gstreamer/NEWS
Tim-Philipp Müller 7cb583bb04 Release 1.18.0
2020-09-08 00:07:47 +01:00

2204 lines
95 KiB
Text
Raw Blame History

This file contains invisible Unicode characters

This file contains invisible Unicode characters that are indistinguishable to humans but may be processed differently by a computer. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

GStreamer 1.18 Release Notes
GStreamer 1.18.0 was originally released on 7 September 2020.
See https://gstreamer.freedesktop.org/releases/1.18/ for the latest
version of this document.
Last updated: Monday 7 September 2020, 10:30 UTC (log)
Introduction
The GStreamer team is proud to announce a new major feature release in
the stable 1.x API series of your favourite cross-platform multimedia
framework!
As always, this release is again packed with many new features, bug
fixes and other improvements.
Highlights
- GstTranscoder: new high level API for applications to transcode
media files from one format to another
- High Dynamic Range (HDR) video information representation and
signalling enhancements
- Instant playback rate change support
- Active Format Description (AFD) and Bar Data support
- ONVIF trick modes support in both GStreamer RTSP server and client
- Hardware-accelerated video decoding on Windows via DXVA2 /
Direct3D11
- Microsoft Media Foundation plugin for video capture and
hardware-accelerated video encoding on Windows
- qmlgloverlay: New overlay element that renders a QtQuick scene over
the top of an input video stream
- New imagesequencesrc element to easily create a video stream from a
sequence of jpeg or png images
- dashsink: Add new sink to produce DASH content
- dvbsubenc: DVB Subtitle encoder element
- TV broadcast compliant MPEG-TS muxing with constant bitrate muxing
and SCTE-35 support
- rtmp2: new RTMP client source and sink element implementation
- svthevcenc: new SVT-HEVC-based H.265 video encoder
- vaapioverlay compositor element using VA-API
- rtpmanager support for Googles Transport-Wide Congestion Control
(twcc) RTP extension
- splitmuxsink and splitmuxsrc gained support for auxiliary video
streams
- webrtcbin now contains some initial support for renegotiation
involving stream addition and removal
- New RTP source and sink elements to easily set up RTP streaming via
rtp:// URIs
- New Audio Video Transport Protocol (AVTP) plugin for Time-Sensitive
Applications
- Support for the Video Services Forums Reliable Internet Stream
Transport (RIST) TR-06-1 Simple Profile
- Universal Windows Platform (UWP) support
- rpicamsrc element for capturing from the Raspberry Pi camera
- RTSP Server TCP interleaved backpressure handling improvements as
well as support for Scale/Speed headers
- GStreamer Editing Services gained support for nested timelines,
per-clip speed rate control and the OpenTimelineIO format.
- Autotools build system has been removed in favour of Meson
Major new features and changes
Noteworthy new features and API
Instant playback rate changes
Changing the playback rate as quickly as possible so far always required
a flushing seek. This generally works, but has the disadvantage of
flushing all data from the playback pipeline and requiring the demuxer
or parser to do a full-blown seek including resetting its internal state
and resetting the position of the data source. It might also require
considerable decoding effort to get to the right position to resume
playback from at the higher rate.
This release adds a new mechanism to achieve quasi-instant rate changes
in certain playback pipelines without interrupting the flow of data in
the pipeline. This is activated by sending a seek with the
GST_SEEK_FLAG_INSTANT_RATE_CHANGE flag and start_type = stop_type =
GST_SEEK_TYPE_NONE. This flag does not work for all pipelines, in which
case it is necessary to fall back to sending a full flushing seek to
change the playback rate. When using this flag, the seek event is only
allowed to change the current rate and can modify the trickmode flags
(e.g. keyframe only or not), but it is not possible to change the
current playback position, playback direction or do a flush.
This is particularly useful for streaming use cases like HLS or DASH
where the streaming download should not be interrupted when changing
rate.
Instant rate changing is handled in the pipeline in a specific sequence
which is detailed in the seeking design docs. Most elements dont need
to worry about this, only elements that sync to the clock need some
special handling which is implemented in the GstBaseSink base class, so
should be taken care of automatically in most normal playback pipelines
and sink elements.
See Jans GStreamer Conference 2019 talk “Changing Playback Rate
Instantly” for more information.
You can try this feature by passing the -i command line option to
gst-play-1.0. It is supported at least by qtdemux, tsdemux, hlsdemux,
and dashdemux.
Google Transport-Wide Congestion Control
rtpmanager now supports the parsing and generating of RTCP messages for
the Google Transport-Wide Congestion Control RTP Extension, as described
in:
https://tools.ietf.org/html/draft-holmer-rmcat-transport-wide-cc-extensions-01.
This “just” provides the required plumbing/infrastructure, it does not
actually make effect any actual congestion control on the sender side,
but rather provides information for applications to use to make such
decisions.
See Håvards “Google Transport-Wide Congestion Control” talk for more
information about this feature.
GstTranscoder: a new high-level transcoding API for applications
The new GstTranscoder library, along with transcodebin and
uritranscodebin elements, provides high level API for applications to
transcode media files from one format to another. Watch Thibaults talk
“GstTranscoder: A High Level API to Quickly Implement Transcoding
Capabilities in your Applications” for more information.
This also comes with a gst-transcoder-1.0 command line utility to
transcode one URI into another URI based on the specified encoding
profile.
Active Format Description (AFD) and Bar Data support
The GstVideo Ancillary Data API has gained support for Active Format
Description (AFD) and Bar data.
This includes various two new buffer metas: GstVideoAFDMeta and
GstVideoBarMeta.
GStreamer now also parses and extracts AFD/Bar data in the h264/h265
video parsers, and supports both capturing them and outputting them in
the decklink elements. See Aarons lightning talk at the GStreamer
Conference for more background.
ONVIF trick modes support in both GStreamer RTSP server and client
- Support for the various trick modes described in section 6 of the
ONVIF streaming spec has been implemented in both gst-rtsp-server
and rtspsrc.
- Various new properties in rtspsrc must be set to take advantage of
the ONVIF support
- Examples are available here: test-onvif-server.c and
test-onvif-client.c
- Watch Mathieu Duponchelles talk “Implementing a Trickmode Player
with ONVIF, RTSP and GStreamer” for more information and a live
demo.
GStreamer Codecs library with decoder base classes
This introduces a new library in gst-plugins-bad which contains a set of
base classes that handle bitstream parsing and state tracking for the
purpose of decoding different codecs. Currently H264, H265, VP8 and VP9
are supported. These bases classes are meant primarily for internal use
in GStreamer and are used in various decoder elements in connection with
low level decoding APIs like DXVA, NVDEC, VAAPI and V4L2 State Less
decoders. The new library is named gstreamer-codecs-1.0 /
libgstcodecs-1.0 and is not yet guaranteed to be API stable across major
versions.
MPEG-TS muxing improvements
The GStreamer MPEG-TS muxer has seen major improvements on various
fronts in this cycle:
- It has been ported to the GstAggregator base class which means it
can work in defined-latency mode with live input sources and
continue streaming if one of the inputs stops producing data.
- atscmux, a new ATSC-specific tsmux subclass
- Constant Bit Rate (CBR) muxing support via the new bitrate property
which allows setting the target bitrate in bps. If this is set the
muxer will insert null packets as padding to achieve the desired
multiplex-wide constant bitrate.
- compliance fixes for TV broadcasting use cases (esp. ATSC). See
Jans talk “TV Broadcast compliant MPEG-TS” for details.
- Streams can now be added and removed at runtime: Until now, any
streams in tsmux had to be present when the element started
outputting its first buffer. Now they can appear at any point during
the stream, or even disappear and reappear later using the same PID.
- new pcr-interval property allows applications to configure the
desired interval instead of hardcoding it
- basic SCTE-35 support. This is enabled by setting the scte-35-pid
property on the muxer. Sending SCTE-35 commands is then done by
creating the appropriate SCTE-35 GstMpegtsSection and sending them
on the muxer.
- MPEG-2 AAC handling improvements
New elements
- New qmlgloverlay element for rendering a QtQuick scene over the top
of a video stream. qmlgloverlay requires that Qt support adopting an
external OpenGL context and is known to work on X11 and Windows.
Wayland is known not to work due to limitations within Qt. Check out
the example to see how it works.
- The clocksync element is a generic element that can be placed in a
pipeline to synchronise passing buffers to the clock at that point.
This is similar to identity sync=true, but because it isnt
GstBaseTransform-based, it can process GstBufferLists without
breaking them into separate GstBuffers. It is also more discoverable
than the identity option. Note that you do not need to insert this
element into your pipeline to make GStreamer sync to the pipeline
clock, this is usually handled automatically by the elements in the
pipeline (sources and sinks mostly). This element is useful to feed
non-live input such as local files into elements that expect live
input such as webrtcbin.`
- New imagesequencesrc element to easily create a video stream from a
sequence of JPEG or PNG images (or any other encoding where the type
can be detected), basically a multifilesrc made specifically for
image sequences.
- rpicamsrc element for capturing raw or encoded video (H.264, MJPEG)
from the Raspberry Pi camera. This works much like the popular
raspivid command line utility but outputs data nicely timestamped
and formatted in order to integrate nicely with other GStreamer
elements. Also comes with a device provider so applications can
discover the camera if available.
- aatv and cacatv video filters that transform video ASCII art style
- avtp: new Audio Video Transport Protocol (AVTP) plugin for Linux.
See Andre Guedes talk “Audio/Video Bridging (AVB) support in
GStreamer” for more details.
- clockselect: a pipeline element that enables clock selection/forcing
via gst-launch pipeline syntax.
- dashsink: Add new sink to produce DASH content. See Stéphanes talk
or blog post for details.
- dvbsubenc: a DVB subtitle encoder element
- microdns: a libmicrodns-based mdns device provider to discover RTSP
cameras on the local network
- mlaudiosink: new audio sink element for the Magic Leap platform,
accompanied by an MLSDK implementation in the amc plugin
- msdkvp9enc: VP9 encoder element for the Intel MediaSDK
- rist: new plugin implementing support for the Video Services Forums
Reliable Internet Stream Transport (RIST) TR-06-1 Simple Profile.
