1354813 - Page distorted vertically when scrolling

Status: RESOLVED FIXED

(Core :: Graphics, defect, P3)

Description

[pietro.brenna]

Attachment: scrolling stretch.mkv

User Agent: Mozilla/5.0 (X11; Linux x86_64; rv:54.0) Gecko/20100101 Firefox/54.0
Build ID: 20170408074747

Steps to reproduce:

1. visit http://ticki.github.io/blog/seahash-explained/
2. scroll vertically a few times


Actual results:

After some scrolling, at some point, the top of the viewport continues to scroll upward but the bottom doesn't; this result in a vertically stretched version of the web page.


Expected results:

Regular scrolling.

I attach a screen recording showing the issue.

Comment 1

[Timothy Nikkel (:tnikkel)]

I couldn't reproduce on mac. Does it reproduce in safemode or a fresh profile? Or nightly or release?

Comment 2

[pietro.brenna]

Can't reproduce in new profile/safe mode. 
I think this is caused by layers.acceleration.force-enabled = true,
which I had set on my profile since it provides some performance
gain, and then forgot about since everithing worked fine.

Gpu:             Mesa DRI Intel(R) 965GM 
Driver version:  2.1 Mesa 17.0.3

HW_COMPOSITING	
blocked by default: Acceleration blocked by platform
force_enabled by user: Force-enabled by pref
OPENGL_COMPOSITING	
force_enabled by user: Force-enabled by pref

This does not happen on 52.0.2 as packaged by Arch Linux (even
with force-enabled = true).

And perhaps more interestingly: can't reproduce in nightly 55,
with the very same profile that causes the bug in version 54.

Comment 3

[Timothy Nikkel (:tnikkel)]

Ok, so I guess this is fixed on trunk. Since you force enabled that pref hopefully only users who also force enabled will see this problem. If you are curious to see what fixed (or broke) this you could use mozregression to track that down. Otherwise we'll probably just wait for whatever patch fixed this in 55 to make its way to aurora/beta/release.

Comment 4

[pietro.brenna]

Just for info, mozregression shows the bug appearing here:

https://hg.mozilla.org/releases/mozilla-aurora/pushloghtml?fromchange=24a81d93e07cc96300f8e1f5c69034dd4743bd63&tochange=4b053b4106a9b99268312c5fcf8ac1048cc80430

Can't tell when it was fixed since I can't manage to compare the aurora and nightly channels with mozregression.

Comment 5

[Timothy Nikkel (:tnikkel)]

(In reply to pietro.brenna from comment #4)
> Just for info, mozregression shows the bug appearing here:
> 
> https://hg.mozilla.org/releases/mozilla-aurora/
> pushloghtml?fromchange=24a81d93e07cc96300f8e1f5c69034dd4743bd63&tochange=4b05
> 3b4106a9b99268312c5fcf8ac1048cc80430

That is when nightly merged to aurora, so it includes all changes made to nightly between 2016-11-14 and 2017-01-23. Which is a lot.

Try doing the regression only on the nightly channel, you might have to try a few dates on nightly to find a date before the problem appeared, one with the problem, and one with the problem fixed.

Comment 6

[pietro.brenna]

Thanks for the help

Introduced: https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=f8e550969421c82a0ab636ffeac3a677b4959f41&tochange=f5c0b7682e2ff43f9bbe7400dc0ec7d397f256b3

Fixed: https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=25e9dfe731cc910d4f0dd8fe4af10e62baa84f45&tochange=c9f09b670a27bd01d89193b0d4c79f72001e7987

Comment 7

[Timothy Nikkel (:tnikkel)]

Anything you want to uplift here?

Comment 8

[Miko Mynttinen]

Judging by the regression range and fix, this might have been introduced/fixed by the texture clamping/repeat change. I'm trying to see if I can get hardware to verify that this is the case.

Unfortunately, the bug 1286716 that introduced this is in FF53, and the bug 1323791 that fixes this is in FF55.

It would be possible to uplift the specific part 1 of the patch with minimal effort, but the release day for FF53 is very soon. What makes this decision tricky is that this bug happens only with hardware acceleration force-enabled.

Comment 9

[Jet Villegas (inactive)]

(In reply to Miko Mynttinen [:miko] from comment #8)
> Judging by the regression range and fix, this might have been
> introduced/fixed by the texture clamping/repeat change. I'm trying to see if
> I can get hardware to verify that this is the case.
> 
> Unfortunately, the bug 1286716 that introduced this is in FF53, and the bug
> 1323791 that fixes this is in FF55.

1323791 has already been deemed too risky to uplift.

> What makes this decision tricky is that this bug happens only with hardware acceleration
> force-enabled.

Indeed, and doesn't help the argument for uplifting this late.

Comment 10

[Miko Mynttinen]

Marking this fixed based on comment 2.