1422274 - Images and text go missing and disappears after landing patch from bug #1418945

Status: VERIFIED WORKSFORME

(Core :: Web Painting, defect, P3)

Description

[Virtual_ManPL [:Virtual] 🇵🇱 - (please needinfo? me - so I will see your comment/reply/question/etc.)]

[Tracking Requested - why for this release]: Regression

STR:
1. Open some website page (preferable long and with many images)
2. Scroll it down
3. Scroll it up sometimes
and see that images and text go missing and disappear

Workaround:
Images and text will be visible and reappear, when you move mouse cursor/pointer in that place

Comment 1

[Virtual_ManPL [:Virtual] 🇵🇱 - (please needinfo? me - so I will see your comment/reply/question/etc.)]

I suspect, that it's new regression (so probably from only 1 or 2 days)
and I will try to find specific regression range soon.

Comment 2

[YF (Yang)]

STR:
I can always reproduce a black screen when viewing a message with "Checker Plus for Gmail" and scrolling. I also occasionally encountered screen content disappears in Nightly 59.


Regression range:
https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=4ad2150e73c0b19f1f6264a4b9ad6927986d43ce&tochange=37e0bd919af057d44c5c1410458c0f00a3653c11

Comment 3

[Virtual_ManPL [:Virtual] 🇵🇱 - (please needinfo? me - so I will see your comment/reply/question/etc.)]

@ YF (Yang) - Thank you very much for specific regression range!

@ Jared Wein [:jaws] &  Emilio Cobos Álvarez [:emilio]  - Any ideas which bug could be the cause?

Comment 4

[Emilio Cobos Álvarez (:emilio)]

Is there any particular URL this happens with / or add-on you use?

It could be because of @-moz-document, though it looks kinda random for it to be...

I haven't seen the same on Linux.

Comment 5

[Virtual_ManPL [:Virtual] 🇵🇱 - (please needinfo? me - so I will see your comment/reply/question/etc.)]

(In reply to Emilio Cobos Álvarez [:emilio] from comment #4)
> Is there any particular URL this happens with / or add-on you use?

It happens on https://www.wykop.pl/mikroblog/

About extensions, I use:
- CanvasBlocker 0.4.2
- Cookie AutoDelete 2.0.1
- Decentraleyes 2.0.1
- Don't touch my tabs! (rel=noopener) 1.0
- Don't track me Google 4.19
- Download Star 1.0.1
- Firefox Pioneer 1.0
- h264ify 1.0.6
- The ESPER experiment - Firefox Pioneer Study 1.0.1
- uBlock Origin 1.14.21b0
but it should be also reproducible with fresh profile.

Comment 6

[Jared Wein [:jaws] (please needinfo? me)]

My commit in that regression range woudln't affect page content. I suspect the regression range is incorrect.

Comment 7

[YF (Yang)]

(In reply to Jared Wein [:jaws] (please needinfo? me) from comment #6)
> My commit in that regression range woudln't affect page content. I suspect
> the regression range is incorrect.

I got the same reproducible range, and this problem can be reproduced with Checker Plus for Gmail on clean profile.

Comment 8

[Emilio Cobos Álvarez (:emilio)]

(In reply to YF (Yang) from comment #7)
> (In reply to Jared Wein [:jaws] (please needinfo? me) from comment #6)
> > My commit in that regression range woudln't affect page content. I suspect
> > the regression range is incorrect.

Yeah, I checked that page out locally and there's no @-moz-document rules at all... And that add-on seems to be a web-extension that also doesn't use them... Anyway, does it reproduce with layout.css.moz-document.content.enabled turned on, just in case?

Comment 9

[Virtual_ManPL [:Virtual] 🇵🇱 - (please needinfo? me - so I will see your comment/reply/question/etc.)]

(In reply to Emilio Cobos Álvarez [:emilio] from comment #8)
> does it reproduce with
> layout.css.moz-document.content.enabled turned on, just in case?

Unfortunately, it's still reproducible,
so I suspect, that regression range could be different in my case.

Comment 10

[Emilio Cobos Álvarez (:emilio)]

Ok, that means that regression range is more than probably not correct... If it's windows only I'd look at windows-related graphics changes, there have been a few.

Comment 11

[Liz Henry (:lizzard) (relman/hg->git project)]

Andrei, can your team try again to find a regression range? I'm not sure we have anything actionable here, yet.

Comment 12

[Timea Zsoldos [:zstimi/tzsoldos], Desktop QA]

I have reproduced this issue using Firefox 59.0a1 (2017.11.30) on Win 8.1 x64. Reproduced for me using "Checker Plus for Gmail", viewing a message, then scrolling.

[Regression range]:
Last good: build id=20171129220149
First bad: build id=20171130220131
https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=c2248f85346939d3e0b01f57276c440ccb2d16a1&tochange=ea747bb2ffb77d1fd62b5fa6217cbee15b73d31f

Comment 13

[Virtual_ManPL [:Virtual] 🇵🇱 - (please needinfo? me - so I will see your comment/reply/question/etc.)]