See Nicolas blog post “GStreamer support for the RIST
Specification” for more details.
- rtmp2: new RTMP client source and sink elements with fully
asynchronous network operations, better robustness and additional
features such as handling ping and stats messages, and adobe-style
authentication. The new rtmp2src and rtmp2sink elements should be
API-compatible with the old rtmpsrc / rtmpsink elements and should
work as drop-in replacements.
- new RTP source and sink elements to easily set up RTP streaming via
rtp:// URIs: The rtpsink and rtpsrc elements add an URI interface so
that streams can be decoded with decodebin using rtp:// URIs. These
can be used as follows: ``` gst-launch-1.0 videotestsrc ! x264enc !
rtph264pay config-interval=3 ! rtpsink uri=rtp://239.1.1.1:1234
gst-launch-1.0 videotestsrc ! x264enc ! rtph264pay config-interval=1
! rtpsink uri=rtp://239.1.2.3:5000 gst-launch-1.0 rtpsrc
uri=rtp://239.1.2.3:5000?encoding-name=H264 ! rtph264depay !
avdec_h264 ! videoconvert ! xvimagesink
gst-launch-1.0 videotestsrc ! avenc_mpeg4 ! rtpmp4vpay
config-interval=1 ! rtpsink uri=rtp://239.1.2.3:5000 gst-launch-1.0
rtpsrc uri=rtp://239.1.2.3:5000?encoding-name=MP4V-ES ! rtpmp4vdepay
! avdec_mpeg4 ! videoconvert ! xvimagesink ```
- svthevcenc: new SVT-HEVC-based H.265 video encoder
- switchbin: new helper element which chooses between a set of
processing chains (paths) based on input caps, and changes the
active chain if new caps arrive. Paths are child objects, which are
accessed by the GstChildProxy interface. See the switchbin
documentation for a usage example.
- vah264dec: new experimental va plugin with an element for H.264
decoding with VA-API using GStreamers new stateless decoder
infrastructure (see Linux section below).
- v4l2codecs: introduce an V4L2 CODECs Accelerator supporting the new
CODECs uAPI in the Linux kernel (see Linux section below)
- zxing new plugin to detect QR codes and barcodes, based on libzxing
- also see the Rust plugins section below which contains plenty of new
exciting plugins written in Rust!
New element features and additions
GStreamer core
- filesink: Add a new “full” buffer mode. Previously the default and
full modes were the same. Now the default mode is like before: it
accumulates all buffers in a buffer list until the threshold is
reached and then writes them all out, potentially in multiple
writes. The new full mode works by always copying memory to a single
memory area and writing everything out with a single write once the
threshold is reached.
- multiqueue: Add stats property and
current-level-{buffers, bytes, time} pad properties to query the
current levels of the corresponding internal queue.
Plugins Base
- alsa: implement a device provider
- alsasrc: added use-driver-timestamp property to force use of
pipeline timestamps (and disable driver timestamps) if so desired
- audioconvert: fix changing the mix-matrix property at runtime
- appsrc: added support for segment forwarding or custom GstSegments
via GstSample, enabled via the handle-segment-change property. This
only works for segments in TIME format for now.
- compositor: various performance optimisations, checkerboard drawing
fixes, and support for VUYA format
- encodebin: Fix and refactor smart encoding; ensure that a single
segment is pushed into encoders; improve force-key-unit event
handling.
- opusenc: Add low delay option (audio-type=restricted-lowdelay) to
disable the SILK layer and achieve only 5ms delay.
- opusdec: add stats property to retrieve various decoder statistics.
- uridecodebin3: Let decodebin3 do its stream selection if no one
answers
- decodebin3: Avoid overriding explicit user selection of streams
- playbin: add flag to force use of software decoders over any
hardware decoders that might also be available
- playbin3, playbin: propagate sink context
- rawvideoparse: Fix tiling support, allow setting colorimetry
- subparse: output plain utf8 text instead of pango-markup formatted
text if downstream requires it, useful for interop with elements
that only accept utf8-formatted subtitles such as muxers or closed
caption converters.
- tcpserversrc, tcpclientsrc: add stats property with TCP connection
stats (some are only available on Linux though)
- timeoverlay: add show-times-as-dates, datetime-format and
datetime-epoch properties to display times with dates
- videorate: Fix changing rate property during playback; reverse
playback fixes; update QoS events taking into account our rate
- videoscale: pass through and transform size sensitive metas instead
of just dropping them
Plugins Good
- avidemux can handle H.265 video now. Our advice remains to
immediately cease all contact and communication with anyone who
hands you H.265 video in an AVI container, however.
- avimux: Add support for S24LE and S32LE raw audio and v210 raw video
formats; support more than 2 channels of raw audio.
- souphttpsrc: disable session sharing and cookie jar when the cookies
property is set; correctly handle seeks past the end of the content
- deinterlace: new YADIF deinterlace method which should provide
better quality than the existing methods and is LGPL licensed;
alternate fields are supported as input to the deinterlacer as well
now, and there were also fixes for switching the deinterlace mode on
the fly.
- flvmux: in streamable mode allow adding new pads even if the initial
header has already been written. Old clients will only process the
initial stream, new clients will get a header with the new streams.
The skip-backwards-streams property can be used to force flvmux to
skip and drop a few buffers rather than produce timestamps that go
backward and confuse librtmp-based clients. Theres also better
handling for timestamp rollover when streaming for a long time.
- imagefreeze: Add live mode, which can be enabled via the new is-live
property. In this mode frames will only be output in PLAYING state
according to the negotiated framerate, skipping frames if the output
cant keep up (e.g. because its blocked downstream). This makes it
possible to actually use imagefreeze in live pipelines without
having to manually ensure somehow that it starts outputting at the
current running time and without still risking to fall behind
without recovery.
- matroskademux, qtdemux: Provide audio lead-in for some lossy formats
when doing accurate seeks, to make sure we can actually decode
samples at the desired position. This is especially important for
non-linear audio/video editing use-cases.
- matroskademux, matroskamux: Handle interlaced field order (tff, bff)
- matroskamux:
- new offset-to-zero property to offset all streams to start at
zero. This takes the timestamp of the earliest stream and
offsets it so that it starts at 0. Some software (VLC,
ffmpeg-based) does not properly handle Matroska files that start
at timestamps much bigger than zero, which could happen with
live streams.
- added a creation-time property to explicitly set the creation
time to write into the file headers. Useful when remuxing, for
example, but also for live feeds where the DateUTC header can be
set a UTC timestamp corresponding to the beginning of the file.
- the muxer now also always waits for caps on sparse streams, and
warns if caps arrive after the header has already been sent,
otherwise the subtitle track might be silently absent in the
final file. This might affect applications that send sparse data
into matroskamux via an appsrc element, which will usually not
send out the initial caps before it sends out the first buffer.
- pulseaudio: device provider improvements: fix discovery of
newly-added devices and hide the alsa device provider if we provide
alsa devices
- qtdemux: raw audio handling improvements, support for AC4 audio, and
key-units trickmode interval support
- qtmux:
- was ported to the GstAggregator base class which allows for
better handling of live inputs, but might entail minor
behavioural changes for sparse inputs if inputs are not live.
- has also gained a force-create-timecode-trak property to create
a timecode trak in non-mov flavors, which may not be supported
by Apple but is supported by other software such as Final Cut
Pro X
- also a force-chunks property to force the creation of chunks
even in single-stream files, which is required for Apple ProRes
certification.
- also supports 8k resolutions in prefill mode with ProRes.
- rtpbin gained a request-jitterbuffer signal which allows
applications to plug in their own jitterbuffer implementation such
as the threadsharing jitterbuffer from the Rust plugins, for
example.
- rtprtxsend: add clock-rate-map property to allow generic RTP input
caps without a clock-rate whilst still supporting the max-size-time
property for bundled streams.
- rtpssrcdemux: introduce max-streams property to guard against
attacks where the sender changes SSRC for every RTP packet.
- rtph264pay, rtph264pay: implement STAP-A and various aggregation
modes controled by the new aggegrate-mode property: none to not
aggregate NAL units (as before), zero-latency to aggregate NAL units
until a VCL or suffix unit is included, or max to aggregate all NAL
units with the same timestamp (which adds one frame of latency). The
default has been kept at none for backwards compatibility reasons
and because various RTP/RTSP implementions dont handle aggregation
well. For WebRTC use cases this should be set to zero-latency,
however.
- rtpmp4vpay: add support for config-interval=-1 to resend headers
with each IDR keyframe, like other video payloaders.
- rtpvp8depay: Add wait-for-keyframe property for waiting until the
next keyframe after packet loss. Useful if the video stream was not
encoded with error resilience enabled, in which case packet loss
tends to cause very bad artefacts when decoding, and waiting for the
next keyframe instead improves user experience considerably.
- splitmuxsink and splitmuxsrc can now handle auxiliary video streams
in addition to the primary video stream. The primary video stream is
still used to select fragment cut points at keyframe boundaries.
Auxilliary video streams may be broken up at any packet - so
fragments may not start with a keyframe for those streams.
- splitmuxsink:
- new muxer-preset and sink-preset properties for setting
muxer/sink presets
- a new start-index property to set the initial fragment id
- and a new muxer-pad-map property which explicitly maps
splitmuxsink pads to the muxer pads they should connect to,
overriding the implicit logic that tries to match pads but
yields arbitrary names.
- Also includes the actual sink element in the fragment-opened and
fragment-closed element messages now, which is especially useful
for sinks without a location property or when finalisation of
the fragments is done asynchronously.
