61630 - Not able to render some web sites - causes problems in performance comparision test runs.

Status: VERIFIED FIXED

(SeaMonkey :: General, defect, P2)

Description

[chris hofmann]

Version: Mozilla/5.0 (Windows; U; Win98; en-US; m18) Gecko/20001031

This is a browser-general bug, Mozilla is not able to render the following 
sites, eventually timeout with missing images. This does not occur to Netscape 
4.7 or IE, so it could be problem with the Gecko rendering engine.

URLs:
http://www.target.com
http://www.hitbox.com
http://www.moviefone.com
http://www.lowestfare.com
http://www.biography.com



------- Additional Comments From [e-mail] 2000-11-27 23:33:34 ----

hmm, I just tried this on the Netscape 6 rtm release on WinNT and did not have 
any problems with rendering images on any of the above sites except for 
http://www.biography.com.

This bug should be moved to Bugzilla.



------- Additional Comments From [e-mail] 2000-11-29 14:53:02 ----

Some images are not rendering.



------- Additional Comments From [e-mail] 2000-11-30 08:08:19 ----

I just tried this with netscape6 rtm on win98.
all the sites appeared to load ok except for
http://www.target.com

the exact error message I received was
"operation timed out when attempting to contact www.target.com"

I just tried 4.7x and also got the error response back that
"server could be down or not responding"

It sounds like we need to construct a good test case
where with pages where all or parts of the content
served is unavailable or simulate network flakeyness
to make good head way on understanding what might be happening
and protect against error conditions.   Its easy to think
about banner ads served from separated sites going down
and possible confusing page loading.

maybe tever has such a test????




------- Additional Comments From [e-mail] 2000-11-30 11:33:17 ----

and now latter in the day http://www.target.com

a few weeks back while on sabatical I was watching a cnbc
story about a entire reorg of the target web site.  I bet
this flakey behavior might be related to them trying
to keep the site running stable after a lot of hardware
and content reorg...

see folks, valuable thing come out of sabaticals... ;-)

we still need to design some good test case to simulate
flakey sites to make sure the browser behaves correctly,
and we need to make sure any performance
related tests accurately report "site unreachable"
and remove it from page loading time stats.
maybe this bug should be split into two for those
different areas but lets keep it together until
we know more.




------- Bug moved to this database by [e-mail] 2000-11-30 11:49 -------

This bug previously known as bug 3141 at http://bugscape.netscape.com/
http://bugscape.netscape.com/show_bug.cgi?id=3141
Originally filed under the Gecko Embedding product and Komodo component.

Unknown version unspecified in product Browser. Setting version to "other".
The original reporter of this bug does not have
   an account here. Reassigning to the person who moved
   it here, [e-mail].
   Previous reporter was [e-mail].

Comment 1

[Tom Everingham]

Sorry Chris, I don't have such a test at the moment.  I will look into setting 
up a page later today with missing content from various sources.  One question, 
have you tried this on winNT or only win98?  All the sites listed are working 
for me.

Comment 2

[Scott MacGregor]

Several things jump out at me right away:
1) Several of the sites (like www.biography.com) use document.layers. Of course
we don't support this in NS6. That could be causing part of the problems.

2) In all the cases where we failed to fire an on end document load for these
urls, the only remaining channel in the load group was our dummy layout channel.
For some reason it isn't getting removed for these websites. Rick helped cook up
this channel so I'll cc him on this bug too. 

Comment 3

[Scott MacGregor]

I *think* this problem may have to do with the parser. The dummy layout channel
is never being removed from the load group. this is because the parser is never
call DidBuildModel on the html content sink. 

Comment 4

[Scott MacGregor]

Attachment: band aid fix...(not the correct / final solution)

Comment 5

[Scott MacGregor]

I've attached a band aid fix which at least allows the pages to properly load
and more importantly terminate. All I'm doing is allowing the dummy layout
channel to be removed when the # of outstanding remaining requests == 0.
Regardless of the state of the mIsLoading flag.

This flag is cleared only when the parser calls DidBuildModel. For these web
pages it isn't calling this method. The right fix is to keep digging and figure
out why this notification isn't going out.

I'll keep poking but I'm a little lost in the layout / parser code. 

Comment 6

[Scott MacGregor]

I'm going to try to bring a big gun for this puppy. cc'ing rickg.

Hey Rick, so I'm having trouble loading some urls: www.target.com,
www.biography.com are a couple of examples.

Everything in the page lays out correctly. However, the html parser is never
firing a DidBuildModel notification on the content sink. I've stepped through
the debugger and the parser does receive an OnStopRequest notification for the
over-all document but that doesn't turn into a call to DidBuildModel for these
particular web pages.

I'm still debugging but if you have any insights on what kind of conditions
could prevent the htmlparser from firing a DidBuildModel it would be helpful.
Thanks!

Comment 7

[chris hofmann]

Here is the latest list of urls where
the 'timeout' problem was seen in a recent test run. 
      http://www.ACCOUNTONLINE.COM 
      http://www.GOVERNMENTGUIDE.COM 
      http://events.ticketmaster.com 
      http://www.AUTOBYTEL.COM 
      http://www.DRKOOP.COM 
      http://www.SEARS.COM 
      http://www.WELLSFARGO.COM 
      http://www.WINVITE.COM 
      http://www.INSIDETHEWEB.COM 
      http://www.BIZRATE.COM 
      http://www.HITBOX.COM 
      http://www.ACCUWEATHER.COM 
      http://www.NAPSTER.COM 
      http://www.NETSETTER.COM 
      http://www.REGISTER-ONCE.COM 
      http://www.TARGET.COM 
      http://www.sciam.com 
      http://www.TICKETMASTER.com 

Comment 8

[harishd]

mscott wrote:

I'm still debugging but if you have any insights on what kind of conditions
could prevent the htmlparser from firing a DidBuildModel it would be helpful.

(A) DidBuildModel() may not get fired when the parser is blocked ( on loading 
JS,CSS, etc.. ).

Comment 9

[Scott MacGregor]

When I was last looking at this, I tried loading each of the individual parts of
www.target by hand and didn't see anything we couldn't load or choked on
individually. I also looked at an http log and it appeared that we were getting
responses for each part when loading the overall document.

Not sure if that helps....

Comment 10

[Phil Peterson]

Add nsbeta1, embed keywords. This appears to be important for embedding.

Comment 11

[harishd]

Attachment: Testcase.

Comment 12

[harishd]

CC ing myself. 

Btw, I'm working on tracking down the problem.

Comment 13

[harishd]

mscott was right. I did not hit did build model at all!!!!!! 

Investigating.

Comment 14

[harishd]

Reassigning bug to myself. 

Comment 15

[harishd]

Moving to my plate for REAL!

Comment 16

[harishd]

Ok...the good news is that I'm very close to a fix. The bad news, however, is 
that I'm not 100% sure if it is the correct fix! Will update soon.

The bottom level is that we are ending up with a wrong parser context which 
messes up the timing.

Comment 17

[harishd]

Attachment: Proposed patch v1.1 [ I haven't done all the necessary testings yet ]

Comment 18

[Heikki Toivonen (remove -bugzilla when emailing directly)]

r=heikki for the patch harishd showed me. It is not the one in v1.1, right?
Please add the correct patch here as well.

Comment 19

[harishd]

Attachment: Patch v1.2 ( Much simpler )

Comment 20

[harishd]

Fix is in. Marking FIXED.

Comment 21

[harishd]

Ummm. ticketmaster.com & accuweather still seem to have the problem. Reopening
the bug until I fix all the sites.

Comment 22

[harishd]

Attachment: Reduced case derived from ticketmaster.com.

Comment 23

[harishd]

Ok, the problem with ticketmaster.com ( refer the reduced case id=24223 ) seems 
to be different. It's unable to load the image. However, if I replace the image 
src with an image in my local server 
( http://ortcloud.mcom.com/harishd/Bugs/Images/aim.gif) I don't see the problem.

Marking FIXED again.

Comment 24

[harishd]

Attachment: Reduced case derived from www.accuweather.com [ same problem as ticketmaster.com - Ref. my previous comment ]

Comment 25

[Grace Bush]

verified fixed