754000 - Javascript/SVG app updates less frequently on Linux

Status: RESOLVED INCOMPLETE

(Core :: SVG, defect)

Description

[Timvde]

User Agent: Mozilla/5.0 (X11; Linux x86_64; rv:15.0) Gecko/15.0 Firefox/15.0a1
Build ID: 20120510030517

Steps to reproduce:

Go to http://mbostock.github.com/d3/ex/voronoi.html and hover with your mouse over the applet. Happens in Linux x86 and Linux x64.


Actual results:

Bad performance, stuttering


Expected results:

It should've been a fluent experience like it is in Firefox 12

Comment 1

[Boris Zbarsky [:bzbarsky]]

This worksforme on Mac, so may have to do with the usual path rendering mess.  Need a regression range....

Requesting tracking for the regression.

Comment 2

[Timvde]

Requested by Mardeg, this is from the x64 pc (Linux Mint Lisa):

  Graphics

        Adapter Description
        NVIDIA Corporation -- GeForce 9600M GT/PCIe/SSE2

        Vendor ID
        NVIDIA Corporation

        Device ID
        GeForce 9600M GT/PCIe/SSE2

        Driver Version
        3.3.0 NVIDIA 295.20

        WebGL Renderer
        NVIDIA Corporation -- GeForce 9600M GT/PCIe/SSE2 -- 3.3.0 NVIDIA 295.20

        GPU Accelerated Windows
        1/1 OpenGL

        AzureBackend
        skia


Also tested with HWA disabled (which it is by default btw, I force enabled it), but I still get the same results.

Comment 3

[Matthias Versen [:Matti]]

wfm with  Mozilla/5.0 (Windows NT 6.1; rv:15.0) Gecko/15.0 Firefox/15.0a1 SeaMonkey/2.12a1 (great build ID in the UA string :-( )

Comment 4

[Robert Longson [:longsonr]]

For me, with Firefox 12 on Windows the yellow shape resizes smoothly as the mouse moves about. With trunk the yellow shape resizes less frequently and therefore the movement appears somewhat jerky.

Comment 5

[Boris Zbarsky [:bzbarsky]]

Aha!  Robert, do you think you can hunt down a regression range?

Comment 6

[Timvde]

bz: I am willing to hunt down the regression range too, but I don't have time until tomorrow. Sorry for not mentioning that before :)

Comment 7

[Robert Longson [:longsonr]]

There's no performance difference. It's an illusion...

The application works quite differently in trunk and on 12. On trunk the yellow shape snaps to where the black lines on the diagram intersect each other on 12 it attaches to the lines rather than the intersection of lines and resizes smoothly along the lines.

Do you not see this on a Mac? If not do you see the 12 or the trunk behaviour I'm describing.

Comment 8

[Robert Longson [:longsonr]]

I'll leave it to Tim to find the regression range. A reduced testcase would be nice to since it's a functional change rather than a performance issue.

Comment 9

[Robert Longson [:longsonr]]

FWIW Opera matches Firefox 12 for me.

Comment 10

[Boris Zbarsky [:bzbarsky]]

On Mac, with both a 13 beta and a trunk nightly build, I see the behavior you describe as seeing in 12.

Comment 11

[Timvde]

I think that is not true. If you move your mouse slowly on trunk, you'll see it intersects everywhere on the line. It might appear like it's only on the intersections, but that's because it creates new ones.

Comment 12

[Robert Longson [:longsonr]]

Perhaps you're right, if you move the mouse really slowly it does seem to go on the lines. It's not CPU bound on my machine though, it just doesn't seem to update very often.

Comment 13

[Timvde]

Using mozregression to bisect, I found this to contain the regression:

Last good nightly: 2012-03-21
First bad nightly: 2012-03-22

Pushlog:
http://hg.mozilla.org/mozilla-central/pushloghtml?startdate=2012-03-21&enddate=2012-03-22

So the regression is not in Firefox 13, but in Firefox 14. The reason why I thought I was seeing it in 13 too, is because somehow the build Ubuntu/Mint deliver in their repositories appears to be faster than Mozilla's (stable Firefox 12) download. So it's faster than the 13 beta too, which is about the same speed as FF 12. But there also appears to be a real regression.

However, it is very hard to compare performance in a visual way if I can't view them side by side. But I switched back and forth multiple times and I actually think there is a performance difference between those builds.
Robert: as you appear to be the only one who could reproduce, can you confirm this?

Comment 14

[Robert Longson [:longsonr]]

Well that makes a lot more sense. It must be bug 734079

Comment 15

[Jonathan Watt [:jwatt]]

Looks like it from that bisect range, although that bug doesn't actually show in it. The builds from before and after bug 734079 landed are (respectively):

https://ftp.mozilla.org/pub/mozilla.org/firefox/nightly/2012/03/2012-03-21-03-11-51-mozilla-central/

https://ftp.mozilla.org/pub/mozilla.org/firefox/nightly/2012/03/2012-03-22-03-12-20-mozilla-central/

Tim, can you confirm that things work in the former, but not in the latter?