- videocrop: add support for Y444, Y41B and Y42B pixel formats
- vp8enc, vp9enc: change default value of VP8E_SET_STATIC_THRESHOLD
from 0 to 1 which matches what Google WebRTC does and results in
lower CPU usage; also added a new bit-per-pixel property to select a
better default bitrate
- v4l2: add support for ABGR, xBGR, RGBA, and RGBx formats and for
handling interlaced video in alternate fields interlace mode (one
field per buffer instead of one frame per picture with both fields
interleaved)
- v4l2: Profile and level probing support for H264, H265, MPEG-4,
MPEG-2, VP8, and VP9 video encoders and decoders
Plugins Ugly
- asfdemux: extract more metadata: disc number and disc count
- x264enc:
- respect YouTube bitrate recommendation when user sets the
YouTube profile preset
- separate high-10 video formats from 8-bit formats to improve
depth negotiation and only advertise suitable input raw formats
for the desired output depth
- forward downstream colorimetry and chroma-site restrictions to
upstream elements
- support more color primaries/mappings
Plugins Bad
- av1enc: add threads, row-mt and tile-{columns,rows} properties for
this AOMedia AV1 encoder
- ccconverter: implement support for CDP framerate conversions
- ccextractor: Add remove-caption-meta property to remove caption
metas from the outgoing video buffers
- decklink: add support for 2K DCI video modes, widescreen NTSC/PAL,
and for parsing/outputting AFD/Bar data. Also implement a simple
device provider for Decklink devices.
- dtlsrtpenc: add rtp-sync property which synchronises RTP streams to
the pipeline clock before passing them to funnel for merging with
RTCP.
- fdkaac: also decode MPEG-2 AAC; encoder now supports more
multichannel/surround sound layouts
- hlssink2: add action signals for custom playlist/fragment handling:
Instead of always going through the file system API we allow the
application to modify the behaviour. For the playlist itself and
fragments, the application can provide a GOutputStream. In addition
the sink notifies the application whenever a fragment can be
deleted.
- interlace: can now output data in alternate fields mode; added field
switching mode for 2:2 field pattern
- iqa: Add a mode property to enable strict mode that checks that all
the input streams have the exact same number of frames; also
implement the child proxy interface
- mpeg2enc: add disable-encode-retries property for lower CPU usage
- mpeg4videoparse: allow re-sending codec config at IDR via
config-interval=-1
- mpegtsparse: new alignment property to determine number of TS
packets per output buffer, useful for feeding an MPEG-TS stream for
sending via udpsink. This can be used in combination with the
split-on-rai property that makes sure to start a new output buffer
for any TS packet with the Random Access Indicator set. Also set
delta unit buffer flag on non-random-access buffers.
- mpegdemux: add an ignore-scr property to ignore the SCR in
non-compliant MPEG-PS streams with a broken SCR, which will work as
long as PTS/DTS in the PES header is consistently increasing.
- tsdemux:
- add an ignore-pcr property to ignore MPEG-TS streams with broken
PCR streams on which we cant reliably recover correct
timestamps.
- new latency property to allow applications to lower the
advertised worst-case latency of 700ms if they know their
streams support this (must have timestamps in higher frequency
than required by the spec)
- support for AC4 audio
- msdk - Intel Media SDK plugin for hardware-accelerated video
decoding and encoding on Windows and Linux:
- mappings for more video formats: Y210, Y410, P012_LE, Y212_LE
- encoders now support bitrate changes and input format changes in
playing state
- msdkh264enc, msdkh265enc: add support for CEA708 closed caption
insertion
- msdkh264enc, msdkh265enc: set Region of Interest (ROI) region
from ROI metas
- msdkh264enc, msdkh265enc: new tune property to enable low-power
mode
- msdkh265enc: add support 12-bit 4:2:0 encoding and 8-bit 4:2:2
encoding and VUYA, Y210, and Y410 as input formats
- msdkh265enc: add support for screen content coding extension
- msdkh265dec: add support for main-12/main-12-intra,
main-422-10/main-422-10-intra 10bit,
main-422-10/main-422-10-intra 8bit,
main-422-12/main-422-12-intra, main-444-10/main-444-10-intra,
main-444-12/main-444-12-intra, and main-444 profiles
- msdkvp9dec: add support for 12-bit 4:4:4
- msdkvpp: add support for Y410 and Y210 formats, cropping via
properties, and a new video-direction property.
- mxf: Add support for CEA-708 CDP from S436 essence tracks. mxfdemux
can now handle Apple ProRes
- nvdec: add H264 + H265 stateless codec implementation nvh264sldec
and nvh265sldec with fewer features but improved latency. You can
set the environment variable GST_USE_NV_STATELESS_CODEC=h264 to use
the stateless decoder variant as nvh264dec instead of the “normal”
NVDEC decoder implementation.
- nvdec: add support for 12-bit 4:4:4/4:2:0 and 10-bit 4:2:0 decoding
- nvenc:
- add more rate-control options, support for B-frame encoding (if
device supports it), an aud property to toggle Access Unit
Delimiter insertion, and qp-{min,max,const}-{i,p,b} properties.
- the weighted-pred property enables weighted prediction.
- support for more input formats, namely 8-bit and 10-bit RGB
formats (BGRA, RGBA, RGB10A2, BGR10A2) and YV12 and VUYA.
- on-the-fly resolution changes are now supported as well.
- in case there are multiple GPUs on the system, there are also
per-GPU elements registered now, since different devices will
have different capabilities.
- nvh265enc can now support 10-bit YUV 4:4:4 encoding and 8-bit
4:4:4 / 10-bit 4:2:0 formats up to 8K resolution (with some
devices). In case of HDR content HDR related SEI nals will be
inserted automatically.
- openjpeg: enable multi-threaded decoding and add support for
sub-frame encoding (for lower latency)
- rtponviftimestamp: add opt-out “drop-out-of-segment” property
- spanplc: new stats property
- srt: add support for IPv6 and for using hostnames instead of IP
addresses; add streamid property, but also allow passing the id via
the stream URI; add wait-for-connection property to srtsink
- timecodestamper: this element was rewritten with an updated API
(properties); it has gained many new properties, seeking support and
support for linear timecode (LTC) from an audio stream.
- uvch264src now comes with a device provider to advertise available
camera sources that support this interface (mostly Logitech C920s)
- wpe: Add software rendering support and support for mouse scroll
events
- x265enc: support more 8/10/12 bits 4:2:0, 4:2:2 and 4:4:4 profiles;
add support for mastering display info and content light level
encoding SEIs
gst-libav
- Add mapping for SpeedHQ video codec used by NDI
- Add mapping for aptX and aptX-HD
- avivf_mux: support VP9 and AV1
- avvidenc: shift output buffer timestamps and output segment by 1h
just like x264enc does, to allow for negative DTS.
- avviddec: Limit default number of decoder threads on systems with
more than 16 cores, as the number of threads used in avdec has a
direct impact on the latency of the decoder, which is of as many
frames as threads, so a large numbers of threads can make for
latency levels that can be problematic in some applications.
- avviddec: Add thread-type property that allows applications to
specify the preferred multithreading method (auto, frame, slice).
Note that thread-type=frame may introduce additional latency
especially in live pipelines, since it introduces a decoding delay
of number of thread frames.
Plugin and library moves
- There were no plugin moves or library moves in this cycle.
- The rpicamsrc element was moved into -good from an external
repository on github.
Plugin removals
The following elements or plugins have been removed:
- The yadif video deinterlacing plugin from gst-plugins-bad, which was
one of the few GPL licensed plugins, has been removed in favour of
deinterlace method=yadif.
- The avdec_cdgraphics CD Graphics video decoder element from
gst-libav was never usable in GStreamer and we now have a cdgdec
element written in Rust in gst-plugins-rs to replace it.
- The VDPAU plugin has been unmaintained and unsupported for a very
long time and does not have the feature set we expect from
hardware-accelerated video decoders. Its been superseded by the
nvcodec plugin leveraging NVIDIAs NVDEC API.
Miscellaneous API additions
GStreamer core
- gst_task_resume(): This new API allows resuming a task if it was
paused, while leaving it in stopped state if it was stopped or not
started yet. This can be useful for callback-based driver workflows,
where you basically want to pause and resume the task when buffers
are notified while avoiding the race with a gst_task_stop() coming
from another thread.
- info: add printf extensions GST_TIMEP_FORMAT and GST_STIMEP_FORMAT
for printing GstClockTime/GstClockTimeDiff pointers, which is much
more convenient to use in debug log statements than the usual
GST_TIME_FORMAT-followed-by-GST_TIME_ARGS dance. Also add an
explicit GST_STACK_TRACE_SHOW_NONE enum value.
- gst_element_get_current_clock_time() and
gst_element_get_current_running_time(): new helper functions for
getting an element clocks time, and the clock time minus base time,
respectively. Useful when adding additional input branches to
elements such as compositor, audiomixer, flvmux, interleave or
input-selector to determine initial pad offsets and such.
- seeking: Add GST_SEEK_FLAG_TRICKMODE_FORWARD_PREDICTED to just skip
B-frames during trick mode, showing both keyframes + P-frame, and
add support for it in h264parse and h265parse.
- elementfactory: add GST_ELEMENT_FACTORY_TYPE_HARDWARE to allow
elements to advertise that they are hardware-based or interact with
hardware. This has multiple applications:
- it makes it possible to easily differentiate hardware and
software based element implementations such as audio or video
encoders and decoders. This is useful in order to force the use
of software decoders for specific use cases, or to check if a
selected decoder is actually hardware-accelerated or not.
- elements interacting with hardware and their respective drivers
typically dont know the actually supported capabilities until
the element is set into at least READY state and can open a
device handle and probe the hardware.
- gst_uri_from_string_escaped(): identical to gst_uri_from_string()
except that the userinfo and fragment components of the URI will not
be unescaped while parsing. This is needed for correctly parsing
usernames or passwords with : in them .
- paramspecs: new GstParamSpec flag GST_PARAM_CONDITIONALLY_AVAILABLE
to indicate that a property might not always exist.
- gst_bin_iterate_all_by_element_factory_name() finds elements in a
bin by factory name
- pad: gst_pad_get_single_internal_link() is a new convenience
function to return the single internal link of a pad, which is
useful e.g. to retrieve the output pad of a new multiqueue request
pad.
- datetime: Add constructors to create datetimes with timestamps in
microseconds, gst_date_time_new_from_unix_epoch_local_time_usecs()
and gst_date_time_new_from_unix_epoch_utc_usecs().
- gst_debug_log_get_lines() gets debug log lines formatted in the same
way the default log handler would print them
- GstSystemClock: Add GST_CLOCK_TYPE_TAI as GStreamer abstraction for
CLOCK_TAI, to support transmission offloading features where network
packets are timestamped with the time they are deemed to be actually
transmitted. Useful in combination with the new AVTP plugin.