This regression range is too vast, as looks like whole one day Nightly release and it makes mostly impossible to deduce, which patch is cause of this issue.
What's more, neary same big regression range was pointed alredy in comment #1.
Search for regression range was done with mozregression or mozregression-gui?

Comment 14

[Timea Zsoldos [:zstimi/tzsoldos], Desktop QA]

(In reply to Virtual_ManPL [:Virtual] - (please needinfo? me - so I will see your comment/reply/question/etc.) from comment #13)
> This regression range is too vast, as looks like whole one day Nightly
> release and it makes mostly impossible to deduce, which patch is cause of
> this issue.
> What's more, neary same big regression range was pointed alredy in comment
> #1.
> Search for regression range was done with mozregression or mozregression-gui?

This issue, as described in comment 5, is not reproducible for me using latest Nightly 59.0a1 on Windows 8.1 and Windows 10 x64 (ATI Radeon 3000 Graphics; Intel HD 2500). Are there any other links I could use to try to reproduce this?

(In reply to Emilio Cobos Álvarez [:emilio] from comment #8)
> (In reply to YF (Yang) from comment #7)
> Anyway, does it reproduce with layout.css.moz-document.content.enabled turned on, just in case?

The issue I was seeing in comment 12 is different from this bug and is only related to "Checker Plus for Gmail" add-on - using mozregression-gui I could found the culprit for it: Bug 1035091: Disable @-moz-document on author sheets on nightly and early beta. r=xidorn
The issue goes away after modifying the pref layout.css.moz-document.content.enabled to true.
Emilio, should this bug be tracked separately?

Comment 15

[Emilio Cobos Álvarez (:emilio)]

(In reply to Timea Zsoldos from comment #14)
> The issue I was seeing in comment 12 is different from this bug and is only
> related to "Checker Plus for Gmail" add-on - using mozregression-gui I could
> found the culprit for it: Bug 1035091: Disable @-moz-document on author
> sheets on nightly and early beta. r=xidorn
> The issue goes away after modifying the pref
> layout.css.moz-document.content.enabled to true.
> Emilio, should this bug be tracked separately?

Yes, please, let's file a new bug for this.

Comment 16

[Virtual_ManPL [:Virtual] 🇵🇱 - (please needinfo? me - so I will see your comment/reply/question/etc.)]

FYI - I split other issue to bug #1425455

I'm also closer (or at least I think that) to finding my bug regression range, which for now seems:
2017-11-21 = good
2017-11-22 = bad
I will probably find in detail which patch caused this in this weekend.

Comment 17

[Virtual_ManPL [:Virtual] 🇵🇱 - (please needinfo? me - so I will see your comment/reply/question/etc.)]

[Tracking Requested - why for this release]: Regression



OK, I found the cause, mozregression-gui pointing this:

Regression caused by:
bug #1418945

Regression range pushlog:
https://hg.mozilla.org/integration/mozilla-inbound/rev/9f3615964530


and looks like now, per bug #1418945 comment #11, Firefox 58 is also affected.

Comment 18

[Virtual_ManPL [:Virtual] 🇵🇱 - (please needinfo? me - so I will see your comment/reply/question/etc.)]

Attachment: screencast.mp4

Attaching screencast to show how issue looks like.


Updated STR:
1. Open some website page (preferable long and with many images, like for example - https://www.wykop.pl/mikroblog/ )
2. Scroll it down
3. While scrolling down, do circular movements with mouse cursor/pointer
Hovering these buttons hyperlinks:
>( "odpowiedz" - "ulubiony" - "zgłoś" - "udostępnij" )
will easier time to reproduce issue
4. Scroll it up sometimes
and see that images and text go missing and disappear, like it shown in attachment

Comment 19

[Mike Taylor [:miketaylr]]

marking as fix-optional for 58 given P3 status.

Comment 20

[Julien Cristau [:jcristau]]

Is this actually reproducible in beta (with layout.display-list.retain=false)?

Comment 21

[Virtual_ManPL [:Virtual] 🇵🇱 - (please needinfo? me - so I will see your comment/reply/question/etc.)]

(In reply to Julien Cristau [:jcristau] from comment #20)
> Is this actually reproducible in beta (with
> layout.display-list.retain=false)?

It's reproducible with not default configuration (with "layout.display-list.retain" preference set to "true" in about"config). Basing on your comment and on bug #1416055, Tracking Flags were properly changed by you, as this feature (Retained Display Lists) is disabled by default on Firefox 58. Thank you very much for notice and correction!

Comment 22

[Miko Mynttinen]

I am unable to reproduce this with Nightly on macOS 10.13 and W10.

Virtual, could you please see if this problem still persists?

Comment 23

[Virtual_ManPL [:Virtual] 🇵🇱 - (please needinfo? me - so I will see your comment/reply/question/etc.)]

(In reply to Miko Mynttinen [:miko] from comment #22)
> I am unable to reproduce this with Nightly on macOS 10.13 and W10.
> 
> Virtual, could you please see if this problem still persists?

