va: basedec: Should unmap src frame when dst frame map failing.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1692>

va: bufferpool: use release_buffer to clean the mem.

The current bufferpool wastes all pre-allocate buffers when the
buffer pool is actived.
The pool->priv->size is 0 for va buffer pool. And every time, the
reset_buffer() will clean all mem and make the buffer size 0, that
can cache the gst_buffer in the buffer pool.
But when the buffer pool is activing, the default_start() just
allocate the buffer and release_buffer() immediately, all the pre
allocated buffers and surfaces are destroyed because of
gst_buffer_get_size (buffer) != pool->priv->size.
We need to use release_buffer() to do the clean job at the pool
start time.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1686>

h265parse: Don't enable passthrough by default

SEI messages contain various information which wouldn't be conveyed
by using upstream CAPS (HDR, timecode for example).

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1639>

cameracalibrate: Improve gst-inspect documentation

Thanks to @kazz_naka on Twitter

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1691>

wpesrc: add some debug logging around WPEView creation/destruction

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1663>

wpesrc: fix a memory leak of the bytes

free the previous GBytes if load-bytes is called multiple times
before view creation.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1663>

wpesrc: only create webview if not already created

e.g. _decide_allocation() can be called multiple times throughout the
element's lifetime and we only want to create the view once rather than
overwriting.

Fixes a leak of the WPEView under certain circumstances.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1663>

wpe: free a previous pending image/shm buffer

Don't blindly overwrite a possibly previously set buffer.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1663>

srtsrc: Prevent `delay` from being negative

`delay` should be a GstClockTimeDiff since SRT time is int64_t.

All values are in local time so we should never see a srctime that's in
the future. If we do, clamp the delay to 0 and warn about it.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1674>

srtsrc: Don't calculate a delay if the srctime is 0

A zero srctime is a missing srctime. Apparently this can happen when
["the connection is not between SRT peers or if Timestamp-Based Packet
Delivery mode (TSBPDMODE) is not enabled"][1] so it may not apply to us,
but it's best to be defensive.

[1]: https://github.com/Haivision/srt/blob/v1.4.2/docs/API.md#sending-and-receiving

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1674>

srtsrc: Defend against missing clock

If we don't have a clock, stop the source instead of asserting in
gst_clock_get_time. This can happen when the element is removed from the
pipeline while it's playing.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1674>

rtpmanagerbad: remove duplicate parent declaration

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1689>

rtmp2sink: fix since marker on new "stop-commands" property

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1687>

va: basedec: copy frames logic to decide_allocation()

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1673>

va: basedec: refactor the other video pool instantiation

Just a code clean up

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1673>

va: basedec: add gstvabasedec helper

This is a helper for all decoders.

It is not an abstract subclass, just merely a helper that avoids code
duplication among the decoders.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1673>

va: vp8dec: add element documentation

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1673>

va: h264dec: set video alignment definition earlier

This patch renames need_cropping variable to need_videoalign which is clearer
with its function. And now GstVideoAlignment is part of GstVaH264Dec structure,
so it can be set earlier.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1673>

dtlsconnection: Ignore OpenSSL system call errors

OpenSSL shouldn't be making real system calls, so we can safely
ignore syscall errors. System interactions should happen through
our BIO. So especially don't look at the system's errno, as it
should be meaningless.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1656>

tests: svthevcenc: Fix test_encode_simple

Pick the same I420 format the other test use. Without this, the source
picks AYUV64, which fails.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1573>

va: allocator: add _set_format() and _get_format()

Since allocators keep an available memory queue to reuse, video format and usage
hint are now persistant while allocator's memories are around.

This patch adds _set_format() and _get_format() for both VA allocators.

_set_format() validates if given format can be used or reused. If no allocated
surface previously it creates a dummy one to fetch its offsets and
strides. Updated info is returned to callee.

GstVaPool uses _set_format() at config to verify the allocator capacity and to
get the surfaces offsets and strides, which are going to be used by the video
meta.

