703872 - Firefox poor performance/hang on svg replicate with filters. window shows "not responding"

Status: RESOLVED FIXED

(Core :: SVG, defect)

Description

[david.dailey]

User Agent: Mozilla/5.0 (Windows NT 6.1; WOW64; rv:8.0) Gecko/20100101 Firefox/8.0
Build ID: 20111104165243

Steps to reproduce:

http://srufaculty.sru.edu/david.dailey/svg/replicate/repPathTurb2.svg


Actual results:

Firefox crashes


Expected results:

It should have rendered the file as in Opera

Comment 1

[Robert Longson [:longsonr]]

Do you really mean crash. If so please type about:crashes in the URL bar and post the crash report Id here. Or do you mean takes a really really really really long time to render the drawing.

Comment 2

[david.dailey]

https://crash-stats.mozilla.com/report/index/bp-a4ee3251-2fd5-47d0-a36f-a10032111118

Actually, this time I waited a bit longer and eventually, (maybe after a minute) the browser actually rendered the image -- then it crashed -- loss of control via address box, back button, and ability to close with window close box --I had to use task manager to stop the process.

Comment 3

[Robert Longson [:longsonr]]

I can't get it to crash. It did chew up the CPU and made the browser unresponsive for as long as I could be bothered to run it though.

Comment 4

[Jonathan Watt [:jwatt]]

(In reply to david.dailey from comment #2)
> then it crashed -- loss of control via address box, back button, and ability
> to close with window close box --I had to use task manager to stop the process.

What you're describing there is what we'd call a hang, not a crash. A crash is when the process dies unexpectedly of its own accord, in which case there would be nothing for you to stop using the task manager.

Can you clarify whether the other times that you experienced this bug the process died, or whether it hung?

The crash with the crash-stats.mozilla.com URL that you've given doesn't give any indication of being anything to do with SVG, so I suspect that may be a report from an earlier crash that you experienced that's unrelated to this bug? (The link in about:crashes should have a date and time that should give you a clue.)

Of course a hang is even more serious than a crash, since some users don't even know how to end task. I'm just trying to clarify what you've experienced.

Comment 5

[david.dailey]

Well, it is a bit hard to tell quite what is going on. For example, when I tried it this time, after a minute or so it finally rendered (like I expect it to). Then the browser window seems to lose all responsiveness. The cursor turns to a two directional vertical arrow. If I click on title bar, then the message "not responding" appears. As I go back and forth from this window (IE) to that window (FF), by clicking on it,  it is sluggish to re-emerge back on top, and then the cursor becomes a rotating circle. Trying to give focus to the FF window sometimes takes two mouse clicks.

 Clicking on the close box doesn't work properly. I click on the close, and get a message "Firefox is not responding"  which then disappears (sometimes) before I can click on it. I tried again though and it did after saying "Firefox is not responding" let me terminate the process (without having to use the task manager).

Restarting the browser, it tris to go to the same page that caused the "hang/crash". After another minute, it indeed loads the page and places me right back where I was.

The second time through, I get a "well this is embarrassing" message. from there I chose not to restore the tab, but to start a new session.

typing about:crashes in the new window, just shows the same early morning crash report I tried offering earlier.

Hope this helps.

By the way, there are two or three other examples at http://srufaculty.sru.edu/david.dailey/svg/SVGOpen2010/replicate.htm that seem to nag FF -- I've flagged them with a note of some sort about hanging FF (4) though this is still true of FF8.  I'm not sure what these examples might have in common.

(sorry they aren't simpler)
David

Comment 6

[Boris Zbarsky [:bzbarsky]]

Site linked in comment 0 is fast for me on Mac.  Is this the cairo path code biting us again?

Comment 7

[Wayne Mery (:wsmwk)]

http://srufaculty.sru.edu/david.dailey/svg/replicate/repPathTurb2.svg totally fried my firefox.  Had to kill it. Mozilla/5.0 (Windows NT 6.1; Win64; x64; rv:16.0) Gecko/16.0 Firefox/16.0  06-24 build

but cpu is only ~15%



  Graphics

        Adapter Description
        ATI Radeon HD 4550

        Vendor ID
        0x1002

        Device ID
        0x9540

        Adapter RAM
        512

        Adapter Drivers
        aticfx64 aticfx64 aticfx32 aticfx32 atiumd64 atidxx64 atiumdag atidxx32 atiumdva atiumd6a atitmm64

        Driver Version
        8.881.0.0

        Driver Date
        7-28-2011

        Direct2D Enabled
        true

        DirectWrite Enabled
        true (6.1.7601.17789)

        ClearType Parameters
        DISPLAY1 [ Gamma: 2200 Pixel Structure: RGB ClearType Level: 100 Enhanced Contrast: 200 ] DISPLAY2 [ Gamma: 2200 Pixel Structure: RGB ClearType Level: 50 Enhanced Contrast: 300 ]

        WebGL Renderer
        Google Inc. -- ANGLE (ATI Radeon HD 4550   ) -- OpenGL ES 2.0 (ANGLE 1.0.0.1041)

        GPU Accelerated Windows
        0

        AzureBackend
        direct2d

Comment 8

[Jonathan Watt [:jwatt]]

Switching to a tab containing the SVG from comment 0 still hangs for a bit for me using latest Nightly. The profiler shows 98% of the time being spent in nsSVGFilterFrame::PaintFilteredFrame(). We really need bug 869496 to be fixed to get better perf here.

Comment 9

[Markus Stange [:mstange]]

Attachment: v1

The testcase has transform="scale(10, 0.5)", and the current code uses 10 as the scale factor for both dimensions, so we end up processing 20x the pixels we need to.

Comment 10

[Markus Stange [:mstange]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/f8e43006b159

Comment 11

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/f8e43006b159