Yes. It's still reproducible for me with STR from Comment #18 in latest Mozilla Firefox Nightly 59.0a1 (2018-01-02) (64-bit) on Windows 7 (64-bit).

Comment 24

[Miko Mynttinen]

For some reason STR do not work on my machines. Instead, what seems to trigger this is the new posts counter on the page changing. The change triggers a display list rebuild, and for some reason a subdocument item and it's descendants are missing from the merged list.

Retained DL verifier output:
**** Differences between retained-after-merged (AM) and non-retained (NR) display lists:
NR#107 > NR#111 > NR#1246 > SubDocument NR#1250 p=0x125338aa0 f=0x124de4020() key=50
 * Cannot find corresponding item under AM#107 > AM#111 > AM#1246
NR#107 > NR#111 > NR#1309 > Border NR#1311 p=0x124c93a20 f=0x124d23690( class:sub-menu inline-list) key=7
 * Cannot find corresponding item under AM#107 > AM#111 > AM#1250
NR#107 > NR#111 > NR#1309 > Text NR#1312 p=0x124d7b920 f=0x1256261b8() key=63
 * Cannot find corresponding item under AM#107 > AM#111 > AM#1250
[..]

This might be related to bug 1427914 (since subdocument is inside OOF frame). Bug 1418840 has a patch for other subdocument problems, but it does not fix the problem.

Comment 25

[Miko Mynttinen]

At least the part of the problem I was able to reproduce was fixed by bug 1427914.

Virtual_ManPL, could you please see if you are still able to reproduce this with the latest Nightly?

Comment 26

[Virtual_ManPL [:Virtual] 🇵🇱 - (please needinfo? me - so I will see your comment/reply/question/etc.)]

Yes. It's still reproducible for me with STR from Comment #18 in latest Mozilla Firefox Nightly 59.0a1 (2018-01-10) (64-bit) [ rev: https://hg.mozilla.org/mozilla-central/rev/d5f42a23909eb181274731b07e4984bfbd18557d ] on Windows 7 (64-bit).

Comment 27

[Miko Mynttinen]

Sorry for not having made much progress with this bug. I failed to reproduce this on Windows 7 running on a virtual machine, with and without Checker Plus for Gmail -addon.

We do have landed some patches that might help with this: bug 1428993 and bug 1420737. There is also some follow-up work coming for both bugs.

Comment 28

[YF (Yang)]

I can not reproduce this problem with Checker Plus for Gmail in Fx58.0, but the text disappearance still appears occasionally in different forms.

Comment 29

[Virtual_ManPL [:Virtual] 🇵🇱 - (please needinfo? me - so I will see your comment/reply/question/etc.)]

STR is in comment #18 and there isn't any step with "Checker Plus for Gmail" extension, as it's different bug - see bug #1425455.

Comment 30

[Alice0775 White]

:Virtual,
Why was status-firefox60 to disabled?

layout.display-list.retain is true by default on Beta 60.0b6.

Comment 31

[Virtual_ManPL [:Virtual] 🇵🇱 - (please needinfo? me - so I will see your comment/reply/question/etc.)]

@  Alice0775 White- Oh, true, my bad, I'm very sorry. Thank you very much for noticing this and informing me.

Requesting tracking, since it's enabled by default even in release channels now per bug #1440144.

Comment 32

[Ryan VanderMeulen [:RyanVM]]

We just shipped a dot-release for Fx59 for rendering issues that looked very similar to this (different culprit, similar results). I suspect that if we ship 60 with this bug, there's a good chance we'll end up eventually needing to do the same. Miko or Matt, what do we need to better investigate this?

Comment 33

[Matt Woodrow (:mattwoodrow)]

Miko, do you want to have another go at reproducing this?

Can anyone else still reproduce this issue?

We definitely don't want to ship with a regression like this, but we haven't been able to reproduce locally as of yet.

Comment 34

[Miko Mynttinen]

I unsuccessfully tried reproducing this today on OSX, Win7 (on VM) and Win10. I tried with and without scroll wheel paging.

Comment 35

[Ryan VanderMeulen [:RyanVM]]

Maybe the graphics information from about:support would be useful from the people who can reproduce?

Comment 36

[YF (Yang)]

I don't have the steps to reproduce for this issue now.

Comment 37

[Virtual_ManPL [:Virtual] 🇵🇱 - (please needinfo? me - so I will see your comment/reply/question/etc.)]

I'm looking on it, as I have some trouble with reproduction. I will report back results this week.

Comment 38

[Virtual_ManPL [:Virtual] 🇵🇱 - (please needinfo? me - so I will see your comment/reply/question/etc.)]

I can't reproduce this issue anymore, so I'm marking this bug as WORKSFORME.
What's more, even when I used mozregression-gui to find fix, I couldn't even reproduce issue on reproducible builds.
Basing on this, I suspect that it could be NVIDIA GPU driver bug which was fixed in recent drivers, now I'm using 391.35 version.

P.S. But if someone can still reproduce this issue, please reopen it. Thanks.