Comment 16

[Robert Longson [:longsonr]]

I can't see any regression at all even with the latest trunk if I use nightly builds. The only difference I see is when I compare with a self built debug trunk.

Comment 17

[Jonathan Watt [:jwatt]]

That's not entirely unexpected, since I added a fair bit of debug code in the patches for bug 734079. We only really care about the release build performance of course.

If Tim can't reproduce using the builds linked in comment 15, then maybe we're looking at the wrong range.

Comment 18

[Robert Longson [:longsonr]]

it's quite possible, bug 734079 is just an educated guess at this point.

Comment 19

[Timvde]

Well, I think there is a small difference indeed, but I can only confirm my last comment. It is hard to see a difference between those builds, but I switched between them multiple times and I think it is there. It is, however, not "the big regression". On exactly the same (clean) profile, Firefox downloaded from http://www.mozilla.org/en-US/firefox/fx/ is *very* noticeably slower than the one that comes preinstalled by Ubuntu/Mint, and that's why I thought this was such a big regression, which it isn't.

Comment 20

[Robert Longson [:longsonr]]

Perhaps there's some difference in compiler flags such that the preinstalled version is more tailored to your cpu? Are there differences in what about:buildconfig says about your Ubuntu installation compared to the standard Mozilla build?

Comment 21

[Robert Longson [:longsonr]]

Or maybe the ubuntu build is running hardware accelerated and the mozilla build isn't. Is there any difference in the graphics stuff at the end about:support

Comment 22

[Alex Keybl [:akeybl]]

Adding qawanted - we should use the STR in comment 0 across the builds in comment 15 (where it broke) and builds around https://bugzilla.mozilla.org/show_bug.cgi?id=734079#c36 (where it may be fixed) on Mac/Linux. We'd like to understand whether this performance regression is visible in any way. Thanks!

Comment 23

[Paul Silaghi, QA [:pauly]]

http://mbostock.github.com/d3/ex/voronoi.html is working fine on FF 14b8 on Mac OS X 10.6.

On Ubuntu 12.04 I see a very small deterioration between the builds from comment 15, and definetely a bigger one between nightly 2012-03-22 and FF 14b8. Anyway, I don't see a 100% smooth move neither in FF 12 (compared with FF 14b8 on Mac OS X 10.6 for e.g).

Comment 24

[Alex Keybl [:akeybl]]

(In reply to Paul Silaghi [QA] from comment #23)
> http://mbostock.github.com/d3/ex/voronoi.html is working fine on FF 14b8 on
> Mac OS X 10.6.
> 
> On Ubuntu 12.04 I see a very small deterioration between the builds from
> comment 15, and definetely a bigger one between nightly 2012-03-22 and FF
> 14b8. Anyway, I don't see a 100% smooth move neither in FF 12 (compared with
> FF 14b8 on Mac OS X 10.6 for e.g).

Can we get a regression window on Linux please? That'll help us resolve the regression.

Comment 25

[Paul Silaghi, QA [:pauly]]

(In reply to Alex Keybl [:akeybl] from comment #24)
> (In reply to Paul Silaghi [QA] from comment #23)
> > http://mbostock.github.com/d3/ex/voronoi.html is working fine on FF 14b8 on
> > Mac OS X 10.6.
> > 
> > On Ubuntu 12.04 I see a very small deterioration between the builds from
> > comment 15, and definetely a bigger one between nightly 2012-03-22 and FF
> > 14b8. Anyway, I don't see a 100% smooth move neither in FF 12 (compared with
> > FF 14b8 on Mac OS X 10.6 for e.g).
> 
> Can we get a regression window on Linux please? That'll help us resolve the
> regression.

Ever since FF 4. Tested on Ubuntu 11.10 and 12.04 32-bit.

Comment 26

[Paul Silaghi, QA [:pauly]]

And it's getting worse between the builds from comment 15 and still deteriorating after that until the latest nightly (I couldn't find a regression range for this, I think it's getting worse progressively in several times)

Comment 27

[u279076]

Mozilla/5.0 (X11; Ubuntu; Linux x86_64; rv:13.0) Gecko/20100101 Firefox/13.0.1
> Result is smooth with no lag whatsoever

Mozilla/5.0 (X11; Linux x86_64; rv:14.0) Gecko/14.0 Firefox/14.0a1 ID:20120314031139
> Lag is somewhat noticeable, this is the first 14.0 Nightly

Mozilla/5.0 (X11; Linux x86_64; rv:14.0) Gecko/14.0 Firefox/14.0a1 ID:20120401041203
> Lag is fairly evident at this point

Given I can reproduce the lag in the first 14.0 Nightly and 13.0 is unaffected, I'm not sure where else to debug this.

