Closed Bug 1224976 Opened 4 years ago Closed 4 years ago

awful repaint performance on amazon.com

Categories

(Core :: General, defect)

x86_64
Linux
defect
Not set

Tracking

()

RESOLVED FIXED
mozilla45
Tracking Status
firefox43 + unaffected
firefox44 - unaffected
firefox45 + unaffected
firefox46 --- fixed

People

(Reporter: froydnj, Assigned: roc)

References

(Depends on 1 open bug)

Details

(Keywords: regression)

Attachments

(1 file)

STR:

Go to any Amazon.com page; I used http://www.amazon.com/The-Theology-Augustine-Introductory-Important/dp/0801048486/

Actual results:

The page seems to start loading OK; the main image for the book comes in, followed by "frequently bought together", "customers who bought this item also bought", etc.  Then, after waiting, most of the images on the page repaint with blank white, then repaint again with the intended images.  This happens again, but the second time, the main image for the book no longer appears.

I should note that this is on a 2560x1440 monitor, with Firefox in full screen, in case it's difficult to get all the stuff on the screen at once.
(In reply to Nathan Froyd [:froydnj] from comment #0)
> I should note that this is on a 2560x1440 monitor, with Firefox in full
> screen, in case it's difficult to get all the stuff on the screen at once.

I can also see this on smaller windows, but I don't see it on my Windows machine at 1920x1080, FWIW.
I see this in my Linux build.
I get a ton of:
[GFX2-]: DrawTargetCairo context in error state: invalid matrix (not invertible)(5)
from the chrome process BasicCompositor.

As far as I can tell, it might just be that DrawTargetCairo goes into an error state as soon as we try to draw with a singular transform, and stays there :-(. For example, we may be drawing a scale(0) transform element and then need to draw something else, but that next thing won't get drawn. I don't see any code that tries to avoid this problem.
Nathan, can you try that patch?
Flags: needinfo?(nfroyd)
(In reply to Robert O'Callahan (:roc) (Mozilla Corporation) from comment #5)
> Nathan, can you try that patch?

It doesn't help. :(  The repainting seems to happen a little faster, perhaps..? but there's still flashing and every image on the page getting repainted several times.
Flags: needinfo?(nfroyd)
The 2015-09-18 nightly is good, the 2015-09-19 nightly is bad.  Pushlog:

https://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=11dc79e232110ba6de5179e46dfbda77b52a88c3&tochange=4313752f69956ae248bd4e7ff3913c8dd4252698

Nothing immediately jumps out; bug 1205970 looks like it might be sort of in the right area?
[Tracking Requested - why for this release]: Bad rendering of amazon.com pages.

Better pushlog:

https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=11dc79e232110ba6de5179e46dfbda77b52a88c3&tochange=4313752f69956ae248bd4e7ff3913c8dd4252698

Looks like this is bug 1097464.

In a working build, you can see the cover of the book tilt as though showing you it has a side and you can open it up.  I guess that's done with a 3d transform?

I don't know if this entirely solves things; it looks like post-bug 1097464, we only repaint everything once, so we might have introduced another regression in there somewhere?
Blocks: 1097464
Flags: needinfo?(tlee)
Keywords: regression
It seems a dup of bug 1224433.  I make this bug depending on bug 1224433 now.  And, I will check it later to make sure if this is a duplication.
Depends on: 1224433
Flags: needinfo?(tlee)
(In reply to Nathan Froyd [:froydnj] from comment #6)
> (In reply to Robert O'Callahan (:roc) (Mozilla Corporation) from comment #5)
> > Nathan, can you try that patch?
> 
> It doesn't help. :(  The repainting seems to happen a little faster,
> perhaps..? but there's still flashing and every image on the page getting
> repainted several times.

I saw such symptoms without my patch, and I definitely don't see them with the patch.

Try results look good so I think we should take the patch anyway.
Bug 1224976. Recover from singular-matrix cairo errors. r=mattwoodrow
Attachment #8690628 - Flags: review?(matt.woodrow)
Attachment #8690628 - Flags: review?(matt.woodrow) → review+
Comment on attachment 8690628 [details]
MozReview Request: Bug 1224976. Recover from singular-matrix cairo errors. r=mattwoodrow

https://reviewboard.mozilla.org/r/25895/#review23273
Robert, I just update patches at bug 1224433.  I find it solves the symptoms here too.
(In reply to Robert O'Callahan (:roc) (Mozilla Corporation) from comment #11)
> Created attachment 8690628 [details]
> MozReview Request: Bug 1224976. Recover from singular-matrix cairo errors.
> r=mattwoodrow
> 
> Bug 1224976. Recover from singular-matrix cairo errors. r=mattwoodrow

Is this patch relative to bug 1212483?
https://hg.mozilla.org/mozilla-central/rev/8684df98e649
Status: NEW → RESOLVED
Closed: 4 years ago
Resolution: --- → FIXED
Target Milestone: --- → mozilla45
Assignee: nobody → roc
Depends on: 1230033
Wontfix for 43, though this could potentially still make it into 44.
Ryan, why is this FF44 unaffected? A typo perhaps?
Flags: needinfo?(ryanvm)
This was a regression from bug 1097464, which was backed out from 44. Therefore, 44 is unaffected.
Flags: needinfo?(ryanvm)
(In reply to Ryan VanderMeulen [:RyanVM] from comment #23)
> This was a regression from bug 1097464, which was backed out from 44.
> Therefore, 44 is unaffected.

Thanks Ryan! Since 44 is unaffected, there is no need to track it.
This along with the change that caused this regression were backed out from Firefox 45.

https://hg.mozilla.org/releases/mozilla-aurora/rev/64ec448f156d
You need to log in before you can comment on or make changes to this bug.