- miscellaneous utility functions: gst_clear_uri(),
gst_structure_take().
- harness: Added gst_harness_pull_until_eos()
- GstBaseSrc:
- gst_base_src_new_segment() allows subclasses to update the
segment to be used at runtime from the ::create() function. This
deprecates gst_base_src_new_seamless_segment()
- gst_base_src_negotiate() allows subclasses to trigger format
renegotiation at runtime from inside the ::create() or ::alloc()
function
- GstBaseSink: new stats property and gst_base_sink_get_stats() method
to retrieve various statistics such as average frame rate and
dropped/rendered buffers.
- GstBaseTransform: gst_base_transform_reconfigure() is now public
API, useful for subclasses that need to completely re-implement the
::submit_input_buffer() virtual method
- GstAggregator:
- gst_aggregator_update_segment() allows subclasses to update the
output segment at runtime. Subclasses should use this function
rather than push a segment event onto the source pad directly.
- new sample selection API:
- subclasses should now call gst_aggregator_selected_samples()
from their ::aggregate() implementation to signal that they
have selected the next samples they will aggregate
- GstAggregator will then emit the samples-selected signal
where handlers can then look up samples per pad via
gst_aggregator_peek_next_sample().
- This is useful for example to atomically update input pad
properties in mixer subclasses such as compositor.
Applications can now update properties with precise control
of when these changes will take effect, and for which input
buffer(s).
- gst_aggregator_finish_buffer_list() allows subclasses to push
out a buffer list, improving efficiency in some cases.
- a ::negotiate() virtual method was added, for consistency with
other base classes and to allow subclasses to completely
override the negotiation behaviour.
- the new ::sink_event_pre_queue() and ::sink_query_pre_queue()
virtual methods allow subclasses to intercept or handle
serialized events and queries before theyre queued up
internally.
GStreamer Plugins Base Libraries
Audio library
- audioaggregator, audiomixer: new output-buffer-duration-fraction
property which allows use cases such as keeping the buffers output
by compositor on one branch and audiomixer on another perfectly
aligned, by requiring the compositor to output a n/d frame rate, and
setting output-buffer-duration-fraction to d/n on the audiomixer.
- GstAudioDecoder: new max-errors property so applications can
configure at what point the decoder should error out, or tell it to
just keep going
- gst_audio_make_raw_caps() and gst_audio_formats_raw() are
bindings-friendly versions of the GST_AUDIO_CAPS_MAKE() C macro.
- gst_audio_info_from_caps() now handles encoded audio formats as well
PbUtils library
- GstEncodingProfile:
- Do not restrict number of similar profiles in a container
- add GstValue serialization function
- codec utils now support more H.264/H.265 profiles/levels and have
improved extension handling
RTP library
- rtpbasepayloader: Add scale-rtptime property for scaling RTP
timestamp according to the segment rate (equivalent to RTSP speed
parameter). This is useful for ONVIF trickmodes via RTSP.
- rtpbasepayload: add experimental property for embedding twcc
sequencenumbers for Transport-Wide Congestion Control (gated behind
the GST_RTP_ENABLE_EXPERIMENTAL_TWCC_PROPERTY environment
variable) - more generic API for enabling this is expected to land
in the next development cycle.
- rtcpbuffer: add RTPFB_TYPE_TWCC for Transport-Wide Congestion
Control
- rtpbuffer: add
gst_rtp_buffer_get_extension_onebyte_header_from_bytes()``, so that one can parse theGBytes`
returned by gst_rtp_buffer_get_extension_bytes()
- rtpbasedepayload: Add max-reorder property to make the
previously-hardcoded value when to consider a sender to have
restarted configurable. In some scenarios its particularly useful
to set max-reorder=0 to disable the behaviour that the depayloader
will drop packets: when max-reorder is set to 0 all
reordered/duplicate packets are considered coming from a restarted
sender.
RTSP library
- add gst_rtsp_url_get_request_uri_with_control() to create request
uri combined with control url
- GstRTSPConnection: add the possibility to limit the Content-Length
for RTSP messages via
gst_rtsp_connection_set_content_length_limit(). The same
functionality is also exposed in gst-rtsp-server.
SDP library
- add support for parsing the extmap attribute from caps and storing
inside caps The extmap attribute allows mapping RTP extension header
IDs to well-known RTP extension header specifications. See RFC8285
for details.
Tags library
- update to latest iso-code and support more languages
- add tags for acoustid id & acoustid fingerprint, plus MusicBrainz ID
handling fixes
Video library
- High Dynamic Range (HDR) video information representation and
signalling enhancements:
- New APIs for HDR video information representation and
signalling:
- GstVideoMasteringDisplayInfo: display color volume info as
per SMPTE ST 2086
- GstVideoContentLightLevel: content light level specified in
CEA-861.3, Appendix A.
- plus functions to serialise/deserialise and add them to or
parse them from caps
- gst_video_color_{matrix,primaries,transfer}_{to,from}_iso():
new utilility functions for conversion from/to ISO/IEC
23001-8
- add ARIB STD-B67 transfer chracteristic function
- add SMPTE ST 2084 support and BT 2100 colorimetry
- define bt2020-10 transfer characteristics for clarity:
bt707, bt2020-10, and bt2020-12 transfer characteristics are
functionally identical but have their own unique values in
the specification.
- h264parse, h265parse: Parse mastering display info and content
light level from SEIs.
- matroskademux: parse HDR metadata
- matroskamux: Write MasteringMetadata and Max{CLL,FALL}. Enable
muxing with HDR meta data if upstream provided it
- avviddec: Extract HDR information if any and map bt2020-10, PQ
and HLG transfer functions
- added bt601 transfer function (for completeness)
- support for more pixel formats:
- Y412 (packed 12 bits 4:4:4:4)
- Y212 (packed 12 bits 4:2:2)
- P012 (semi-planar 4:2:0)
- P016_{LE,BE} (semi-planar 16 bits 4:2:0)
- Y444_16{LE,BE} (planar 16 bits 4:4:4)
- RGB10A2_LE (packed 10-bit RGB with 2-bit alpha channel)
- NV12_32L32 (NV12 with 32x32 tiles in linear order)
- NV12_4L4 (NV12 with 4x4 tiles in linear order)
- GstVideoDecoder:
- new max-errors property so applications can configure at what
point the decoder should error out, or tell it to just keep
going
- new qos property to disable dropping frames because of QoS, and
post QoS messages on the bus when dropping frames. This is
useful for example in a scenario where the decoded video is
tee-ed off to go into a live sink that syncs to the clock in one
branch, and an encoding and save to file pipeline in the other
branch. In that case one wouldnt want QoS events from the video
sink make the decoder drop frames because that would also leave
gaps in the encoding branch then.
- GstVideoEncoder:
- gst_video_encoder_finish_subframe() is new API to push out
subframes (e.g. slices), so encoders can split the encoding into
subframes, which can be useful to reduce the overall end-to-end
latency as we no longer need to wait for the full frame to be
encoded to start decoding or sending out the data.
- new min-force-key-unit-interval property allows configuring the
minimum interval between force-key-unit requests and prevents a
big bitrate increase if a lot of key-units are requested in a
short period of time (as might happen in live streaming RTP
pipelines when packet loss is detected).
- various force-key-unit event handling fixes
- GstVideoAggregator, compositor, glvideomixer: expose
max-last-buffer-repeat property on pads. This can be used to have a
compositor display either the background or a stream on a lower
zorder after a live input stream freezes for a certain amount of
time, for example because of network issues.
- gst_video_format_info_component() is new API to find out which
components are packed into a given plane, which is useful to prevent
us from assuming a 1-1 mapping between planes and components.
- gst_video_make_raw_caps() and gst_video_formats_raw() are
bindings-friendly versions of the GST_VIDEO_CAPS_MAKE() C macro.
- video-blend: Add support for blending on top of 16 bit per component
formats, which makes sure we can support every currently supported
raw video format for blending subtitles or logos on top of video.
- GST_VIDEO_BUFFER_IS_TOP_FIELD() and
GST_VIDEO_BUFFER_IS_BOTTOM_FIELD() convenience macros to check
whether the video buffer contains only the top field or bottom field
of an interlaced picture.
- GstVideoMeta now includes an alignment field with the
GstVideoAlignment so buffer producers can explicitly specify the
exact geometry of the planes, allowing users to easily know the
padded size and height of each plane. Default values will be used if
this is not set.
Use gst_video_meta_set_alignment() to set the alignment and
gst_video_meta_get_plane_size() or gst_video_meta_get_plane_height()
to compute the plane sizes or plane heights based on the information
in the video meta.
- gst_video_info_align_full() works like gst_video_info_align() but
also retrieves the plane sizes.
MPEG-TS library
- support for SCTE-35 sections
- extend support for ATSC tables:
- System Time Table (STT)
- Master Guide Table (MGT)
- Rating Region Table (RRT)
Miscellaneous performance, latency and memory optimisations
As always there have been many performance and memory usage improvements
across all components and modules. Some of them have already been
mentioned elsewhere so wont be repeated here.
The following list is only a small snapshot of some of the more
interesting optimisations that havent been mentioned in other contexts
yet:
- caps negotiation, structure and GValue performance optimizations
- systemclock: clock waiting performance improvements (moved from
GstPoll to GCond for waiting), especially on Windows.
- rtpsession: add support for buffer lists on the recv path for better
performance with higher packet rate streams.
- rtpjitterbuffer: internal timer handling has been rewritten for
better performance, see Nicolas talk “Revisiting RTP Jitter Buffer
Timers” for more details.
- H.264/H.265 parsers and RTP payloaders/depayloaders have been
optimised for latency to make sure data is processed and pushed out
as quickly as possible
- video-scaler: correctness and performance improvements, esp. for
interlaced formats and GBRA
- GstVideoEncoder has gained new API to push out subframes
(e.g. slices), so encoders can split the encoding into subframes,
which can be useful to reduce the overall end-to-end latency as we
no longer need to wait for the full frame to be encoded to start
decoding or sending out the data.
This is complemented by the new GST_VIDEO_BUFFER_FLAG_MARKER which
is a video-specific buffer flag to mark the end of a video frame, so
elements can know that they have received all data for a frame
without waiting for the beginning of the next frame. This is similar
to how the RTP marker flag is used in many RTP video mappings.