Comment 28

[u279076]

Interestingly, I can also detect a very small lag in Mozilla/5.0 (X11; Linux x86_64; rv:13.0) Gecko/13.0 Firefox/13.0a1 ID:20120301031135.

Comment 29

[juan becerra [:juanb]]

Firefox 13.0.1 comes pre-installed in my Ubuntu machine, and the example in comment #0 works really well. 

All versions I downloaded from our ftp servers and installed on the desktop, 4, 5, 10, 11, 12, 13, 13.0.1, 14 (as in comment #15), 14.0b11, or latest nightly, all 64bit versions, were noticeably slow.

I don't know how to install previous versions of Fx on Ubuntu such as they were installed by default with the system, but if anyone knows how to do it, please ping me so I can give it a try.

Comment 30

[Timvde]

Sorry for not replying here anymore, I was busy with exams and doing stuff for my youth movement :)

The biggest difference is probably that Ubuntu compiles Firefox with GCC 4.6 instead of 4.5.

I can attach the output of about:buildconfig here, but I'm not sure how you would like it (plaintext in a comment or as an attachment, as HTML, ...) and I guess it's not very hard to get it anyway :)

Comment 31

[u279076]

Tim, the best way would be a plain-text file attached to this bug (not as a comment), thanks.

Comment 32

[u279076]

Tim, would you be able to assist us in tracking down a regression range for this bug? We're having trouble tracking down a way to get older Ubuntu Firefox versions installed.

Comment 33

[Timvde]

Attachment: about:buildconfig on stock Ubuntu Firefox 13.0.1

Comment 34

[Timvde]

Attachment: about:buildconfig on Firefox 13.0.1 downloaded from Mozilla

As far as I know, Ubuntu only provides the last Firefox version and the one that was the default on release date in its repository. So if you download Ubuntu 11.10, you can try Firefox 7. I think 12.04 packed Firefox 12.

Comment 35

[u279076]

Given my comment 27, I'm not convinced it's compiler related because all of our Linux builds use the same compiler. I think we need to work to narrow down the regression range between the very first Firefox 14.0 Nightly build and Firefox 13.0. I suppose it's possible that a regression got merged in from inbound, but I don't know how to look into this.

I think we need to hand the ball off to a developer who knows this code.

Comment 36

[Virgil Dicu [:virgil] [QA]]

Mozilla/5.0 (X11; Ubuntu; Linux i686; rv:17.0) Gecko/17.0 Firefox/17.0

I've downloaded the daily firefox ppa from Ubuntu (Nightly equivalent) and the app works as slow there as on any Firefox downloaded from ftp. It's installed in the same manner as the default Firefox that comes with Ubuntu, where the app works without problems. 
https://launchpad.net/~ubuntu-mozilla-daily/+archive/ppa

Unless I'm missing something, this would mean that the Ubuntu versions of Firefox regressed somewhere as well, between 14 and 17. 

(In reply to Anthony Hughes, Mozilla QA (:ashughes) [OoO until July 31] from comment #35)
> Given my comment 27, I'm not convinced it's compiler related because all of
> our Linux builds use the same compiler. I think we need to work to narrow
> down the regression range between the very first Firefox 14.0 Nightly build
> and Firefox 13.0. I suppose it's possible that a regression got merged in
> from inbound, but I don't know how to look into this.

I've loaded the app with both Firefox 14.0.1 and 13.0.1, but can't spot a big difference between performance for those builds. Is the 13.0.1 mentioned in comment 27 the Ubuntu packed Firefox?

Comment 37

[u279076]

Virgil, can you please do some more comparative testing? What is your UX with the following builds:

 * Firefox 13.0.1, 14.0.1, 15.0b3, 17.0a1 from ftp.mo
 * Firefox 13.0.1, 14.0.1, 15.0b3, 17.0a1 from Ubuntu ppa

Comment 38

[Virgil Dicu [:virgil] [QA]]

>  * Firefox 13.0.1, 14.0.1, 15.0b3, 17.0a1 from ftp.mo

I can't see any noticeable difference between these builds. There is a big lag when moving the pointer over the diagram in all of them.

>  * Firefox 13.0.1, 14.0.1, 15.0b3, 17.0a1 from Ubuntu ppa

*Firefox 13.0.1 and 14.0.1 from Ubuntu PPA work ok - the planes are highlighted and updated, don't work as well as Chrome and Opera when the mouse is moved faster along the diagram though
*Firefox 15beta, Aurora and Nightly - all behave badly, as all the ftp.mozilla builds

http://vid.ly/5c3t4p - screencast showing comparative behavior of Chrome and 14.0.1 from Ubuntu
http://vid.ly/5f6b1m - screencast showing behavior comparative behavior of Chrome and 15 beta from Ubuntu.