Allocator extracted caps are compared with caps from config and if they have
different strides or offsets, force_videometa is set.

A new bufferpool method gst_va_pool_requires_video_meta() is added return the
value of force_videometa. This value is checked in order to know if decoders
need to copy the surface if downstream doesn't announce video meta support.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1667>

va: caps: added gst_caps_is_raw()

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1667>

va: pool: call parent's start() method

Without preallocating buffers and memories a deadlock in pool allocator is
highly probably since it might hit the case were buffer is returned to the pool
but their memories are still hold by a copy downstream, without other
preallocated buffers available.

This kind of a hack, where buffer_reset() follow the normal path if it's called
from start().

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1667>

va: pool: fix log's object

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1667>

va: allocator: remove noisy log message

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1667>

va: allocator: add a todo for dmabuf_memories_setup()

It would be nice to add a surface pool for this type of surface allocation in
order to have a better control of them.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1667>

va: allocator: add a surface counter

Every time a new surface is created the counter increases by one, and when it is
destroyed (or will be destroyed in case of GstVaAllocator), the counter is
decreased. Then, at allocator dispose, it is warning if the counter is not zero.

This counter will be also used to check if the allocator can change its
configuration if the counter is zero or can not.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1667>

va: allocator: remove GstVideoInfo from GstVaBufferSurface

Don't store it them anymore since it is related with the negotiated stream and
not the concrete buffer.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1667>

va: remove GstVideoInfo parameter from _get_surface() functions

There shouldn't be need to retrieve GstVideoInfo per buffer or memory since it
is the same for all the negotiated stream.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1667>

va: vpp: don't fetch video info from buffer

Instead of fetching video info from the buffer, use the already set ones.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1667>

va: dec, vpp: don't get buffer size from allocators

Since buffer size is now ignored by bufferpool there's no need to get tha value
from the allocator.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1667>

va: pool: ignore size in config

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1667>

h265parse: set interlace-mode=interleaved on interlaced content

interlace-mode=alternate is a special case of interlace-mode=interleaved
where the fields are split using two different buffers.

We should use the latter instead of the former to no break compat with
elements supporting only 'interleaved'.
Decoders producing alternate, such as OMX on the Zynq, should change the
interlace-mode on their output caps.

Fix https://gitlab.freedesktop.org/gstreamer/gst-plugins-base/-/issues/825

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1655>

Replace LGPL v2 with LGPL v2.1 in COPYING and remove COPYING.LIB

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1631>

Replace GPL v2 with LGPL v2 in COPYING file

Fixes #1422
https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/issues/1422

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1631>

srt: Consume the error from gst_srt_object_write

Instead of leaking it.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1668>

srt: Check socket state before retrieving payload size

The connection might be broken, which we should detect instead of just
aborting the write.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1669>

x265enc: fix deadlock on reconfig

Don't attempt to obtain encoder lock that is already held by
gst_x265_enc_encode_frame().

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1671>

webrtc: Require gstreamer-sdp in the pkg-config file

Some headers include API from it.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1660>

srtsrc: Fix timestamping

SRT provides the original timestamp of a packet (with drift/skew corrected for
local clock), which is what should be used for timestamping the outgoing
buffers. This ensures that we output the packets with the same timestamp (and by
extension rate) as the original feed.

Also detect if packets were dropped (by checking the sequence number) and
properly set DISCONT flag on the outgoing buffer.

Finally answer the latency queries

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1658>

mfvideosrc: Use only the first video stream per device

Non-first video stream might not be working with current
implementation. It could be non-video (e.g., photo source) and then
ReadSample() might be blocked forever.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1661>

decklink: Update doc

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1665>

decklink: Update Windows headers with SDK 11.2

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1665>

decklink: Update OSX headers with SDK 11.2

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1665>

decklink: Updated DeckLink SDK to 11.2 to support DeckLink 8K Pro

Updated Decklink SDK to version 11.2 in order to support newer cards like the Decklink 8K Pro.
This required to replace the duplex property by a profile property.