The video encoder base class now also releases the internal stream
lock before pushing out data, so as to not block the input side of
things from processing more data in the meantime.
Miscellaneous other changes and enhancements
- it is now possible to modify the initial rank of plugin features
without modifying the source code or writing code to do so
programmatically via the GST_PLUGIN_FEATURE_RANK environment
variable. Users can adjust the rank of plugin(s) by passing a
comma-separated list of feature:rank pairs where rank can be a
numerical value or one of NONE, MARGINAL, SECONDARY, PRIMARY, and
MAX. Example: GST_PLUGIN_FEATURE_RANK=myh264dec:MAX,avdec_h264:NONE
sets the rank of the myh264dec element feature to the maximum and
that of avdec_h264 to 0 (none), thus ensuring that myh264dec is
prefered as H264 decoder in an autoplugging context.
- GstDeviceProvider now does a static probe on start as fallback for
providers that dont support dynamic probing to make things easier
for users
WebRTC
- webrtcbin now contains initial support for renegotiation involving
stream addition and removal. There are a number of caveats to this
initial renegotiation support and many complex scenarios are known
to require some work.
- webrtcbin now exposes the internal ICE object for advanced
configuration options. Using the internal ICE object, it is possible
to toggle UDP or TCP connection usage as well as provide local
network addresses.
- Fix a number of call flows within webrtcbins GstPromise handling
where a promise was never replied to. This has been fixed and now a
promise will always receive a reply.
- webrtcbin now exposes a latency property for configuring the
internal rtpjitterbuffer latency and buffering when receiving
streams.
- webrtcbin now only synchronises the RTP part of a stream, allowing
RTCP messages to skip synchronisation entirely.
- Fixed most of the webrtcbin state properties (connection-state,
ice-connection-state, signaling-state, but not ice-gathering-state
as that requires newer API in libnice and will be fixed in the next
release series) to advance through the state values correctly. Also
implemented DTLS connection states in the DTLS elements so that
peer-connection-state is not always new.
- webrtcbin now accounts for the a=ice-lite attribute in a remote SDP
offer and will configure the internal ICE implementation
accordingly.
- webrtcbin will now resolve .local candidate addresses using the
system DNS resolver. .local candidate addresses are now produced by
web browsers to help protect the privacy of users.
- webrtcbin will now add candidates found in the SDP to the internal
ICE agent. This was previously unsupported and required using the
add-ice-candidate signal manually from the application.
- webrtcbin will now correctly parse a TURN URI that contains a
username or password with a : in it.
- The GStreamer WebRTC library gained a GstWebRTCDataChannel object
roughly matching the interface exposed by the WebRTC specification
to allow for easier binding generation and use of data channels.
OpenGL integration
GStreamer OpenGL bindings/build related changes
- The GStreamer OpenGL library (libgstgl) now ships pkg-config files
for platform-specific API where libgstgl provides a public
integration interface and a pkg-config file for a dependency on the
detected OpenGL headers. The new list of pkg-config files in
addition to the original gstreamer-gl-1.0 are gstreamer-gl-x11-1.0,
gstreamer-gl-wayland-1.0, gstreamer-gl-egl-1.0, and
gstreamer-gl-prototypes-1.0 (for OpenGL headers when including
gst/gl/gstglfuncs.h).
- GStreamer OpenGL now ships some platform-specific introspection data
for platforms that have a public interface. This should allow for
easier integration with bindings involving platform specific
functionality. The new introspection data files are named
GstGLX11-1.0, GstGLWayland-1.0, and GstGLEGL-1.0.
GStreamer OpenGL Features
- The iOS implementation no longer accesses UIKit objects off the main
thread fixing a loud warning message when used in iOS applications.
- Support for mouse and keyboard handling using the GstNavigation
interface was added for the wayland implementation complementing the
already existing support for the X11 and Windows implementations.
- A new helper base class for source elements, GstGLBaseSrc is
provided to ease writing source elements producing OpenGL video
frames.
- Support for some more 12-bit and 16-bit video formats (Y412_LE,
Y412_BE, Y212_LE, Y212_BE, P012_LE, P012_BE, P016, NV16, NV61) was
added to glcolorconvert.
- glupload can now import dma-bufs into external-oes textures.
- A new display type for EGLDevice-based systems was added. It is
currently opt-in by using either the GST_GL_PLATFORM=egl-device
environment variable or manual construction
(gst_gl_display_egl_device_new*()) due to compatibility issues with
some platforms.
- Support was added for WinRT/UWP using the ANGLE project for running
OpenGL-based pipelines within a UWP application.
- Various elements now support changing the GstGLDisplay to be used at
runtime in simple cases. This is primarily helpful for changing or
adding an OpenGL-based video sink that must share an OpenGL context
with an external source to an already running pipeline.
GStreamer Vulkan integration
- There is now a GStreamer Vulkan library to provide integration
points and helpers with applications and external GStreamer Vulkan
based elements. The structure of the library is modelled similarly
to the already existing GStreamer OpenGL library. Please note that
the API is still unstable and may change in future releases,
particularly around memory handling. The GStreamer Vulkan library
contains objects for sharing the vkInstance, vkDevice, vkQueue,
vkImage, VkMemory, etc with other elements and/or the application as
well as some helper objects for using Vulkan in an application or
element.
- Added support for building and running on/for the Android and
Windows systems to complement the existing XCB, Wayland, MacOS, and
iOS implementations.
- XCB gained support for mouse/keyboard events using the GstNavigation
API.
- New vulkancolorconvert element for converting between color formats.
vulkancolorconvert can currently convert to/from all 8-bit RGBA
formats as well as 8-bit RGBA formats to/from the YUV formats AYUV,
NV12, and YUY2.
- New vulkanviewconvert element for converting between stereo view
layouts. vulkanviewconvert can currently convert between all of the
single memory formats (side-by-side, top-bottom, column-interleaved,
row-interleaved, checkerboard, left, right, mono).
- New vulkanimageidentity element for a blit from the input vulkan
image/s to a new vulkan image/s.
- The vulkansink element can now scale the input image to the output
window/surface size where that information is available.
- The vulkanupload element can now configure a transfer from system
memory to VulkanImage-based memory. Previously, this required two
vulkanupload elements.
Tracing framework and debugging improvements
- gst_tracing_get_active_tracers() returns a list of active tracer
objects. This can be used to interact with tracers at runtime using
GObject API such as action signals. This has been implemented in the
leaks tracer for snapshotting and retrieving leaked/active objects
at runtime.
- The leaks tracer can now be interacted with programmatically at
runtime via GObject action signals:
- get-live-object returns a list of live (allocated) traced
objects
- log-live-objects logs a list of live objects into the debug log.
This is the same as sending the SIGUSR1 signal on unix systems,
but works on all operating systems including Windows.
- activity-start-tracking, activity-get-checkpoint,
activity-log-checkpoint, activity-stop-tracking: add support for
tracking and checkpointing objects, similar to what was
previously available via SIGUSR2 on unix systems, but works on
all operating systems including Windows.
- various GStreamer gdb debug helper improvements:
- new gst-pipeline-tree command
- more gdb helper functions: gst_element_pad(), gst_pipeline() and
gst_bin_get()
- support for queries and buffers
- print more info for segment events, print event seqnums, object
pointers and structures
- improve gst-print command to show more pad and element
information
Tools
gst-launch-1.0
- now prints the pipeline position and duration if available when the
pipeline is advancing. This is hopefully more user-friendly and
gives visual feedback on the terminal that the pipeline is actually
up and running. This can be disabled with the --no-position command
line option.
- the parse-launch pipeline syntax now has support for presets:
use@preset=<preset-name>" after an element to load a preset.
gst-inspect-1.0
- new --color command line option to force coloured output even if not
connected to a tty
gst-tester-1.0 (new)
- gst-tester-1.0 is a new tool for plugin developers to launch
.validatetest files with TAP compatible output, meaning it can
easily and cleanly be integrated with the meson test harness. It
allows you to use gst-validate (from the gst-devtools module) to
write integration tests in any GStreamer repository whilst keeping
the tests as close as possible to the code. The tool transparently
handles gst-validate being installed or not: if it is not installed
those integration tests will simply be skipped.
gst-play-1.0
- interactive keyboard controls now also work on Windows
gst-transcoder-1.0 (new)
- gst-transcoder-1.0 is a new command line tool to transcode one URI
into another URI based on the specified encoding profile using the
new GstTranscoder API (see above).
GStreamer RTSP server
- Fix issue where the first few packets (i.e. keyframes) could
sometimes be dropped if the rtsp media pipeline had a live input.
This was a regression from GStreamer 1.14. There are more fixes
pending for that which will hopefully land in 1.18.1.
- Fix backpressure handling when sending data in TCP interleave mode
where RTSP requests and responses and RTP/RTCP packets flow over the
same RTSP TCP connection: The previous implementation would at some
point stop sending data to other clients when a single client
stopped consuming data or did not consume data fast enough. This
obviously created problems for shared media, where the same stream
from a single producer pipeline is sent to multiple clients. Instead
we now manage a backlog in the servers stream-transport component
and remove slow clients once this backlog exceeds a maximum duration
(which is currently hardcoded).
- Onvif Streaming Specification trick modes support (see section at
the beginning)
- Scale/Speed header support: Speed will deliver the data at the
requested speed, which means increasing the data bandwidth for
speeds > 1.0. Scale will attempt to do the same without affecting
the overall bandwidth requirement vis-a-vis normal playback speed
(e.g. it might drop data for fast-forward playback).
- rtspclientsink: send buffer lists in one go for better performance
GStreamer VAAPI
- A lot of work was done adding support for media-driver (iHD), the
new VAAPI driver for Intel, mostly for Gen9 onwards.
- Available color formats and frame sizes are now detected at run-time
according to the context configuration.
- Gallium drivers have been re-enabled in the allowed drivers list
- Improved the mapping between VA formats and GStreamer formats by
generating a mapping table at run-time since even among different
drivers the mapping might be different, particularly for RGB with
little endianness.
- The experimental Flexible Encoding Infrastructure (FEI) elements
have been removed since they were not really actively maintained or
tested.
