316811 - Hang with spinning cursor and over 90% CPU usage (_NSAppKitLock hang?)

Status: RESOLVED WORKSFORME

(Camino Graveyard :: General, defect)

Description

[Ben Fowler]

User-Agent:       Mozilla/5.0 (Macintosh; U; PPC Mac OS X Mach-O; en-US; rv:1.8) Gecko/20051107 Camino/1.0b1
Build Identifier: Mozilla/5.0 (Macintosh; U; PPC Mac OS X Mach-O; en-US; rv:1.8) Gecko/20051107 Camino/1.0b1

Whilst typing into the text box, I scrolled the page up and Camino 
went into a hang with beachball for 3 hours.

Reproducible: Didn't try

Actual Results:  
Hang

Expected Results:  
Allow me to type into box, read what I had written and scroll the
page; not beachball

There are few if any performance bugs still open for Camino: Bug 288894 "Floating
 ad locks up Camino (except for dragging the browser window title bar)" is one,
but I don't think that this is the same.

I have also read 
http://forums.mozillazine.org/viewtopic.php?t=341745&sid=4fc0798af171765b875ee4285deeacee
which is possibly related (there is some javascript for counting characters).

I will attach a graph from sampler.

Comment 1

[Ben Fowler]

Attachment: Graph from sampler

Comment 2

[Smokey Ardisson (offline for a while; not following bugs - do not email)]

I can't repro this.  There is s flash ad on the page if you turn of ad-blocking, but this character-counter seems well-behaved and I don't notice a perf hit at all, let alone a SBOD.

How fast is your Mac, and what OS?

Comment 3

[Ben Fowler]

(In reply to comment #2)
> I can't repro this.  There is a flash ad on the page if you turn of
> ad-blocking, but this character-counter seems well-behaved and I don't notice a
> perf hit at all, let alone a SBOD.
> 
> How fast is your Mac, and what OS?

15" Powerbook G4 = 500Mz 10.2.6

I haven't specificially tried to reproduce it, but it hasn't happened since.
It is a puzzle because I was under the impression that there were no unknown
performance problems with Camino.

I wondered if there were some sort of loop in networking code; I turned add 
blocking about a week ago because a site I was using was plastered with 
multiple copies of an animated ad and the page was unusable.

Comment 4

[Smokey Ardisson (offline for a while; not following bugs - do not email)]

Flash perf is still bad in some cases (particularly on older, slower Macs); if you run a search for open Camino bugs with perf in the keyword field, you'll see several are Flash bugs (and there's another Flash-causes-hang bug with a trunk build that doesn't show up on the list).  The gist of most of them is that we're hitting our heads against the QuickDraw ceiling....

Mike, Simon, anything interesting in this sample?

Comment 5

[Simon Fraser [no longer active]]

The sample doesn't show flash eating CPU; it shows lots of CPU being used for scrolling.
The test URL takes me to a login page. Is that the correct page?

Comment 6

[Smokey Ardisson (offline for a while; not following bugs - do not email)]

It takes me to a page that lets you enter a post (but also has verification and username/password fields and options for subscribing).

Comment 7

[Simon Fraser [no longer active]]

No more pointless acronyms please.

I think this is hanging in _NSAppKitLock, which we've seen before.

Comment 8

[Ben Fowler]

The page should be "Post a comment" and there should be a text box labelled
"Message".

If there is not, then you are not seeing the same page, sorry.

I can't really improve on my comment 0 , so I repeat:

1. The problem occurred during (and I would guess was evinced by) scrolling 
the page.
2. The hang lasted 3 hours and didn't look as though it was going to
end spontaneously.
3. There are few, if any, open performance issues for Camino; I was
surprised to see even one and thought to file a bug about it.
4. The other issues which I located whilst searching for Dups (and mentioned
for completeness) are likely to be unrelated; but I can't be sure.

Acronyms: See 
http://www.hyperborea.org/journal/archives/2005/06/21/spinning-beach-balls-of-death/
and http://vjarmy.com/archives/2003/08/beach_ball_myst.php .

As I understand it, 
http://www.cocoabuilder.com/archive/message/cocoa/2004/2/7/96405 
some types of drawing can cause deadlock under particular circumstances, and
there may be no specific recommended action; but one can use a semaphore.

NSAppKitLock doesn't make many appearances in this Bugzilla, so I can't be
definite, but it is possible that we may have to disclaim support for Jaguar
less than 10.2.8, and this report, if not reproduced, becomes WFM (unsupported
OS version)

Comment 9

[Smokey Ardisson (offline for a while; not following bugs - do not email)]

Torben, since you're the only other person we know with a 10.2.8 Mac, can you see if you can repro this?

Comment 10

[Torben]

(In reply to comment #9)
> Torben, since you're the only other person we know with a 10.2.8 Mac, can you
> see if you can repro this?

The site doesn't give me any problems, I can type and scroll as much as I like. 

Comment 11

[Smokey Ardisson (offline for a while; not following bugs - do not email)]

Ben, have you seen this again?  If not, I'm tempted to WFM it since Torben can't repro it.

Comment 12

[Ben Fowler]

No.

I do have another NSAppKitLock hang with sample (which I can attach here or
start a new bug report).

I suspect that this is some sort of race condition, and that several things 
have to happen perhaps in a particular order, for the problem to occur. In
the instance that I reported, I had many pages open, some for several days.

It is possible that this is Apple's bug, and they have fixed it in later 
revisions of the O.S.

I am currently on 10.2.6, and in light of light of what I know, and Simon's 
comments, I am perfectly prepared to accept that this problem will not be
evident in versions later that this. Obviously only Apple can state whether
this is so, and I don't expect them to be interested, or indeed to confirm
bugs. Later this week, I expect to upgrade to Panther and I may therefore
have few, if any, opportunities to check Jaguar-specific problems.

I would be going out on a limb in the sense that I have no information
to confirm this, but we may have to say that support for Mac OS less
than 10.2.8 cannot be verified for Firefox and is not guaranteed. It won't
be long before Firefox cannot guarantee support Mac OS earlier than
Panther, of course!

It might be a little early to say WFM, unless this is the resolution for
bugs that can't be readily reproduced (irrespective of whether the feature
'works').

You can resolve it as INVALID - unsupported OS, provided that you are
prepared to specify in the documentation the OS versions concerned.