webrtc: Fix documentaton moving symbols in the right pages

As those symbols are documented in a 'fwd' header smart indexing in
hotdoc wasn't working.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2891>

webrtc:ice: Avoid using GArray in public API

And use plain null terminated C arrays instead.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2891>

webrtc:ice: Fix some annotations

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2891>

webrtc: Make GstWebRTCIceCandidateStats a Boxed type

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2891>

webrtc:ice: Add padding to structures

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2891>

cuda: Hide runtime compiler related header and symbols

That's already abstracted via gst_cuda_nvrtc_compile() method
and therefore, we do not need to expose such symbols yet.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2884>

aggregator: Improve debug output to better understand why pads are not ready or can't accept more data

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2907>

vajpegdec: Fix memory leak.

Free value content.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2899>

vajpegdec: Fix 4:2:2 for i965.

Since i965 uses NV12 either for 4:2:0 and 4:2:2 chroma (using an
internal color conversion) the chroma validation has to be shortcut.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2899>

va: libs: Add Y42B format.

Enable 4:2:2 JPEG decoding in iHD.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2899>

va: Validate JPEG subsampling configurations.

There are Mesa Gallium drivers that report subsampling but without
any pixel format. So, strip out these subsamplings.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2899>

va: gst_va_create_raw_caps_from_config() may return NULL.

This patch verifies if the function returns NULL in the caller.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2899>

decklink: Validate video input format

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2892>

decklink: Auto-detect 10-bit YUV

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2892>

examples: gtk: Fix MSVC build

gtk-play.c(370): error C2375: 'rewind_button_clicked_cb': redefinition;
different linkage

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2871>

base: video-converter: add accessors for input and output formats

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2905>

vaapi: plugin: Handle when no encoders/decoders available.

Nouveau driver currently only exposes postproc entry. But
vaapidecodebin is registered independent if there are decoders or not,
exposing a segmentation fault.

This patch removes the encoder/decoder/codec arrays if no entries are
found, and if no decoders are found vaapidecodebin is not
registered. Also for vaapipostproc if no postproc entry is found.

Also, if general decoder, used by vaapidecodebin, doesn't have a sink
pad string, don't register the glib type.

Fixes: #1349
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2865>

doc: Fix typo

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2903>

gst-play: missing cleanup for g_autoptr

Without this change cleanup function for g_autoptr is not defined for
GstPlayMediaInfo, GstPlaySignalAdapter, GstPlayVideoRenderer,
GstPlayVideoOverlayVideoRenderer and GstPlayVisualization. Cleanup
function was defined in gstplay.h, but missing in other header files.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2888>

splitmuxsrc: Stop pad task before cleanup

When stopping the element, make sure the pad task
is stopped before destroying the part readers.

Closes a race where the pad task might access
a freed pointer.

Also add a guard against this sort of thing
by holding a ref to the reader in the pad loop.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2901>

qtdemux: Avoid crash on reconfiguring.

When reconfiguring a stream that never created
an output pad, don't access a NULL GstPad pointer.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2869>

qtdemux: Set parsed=true on ONVIF Timed Metadata caps

Inside MP4 the metadata must be properly parsed into frames and in
order.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2897>

meson: d3d11: Fix missing header install

Install gstd3d11compile.h header file

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2896>

v4l2bufferpool: Fix debug trace

The tracing of the buffer pointer was done on the secondary pointer, which
does not match with other traces of the same buffer. This made the trace
confusing and less useful.

Fixes #1379

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2893>

adaptivedemux2: Fix for period switching in live streams

When playing live, it's possible that one stream reaches
the end of the available playback window and goes to sleep
waiting for a manifest update, and the manifest update
introduces a new period. In that case, the sleeping
stream needs to wake up and go 'properly' EOS before we
can advance the input to the new period.

Accordingly, make sure that a stream's last_ret value
is not marked as EOS if it's just sleeping waiting for a live
manifest update.