- Enhanced the juggling of DMABuf buffers and VASurface metas
- New vaapioverlay element: a compositor element using VA VPP blend
capabilities to accelerate overlaying and compositing. Example
pipeline:
gst-launch-1.0 -vf videotestsrc ! vaapipostproc ! tee name=testsrc ! queue \
! vaapioverlay sink_1::xpos=300 sink_1::alpha=0.75 name=overlay ! vaapisink \
testsrc. ! queue ! overlay.
vaapipostproc
- added video-orientation support, supporting frame mirroring and
rotation
- added cropping support, either via properties (crop-left,
crop-right, crop-bottom and crop-top) or buffer meta.
- new skin-tone-enhancenment-level property which is the iHD
replacement of the i965 drivers sink-tone-level. Both are
incompatible with each other, so both were kept.
- handle video colorimetry
- support HDR10 tone mapping
vaapisink
- resurrected wayland backend for non-weston compositors by extracting
the DMABuf from the VASurface and rendering it.
- merged the video overlay API for wayland. Now applications can
define the “window” to render on.
- demoted the vaapisink element to secondary rank since libva
considers rendering as a second-class feature.
VAAPI Encoders
- new common target-percentage property which is the desired target
percentage of bitrate for variable rate control.
- encoders now extract their caps from the driver at registration
time.
- vaapivp9enc: added support for low power mode and support for
profile 2 (profile 0 by default)
- vaapih264enc: new max-qp property that sets the maximum quantization
value. Support for ICQ and QBVR bitrate control mode, adding a
quality-factor property for these modes. Support baseline profile as
constrained-baseline
- vaapih265enc:
- support for main-444 and main-12 encoding profiles.
- new max-qp property that sets the maximum quantization value.
- support for ICQ and QBVR bitrate control mode, adding a
quality-factor property for these modes.
- handle SCC profiles.
- num-tile-cols and num-tile-row properties to specify the number
of tiles to use.
- the low-delay-b property was deprecated and is now determined
automatically.
- improved profile selection through caps.
VAAPI Decoders
- Decoder surfaces are not bound to their context any longer and can
thus be created and used dynamically, removing the deadlock
headache.
- Reverse playback is now fluid
- Forward Region-of-Interest (ROI) metas downstream
- GLTextureUploadMeta uses DMABuf when GEM is not available. Now
Gallium drivers can use this meta for rendering with EGL.
- vaapivp9dec: support for 4:2:2 and 4:4:4 chroma type streams
- vaapih265dec: skip all pictures prior to the first I-frame. Enable
passing range extension flags to the driver. Handle SCC profiles.
- vaapijpegdec: support for 4:0:0, 4:1:1, 4:2:2 and 4:4:4 chroma types
pictures
- vaapih264dec: handle baseline streams as constrained-baseline if
possible and make it more tolerant when encountering unknown NALs
GStreamer OMX
- omxvideoenc: use new video encoder subframe API to push out slices
as soon as theyre ready
- omxh264enc, omxh265enc: negotiate subframe mode via caps. To enable
it, force downstream caps to video/x-h264,alignment=nal or
video/x-h265,alignment=nal.
- omxh264enc: Add ref-frames property
- Zynq ultrascale+ specific video encoder/decoder improvements:
- GRAY8 format support
- support for alternate fields interlacing mode
- video encoder: look-ahead, long-term-ref, and long-term-freq
properties
GStreamer Editing Services and NLE
- Added nested timelines and subproject support so that GES projects
can be used as clips, potentially serializing nested projects in the
main file or referencing external project files.
- Implemented an OpenTimelineIO GES formatter. This means GES and
GStreamer can now load and save projects in all the formats
supported by otio.
- Implemented a GESMarkerList object which allow setting timed
metadata on any GES object.
- Fixed audio rendering issues during clip transition by ensuring that
a single segment is pushed into encoders.
- The GESUriClipAsset API is now MT safe.
- Added ges_meta_container_register_static_meta() to allow fixing a
type for a specific metadata without actually setting a value.
- The framepositioner element now handles resizing the project and
keeps the same positioning when the aspect ratio is not changed .
- Reworked the documentation, making it more comprehensive and much
more detailed.
- Added APIs to retrieve natural size and framerate of a clip (for
example in the case of URIClip it is the framerate/size of the
underlying file).
- ges_container_edit() is now deprecated and GESTimelineElement gained
the ges_timeline_element_edit() method so the editing API is now
usable from any element in the timeline.
- GESProject::loading was added so applications can be notified about
when a new timeline starts loading.
- Implemented the GstStream API in GESTimeline.
- Added a way to add a timeoverlay inside the test source (potentially
with timecodes).
- Added APIs to convert times to frame numbers and vice versa:
- ges_timeline_get_frame_time()
- ges_timeline_get_frame_at()
- ges_clip_asset_get_frame_time()
- ges_clip_get_timeline_time_from_source_frame()
Quite a few validate tests have been implemented to check the
behavior for various demuxer/codec formats
- Added ges_layer_set_active_for_tracks() which allows muting layers
for the specified tracks
- Deprecated GESImageSource and GESMultiFileSource now that we have
imagesequencesrc which handles the imagesequence “protocol”
- Stopped exposing deinterlacing children properties for clip types
where they do not make sense.
- Added support for simple time remapping effects
GStreamer validate
- Introduced the concept of “Test files” allowing to implement “all
included” test cases, meaning that inside the file the following can
be defined:
- The application arguments
- The validate configurations
- The validate scenario
This replaces the previous big dictionary file in
gst-validate-launcher to implement specific test cases.
We set several variables inside the files (as well as inside
scenarios and config files) to make them relocatable.
The file format has been enhanced so it is easier to read and write,
for example line ending with a coma or (curly) brackets can now be
used as continuation marker so you do not need to add \ at the end
of lines to write a structure on several lines.
- Support the imagesequence “protocol” and added integration tests for
it.
- Added action types to allow the scenario to run the Test Clock for
better reproducibility of tests.
- Support generating tests to check that seeking is frame accurate
(base on ssim).
- Added ways to record buffers checksum (in different ways) in the
validateflow module.
- Added vp9 encoding tests.
- Enhanced seeking action types implementation to allow support for
segment seeks.
- Output improvements:
- Logs are now in markdown formats (and bat is used to dump them
if available).
- File format issues in scenarios/configs/tests files are nicely
reported with the line numbers now.
GStreamer Python Bindings
- Python 2.x is no longer supported
- Support mapping buffers without any memcpy:
- Added a ContextManager to make the API more pythonic
with buf.map(Gst.MapFlags.READ | Gst.MapFlags.WRITE) as info:
info.data[42] = 0
- Added high-level helper API for constructing pipelines:
- Gst.Bin.make_and_add(factory_name, instance_name=None)
- Gst.Element.link_many(element, ...)
GStreamer C# Bindings
- Bind gst_buffer_new_wrapped() manually to fix memory handling.
- Fix gst_promise_new_with_change_func() where bindgen didnt properly
detect the func as a closure.
- Declare GstVideoOverlayComposition and GstVideoOverlayRectangle as
opaque type and subclasses of Gst.MiniObject. This changes the API
but without this all usage will cause memory corruption or simply
not work.
- on Windows, look for gstreamer, glib and gobject DLLs using the MSVC
naming convention (i.e. gstvideo-1.0-0.dll instead of
libgstvideo-1.0-0.dll).
The names of these DLLs have to be hardcoded in the bindings, and
most C# users will probably be using the Microsoft toolchain anyway.
This means that the MSVC compiler is now required to build the
bindings, MingW will no longer work out of the box.
GStreamer Rust Bindings and Rust Plugins
The GStreamer Rust bindings are released separately with a different
release cadence thats tied to gtk-rs, but the latest release has
already been updated for the new GStreamer 1.18 API, so theres
absolutely no excuse why your next GStreamer application cant be
written in Rust anymore.
gst-plugins-rs, the module containing GStreamer plugins written in Rust,
has also seen lots of activity with many new elements and plugins.
What follows is a list of elements and plugins available in
gst-plugins-rs, so people dont miss out on all those potentially useful
elements that have no C equivalent.
Rust audio plugins
- audiornnoise: New element for audio denoising which implements the
noise removal algorithm of the Xiph RNNoise library, in Rust
- rsaudioecho: Port of the audioecho element from gst-plugins-good
rsaudioloudnorm: Live audio loudness normalization element based on
the FFmpeg af_loudnorm filter
- claxondec: FLAC lossless audio codec decoder element based on the
pure-Rust claxon implementation
- csoundfilter: Audio filter that can use any filter defined via the
Csound audio programming language
- lewtondec: Vorbis audio decoder element based on the pure-Rust
lewton implementation
Rust video plugins
- cdgdec/cdgparse: Decoder and parser for the CD+G video codec based
on a pure-Rust CD+G implementation, used for example by karaoke CDs
- cea608overlay: CEA-608 Closed Captions overlay element
- cea608tott: CEA-608 Closed Captions to timed-text (e.g. VTT or SRT
subtitles) converter
- tttocea608: CEA-608 Closed Captions from timed-text converter
- mccenc/mccparse: MacCaption Closed Caption format encoder and parser
- sccenc/sccparse: Scenarist Closed Caption format encoder and parser
- dav1dec: AV1 video decoder based on the dav1d decoder implementation
by the VLC project
- rav1enc: AV1 video encoder based on the fast and pure-Rust rav1e
encoder implementation
- rsflvdemux: Alternative to the flvdemux FLV demuxer element from
gst-plugins-good, not feature-equivalent yet
- rsgifenc/rspngenc: GIF/PNG encoder elements based on the pure-Rust
implementations by the image-rs project
Rust text plugins
- textwrap: Element for line-wrapping timed text (e.g. subtitles) for
better screen-fitting, including hyphenation support for some
languages
Rust network plugins
- reqwesthttpsrc: HTTP(S) source element based on the Rust
reqwest/hyper HTTP implementations and almost feature-equivalent
with the main GStreamer HTTP source souphttpsrc
- s3src/s3sink: Source/sink element for the Amazon S3 cloud storage
- awstranscriber: Live audio to timed text transcription element using
the Amazon AWS Transcribe API
Generic Rust plugins
- sodiumencrypter/sodiumdecrypter: Encryption/decryption element based
on libsodium/NaCl
- togglerecord: Recording element that allows to pause/resume
recordings easily and considers keyframe boundaries
- fallbackswitch/fallbacksrc: Elements for handling potentially
failing (network) sources, restarting them on errors/timeout and
showing a fallback stream instead
- threadshare: Set of elements that provide alternatives for various
existing GStreamer elements but allow to share the streaming threads
between each other to reduce the number of threads
- rsfilesrc/rsfilesink: File source/sink elements as replacements for
the existing filesrc/filesink elements
Build and Dependencies
- The Autotools build system has finally been removed in favour of the
Meson build system. Developers who currently use gst-uninstalled
should move to gst-build.
