[PATCH v3 3/3] winegstreamer: Replace source pad interface with GstAppSrc.

[Zebediah Figura]

I got this running tests:

leslie at terabithia:~/git/wine64/dlls/quartz/tests$ make && 
~/git/wine/tools/runtest waveparser
make[1]: Entering directory '/home/leslie/git/wine64'
make[1]: Nothing to be done for 'dlls/quartz/tests/all'.
make[1]: Leaving directory '/home/leslie/git/wine64'
waveparser.c:154: Test marked todo: Got hr 0x80004002, expected 0.
0118:fixme:ole:CoCreateInstanceEx no instance created for interface 
{56a86895-0ad4-11ce-b03a-0020af0ba770} of class 
{d51bd5a1-7548-11cf-a520-0080c77ef58a}, hr 0x80004002.

(wine:329482): GLib-GObject-WARNING **: 15:30:24.992: instance with 
invalid (NULL) class pointer

(wine:329482): GLib-GObject-CRITICAL **: 15:30:24.992: 
g_signal_emit_valist: assertion 'G_TYPE_CHECK_INSTANCE (instance)' failed
quartz\tests\quartz_test.exe: 
../wine/dlls/winegstreamer/wg_parser.c:584: wg_parser_push_data: 
Assertion `ret == GST_FLOW_OK || ret == GST_FLOW_FLUSHING' failed.


It can be reproduced every time if you add a g_usleep(100000) right 
before the gst_buffer_new_and_alloc() call.

The problem is that you're (implicitly) destroying the appsrc in 
wg_parser_disconnect(), but the read thread isn't shut down until later.

There are a few ways to solve this. We have to keep the disconnect -> 
shutdown read thread -> destroy order as-is (because the read thread 
needs to be unblocked if it's blocked in wg_parser_get_next_read_offset, 
and we can't destroy the parser while anything still has a pointer to 
it), but given that, I think the right solution here is probably to 
return from wg_parser_push_data() if !parser->sink_connected.