Also fix the output loop to go back and re-check if it's
time to switch to the next period after dequeuing and
discarding an EOS event.

https://livesim.dashif.org/livesim/periods_20/testpic_2s/Manifest.mpd

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2895>

videoconvert,videoscale: Do conversion in videoconvert and scaling in videoscale

Keep behaving the same as before videoconvertscale port

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2778>

Revert "videoconvertscale: Add properties to disable scaling/converting in videoconvert/videoscale"

This reverts commit cd7a91cef1c6a2e24d440126b7f2ab543fb205c5.

Reverting properties, scaling in videoconvert and converting in
videoscale will be disabled by the other commit

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2778>

mssdemux2: Use gsturi structure to form fragment urls

Utilize gsturi to form fragment url paths.
A token query may contain the string "manifest" and this would lead
to improper url creations.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2859>

meta: Set the parent refcount of the GstStructure correctly

The parent refcount is of the *transformed* buffer, not the input
buffer.

Also update the docs to clarify that @transbuf is the transformed
buffer, and not the buffer on which a transformation is being
performed.

Due to this bug, modifying the structure of a meta that has been
copied to another buffer fails with:

gst_structure_set: assertion 'IS_MUTABLE (structure) || field == NULL' failed

Add a test for the same.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2890>

d3d11decoder: Do timer based DecoderBeginFrame retry

... instead of retry count based one, because the precision of Sleep()
varies depending on system and application configuration.
Also, don't retry DecoderBeginFrame if decoder is doing flush.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2886>

cudaupload,cudadownload: Don't simplify caps in transform_caps

The simplified caps might not be a subset of filter caps
and basetransform will complain about it.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2883>

d3d11decoder: Do not use miniobject qdata

The miniobject qdata uses global mutex.
Use ID3D11DeviceChild::{Set,Get}PrivateData methods instead.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2885>

ssaparse: include required system headers for isspace() and sscanf() functions

Newer compilers ( clang 15 ) have turned stricter and errors out instead
of warning on implicit function declations

Fixes
gstssaparse.c:297:12: error: call to undeclared library function 'isspace' with type 'int (int)'; ISO C99 and later do not support implicit function declarations [-Wimplicit-function-declaration]
    while (isspace(*t))

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2879>

docs: Remove dxgiscreencapsrc from documentation

The implementation was dropped in
https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1750

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2878>

player: Don't leak wrapped video info

Fixes https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/1373

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2880>

play: Make ownership of video-sink clearer in combination with floating references

And correctly handle the case of VideoRenderer::create_video_sink() not
actually returning a floating reference, which might be tricky for some
bindings.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2880>

play: Fix object construction

Ideally new() functions should simply call g_object_new() and not much
else, so let's do that here and handle all the construction properly in
a GObject way.

Now a play object created via g_object_new() is actually usable.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2880>

player: Fix object construction

Ideally new() functions should simply call g_object_new() and not much
else, so let's do that here and handle all the construction properly in
a GObject way.

Now a player object created via g_object_new() is actually usable.

In addition, also fix the video-renderer property so that reading it
returns an object of the correct type.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2880>

player: Release signal adapter on finalize

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2880>

mxfdemux: Always calculate BlockAlign of raw audio

Workaround for nBlockAlign and nBitsPerSample mismatch. Always
use the formula described in the specification for BlockAlign value

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2864>

rtspsrc: Consider the actual control base URI also in case the connection URI contains a query string

That is, get rid of unnecessary and wrong special-casing.

This could always use gst_rtsp_url_get_request_uri_with_control() but as
we only have the control base URI as string it is easier to just call
gst_uri_join_strings().

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2868>

rtspurl: Use gst_uri_join_strings() in gst_rtsp_url_get_request_uri_with_control() instead of a hand-crafted, wrong version

For example the query string of the base must not be taken over to the
request URL unless there is no control path, and control paths can be
absolute and must not be considered relative if they start with a /.