- API and plugin documentation are no longer built with gtk_doc. The
gtk_doc documentation has been removed in favour of a new unified
documentation module built with hotdoc (also see “Documentation
improvements” section below). Distributors should use the
documentation release tarball instead of trying to package hotdoc
and building the documentation from scratch.
- gst-plugins-bad now includes an internal copy of libusrsctp, as
there are problems in usrsctp with global shared state, lack of API
stability guarantees, and the absence of any kind of release
process. We also cant rely on distros shipping a version with the
fixes we need. Both firefox and Chrome bundle their own copies too.
It is still possible to build against an external copy of usrsctp if
so desired.
- nvcodec no longer needs the NVIDIA NVDEC/NVENC SDKs available at
build time, only at runtime. This allows distributions to ship this
plugin by default and it will just start to work when the required
run-time SDK libraries are installed by the user, without users
needing to build and install the plugin from source.
- the gst-editing-services tarball is now named gst-editing-services
for consistency (used to be gstreamer-editing-services).
- the gst-validate tarball has been superseded by the gst-devtools
tarball for consistency with the git module name.
gst-build
gst-build is a meta-module and serves primarily as our uninstalled
development environment. It makes it easy to build most of GStreamer,
but unlike Cerbero it only comes with a limited number of external
dependencies that can be built as subprojects if they are not found on
the system.
gst-build is based on Meson and replaces the old autotools
gst-uninstalled script.
- The uninstalled target has been renamed to devenv
- Experimental gstreamer-full library containing all built plugins and
their deps when building with -Ddefault_library=static. A monolithic
library is easier to distribute, and may be required in some
environments. GStreamer core, GLib and GObject are always included,
but external dependencies are still dynamically linked. The
gst-full-libraries meson option allows adding other GStreamer
libraries to the gstreamer-full build. This is an experiment for now
and its behaviour or API may still change in future releases.
- Add glib-networking as a subproject when glib is a subproject and
load gio modules in the devenv, tls option control whether to use
openssl or gnutls.
- git-worktree: Allow multiple worktrees for subproject branches
- Guard against meson being run from inside the uninstalled devenv, as
this might have unexpected consequences.
- our ffmpeg and x264 meson ports have been updated to the latest
stable version (you might need to update the subprojects checkout
manually though, or just remove the checkouts so meson checks out
the latest version again; improvements for this are pending in
meson, but not merged yet).
Cerbero
Cerbero is a meta build system used to build GStreamer plus dependencies
on platforms where dependencies are not readily available, such as
Windows, Android, iOS and macOS.
General improvements
- Recipe build steps are done in parallel wherever possible. This
leads to massive improvements in overall build time.
- Several recipes were ported to Meson, which improved build times
- Moved from using both GnuTLS and OpenSSL to only OpenSSL
- Moved from yasm to nasm for all assembly compilation
- Support zsh when running the cerbero shell command
- Numerous version upgrades for dependencies
- Default to xz for tarball binary packages. bz2 can be selected with
the --compress-method option to package.
- Added boolean variant for controlling the optimization level:
-v optimization
- Ship .pc pkgconfig files for all plugins in the binary packages
- CMake and nasm will only be built by Cerbero if the system versions
are unusable
- The nvcodec variant was removed and the nvcodec plugin is built by
default now (as it no longer requires the SDK to be installed at
build time, only at runtime)
macOS / iOS
- Minimum iOS SDK version bumped to 11.0
- Minimum macOS SDK version bumped to 10.11
- No longer need to manually add support for newer iOS SDK versions
- Added Vulkan elements via MoltenVK
- Build times were improved by code-signing all build tools
- macOS framework ships all gstreamer libraries instead of an outdated
subset
- Ship pkg-config in the macOS framework package
- fontconfig: Fix EXC_BAD_ACCESS crash on iOS ARM64
- Improved App Store compatibility by setting LC_VERSION_MIN_MACOSX,
fixing relocations, and improved bitcode support
Windows
- MinGW-GCC toolchain was updated to 8.2. It uses the Universal CRT
instead of MSVCRT which eliminates cross-CRT issues in the Visual
Studio build.
- Require Windows 7 or newer for running binaries produced by Cerbero
- Require Windows x86_64 for running Cerbero to build binary packages
- Cerbero no longer uses C:/gstreamer/1.0 as a prefix when building.
That prefix is reserved for use by the MSI installers.
- Several recipes can now be buit with Visual Studio instead of MinGW.
Ported to meson: opus, libsrtp, harfbuzz, cairo, openh264, libsoup,
libusrsctp. Existing build system: libvpx, openssl.
- Support building using Visual Studio for 32-bit x86. Previously we
only supported building for 32-bit x86 using the MinGW toolchain.
- Fixed annoying msgmerge popups in the middle of cerbero builds
- Added configuration options vs_install_path and vs_install_version
for specifying custom search locations for older Visual Studio
versions that do not support vswhere. You can set these in
~/.cerbero/cerbero.cbc where ~ is the MSYS homedir, not your Windows
homedir.
- New Windows-specific plugins: d3d11, mediafoundation, wasapi2
- Numerous compatibility and reliability fixes when running Cerbero on
Windows, especially non-English locales
- proxy-libintl now exports the same symbols as gettext, which makes
it a drop-in replacement
- New mapping variant for selecting the Visual Studio CRT to use:
-v vscrt=<value>. Valid values are md, mdd, and auto (default). A
separate prefix is used when building with either md (release) or
mdd (debug), and the outputted package will have +debug in the
filename. This variant is also used for selecting the correct Qt
libraries (debug vs release) to use when building with -v qt5 on
Windows.
- Support cross-compile on Windows to Windows ARM64 and ARMv7
- Support cross-compile on Windows to the Universal Windows Platform
(UWP). Only the subset of plugins that can be built entirely with
Visual Studio will be selected in this case. To do so, use the
config/cross-uwp-universal.cbc configuration, which will build
ARM64, x86, and x86_64 binaries linked to the release CRT, with
optimizations enabled, and debugging turned on. You can combine this
with -v vscrt=mdd to produce binaries linked to the debug CRT. You
can turn off optimizations with the -v nooptimization variant.
Windows MSI installer
- Require Windows 7 or newer for running GStreamer
- Fixed some issues with shipping of pkg-config in the Windows
installers
- Plugin PDB debug files are now shipped in the development package,
not the runtime package
- Ship installers for 32-bit binaries built with Visual Studio
- Ship debug and release “universal” (ARM64, X86, and X86_64) tarballs
built for the Universal Windows Platform
- Windows MSI installers now install into separate prefixes when
building with MSVC and MinGW. Previously both would be installed
into C:/gstreamer/1.0/x86 or C:/gstreamer/1.0/x86_64. Now, the
installation prefixes are:
----------------------------------------------------------------------------------------------------------------
Target Path Build options
--------------------------- ------------------------------------ -----------------------------------------------
MinGW 32-bit C:/gstreamer/1.0/mingw_x86 -c config/win32.cbc
MinGW 64-bit C:/gstreamer/1.0/mingw_x86_64 -c config/win64.cbc
MSVC 32-bit C:/gstreamer/1.0/msvc_x86 -c config/win32.cbc -v visualstudio
MSVC 64-bit C:/gstreamer/1.0/msvc_x86_64 -c config/win64.cbc -v visualstudio
MSVC 32-bit (debug) C:/gstreamer/1.0/msvc-debug_x86 -c config/win32.cbc -v visualstudio,vscrt=mdd
MSVC 64-bit (debug) C:/gstreamer/1.0/msvc-debug_x86_64 -c config/win64.cbc -v visualstudio,vscrt=mdd
----------------------------------------------------------------------------------------------------------------
Note: UWP binary packages are tarballs, not MSI installers.
Linux
- Support creating MSI installers using WiX when cross-compiling to
Windows
- Support running cross-windows binaries with Wine when using the
shell and runit cerbero commands
- Added bash-completion support inside the cerbero shell on Linux
- Require a system-wide installation of openssl on Linux
- Added variant -v vaapi to build gstreamer-vaapi and the new gstva
plugin
- Debian packaging was disabled because it does not work. Help in
fixing this is appreciated.
- Trimmed the list of packages needed for bootstrap on Linux
Android
- Updated to NDK r21
- Support Vulkan
- Support Qt 5.14+ binary package layout
Platform-specific changes and improvements
Android
- opensles: Remove hard-coded buffer-/latency-time values and allow
openslessink to handle 48kHz streams.
- photography interface and camera source: Add additional settings
relevant to Android such as: Exposure mode property, extra colour
tone values (aqua, emboss, sketch, neon), extra scene modes
(backlight, flowers, AR, HDR), and missing virtual methods for
exposure mode, analog gain, lens focus, colour temperature, min &
max exposure time. Add new effects and scene modes to Camera
parameters.
macOS and iOS
- vtdec can now output to Vulkan-backed memory for zerocopy support
with the Vulkan elements.
Windows
- d3d11videosink: new Direct3D11-based video sink with support for
HDR10 rendering if supported.
- Hardware-accelerated video decoding on Windows via DXVA2 /
Direct3D11 using native Windows APIs rather than per-vendor SDKs
(like MSDK for Intel or NVCODEC for NVidia). Plus modern Direct3D11
integration rather than the almost 20-year old Direct3D9 from
Windows XP times used in d3dvideosink. Formats supported for
decoding are H.264, H.265, VP8, and VP9, and zero-copy operation
should be supported in combination with the new d3d11videosink. See
Seunghas blog post “Windows DXVA2 (via Direct3D 11) Support in
GStreamer 1.17” for more details.