Profile values can be the following:
-  bmdProfileOneSubDeviceFullDuplex
-  bmdProfileOneSubDeviceHalfDuplex
-  bmdProfileTwoSubDevicesFullDuplex
-  bmdProfileTwoSubDevicesHalfDuplex
-  bmdProfileFourSubDevicesHalfDuplex

Fixes #987

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1665>

mfvideosrc: Fix invalid memory access when outputting jpeg

Don't access unknown-dangerous-nonsense address

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1662>

Revert "webrtc: Save the media kind in the transceiver"

This reverts commit f54d8e99457996303b8477b1f3a710f0fabd1cc6.

It breaks the CI until the C# bindings are fixed.

Revert "rtpsender: Add API to set the priority"

This reverts commit a8b287c76472c8d7fd38800807c482d020ff4a63.

It breaks the CI until the C# bindings are fixed.

Revert "rtptransceiver: Store the SSRC of the current stream"

This reverts commit d1da271f255101dbe95a426d9f5065d300b53e5a.

It breaks the CI until the C# bindings are fixed.

Revert "webrtcbin: Remove unused function"

This reverts commit 39723dbe934186c11f7b2a2b04c0af7932a1509c.

It breaks the CI until the C# bindings are fixed.

Revert "webrtc: Set the DSCP markings based on the priority"

This reverts commit 8ba08598bbe51f3b1f063ae22605f9608865f16b.

It breaks the CI until the C# bindings are fixed.

Revert "webrtc: Document more objects"

This reverts commit ad68c6b1eb7c73c66dc9d1dbf1a8cc47fd489c61.

It breaks the CI until the C# bindings are fixed.

Revert "webrtc: Add hotdoc style since tags"

This reverts commit 63a5fa818c31ecbe43891c077a38b6b162d73c28.

It breaks the CI until the C# bindings are fixed.

webrtc: Add hotdoc style since tags

We're stuck having to add a separate comment for now.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1425>

webrtc: Document more objects

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1425>

webrtc: Set the DSCP markings based on the priority

This matches how the WebRTC javascript API works and the Chrome implementation.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1425>

webrtcbin: Remove unused function

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1425>

rtptransceiver: Store the SSRC of the current stream

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1425>

rtpsender: Add API to set the priority

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1425>

webrtc: Save the media kind in the transceiver

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1425>

srt: Remove unused sa_family tracking

Now that SRT no longer needs the family when creating the socket, this
code has become useless.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1659>

srt: Move off deprecated `srt_socket()`.

See https://github.com/Haivision/srt/blob/73ee1e1a3e3adc2702a9a5057d101ef80447b38c/docs/API-functions.md#srt_socket

`srt_create_socket()` was added in
https://github.com/Haivision/srt/commit/4b897ba92d34f1829a1c6e419eeab17f0763a0fc
and srt `v1.3.0` is the first release that has it.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1659>

srt: Register a log handler

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1659>

srt: Avoid removing invalid sockets from the polls

This would provoke error messages from SRT.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1659>

srt: Fix use of srt_startup

`srt_startup` can also return 1 if it was successful. Avoid warning in
this case.

Avoid a race when checking whether we need to call it at all.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1659>

srt: Fix parameter types used for socket options

The [SRT documentation][1] specifies exact types for the socket options.
Make sure we match these.

This reverts the linger workaround in commit 84f8dbd932029220ee86154dd
and extends srt_constant_params to support other types than int.

[1]: https://github.com/Haivision/srt/blob/master/docs/APISocketOptions.md

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1659>

d3d11upload: Allow passthrough for system memory

... like how d3d11download and gl{upload,download} do.
This should've been part of the commit 9b72b04daddafb1c86cb6ab5923c593a70bc4166
but I missed.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1645>

va: allocator: refactor flush methods for both allocators

Since the logic is the same, it can be generalized in a single common
function.