This is Ubuntu 12.04 i686.
Also tried disabling any other add-ons instaled with the software through the Software center(raster fonts, unity integration, browser plugin, gnome support). That didn't change the way the app behaved in any situation.

Comment 39

[u279076]

Based on comment 38, I would assume the regression in comparing Firefox versions to have occurred within Firefox 15 Nightly. In terms of competitive comparison to Chrome and Opera, the regression likely would have occurred in Firefox 13 Nightly or earlier.

Can you investigate this any further Virgil?

Comment 40

[Alice0775 White]

I can reproduce in Windows7 with HWA disabled

Regression window(m-c)
Good:
http://hg.mozilla.org/mozilla-central/rev/3de967c4624c
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:14.0) Gecko/20120321 Firefox/14.0a1 ID:20120321025041
Bad:
http://hg.mozilla.org/mozilla-central/rev/f10862ac3217
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:14.0) Gecko/20120321 Firefox/14.0a1 ID:20120321033140
Pushlog:
http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=3de967c4624c&tochange=f10862ac3217


Regression window(m-c)
Good:
http://hg.mozilla.org/integration/mozilla-inbound/rev/43cc4d6d160a
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:14.0) Gecko/20120320 Firefox/14.0a1 ID:20120320051441
Bad:
http://hg.mozilla.org/integration/mozilla-inbound/rev/34454de86833
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:14.0) Gecko/20120320 Firefox/14.0a1 ID:20120320051642
Pushlog:
http://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=43cc4d6d160a&tochange=34454de86833

Comment 41

[Virgil Dicu [:virgil] [QA]]

(In reply to Anthony Hughes, Mozilla QA (:ashughes) from comment #39)
> Based on comment 38, I would assume the regression in comparing Firefox
> versions to have occurred within Firefox 15 Nightly. In terms of competitive
> comparison to Chrome and Opera, the regression likely would have occurred in
> Firefox 13 Nightly or earlier.
> 
> Can you investigate this any further Virgil?

This isn't a regression in Firefox builds (on Linux, at least). My comment was a follow-up on what already was investigated by Juan and Paul. To resume, I see the following on Ubuntu (same as Juan in comment 29):

----Firefox downloaded from mozilla ftp:
*All* Firefox versions exhibit a laggy animation when moving the pointer over the diagram. (tested from 4.0.1 to current Nightly, going through 9, 13, 14, 15, 16). There might be very small differences, but the animation is not smooth in any case.

----When using Firefox builds from Ubuntu:
13.0.1 and 14.0.1  work fine (not as perfect as Chrome, but fine - see videos in comment 38)
15 beta, Aurora 16 and Nightly are as laggy as the Firefox versions 

If there isn't any other installation issue involved here, this issue is a regression in Firefox Ubuntu builds starting with 15. So that's the only regression involved here.

Comment 42

[Virgil Dicu [:virgil] [QA]]

Maybe we should split the Linux and windows bugs...

This never worked with Firefox builds.
The only way to go would be to investigate what exact problem triggered the issue in Ubuntu builds, but that won't be as easy as bisecting with Firefox. 
After asking around on Ubuntu's chat, managed to get some links to archived versions:
http://archive.ubuntu.com/ubuntu/pool/main/f/firefox/
http://ppa.launchpad.net/ubuntu-mozilla-daily/ppa/ubuntu/pool/main/f/firefox-trunk/

But there don't seem to be archived versions of 14 Nightlies, though. 
If there is any directions to go from here, please advise. At the moment, there isn't much left to do.

Comment 43

[Robert Longson [:longsonr]]

Is there any difference in about:buildconfig in the compiler/configure options between Ubuntu 13/14 and 15/16? Note that about:buildconfig from Ubuntu 13/14 is attached to this bug already.

Comment 44

[Virgil Dicu [:virgil] [QA]]

Attachment: about:builconfig of 15 beta Ubuntu version

Can't see an y difference between the buildcongfig information. Exact asame arguments and compiler version.

Comment 45

[Jonathan Watt [:jwatt]]

Definitely not going to get to this for 15.

Comment 46

[Lukas Blakk [:lsblakk] use ?needinfo]

(In reply to Jonathan Watt [:jwatt] from comment #45)
> Definitely not going to get to this for 15.

Un-tracking for 15 then, we'll keep this on our radar for 16.

Comment 47

[Jonathan Watt [:jwatt]]

Spun off bug 754000 to possibly continue investigating our performance.

Comment 48

[Jonathan Watt [:jwatt]]

(In reply to Jonathan Watt [:jwatt] from comment #47)
> Spun off bug 754000

I mean I span off bug 1348781.