819205 - [Gaia::Browser] Large empty regions appear for several seconds when panning through a webpage

Status: RESOLVED DUPLICATE of bug 986694

(Firefox OS Graveyard :: Performance, defect, P3)

Description

[Peter La]

Attachment: Video

What makes it feel slow/broken?

When panning through a webpage, the portion of the page that you just scrolled to will display as a white rectangle, and it can take roughly 1-3 seconds before that portion is rendered.  This creates a very disjointed browsing experience.  It occurs even when the page has loaded.

Did it prevent you from doing what you wanted? Why?

No, but it makes browsing an unpleasant, disjointed feeling activity.  It makes the page feel... not solid, nor continuous.

How does this make you feel?

[ ]  :)  I feel happy about it
[ ]  :|  Meh
[X]  :(  I'm upset
[ ] >:O  I'm angry

Device: Unagi running Dec. 6th nightly.

Details: (technical factors, FPS, app startup time, ms elapsed, etc)

Bonus: can you attach a video of the problem?
Attached.

Comment 1

[Peter La]

Additional info:

The webpage accessed in this example was www.thestar.com, a Toronto Newspaper.

Comment 2

[Chris Lee [:clee]]

Do we have other bugs that block this bug or can dupe against?

Comment 3

[Ben Francis [:benfrancis]]

This seems to be Gecko being slow to paint the content of this web site. Chris Jones might have a better idea of whether there's any more work we have planned that could improve this, but I'm not currently aware of any further changes we can make on the Gaia side to help.

Is Core -> Graphics a better component for this?

Comment 4

[Josh Carpenter [:jcarpenter]]

Chris, do you anticipate this improving for v1?

Comment 5

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

We always have irons on the fire ;).

ajones, do either of these dep bugs make a significant difference for this testcase?

Comment 6

[Anthony Jones (:ajones, :kentuckyfriedtakahe, :k17e)]

What's with these slow news web sites. nzherald, nytimes, thestar. All of them have mobile versions of the site but don't detect Firefox 

822402 makes fling faster without making the blankness issue considerably worse. I'm not sure about 825809 but I don't expect it will make a difference. Fennec uses tiling and performs a lot better so perhaps the best approach would be to get everything working the same.

Comment 7

[leo.bugzilla.gecko]

This issue is reproduced until noe in leo device.
According to upper memtion, many portal and news site give the pc site to FFOS.
User has to do zoom in and panning frequently.
In that case, User will feel that the browser has low performance.

Comment 8

[Jet Villegas (inactive)]

I'm not sure what is leo+ here, the UA detection outreach or code changes to improve painting on pan/zoom. Comment 6 suggests moving to tiling vs. buffer rotation. Would we leo+ such a patch?

Comment 9

[Jason Smith [:jsmith]]

(In reply to Jet Villegas (:jet) from comment #8)
> I'm not sure what is leo+ here, the UA detection outreach or code changes to
> improve painting on pan/zoom. Comment 6 suggests moving to tiling vs. buffer
> rotation. Would we leo+ such a patch?

leo+ means this is a blocking bug for 1.1. Are you implying we shouldn't block on this?

Comment 10

[Jet Villegas (inactive)]

I'm stating that I don't understand the risk profile for leo+ at this point. 

Calling up a handful of news sites to detect our UA and serve m.content is very low risk, but that's not this bug. Changing our painting scheme is high risk (though work is in progress) but it's not clear to me that release-drivers should leo+ such a change at this point.

Comment 11

[Jason Smith [:jsmith]]

Sounds good. Makes sense to me.

Setting back to leo? to minus based on comment 10's rationale.

Comment 12

[Gregor Wagner [:gwagner]]

Is this still a valid bug after fixing bug 825809?

Comment 13

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Changing to tiling for 1.1 sounds way too risky to me.

Outreach to Web sites to serve us more mobile versions is way less risky and gives a bigger payoff (although it's more work I guess).

Comment 14

[Anthony Jones (:ajones, :kentuckyfriedtakahe, :k17e)]

I get the same results as in the video. Notice that the tx/rx icons are flashing. The problem appears to be that the content hasn't finished loading but we've scrolled both the url bar and the progress candy cane off the top of the screen.

Perhaps we could keep the candy cane on the screen (as an overlay so nothing jumps when it disappears) until the page load is complete?

Comment 15

[Gregor Wagner [:gwagner]]

(In reply to Robert O'Callahan (:roc) (Mozilla Corporation) from comment #13)
> Changing to tiling for 1.1 sounds way too risky to me.

Is it already on trunk or do we have a time estimate when this will be implemented? 

(In reply to Anthony Jones (:kentuckyfriedtakahe) from comment #14)
> I get the same results as in the video. Notice that the tx/rx icons are
> flashing. The problem appears to be that the content hasn't finished loading
> but we've scrolled both the url bar and the progress candy cane off the top
> of the screen.

It looks like this also happens after the page finished loading. Or does the page never finish loading?

Comment 16

[Robert O'Callahan (:roc) (email my personal email if necessary)]

(In reply to Gregor Wagner [:gwagner] from comment #15)
> (In reply to Robert O'Callahan (:roc) (Mozilla Corporation) from comment #13)
> > Changing to tiling for 1.1 sounds way too risky to me.
> 
> Is it already on trunk or do we have a time estimate when this will be
> implemented?

I don't know.

Comment 17

[Anthony Jones (:ajones, :kentuckyfriedtakahe, :k17e)]

(In reply to Gregor Wagner [:gwagner] from comment #15)
> Is it already on trunk or do we have a time estimate when this will be
> implemented?

Tiling isn't implemented for b2g. It isn't likely to get implemented until after the subframe scrolling stuff is complete. There are still a lot of differences between fennec and b2g that need to be cleaned up first. I'm also less inclined to think tiling will help with this specific issue.

> (In reply to Anthony Jones (:kentuckyfriedtakahe) from comment #14)
> It looks like this also happens after the page finished loading. Or does the
> page never finish loading?

The loading takes about 60 seconds. During this time the CPU is close to 100% busy.

Comment 18

[Alex Keybl [:akeybl]]

This is a risky feature request, and it's too late in the 1.1 timeframe to be taking change of this nature. Given no new perf requirements in 1.1, not a blocker.

Comment 19

[Hubert Figuiere [:hub]]

Bug 915673 could be a related fix / solution.

Comment 20

[Kevin Grandon :kgrandon]

Tagging, but this may be fixed now with APZ.

Comment 21

[Mike Lee [:mlee]]

Mason,

Is this this still a concern after all of the scrolling-related changes we've landed in the past few months?

Mike