splitmuxsink: Forward EOS messages from async fragments.

Re-enable forwarding EOS messages from fragments that are completing
asynchronously, so that splitmuxsink itself won't go EOS until they
are complete. This was disabled to work around a bug in core that
is fixed in
https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/683

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-good/-/merge_requests/798>
This commit is contained in:
Jan Schmidt 2020-10-29 02:36:35 +11:00 committed by Jan Schmidt
parent 1316dd9c65
commit f0c24319de

View file

@ -1178,16 +1178,15 @@ eos_context_async (MqStreamCtx * ctx, GstSplitMuxSink * splitmux)
helper->pad = sinkpad; /* Takes the reference */
ctx->out_eos_async_done = TRUE;
/* HACK: Here, we explicitly unset the SINK flag on the target sink element
* that's about to be asynchronously disposed, so that it no longer
* participates in GstBin EOS logic. This fixes a race where if
* splitmuxsink really reaches EOS before an asynchronous background
* element has finished, then the bin won't actually send EOS to the
* pipeline. Even after finishing and removing the old element, the
* bin doesn't re-check EOS status on removing a SINK element. This
* should be fixed in core, making this hack unnecessary. */
GST_OBJECT_FLAG_UNSET (splitmux->active_sink, GST_ELEMENT_FLAG_SINK);
/* There used to be a bug here, where we had to explicitly remove
* the SINK flag so that GstBin would ignore it for EOS purposes.
* That fixed a race where if splitmuxsink really reaches EOS
* before an asynchronous background element has finished, then
* the bin wouldn't actually send EOS to the pipeline. Even after
* finishing and removing the old element, the bin didn't re-check
* EOS status on removing a SINK element. That bug was fixed
* in core. */
GST_DEBUG_OBJECT (splitmux, "scheduled EOS to pad %" GST_PTR_FORMAT " ctx %p",
sinkpad, ctx);