137667 - Web Page pegs the CPU at 100%

Status: VERIFIED DUPLICATE of bug 123273

(Core :: DOM: Core & HTML, defect)

Description

[Dameon D. Welch-Abernathy]

Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.0rc1) Gecko/20020415

Leaving this page causes the CPU to drop. Seen this on the 0.9.9 release build
as well as the current 1.0 branch builds.

Appears javascript related since disabling Javascript seems to eliminate the
issue. Filed as a Javascript bug.

Comment 1

[Phil Schwartau]

Confirming bug using Mozilla trunk binary 20020415xx WinNT.
Loading the page brings the CPU up to 100%, but note the browser
does not hang. It is still responsive to commands (though slowly).

The site uses DHTML to draw a menu on the left side of the page:
"Equipment", "Rate Plans", etc. This menu does not load in Mozilla.
The allotted space is blank. I will attach a reduced testcase below.


The external JS files in use at the site are:

     http://www.attws.com/mobileinternet/include/global.js
     http://www.attws.com/mobileinternet/include/new_menu.js
     http://www.attws.com/mobileinternet/include/mi_home.js

For some reason, the view-source: protocol does not allow me to see
these files in any browser. I can see them in the Mozilla JavaScript 
Debugger. There, the mi_home.js file states it is based on the DHMTL
tool "AllWebMenus v2.0" 

Comment 2

[Phil Schwartau]

Attachment: Reduced HTML testcase

Comment 3

[Dameon D. Welch-Abernathy]

Test case appears to peg the CPU, albeit at 80% or so. There's still some idle
time left, but pegging it like that can't be good. :-)

Comment 4

[Phil Schwartau]

When I load the reduced testcase in NN4.7 or IE6, I get that little 
DHTML menu in the upper left. In Mozilla, this menu doesn't load.
The throbber stops throbbing, but no menu appears. 

This is the same problem that occurs at the given URL. When I load
the testcase off the Bugzilla server (i.e. not locally), my CPU
is at 100% non-stop.

I get no errors in the JavaScript Console. Will try interrupting
a debug build to get a stack trace. The JS is output from the
AllWebMenus tool, and is obfuscated: it appears with few
carriage returns and opaque function names. 


Here are a couple functions that I've formatted:


function awmCreateMenu (cll,swn,swr,mh,ud,sa,mvb,dft,crn,dx,dy,ss,
                        ct,cs,ts,tn,ttt,ti,tia,dbi,ew,eh,jcoo,jcoc)
{
  if (awmmo>=0)
    awmmo++;
  else 
  {
    awmm=new Array();
    awmmo=0
  }

  var me={
          ind:awmmo,
          nm:awmMenuName,
          cn:new Array(),
          fl:!awmsc[cs].pos,
          cll:cll,
          mvb:mvb,
          dft:dft,
          crn:crn,
          dx:(ct<2)?dx:0,
          dy:dy,
          ss:ss,
          sht:"<STYLE>"+awmsht+"</STYLE>",
          rep:0,mio:0,
          st:awmOptimize?2:3,
          addSubmenu:awmas,
          ght:awmmght,
          whtd:awmmwhttd,
          buildMenu:awmbmm, <<<----- (I printed out this function below)
          cm:awmmcm
         };

  me.pm=me;
  me.addSubmenu(ct,swn,swr,mh,ud,sa,1,cs,ts,tn,ttt,ti,tia,dbi,ew,eh,jcoo,jcoc);
  me.cn[0].pi=null;
  
  if (mvb)
    document.onmousemove=awmotmm;
  
  awmm[awmmo]=me;
  awmsht="";

  return me.cn[0];
}



function awmbmm()
{
  document.onmousedown = awmodmd;
  status = "AllWebMenus: Building Menu #" + (this.ind + 1);
  this.ght();
  this.whtd();
  awmdb(this.ind);
  status = awmDefaultStatusbarText;
  awmdid = setInterval("awmd()", 100);
  awmsoo = awmso + 1;
}



Notice the use of setInterval(); perhaps this plays a role 
in the CPU usage  - (?)

Comment 5

[Phil Schwartau]

Attachment: WinNT stack traces from interrupting page load of testcase

Comment 6

[Phil Schwartau]

The farthest my stack traces ever get are nsAppShell::Run
Reassigning to Embedding:Docshell for help on this.

NOTE: I don't seem to have CPU problems on Linux or Mac9.1,
only WinNT. The DHTML menu doesn't load, but I don't seem to
see such high CPU usage. Can anyone else confirm this? 

NOTE: same behavior for the home site of the "AllWebMenus"
DHTML menu builder, http://www.likno.com/ . Mozilla doesn't
load the DHTML menu into the blue area to the left of the page
on any platform; and on Windows (only?) the CPU goes to 100%.

Comment 7

[Adam Lock]

The mi_home.js has some complicated code at the top that document.write's a
<script> tag to include this js:

http://www.attws.com/include/menus/awmData-mi_home/awmlib2.js

This file is pretty obfuscated but contains a lot of setTimeouts calls. One in
particular looks a likely culprit - awmdb() is called from awmbmm() which is
called awmCreateMenu() which is called during page loading. The source of this
methods is as follows:

function awmdb(mi)
{
 var crc=awmm[mi].cn[0];
 if (document.getElementById(crc.id).offsetWidth>0)
 {
   if (!awmm[mi].cll) crc.show(1);
 }
 else
   setTimeout("awmdb("+mi+")",0);
}

Notice that it calls setTimeout with a duration of 0 which results in it being
repeatedly called.

My guess is our setTimeout implementation in nsGlobalWindow should put a limit
on how short the delay may be so that silly JS that specifies tiny values, e.g.
0, 3, 5 etc. actually results in a larger interval (e.g. 10ms) to stop the CPU
being pegged.

Reassigning to DOM on that basis.

Comment 8

[Adam Lock]

Attachment: Simple testcase

This simple test case has foo() calling setTimeout("foo()", 0).

Comment 9

[Phil Schwartau]

Adam: thanks.

Perhaps this bug is therefore a dupe of bug 123273:
"setTimeout(something, 0) causes 100% CPU constant"

Note: if the "AllWebMenus" template for DHTML menus becomes
a popular tool, lots of sites will have this problem -

Comment 10

[Adam Lock]

Marking it a dupe

*** This bug has been marked as a duplicate of 123273 ***

Comment 11

[Sivakiran Tummala]

marking verified