Also the methods run the common function with a lock and signal the
buffers' conditional.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1626>

va: allocator: refactor GstVaDmabufAllocator

Move code down to group it.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1626>

va: allocator: refactor GstVaBuffersurface

Move code up and add namespace to methods, and renaming
_creating_buffer_surface() to the canonical
gst_va_buffer_surface_new()

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1626>

va: implement pooled allocators

1. Allocators don't implement memory free() methods since all the memories will
   implement dispose() returning FALSE
2. Memory/miniobject dispose() will act as memory release, enqueueing the
   release memory
3. A new allocator's method prepare_buffer() which queries the released memory
   queue and will add the requiered memories to the buffer.
4. Allocators added a GCond to synchronize dispose() and prepare_buffer()
5. A new allocator's method flush() which will free for real the memories.

While the bufferpool will

1. Remove all the memories at reset_buffer()
2. Implement acquire_buffer() calling allocator's prepare_buffer()
3. Implement flush_start() calling allocator's flush()
4. start() is disabled since it pre-allocs buffers but also calls
   our reset_buffer() which will drop the memories and later the
   buffers are ditched, something we don't want. This approach avoids
   buffer pre-allocation.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1626>

va: allocator: user gst_clear_object() for _buffer_surface_unref()

Event if this function is only used by gst_va_dmabuf_memories_setup(), it might
get reused later by GstVaDmabufAllocator's functions. This change makes the
function less fragile.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1626>

va: allocator: renamed gst_va_dmabuf_memory_release()

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1626>

va: allocator: renamed available_mems queue

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1626>

va: allocator: rename gst_va_dmabuf_allocator_setup_buffer()

Since it's related with GstVaDmabufAllocator.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1626>

va: allocator: calculated surface frame internally

Instead of using gst_buffer_get_size() just add the memory sizes reported by
exported fd.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1626>

va: allocator: make GstVaMemory shareable

Renamed the first variable member of GstVaMemory from parent to mem in
order to avoid confusion with GstMemory's parent.

When freeing the structure, memory's parent is check in order to
decide if surfaces has to be destroyed or not, since only the parent
class have to destroy it.

Removed GST_MEMORY_FLAG_NO_SHARE in memory initialization, since it is
deprecated.

Implemented allocator's share virtual method which creates a new
shallow GstVaMemory structure based on the passed one which will be
it's parent.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1626>

va: allocator: remove copy method for GstVaMemory

Since the memory has to be shareable. That will be address in the next
commits.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1626>

va: allocator: remove va allocator mem_is_span() vmethod

Since it is the default by base class.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1626>

d3d11upload/d3d11download: Make use of staging buffer

... instead of direct cpu map for d3d11memory object. In this way,
we don't need per GstD3D11Memory staging texture.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1627>

d3d11: Don't hold staging texture

Staging texture is used for memory transfer between system and
gpu memory. Apart from d3d11{upload,download} elements, however,
it should happen very rarely.

Before this commit, d3d11bufferpool was allocating at least one
staging texture in order to calculate cpu accessible memory size,
and it wasn't freed for later use of the texture unconditionally.
But it will increase system memory usage. Although GstD3D11memory
object is implemented so that support CPU access, most memory
transfer will happen in d3d11{upload,download} elements.

By this commit, the initial staging texture will be freed immediately
once cpu accessible memory size is calculated.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1627>

d3d11: Move to G_DECLARE_FINAL_TYPE

... and remove unnecessary forward declaration.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1627>

va: caps: centralize caps feature discovering

These function were repeated in the different implemented
elements. This patch centralize them.

