Hi, Zebediah.

Thanks for the detailed research.
I've revert the first patch, applied the next one and the test was successful!

Is the code fix as intended? If it is ok, I will send a PATCH v2.

diff --git a/dlls/quartz/systemclock.c b/dlls/quartz/systemclock.c
index 73a923a7b99..0eb339420db 100644
--- a/dlls/quartz/systemclock.c
+++ b/dlls/quartz/systemclock.c
@@ -338,6 +338,7 @@ HRESULT system_clock_create(IUnknown *outer, IUnknown **out)
list_init(&object->sinks);
InitializeCriticalSection(&object->cs);
object->cs.DebugInfo->Spare[0] = (DWORD_PTR)(__FILE__ ": SystemClockImpl.cs");
+ InterlockedIncrement(&object_locks);

TRACE("Created system clock %p.\n", object);
*out = &object->IUnknown_inner;

2021年8月11日(水) 8:56 Zebediah Figura (she/her) <zfigura@codeweavers.com>:

On 8/10/21 6:40 PM, 粟田大樹 wrote:
> Hi, Zebediah. Thank you for your quick reply.
>
> It was a very interesting problem and I logged the changes in object_locks.
> In my tests, object_locks goes to 0 without system_clock_inner_Release
> being called.
> The game I tested does not have a free version, so I apologize for not
> sharing it. The game is clickable to skip video playback, and the crash
> occurs when clicking.
>
> I have attached the log of my test for your reference. Thank you,

From examination of the code I believe the problem is that we call
system_clock_create() from dsound_render_create(), which bypasses the
increment.

Either that should be changed to CoCreateInstance, or we should only
increment object_locks from creation functions. I'm kind of leaning
toward the latter, even though it's more code, since it's more
consistent (along the lines of "always grab and release a resource in
the same place").

I note also that enum_reg_filter_create() and enum_moniker_create()
don't reference the module, and should.