1079996 - Black artifacts when scrolling

Status: RESOLVED WORKSFORME

(Core :: Graphics, defect)

Description

[(no longer active)]

I've been seeing random black rectangles appearing as I scroll on Mac with Nightly as of a few days ago.  I can't reproduce, but there are two distinct cases that I remember, one is in the tab bar when you scroll to the left and right sometimes, and the other was on an airbnb page.

Comment 2

[(no longer active)]

(In reply to Markus Stange [:mstange] from comment #1)
> 
> *** This bug has been marked as a duplicate of bug 1064236 ***

What makes you think this is a dupe?  I have started to see this in the past few days.

Comment 3

[Markus Stange [:mstange]]

Oh, sorry about that then.

Comment 4

[(no longer active)]

This is very noticeable when scrolling the TBPL main view vertically, I can hit it two or three times per minute.

Comment 5

[(no longer active)]

Attachment: Screenshot on gmail

I also get this now pretty reliably in the Gmail compose UI when scrolling.

Comment 6

[(no longer active)]

[Tracking Requested - why for this release]: Black rectangles when scrolling seems like an important regression to track.

Comment 7

[(no longer active)]

I have an email that I would be happy to forward to someone.  Replying to that email and injecting some text inline in it and scrolling enables me to reproduce 100% reliably.

Comment 8

[Lukas Blakk [:lsblakk] use ?needinfo]

Kairo, can you look into this with Ehsan's email and see about finding a regression range?  This looks like something we'd really need to get a potential fix for before Beta 1.

Comment 9

[Robert Kaiser]

(In reply to Lukas Blakk [:lsblakk] use ?needinfo from comment #8)
> Kairo, can you look into this with Ehsan's email and see about finding a
> regression range?  This looks like something we'd really need to get a
> potential fix for before Beta 1.

I don't do webmail anywhere and I boycott centralized nonstandard services like gmail where possible so I probably would not be able to do anything there personally. I also have no experience with and time for finding regression ranges in a decent way. I can try to find someone who can help, though, will email firefox-qe for help with this and other bugs flagged to me.

Comment 10

[Robert Kaiser]

(In reply to :Ehsan Akhgari (not reading bugmail, needinfo? me!, PTO 11/3-11/21) from comment #0)
> I've been seeing random black rectangles appearing as I scroll on Mac with
> Nightly as of a few days ago.

What kind of Mac is that? What OSX version does it run? What graphics hardware does it have? If it has dual graphics, which ones? If you can paste the graphics section from your about:support page would be very helpful.

Also, could you try again on a current Nightly, just to make sure it hasn't been fixed since?


(In reply to :Ehsan Akhgari (not reading bugmail, needinfo? me!, PTO 11/3-11/21) from comment #4)
> This is very noticeable when scrolling the TBPL main view vertically, I can
> hit it two or three times per minute.

I can't reproduce this on TBPL with my Mac Mini (OSX 10.10.1, OpenGL/OMTC acceleration, Intel HD4000 graphics, ID 0x0166) at all. I also tried the compose UI in gmail with a long email quoted for forwarding and didn't see and glitches like what your sceeenshot shows.


(In reply to :Ehsan Akhgari (not reading bugmail, needinfo? me!, PTO 11/3-11/21) from comment #7)
> I have an email that I would be happy to forward to someone.  Replying to
> that email and injecting some text inline in it and scrolling enables me to
> reproduce 100% reliably.

I hoped someone more familiar with Macs, Gmail, and regression-finding would help out, but I guess nobody does, so you you forward it to [e-mail] please? Given the above, I doubt I'll see the issue, but we can try.

Comment 11

[(no longer active)]

(In reply to Robert Kaiser (:[e-mail]) from comment #10)
> (In reply to :Ehsan Akhgari (not reading bugmail, needinfo? me!, PTO
> 11/3-11/21) from comment #0)
> > I've been seeing random black rectangles appearing as I scroll on Mac with
> > Nightly as of a few days ago.
> 
> What kind of Mac is that? What OSX version does it run? What graphics
> hardware does it have? If it has dual graphics, which ones? If you can paste
> the graphics section from your about:support page would be very helpful.

It happens on 10.9 and 10.10 on "MacBook Pro (Retina, 15-inch, Late 2013)".  I have shown this to Jeff Muizelaar in person and he hasn't been able to reproduce unfortunately.

> Also, could you try again on a current Nightly, just to make sure it hasn't
> been fixed since?

I always run the latest Nightly, and it has gotten worse recently.

> (In reply to :Ehsan Akhgari (not reading bugmail, needinfo? me!, PTO
> 11/3-11/21) from comment #4)
> > This is very noticeable when scrolling the TBPL main view vertically, I can
> > hit it two or three times per minute.
> 
> I can't reproduce this on TBPL with my Mac Mini (OSX 10.10.1, OpenGL/OMTC
> acceleration, Intel HD4000 graphics, ID 0x0166) at all. I also tried the
> compose UI in gmail with a long email quoted for forwarding and didn't see
> and glitches like what your sceeenshot shows.

Yeah, not sure what's unique about my environment but Jeff hasn't been able to reproduce either.

> (In reply to :Ehsan Akhgari (not reading bugmail, needinfo? me!, PTO
> 11/3-11/21) from comment #7)
> > I have an email that I would be happy to forward to someone.  Replying to
> > that email and injecting some text inline in it and scrolling enables me to
> > reproduce 100% reliably.
> 
> I hoped someone more familiar with Macs, Gmail, and regression-finding would
> help out, but I guess nobody does, so you you forward it to
> [e-mail] please? Given the above, I doubt I'll see the issue, but
> we can try.

Sure, I will forward you the email, but I'm not very hopeful that you can reproduce based on the above.