Fixes https://gitlab.freedesktop.org/gstreamer/gst-plugins-good/-/issues/971

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2868>

rtspurl: Use fail_unless_equals_string() in tests

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2868>

meson: d3d11: Fix build with GIR enabled

... and remove unused list

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2870>

qtdemux: Add reference timestamp meta with UTC times based on the ONVIF Export File Format CorrectStartTime box to outgoing buffers

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2525>

gstplayer: Plug a memory leak

This was showing up as a memory leak in GTK's
gstreamer media backend:

40 bytes in 1 blocks are definitely lost in loss record 18,487 of 40,868
   at 0x484586F: malloc (vg_replace_malloc.c:381)
   by 0x50D5278: g_malloc (gmem.c:125)
   by 0x50EDBA5: g_slice_alloc (gslice.c:1072)
   by 0x50EFBCC: g_slice_alloc0 (gslice.c:1098)
   by 0x51F2F45: g_type_create_instance (gtype.c:1911)
   by 0x51DAE37: g_object_new_internal (gobject.c:2011)
   by 0x51DC080: g_object_new_with_properties (gobject.c:2181)
   by 0x51DCB20: g_object_new (gobject.c:1821)
   by 0x9855F86: UnknownInlinedFun (gstplayer-wrapped-video-renderer.c:109)
   by 0x9855F86: gst_player_new (gstplayer.c:579)

Fixes https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/1374

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2875>

hlsdemux2: Fix reference leak of variant stream

When switching back to the previous variant stream
in gst_hls_demux_change_playlist(), fix a couple of
paths that would leak a reference to the previous
variant.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2873>

hlsdemux2: Don't leak the datetime in time map structs

Add a function to clean up GstHLSTimeMap structs
and free the ref on the optional associated GDateTime

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2873>

hlsdemux2: Fix typefind leak and invalid memory access

When typefinding aggregates incoming data to a pending
typefind buffer and then succeeds in typefinding, it
leaks the aggregated buffer, and leaves the caller
accessing an unreffed buffer.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2873>

hlsdemux2: m3u8: Fix memory leaks on parsing

Fix memory leaks when parsing of an m3u8 file is
incomplete, with EXTINF or EXT-X-PROGRAM-DATE-TIME
directives, but no segment url.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2873>

hlsdemux2: m3u8: Fix memory leak

Clear the GValue holding intermediate GstStructure field
data.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2873>

adaptivedemux2: Fix uninitialised memory usage in debug

Fix printing uninitialised memory by clearing the
GstAdaptiveDemuxClock structure when allocating.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2873>

rtpst2022-1-fecenc: Drain column packets on EOS

Otherwise we won't send the protection packets for the last few
packets when a stream ends.

Also send EOS on the FEC src row pad immediately, and on the FEC src
column pad after draining is complete. This makes it so that the FEC
src pads on rtpbin behave the same way as the RTCP src pads on rtpbin
when EOS is received on the send_rtp_sink pad.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2863>

basetextoverlay: Don't miscalculate text running times

When a new segment event arrives, it immediately updates
the current stored segment, which was used for calculating
the running time of the current text buffer for every
passing video frame. This means a segment that arrives
after the text buffer might get used to (mis)calculate
the running times subsequently.

Instead, calculate and store the right running time
using the current segment when storing the buffer. Later
the stored segment can get freely updated.

This fixes the case where pieces of video and text streams
are seamlessly concatenated and fed through the text overlay.
Previously, it could lead to the current text buffer suddenly
have a massive running time and blocking all further input.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2802>

d3d11window: fix DXGI_SWAP_CHAIN_FLAG_ALLOW_TEARING swap chain flag for full screen

Fixes #1372 (regression introduced in 5eeec165)

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2866>

qtdemux: Don't use invalid values from failed trex parsing

If parsing the fragment default values (`trex` atom) failed, don't try to
compute a bogus sample_description_id value.

Fixes #1369

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2860>

wpesrc: Switch URI handler to web+... protocols

The web://http:// URIs were not compliant with RFC 3986. Using web+http://
allows us to use the GstUri parser to pass down a valid URI to `wpevideosrc`.

Corresponding change for the CEF source element:
https://github.com/centricular/gstcefsrc/commit/8d499495dd79cc0bf9a38ae82b03e374d78f68ae

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2856>

docs: copy-paste fix

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2857>

nvcodec: fix caps leaks in nvh264/h265encoder

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2850>

gst: Protect initialization state with a recursive mutex.

Otherwise a gst_init() call from a plugin would deadlock if the plugin
is loaded as part of registry updating.

