More comments, try to make it clearer, release p_block and p_pic in only one
place. It also fixes a p_block leak if we were not able to create a p_pic.

According to ISO 14496-10 7.4.1.2.4,
next first NAL of next picture appears when
nal_ref_ids differs with one being equal to 0.

The decoder can now detect and avoid this case. Retaining the
conversion is counter-productive (especially for transcode).

This partially reverts commit 84b23f96,
which is no longer necessary.

This makes checking for video output support before activation a lot
simpler (especially avoiding callbacks).

This effectively reverts commit b667100b.

The decoder can request another format, or even retry the same format
with better luck next time. So do not set b_error. b_error is for
unrecoverable failures, such as if the decoder module is upset.

It's now safe to close and open MediaCodec during playback.

Signed-off-by: Thomas Guillem <thomas@gllm.fr>

And fix locking around fmt_out in case of hwaccel.

The decoder downstream is not allowed to return smaller pictures than
requested. Only alignment can still potentially be an issue.

ffmpeg_NewPictBuf() already takes care of it.

Allow the mouse cursor to be hidden in regular active video windws, in addition
to fullscreen videos.
Signed-off-by: David Fuhrmann <dfuhrmann@videolan.org>

Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

--
stangely a similar patch was merged and is now gone oO
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Add the delay to ts_bound when calling input_clock_ConvertTS (For audio,
i_ts_bound is 3000ms).

Fixes #4847