Comment 12

[(no longer active)]

Just confirmed that this is not e10s specific, by finally reproducing in a non-e10s window.

Comment 13

[(no longer active)]

OK, I finally have a 100% reproducible testcase.  Go to <https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html>, keep Space (or Shift-Space) down to scroll up/down rapidly.  A lot of black artifacts appear on the left side of the page, most of them on the leftmost white vertical bar right before where the text starts.

Comment 14

[(no longer active)]

(BTW the test case reproduces both in e10s and non-e10s mode.)

Comment 15

[Nicolas Silva [:nical]]

There is a good chance that the last patch in bug 1071769 will fix this.

Comment 16

[Lukas Blakk [:lsblakk] use ?needinfo]

I hope the patch in comment 15 does resolve this, but we're too late for FF35 at this stage.  Please nominate for Aurora uplift so that it can get shipped in 36.

Comment 17

[Benoit Girard (:BenWa)]

It's fixed now in nightly so we need to find that caused it so we can backport. I unapplied bug 1071769 locally but it still does not reproduce.

Right now my bisection range is between the 23rd and 25th of December.

Comment 18

[Benoit Girard (:BenWa)]

Range over which the bug was fixed, which includes bug 1071769 but my local testing seems to point towards something else fixing it. It's hard to say since it doesn't reproduce 100% of the time:
http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=0532f2509f3f&tochange=b17e7747d3fb

Comment 19

[Bas Schouten (:bas.schouten)]

(In reply to Benoit Girard (:BenWa) from comment #18)
> Range over which the bug was fixed, which includes bug 1071769 but my local
> testing seems to point towards something else fixing it. It's hard to say
> since it doesn't reproduce 100% of the time:
> http://hg.mozilla.org/mozilla-central/
> pushloghtml?fromchange=0532f2509f3f&tochange=b17e7747d3fb

Hrm, note that the things Ehsan describes are things which use component alpha layers, it might be interesting to keep that in mind when looking for the fixing changeset.

Comment 20

[Benoit Girard (:BenWa)]

Darn, I bisected it to:
https://hg.mozilla.org/mozilla-central/rev/6cf334e6acf8

I must of missed the bug when I tested it on Comment 18.

Nical should we and can you uplift that patch?

Comment 21

[(no longer active)]

I just saw the original bug when scrolling <https://bugzilla.mozilla.org/page.cgi?id=splinter.html&bug=1018320&attachment=8540683>.  I don't think the original issue is completely fixed on trunk.

Comment 22

[(no longer active)]

Attachment: Still happening on trunk

Evidence. :)

Comment 23

[Nicolas Silva [:nical]]

(In reply to Benoit Girard (:BenWa) from comment #20)
> Nical should we and can you uplift that patch?

yes

Comment 24

[Nicolas Silva [:nical]]

(In reply to Nicolas Silva [:nical] from comment #23)
> (In reply to Benoit Girard (:BenWa) from comment #20)
> > Nical should we and can you uplift that patch?
> 
> yes

Nevermind. Bug 1108164 which didn't make it to beta caused the regression that got fixed with that patch. So this fix is not to be uplifted. As Ehsan showed, there is another regression which is on beta but we haven't found out what it is yet.

Comment 25

[Sylvestre Ledru [:Sylvestre]]

Nicolas, are you going to work on this bug? If not, do you know who can work on this? I would like to have an assignee. Thanks

Comment 26

[Nicolas Silva [:nical]]

(In reply to Sylvestre Ledru [:sylvestre] from comment #25)
> Nicolas, are you going to work on this bug? If not, do you know who can work
> on this? I would like to have an assignee. Thanks

I am focusing on other stuff right now, forwarding the question of who should work on this to Milan.

Comment 27

[Milan Sreckovic [:milan] (needinfo for best results)]

Benoit, are you seeing this problem (based on comment 18, it seems like you are)?

Comment 28

[Benoit Girard (:BenWa)]

I am no longer seeing this. We should make sure we don't morph this bug into a generic 'rendering is wrong sometimes'. The remaining issue that Ehsan sees should be investigated in a follow-up bug since it could have 1) different regression range, 2) different root cause, 3) Different fix.

This bug seems to best track an issue that was fixed by 6cf334e6acf8. We should make it it's uplifted to all the affected branch and close this bug.

Comment 29

[(no longer active)]

I made a clone of this for continued investigation: bug 1121164.

Comment 30

[Sylvestre Ledru [:Sylvestre]]

I don't think this is going to be fixed for 36. Wontfixing it and tracking for 38.

Comment 31

[Lawrence Mandel [:lmandel] (use needinfo)]

With cloned bug 1121164, should this bug be resolved?

Comment 32

[Milan Sreckovic [:milan] (needinfo for best results)]

Comment 28 + comment 29 + reminder in comment 31 = Resolved WFM

Comment 33

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

This bug and its friends are still confusing. Comment 24 says not to uplift anything. 

But from comment 28, we should uplift the patch from comment 20 actually for bug 101769. That was checked in 2014-12-23, so we should look at bug 1071769 and make sure that patch it was applied to all affected branches.    So, maybe we should untrack this bug, and track bug 1071769 and try to untangle it, and the bugs it is entangled with. I think we want to make sure we were consistently uplifting connected patches in the same versions.  

And, to see if the test case Ehsan found is still reproducible, in bug 1121164. I will comment there to ask for QE help.

Comment 34

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

^^ Sorry, typo. I meant the patch from bug 1071769.

Comment 35

[Lawrence Mandel [:lmandel] (use needinfo)]

My understanding is that there is nothing to uplift to 37 in this bug. The commit referenced in comment 20 landed on m-c while it was 37. We can follow up on bug 1071769 if that turns out to be important to fix in 37 specifically. We should also follow up on Ehsan's clone.