680786 - Firefox4 and later hangs while displaying large page with many copies of an icon

Status: RESOLVED FIXED

(Core :: Layout, defect, P1)

Description

[Alice0775 White]

Attachment: Saved page as web complete, zipped file

Build Identifier:
http://hg.mozilla.org/mozilla-central/rev/6dc468c41136
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:9.0a1) Gecko/20110821 Firefox/9.0a1 ID:20110821030758

I was encountered a problem while I tested Bug 680767.

The Bug 680767 is mentioned for Firefox3.6 linux.
But this is Firefox4.0 and later version.

Firefox4.0 and later version is almost hang up.
On the other hand, Firefox3.6.20, Initial page is displayed within about 10 sec.

Tested with @2.5GHz Core2Quad.

I found a regression range with attachment.

Regression range(cached m-c hourly)
Fast(initial page is displayed within about 10 sec/@2.5GHz Core2Quad):
http://hg.mozilla.org/mozilla-central/rev/5f857be14db9
Mozilla/5.0 (Windows; Windows NT 6.1; WOW64; rv:2.0b3pre) Gecko/20100727 Minefield/4.0b3pre ID:20100728145944
Very slow(browser almost hang up):
http://hg.mozilla.org/mozilla-central/rev/b8b62b351c09
Mozilla/5.0 (Windows; Windows NT 6.1; WOW64; rv:2.0b3pre) Gecko/20100728 Minefield/4.0b3pre ID:20100728152620
Pushlog:
http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=5f857be14db9&tochange=b8b62b351c09
Triggered by:
Bug 572520 - make all imgIDecoderObserver notifications asynchronous

Comment 1

[Alice0775 White]

Attachment: WinDbg log

I got the log in the following procedures
Open the file
Wait about 20 sec.
Brake debugger

Comment 2

[Alice0775 White]

This also happen on Linux build with attachment 554748 [details] .
http://hg.mozilla.org/mozilla-central/rev/6dc468c41136
Mozilla/5.0 (X11; Linux i686; rv:9.0a1) Gecko/20110821 Firefox/9.0a1 ID:20110821030758

Comment 3

[Boris Zbarsky [:bzbarsky]]

So I think there may be two things going on here:

1)  The async OnStartContainer triggers an unconditional reflow request in
    nsImageFrame::OnStartContainer which requires us to find the right frame,
    etc.  I have a fix for that.
2)  The async OnStopRequest means that we hit FindUnfinishedRequestCallback in
    the docloader for each of the image requests while a bunch of other image
    requests are in the hashtable.  Of course none of these have any status
    message, so the whole operation is pointless.  This is covered by bug
    642551.

Alice0775, are you willing to retest with the patches I'm about to upload here and in bug 642551?

Comment 4

[Boris Zbarsky [:bzbarsky]]

Attachment: nsImageFrame::OnStartContainer should not assume that its intrinsic size has changed.

Comment 5

[Alice0775 White]

(In reply to Boris Zbarsky (:bz) from comment #3)
> 
> Alice0775, are you willing to retest with the patches I'm about to upload
> here and in bug 642551?

I built in locally and tested on Windows7 i686
Built from changeset 1720b28e3115 + attachment 554968 [details] [diff] [review] + attachment 554969 [details] [diff] [review] : it works very well.And initial page(attachment 554748 [details]) is displayed within about 10 sec.

Comment 6

[Boris Zbarsky [:bzbarsky]]

Thank you for checking that!

Comment 7

[Boris Zbarsky [:bzbarsky]]

http://hg.mozilla.org/integration/mozilla-inbound/rev/a0d3c755b289

Comment 8

[Marco Bonardo [:mak]]

http://hg.mozilla.org/mozilla-central/rev/a0d3c755b289