Hang in mshtml/tests/dom.c

[Dan Kegel]

While running mshtml/tests/dom.c on a slow dual processor
under valgrind, I got the following hang:
...
  7 0x04b22c1c GetMessageA+0x60(msg=0x7f21fdc8, hwnd=0x0, first=0x0,
last=0x0) [dlls/user32/message.c:2976] in user32 (0x7f21fda8)
  8 0x04946eb1 run_domtest+0x91(str=0x495fab0, test=0x493daa4)
[dlls/mshtml/tests/dom.c:2597] in mshtml_test (0x7f21fdf8)
  9 0x0494711c func_dom+0x45() [dlls/mshtml/tests/dom.c:2652] in
mshtml_test (0x7f21fe18)
...

The offending GetMessage call is in the following loop in tests/dom.c:

while(!doc_complete && GetMessage(&msg, NULL, 0, 0)) {
        TranslateMessage(&msg);
        DispatchMessage(&msg);
    }

The doc_complete test has been in that file since the test was added;
git blame says it was added by this commit:

commit 1534267ba500156e7965fdc399247ca69eb7c014
Author: Jacek Caban 
Date:   Sun Jun 24 21:09:03 2007 +0200

mshtml: Added beginning DOM tests.

Jacek, could you have a look and see if there's some obvious race you could fix?

Thanks,
Dan

[Jacek Caban]

Hi Dan,

Dan Kegel wrote:
> While running mshtml/tests/dom.c on a slow dual processor
> under valgrind, I got the following hang:
> ...
>   7 0x04b22c1c GetMessageA+0x60(msg=0x7f21fdc8, hwnd=0x0, first=0x0,
> last=0x0) [dlls/user32/message.c:2976] in user32 (0x7f21fda8)
>   8 0x04946eb1 run_domtest+0x91(str=0x495fab0, test=0x493daa4)
> [dlls/mshtml/tests/dom.c:2597] in mshtml_test (0x7f21fdf8)
>   9 0x0494711c func_dom+0x45() [dlls/mshtml/tests/dom.c:2652] in
> mshtml_test (0x7f21fe18)
> ...
>
> The offending GetMessage call is in the following loop in tests/dom.c:
> while(!doc_complete && GetMessage(&msg, NULL, 0, 0)) {
>     TranslateMessage(&msg);
>     DispatchMessage(&msg);
> }
>
> The doc_complete test has been in that file since the test was added;
> git blame says it was added by this commit:
>
> commit 1534267ba500156e7965fdc399247ca69eb7c014
> Author: Jacek Caban 
> Date:   Sun Jun 24 21:09:03 2007 +0200
> mshtml: Added beginning DOM tests.
>
> Jacek, could you have a look and see if there's some obvious race you could fix?

I know about this bug and it's not easy to fix (because it's hard to 
reproduce it). It's on my TODO list.

Jacek

[Dan Kegel]

On Thu, Jul 31, 2008 at 1:50 PM, Jacek Caban  wrote:
> > While running mshtml/tests/dom.c on a slow dual processor
> > under valgrind, I got the following hang:
>
> I know about this bug and it's not easy to fix (because it's hard to
> reproduce it). It's on my TODO list.

OK.  It hasn't happened again; it only happened when I overloaded
a 1GHz single-core machine by doing "make -j2 test" with my
parallel test patch and valgrind.  That was an accident,
I thought I was on a dual core machine.  So I'm quite happy
without a fix for now.
- Dan