop.

Regressed by: 1685563
Has Regression Range: --- → yes

deleted

Flags: needinfo?(jnicol)
Flags: needinfo?(jnicol)
Flags: needinfo?(jnicol)

I've been debugging this today by making try builds and asking Mira to test them. So far I've established that the shader optimization is responsible, specifically the driver doesn't seem to like how the use of the ternary operator is optimized. I have a few things more the test tomorrow, but I think I know what the solution is.

Emilio on the other hand can still reproduce his bug with shader optimization disabled. Therefore the android and linux bugs are separate bugs. I've been unable to reproduce the linux bug unfortunately. A reddit user also reported the same issue, both they and Emilio have an AMD gpu, so I believe that is probably why I cannot. I suspect this bug is to do with texture upload rather than a shader issue, but I could be wrong.

Ryan, how quickly do we need these fixed? Can I have a few days and if I haven't fixed them by then I will land a reverting change before they hit beta? (The patch won't back out cleanly due to a few other patches which have landed since, but it'll be easy to effectively revert the change)

Flags: needinfo?(jnicol) → needinfo?(ryanvm)

From the looks of it, this bug is making things pretty unusable for affected users. I think we should backout for now.

Flags: needinfo?(ryanvm)

Okay, I'll write a patch to do that

Bug 1685563 switched to using R8 textures instead of BGRA8 for
non-subpixel AA glyphs. This caused rendering issues on certain
android and linux devices, so switch back temporarily until those
issues are fixed.

Assignee: nobody → jnicol
Status: NEW → ASSIGNED
Pushed by jnicol@mozilla.com: https://hg.mozilla.org/integration/autoland/rev/a62f91b0e206 Revert to using BGRA8 textures for non-subpixel AA glyphs. r=lsalzman
Status: ASSIGNED → RESOLVED
Closed: 4 years ago
Resolution: --- → FIXED
Target Milestone: --- → 86 Branch

reminder to try the TB builds that Jamie sent me :)

Flags: needinfo?(emilio)
Flags: needinfo?(emilio)

Radeon bug reported upstream here: https://gitlab.freedesktop.org/mesa/mesa/-/issues/4227

You need to log in before you can comment on or make changes to this bug.

Attachment

General

Created:
Updated:
Size: