58276 - bringing up dialog while network test is running freezes app

Status: RESOLVED INCOMPLETE

(Core :: XUL, defect, P2)

Description

[sairuh (rarely reading bugmail)]

tever found this using today's commercial branch bits. also occurs using y'day
bits 2000.10.26.13 (on winNT, sp5).

to repro:
1. load the URL to start test.
2. while test is running, open a modal dialog [eg, prefs or open web location].
3. attempt to dismiss the dialog [click cancel, esc, etc.].

results:
unable to dismiss dialog, and, even though the test runs to completion, the app
is essentially frozen. unable to dismiss dialog even after end of test.

Comment 1

[sairuh (rarely reading bugmail)]

essentially a 'crash'. but no talkback.

unable to surf or do much else, even after test completes. i wonder if this
occurs in a trunk build? cc'ing doron/asa to see if it does, or on other win32
os's. according to tpreston, this doesn't occur on win2k. adding pp kw.

Comment 2

[Asa Dotzler [:asa]]

it sucks to be on the trunk too ;)  This does break us on the trunk mozillla
builds 102704 NT4

Comment 3

[Peter Trudelle]

->danm/M18/P2/Critical, nominate for rtm

Comment 4

[John Morrison]

This zdnet test simply chains a series of pages together by setting
|onload='document.location.href="nextDoc.html"';|. The equivalent testcase, 
which just reloads itself 60 times and will also hang-up the browser is at 
  http://jrgm.mcom.com/bugs/58276/test.html
(I will attach the source if anyone outside the firewall would like to look
at it). 

This will hang on win2k (and I would think any win32). On Linux it does not 
hang, and curiously, exiting the dialog kills the page-loading. On Mac, 
when the dialog is up, all application activity ceases aside from the dialog, 
so no hang there. 

I'm not sure about the rtm for this. While the zdnet tests are high-profile,
this is a narrow set of conditions. (In fact, if I use a .setTimeout to 
delay the next page load by any amount, even "0 msec", this will not lock
up in my tests). It is only when the onload handler immediately initiates
another load that a modal dialog can be hung-up. (Actually, I'd be more 'sure'
about rtm if this was likely to be a simple fix, but I doubt it).

Comment 5

[Doron Rosenberg (IBM)]

bug 56934 might be related, as it is also the ZDNET bench thing. As this is a
crasher, nominating for m0.9 should this not make rtm (wondering why the target
is m18, oh well).

Has anyone tested this on linux/mac?  I'll try to get this tested on linux

Comment 6

[selmer (gone)]

Does the ibench test explicitly create this situation, or does it only happen
why you try to mess with the browser while the test is running?  Would ZDNet
ever do this while the tests are running?  If not, it seems like rtm- is the
right call.

Comment 7

[Doron Rosenberg (IBM)]

pdt - this is a need info from the developer this is assigned to probably.

56934 has been marked fixed btw.

Comment 8

[sairuh (rarely reading bugmail)]

doron, this isn't a problem on Mac or Linux, at least with branch bits.

Comment 9

[Doron Rosenberg (IBM)]

se: I had someone on irc who said he saw this on linux, turned out it was some
profile issue. my mistake.

Comment 10

[Peter Trudelle]

rtm-/future

Comment 11

[Matthias Versen [:Matti]]

i can confirm this on win2k with build 2000102704 (trunk)
After the benchmark test i hit the run button on that page and mozilla
freeze.

Comment 12

[Wayne Mery (:wsmwk)]

none of the test URLs exist anymore