The side effect is dmabuf memory type is no longer checked with the
current VAContext, but assuming that dmabuf is a consequence of caps
negotiation from dynamic generated caps templates, where the context's
memory types are validated, there's no need to validate them twice.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1644>

va: filter: fix counter variable reuse

There was a bug reusing the counter variable i in nested loops. Also
the patch makes the code cleaner.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1643>

curlbasesink: Add curl seek callback

Adding functionality to handle SEEK_SET enables rewinding of sent data.
In the HTTP case, this happens after an HTTP 401 has been received from
the other end. This will result in the sent data being resent.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1616>

decklink: Correctly order the different dependent mode tables

One was forgotten in 309f6187fef890c7ffa49305f38e89beac3b1423.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1617>

wasapisrc: Make sure that wasapisrc produces data in loopback mode

An oddness of wasapi loopback feature is that capture client will not
produce any data if there's no outputting sound to corresponding
render client. In other words, if there's no sound to render,
capture task will stall. As an option to solve such issue, we can
add timeout to wake up from capture thread if there's no incoming data
within given time interval. But it seems to be glitch prone.
Another approach is that we can keep pushing silence data into
render client so that capture client can keep capturing data
(even if it's just silence).

This patch will choose the latter one because it's more straightforward
way and it's likely produce glitchless sound than former approach.

A bonus point of this approach is that loopback capture on Windows7/8
will work with this patch. Note that there's an OS bug prior to Windows10
when loopback capture client is running with event-driven mode.
To work around the bug, event signalling should be handled manually
for read thread to wake up.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1588>

wpesrc: fix some caps leaks using the non-GL output

Always chain up to the parent _stop() implementation as it unrefs some
caps (among other things).

Fixes: https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/issues/1409
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1618>

smoothstreaming: clear live adapter on seek

In live streaming, buffers sent by souphttpsrc are pushed to the live
adapter. The buffers in the adapter are sent out of mssdemux when it
is greater than 4096 bytes.

Occasionally, when seeking in live streams, if seek occurs just
after the last data chunk was received, and if this data chunk is
smaller than 4096 bytes, it will be kept in the live adapter.
This remaining data in the live adapter will be erroneously prepended
to the new data that is downloaded after seek and pushed out.
When qtdemux receives this data, since it does not start with
a moof box, it is impossible to demux the fragment, and bogus
size error will occur.

Clear out the live adapter on seek so that no unnecessary remaining
data is pushed out together with the new fragment after seeking.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1345>

va: simplify VPP detection

Also the previous code failed if VPP was not present blacklisting the
plugin.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1619>

msdk: call vaExportSurfaceHandle() to get DMABuf FD

Compared to vaAcquireBufferHandle(), vaExportSurfaceHandle() may
provide the handle details, so we needn't call vaDeriveImage().

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1353>

tests/avtp: Fix coverity issues

Fixes sign extension issues, unchecked return values and some constant
expression results.

CID: 1465073, 1465074, 1465075, 1465076, 1465077
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1398>

avtp: Change "%lu" for G_GUINT64_FORMAT

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1398>

srtobject: typecast SRTO_LINGER to linger

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1615>

codecparsers: vp8parser: clear the frame_hdr before parsing.

Uninited frame_hdr may have garbage and may contain some wrong
results after the parsing process.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1587>

va: Implement the VA vp8 decoder.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1587>

va: codestyle: Clear all tabs in header files

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1587>

va: util: make the _format_changed a common decoder function.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1587>

va: h264dec: allocate output buffers according DPB size

Instead of allocating the maximal number of references for output
buffers, this patch reduces the memory foot print in many cases by
just allocating the output buffers required for the DPB.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1595>

va: decoder: store output buffer rather than surface

GstVaDecodePicture stored the processed VASurfaceID, under the
assumption that the bufferpool with keep the referenced buffers, but
this approach is fragil.

This patch changes GstVaDecodePicture to store the output buffer,
which already contains its VASurfaceID, and provides a new method to
retrieve the VASurfaceID directly from picture.

Based on He Junyan <junyan.he@intel.com> patches for
https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1587

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1594>