Fixes https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/940

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2740>

registry: Remove dead code

Initialization/updating of the registry can't possible fail and all code
paths always returned TRUE.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2740>

gst: Don't fail gst_init() if updating the registry fails

Everything is already marked as initialized at that point and by failing
no tracers would be loaded or plugin feature rank overrides would be
applied.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2740>

videoflip: Add support for 10/12bit planar formats

Implements support for I420, I422 and Y444 in 10/12 bit LE/BE variants.
I422 is handled separately from the rest, as it needs to consider
the endianness of the current format during most transforms.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2788>

vah264enc: Set codec frame sync point if IDR

This flag is used by GstVideoEncoder base class for certain configurations.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2852>

vah264enc: Packed headers can be zero.

A driver can report back no packed header support (VA_ENC_PACKED_HEADER_NONE).
This patch removes that false verification.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2851>

opusenc: improve inband-fec property documentation

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2853>

alpha: fix stride issue when out buffer has padding on right

if outbuf has padding on right, need jump to next line use stride,
otherwise downstream element will show a wrong picture when use the
same stride

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2842>

d3d11: Use WIN32 API directly for locking with RAII pattern

Such abstraction is unnecessary for this library/plugin.
Use WIN32 API directly instead of GLib wrappers.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2848>

d3d11memory: Remove unnecessary locking

* memory map/unmap is already protected by d3d11 device lock.
  Don't need to take another memory lock.
* Use WIN32 critical section and slim reader/writer lock APIs
  directly instead of GLib wrappers.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2848>

vah264enc: Fix caps for mesa gallium.

Radeon mesa gallium driver has a bug which adds P010_10LE sink caps
format. This patch removes formats which arent 420 chroma.

gst_caps_set_format_array() wasn't used because the fix traverse
several structures with potential different formats.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2844>

vah264dec: Complete profiles in decoder.

Instead of specifying all the H.264 "supported" profiles in the global
hash table (used either by decoders and encoders), just complete them
in the decoder only, since the encoder doesn't support them.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2844>

hlsdemux2: Requeue header buffer when restarting fragment

When returning GST_ADAPTIVE_DEMUX_FLOW_RESTART_FRAGMENT
for the first segment data, we might need to requeue the
header.

This was leading to occasional prerolling stalls on
HLS live streams with renditions.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2849>

hlsdemux2: Fix buffer leak when resynching

Unref the buffer in gst_hls_demux_handle_buffer() when
returning GST_ADAPTIVE_DEMUX_FLOW_LOST_SYNC

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2849>

dmabuf: Always skip modifier if it's linear

Accepting both NV12 and NV12:0x0000000000000000 will make the
intersection code too painful to write.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2617>

d3d11: Don't find global default allocator

We were using global default allocator already. Pass null
allocator object to *_alloc() methods then the method will
use default allocator.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2843>

d3d11memory: Allow null GstD3D11Allocator to alloc methods

Similar to gst_allocator_alloc(), use default GstD3D11Allocator
when caller passes null allocator object

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2843>

d3d11: Use std::call_once()

g_once_init_enter() always takes global mutex for non-GCC build.
Use C++ once call implementation

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2843>

adaptivedemux2: Fix a small race on shutdown

Make sure gst_adaptive_demux_loop_cancel_call()
never tries to operate on an invalidated main context. Make
sure to clear the main context pointer while holding the lock,
and to check it in gst_adaptive_demux_loop_cancel_call()

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2847>

d3d11device: Use WIN32 critical section API directly

GLib's GRecMutex will allocate another heap memory for CRITICAL_SECTION
struct and g_rec_mutex_lock/g_rec_mutex_unlock use WIN32 APIs actually.
We don't need such intermediate function calls and redundant heap allocation.
Just call WIN32 APIs directly.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2845>

vah264enc: Lock properties read/write.

This is a first step for changing properties at runtime.

And add missing bitrate upate and notification.

Fixes: #1258
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2836>

vah264enc: Use guint32 for rc_ctrl as it's for rc_ctrl_mode.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2836>

vah264enc: Split aud property and its usage.

