1230320 - theverge.com paints very slow

Status: RESOLVED WORKSFORME

(Core :: Graphics, defect, P3)

Description

[Kartikaya Gupta (email:[e-mail])]

It's very easy to checkerboard on theverge.com - both desktop and Fennec.

Comment 1

[Mason Chang [Inactive] [:mchang]]

(In reply to Kartikaya Gupta (email:[e-mail]) from comment #0)
> It's very easy to checkerboard on theverge.com - both desktop and Fennec.

What kind of machine? I remember this being a lot better after we did some box shadow improvements. I can't checkerboard right now on a 27" iMac.

Comment 2

[bull500]

(In reply to Mason Chang [:mchang] from comment #1)
> (In reply to Kartikaya Gupta (email:[e-mail]) from comment #0)
> > It's very easy to checkerboard on theverge.com - both desktop and Fennec.
> 
> What kind of machine? I remember this being a lot better after we did some
> box shadow improvements. I can't checkerboard right now on a 27" iMac.

For android i had posted a comment and pictures to describe the problem:

https://bugzilla.mozilla.org/show_bug.cgi?id=1213873#c24

Pics:
https://bugzilla.mozilla.org/attachment.cgi?id=8695401

Comment 3

[Kartikaya Gupta (email:[e-mail])]

(In reply to Mason Chang [:mchang] from comment #1)
> What kind of machine?

I was able to do it on a Macbook Air. Although now that I'm trying more it's a lot easier to do during page load than afterwards. But we can focus on Android if you prefer, that's the main motivation for filing this bug - on a Nexus 4 I can reproduce checkerboarding even after load.

Comment 4

[Jeff Muizelaar [:jrmuizel]]

We're spending a fair amount of time in argb32_image_mark_argb32 being called from PopGroupAndBlend.
This seems like we're scaling an the result of an opacity group which is not great and probably deserves more investigation.

Other than that the painting profile doesn't seem to bad on OS X and suspect we'd get more bang from making sure were painting the right things and not throwing away things unnecessarily.

It would be interesting to know what the story is on Windows.

Comment 5

[bull500]

(In reply to Jeff Muizelaar [:jrmuizel] from comment #4)
> We're spending a fair amount of time in argb32_image_mark_argb32 being
> called from PopGroupAndBlend.
> This seems like we're scaling an the result of an opacity group which is not
> great and probably deserves more investigation.
> 
> Other than that the painting profile doesn't seem to bad on OS X and suspect
> we'd get more bang from making sure were painting the right things and not
> throwing away things unnecessarily.
> 
> It would be interesting to know what the story is on Windows.

I tested this site on windows as well with Nightly 45.0a1 (2015-12-04)
I used my mouse scroll wheel and it does happen there as well.
Its really difficult to reproduce because you really need to scroll super quick.   
Happens more easily while loading but also occurs after the site has been completely loaded. 

Another thing I'd like to add is about Fennec Nightly(Android)
This slow painting thing affects long page sites like http://www.theatlantic.com/world/ and http://techcrunch.com/ or something has a lot of content to scroll like a long article. 
It's not as bad as theverge but it definitely happens there as well.  
Same STR as the verge scroll up and down on those sites and you'll be greeted with patches of missing content or slow text sharpening.

Comment 6

[Gregory Pappas [:gregp]]

Unable to reproduce in current versions.