Problem with regression tests getting stuck

[Paul Millar]

Hi everyone,

I've had the regression tests freeze on me again (see [1] for previous
report). Sylvain has reported problems with the wininet regression tests
(described in [2]). The symptoms are the same, but for me the problem is
only intermittent. This time I caught the problem with gdb and obtained
the following info:

[...]
Loaded symbols for /lib/libresolv.so.2
0x402e7744 in __libc_close () at __libc_close:-1
-1      __libc_close: No such file or directory.
        in __libc_close
(gdb) bt
#0  0x402e7744 in __libc_close () at __libc_close:-1
#1  0x400ec9a4 in __DTOR_END__ ()
   from /home/paulm/WINE-cvs/wine/dlls/libntdll.dll.so
#2  0x400b3df5 in WaitForMultipleObjectsEx (count=0, handles=0x0, 
wait_all=0, 
    timeout=100, alertable=0) at ../../scheduler/synchro.c:257
#3  0x400b3b88 in Sleep (timeout=100) at ../../scheduler/synchro.c:178
#4  0x40358747 in winapi_test (flags=268435456) at tests/http.c:121
#5  0x40358b36 in func_http () at tests/http.c:184
#6  0x40358f9d in run_test (name=0xbffffc63 "http.c") at wtmain.c:244
#7  0x40357037 in __wine_exe_main () at wininet_test.exe.spec.c:133
#8  0x400aff84 in start_process () at ../../scheduler/process.c:528
#9  0x400b3f3f in call_on_thread_stack (func=0x400afd2c)
    at ../../scheduler/sysdeps.c:108

So, to summarise (with some extra bits of information):

The problem occurs when wininet/tests/http.c is testing asynchronous
HttpSendRequest(), so flags > 0. The problem is triggered at line 121 of 
http.c, i.e.:
120    while ((flags)&&(!goon))
121        Sleep(100);

Sleep(), in turn, just calls WaitForMultipleObjectsEx(). From 
scheduler/synchro.c:
176	VOID WINAPI Sleep( DWORD timeout )
177	{
178	    WaitForMultipleObjectsEx( 0, NULL, FALSE, timeout, FALSE );
179	}

WaitForMultipleObjectsEx() has problems when it gets to 257:
256	        SERVER_END_REQ;
257	        if (ret == STATUS_PENDING) ret = wait_reply( &cookie );
258	        if (ret != STATUS_USER_APC) break;

ret is 1079729528, btw.

AFAIK, __DTOR_END__ (see near bottom of stack) is the destructor for a
function call in ntdll. I'm probably missing something obvious, but I
don't see the connection between scheduler/synchro.c:257 and ntdll. The
closest is wait_reply() (in scheduler/client.c) calls wait_reply_data (in
same file), which calls NtCurrentTeb(), which is commented as an NTDLL
function. But there doesn't seem to be an implementation of NtCurrentTeb
in ntdll.

Any ideas?

Cheers,

Paul.

[1] http://www.winehq.com/hypermail/wine-devel/2002/08/0006.html
[2] http://www.winehq.com/hypermail/wine-devel/2002/07/0386.html

PS, I'll create a bugzilla in a bit.

----
Paul Millar