Just as other property variables, it's split for ease it usage,
particularly after adding access locks.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2836>

va: baseenc: Untabbify and format.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2836>

adaptivedemux2-stream: Silence a compiler warning

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2839>

adaptivedemux2: Move internal FLOW_SWITCH return value.

Move the internal-only FLOW_SWITCH custom return value
to GST_FLOW_CUSTOM_SUCCESS+2 to avoid collision with
GST_ADAPTIVE_DEMUX_FLOW_LOST_SYNC

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2839>

adaptivedemux2: Modify custom sync loss flow return

Make it a custom sucess and not an error

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2839>

hlsdemux2: Always check DSN if required

We don't want to consider the candidate as being before the playlist if the DSN
don't match

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2839>

adaptivedemux2/hlsdemux2: Handle loss of sync when dowloading.

Media playlist updates and fragment downloads happen in an interleaved
fashion. When a media playlist update fails *while* a segment is being
downloaded, this means we lost synchronization.

Properly propagate and handle this

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2839>

hlsdemux2: Fix initial playlist setup.

There is now only a single case where we setup the initial playlist to 0, which
is for the very first variant stream.

Rendition streams will have the initial playlist "synchronized" against the
variant stream media playlist.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2839>

hlsdemux2: Handle loss of synchronization in live

Loss of synchronization happens when the updated media playlist has no
relationship to the previous ones. This could happen because of network issues,
server issues, etc...

When this happens, we take no chance and "reset" ourselves so that we can "seek
back to live" against the new updated playlists.

Since this happens at the "media playlist update" level, make sure the custom
flow return is propagated up.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2839>

adaptivedemux2: Handle synchronously to lost sync

We are already in the main scheduler thread, therefore we can do the "seek back
to live" directly. This also avoids other pending actions to take place.

Also handle the loss of sync when doing manifest updates.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2839>

hlsdemux2: Prune time maps when possible

Add a new method to prune unused time mappings (i.e. which aren't used by any
current media playlist).

Do that when doing flushing seeks. Could be used in other places later too.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2839>

hlsdemux2: Allow DSN mismatches when re-syncing playlists

Some providers provide completely incompatible DSN across bitrates/renditions,
but do keep MSN consistent.

If we fail to synchronize playlist with DSN, retry without the DSN taken into
account.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2839>

adaptivedemux2: Fixes for period switching in the output loop

Close some race conditions in switching to the next period,
by ensuring the tracks are completely drained first and by
not outputting EOS events to the output source pad
if there is another period pending.

Fixes Manifest_MultiPeriod_1080p.mpd some more.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2838>

adaptivedemux2: stream: Set period has_next_period flag before EOS

Before sending EOS, update the period's has_next_period
flag and/or create the next period. This closes a race
where the output loop might receive the EOS event
and either push it downstream (causing premature EOS),
or receive it and try and switch to the next period
before that period is completely set up.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2838>

adaptivedemux2: period: Rename 'closed' flag to 'has_next_period'

The flag is used to tell the output loop that a
next period is present, since the output loop
can't call the gst_adaptive_demux_has_next_period()
method.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2838>

adaptivedemux2: Recheck for a pending track on drain

When a track is completely drained and EOS, but
there's a pending track on the slot loop again
to switch to that track.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2838>

adaptivedemux2: Check stream selected instead of state

When combining stream flows, ignore streams that
are not selected, instead of checking whether
the stream state has changed yet.

Fixes another issue with dashdemux2 where it fails to
change to the next period when playing content with
several video, audio and text streams, as with
Manifest_MultiPeriod_1080p.mpd when seeking to 730
just before the end of the first period.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2838>

tracers: leaks: delay type name lookup

Micro optimisation: Store the quark of the type name when tracking
objects and only do the quark to string conversion (hashtable lookup)
later when we actually need the string.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2832>

tracers: leaks: fix potentially invalid memory access when trying to detect object type

The is_gst_mini_object_check would sometimes detect a proper GObject
as a mini object, and then bad things happen.

We know whether a pointer is a proper GObject or a MiniObject here
though, so just pass that information to the right code paths and
avoid the heuristics altogether.

Eliminates all remaining uses of object_is_gst_mini_object().

Fixes #1334

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2832>