Original commit message from CVS:
* tests/check/generic/clock-selection.c: (GST_START_TEST):
set to NULL before unreffing, fixes a valgrind leak.
Why was this not triggering the error that an object needs to
be NULL before unreffing ?
* win32/common/config.h:
update
Original commit message from CVS:
* gst/playback/gststreamselector.c:
(gst_stream_selector_set_property),
(gst_stream_selector_bufferalloc):
Preserve the existing buggy streamselector behaviour by performing
a fallback buffer allocation when downstream isn't linked yet.
This should really be fixed in playbin by blocking pads until it's
linked them.
Also, use gst_pad_alloc_buffer instead of
gst_pad_alloc_buffer_and_set.