- Microsoft Media Foundation plugin for hardware-accelerated video
encoding on Windows using native Windows APIs rather than per-vendor
SDKs. Formats supported for encoding are H.264, H.265 and VP9. Also
includes audio encoders for AAC and MP3. See Seunghas blog post
“Bringing Microsoft Media Foundation to GStreamer” for some more
details about this.
- new mfvideosrc video capture source element using the latest Windows
APIs rather than ancient APIs used by ksvideosrc/winks. ksvideosrc
should be considered deprecated going forward.
- d3d11: add d3d11convert, a color space conversion and rescaling
element using shaders, and introduce d3d11upload and d3d11download
elements that work just like glupload and gldownload but for D3D11.
- Universal Windows Platform (UWP) support, including official
GStreamer binary packages for it. Check out Nirbheeks latest blog
post “GStreamer 1.18 supports the Universal Windows Platform” for
more details.
- systemclock correctness and reliability fixes, and also dont start
the system clock at 0 any longer (which shouldnt make any
difference to anyone, as absolute clock time values are supposed to
be meaningless in themselves, only the rate of increase matters).
- toolchain specific plugin registry: the registry cache is now named
differently for MSVC and MinGW toolchains/packages, which should
avoid problems when switching between binaries built with a
different toolchain.
- new wasapi2 plugin mainly to support UWP applications. The core
logic of this plugin is almost identical to existing wasapi plugin,
but the main target is Windows 10 and UWP. This plugin uses WinRT
APIs, so will likely not work on Windows 8 or older. Unlike the
existing wasapi plugin, this plugin supports automatic stream
routing (auto fallback when device was removed) and device level
mute/volume control. Exclusive streaming mode is not supported,
however, and loopback features are not implemented yet. It is also
only possible to build this plugin with MSVC and the Windows 10 SDK,
it cant be cross-compiled with the MingW toolchain.
- new dxgiscreencapsrc element which uses the Desktop Duplication API
to capture the desktop screen at high speed. This is only supported
on Windows 8 or later. Compared to the existing elements
dxgiscreencapsrc offers much better performance, works in High DPI
environments and draws an accurate mouse cursor.
- d3dvideosink was downgraded to secondary rank, d3d11videosink is
preferred now. Support OverlayComposition for GPU overlay
compositing of subtitles and logos.
- debug log output fixes, esp. with a non-UTF8 locale/codepage
- speex, jack: fixed crashes on Windows caused by cross-CRT issues
- gst-play-1.0 interactive keyboard controls now also work on Windows
Linux
- kmssink: Add support for P010 and P016 formats
- vah264dec: new experimental va plugin with an element for H.264
decoding with VA-API. This novel approach, different from
gstreamer-vaapi, uses the gstcodecs library for decoder state
handling, which it is hoped will make for cleaner code because it
uses VA-API without further layers or wrappers. Check out Víctors
blog post “New VA-API H.264 decoder in gst-plugins-bad” for the full
lowdown and the limitations of this new plugin, and how to give it a
spin.
- v4l2codecs: introduce a V4L2 CODECs Accelerator. This plugin will
support the new CODECs uAPI in the Linux kernel, which consists of
an accelerator interface similar to DXVA, NVDEC, VDPAU and VAAPI. So
far H.264 and VP8 are supported. This is used on certain embedded
systems such as i.mx8m, rk3288, rk3399, Allwinner H-series SoCs.
Documentation improvements
- unified documentation containing tutorials, API docs, plugin docs,
etc. all under one roof, shipped in form of a documentation release
tarball containing both devhelp and html documentation.
- all documentation is now generated using hotdoc, gtk-doc is no
longer used. Distributors should use the above-mentioned
documentation release tarball instead of trying to package hotdoc
and building the documentation from scratch.
- there is now documentation for wrapper plugins like gst-libav and
frei0r, as well as tracer plugins.
- for more info, check out Thibaults “GStreamer Documentation”
lightning talk from the 2019 GStreamer Conference.
- new API for plugins to support the documentation system:
- new GParamSpecFlag GST_PARAM_DOC_SHOW_DEFAULT to make
gst-inspect-1.0 (and the documentation) show the paramspecs
default value rather than the actually set value as default
- GstPadTemplate getter and setter for “documentation caps”,
gst_pad_template_set_documentation_caps() and
gst_pad_template_get_documentation_caps(): This can be used in
elements where the caps of pad templates are dynamically
generated and/or dependent on the environment, to override the
caps shown in the documentation (usually to advertise the full
set of possible caps).
- gst_type_mark_as_plugin_api() for marking types as plugin API,
used for plugin-internal types like enums, flags, pad
subclasses, boxed types, and such.
Possibly Breaking Changes
- GstVideo: the canonical list of raw video formats (for use in caps)
has been reordered, so video elements such as videotestsrc or
videoconvert might negotiate to a different format now than before.
The new format might be a higher-quality format or require more
processing overhead, which might affect pipeline performance.
- mpegtsdemux used to wrongly advertise H.264 and H.265 video
elementary streams as alignment=nal. This has now been fixed and
changed to alignment=none, which means an h264parse or h265parse
element is now required after tsdemux for some pipelines where there
wasnt one before, e.g. in transmuxing scenarios (tsdemux ! tsmux).
Pipelines without such a parser may now fail to link or error out at
runtime. As parsers after demuxers and before muxers have been
generally required for a long time now it is hoped that this will
only affect a small number of applications or pipelines.
- The Android opensles audio source and sink used to have hard-coded
buffer-/latency-time values of 20ms. This is no longer needed with
newer Android versions and has now been removed. This means a higher
or lower value might now be negotiated by default, which can affect
pipeline performance and latency.
Known Issues
- None in particular
Contributors
Aaron Boxer, Adam Duskett, Adam x Nilsson, Adrian Negreanu, Akinobu
Mita, Alban Browaeys, Alcaro, Alexander Lapajne, Alexandru Băluț, Alex
Ashley, Alex Hoenig, Alicia Boya García, Alistair Buxton, Ali Yousuf,
Ambareesh “Amby” Balaji, Amr Mahdi, Andoni Morales Alastruey, Andreas
Frisch, Andre Guedes, Andrew Branson, Andrey Sazonov, Antonio Ospite,
aogun, Arun Raghavan, Askar Safin, AsociTon, A. Wilcox, Axel Mårtensson,
Ayush Mittal, Bastian Bouchardon, Benjamin Otte, Bilal Elmoussaoui,
Brady J. Garvin, Branko Subasic, Camilo Celis Guzman, Carlos Rafael
Giani, Charlie Turner, Cheng-Chang Wu, Chris Ayoup, Chris Lord,
Christoph Reiter, cketti, Damian Hobson-Garcia, Daniel Klamt, Daniel
Molkentin, Danny Smith, David Bender, David Gunzinger, David Ing, David
Svensson Fors, David Trussel, Debarshi Ray, Derek Lesho, Devarsh
Thakkar, dhilshad, Dimitrios Katsaros, Dmitriy Purgin, Dmitry Shusharin,
Dominique Leuenberger, Dong Il Park, Doug Nazar, dudengke, Dylan McCall,
Dylan Yip, Ederson de Souza, Edward Hervey, Eero Nurkkala, Eike Hein,
ekwange, Eric Marks, Fabian Greffrath, Fabian Orccon, Fabio DUrso,
Fabrice Bellet, Fabrice Fontaine, Fanchao L, Felix Yan, Fernando
Herrrera, Francisco Javier Velázquez-García, Freyr, Fuwei Tang, Gaurav
Kalra, George Kiagiadakis, Georgii Staroselskii, Georg Lippitsch, Georg
Ottinger, gla, Göran Jönsson, Gordon Hart, Gregor Boirie, Guillaume
Desmottes, Guillermo Rodríguez, Haakon Sporsheim, Haihao Xiang, Haihua
Hu, Havard Graff, Håvard Graff, Heinrich Kruger, He Junyan, Henry
Wilkes, Hosang Lee, Hou Qi, Hu Qian, Hyunjun Ko, ibauer, Ignacio Casal
Quinteiro, Ilya Smelykh, Jake Barnes, Jakub Adam, James Cowgill, James
Westman, Jan Alexander Steffens, Jan Schmidt, Jan Tojnar, Javier Celaya,
Jeffy Chen, Jennifer Berringer, Jens Göpfert, Jérôme Laheurte, Jim
Mason, Jimmy Ohn, J. Kim, Joakim Johansson, Jochen Henneberg, Johan
Bjäreholt, Johan Sternerup, John Bassett, Jonas Holmberg, Jonas Larsson,
Jonathan Matthew, Jordan Petridis, Jose Antonio Santos Cadenas, Josep
Torra, Jose Quaresma, Josh Matthews, Joshua M. Doe, Juan Navarro,
Juergen Werner, Julian Bouzas, Julien Isorce, Jun-ichi OKADA, Justin
Chadwell, Justin Kim, Keri Henare, Kevin JOLY, Kevin King, Kevin Song,
Knut Andre Tidemann, Kristofer Björkström, krivoguzovVlad, Kyrylo
Polezhaiev, Lenny Jorissen, Linus Svensson, Loïc Le Page, Loïc Minier,
Lucas Stach, Ludvig Rappe, Luka Blaskovic, luke.lin, Luke Yelavich,
Marcin Kolny, Marc Leeman, Marco Felsch, Marcos Kintschner, Marek
Olejnik, Mark Nauwelaerts, Markus Ebner, Martin Liska, Martin Theriault,
Mart Raudsepp, Matej Knopp, Mathieu Duponchelle, Mats Lindestam, Matthew
Read, Matthew Waters, Matus Gajdos, Maxim Paymushkin, Maxim P.
Dementiev, Michael Bunk, Michael Gruner, Michael Olbrich, Miguel París
Díaz, Mikhail Fludkov, Milian Wolff, Millan Castro, Muhammet Ilendemli,
Nacho García, Nayana Topolsky, Nian Yan, Nicola Murino, Nicolas
Dufresne, Nicolas Pernas Maradei, Niels De Graef, Nikita Bobkov, Niklas
Hambüchen, Nirbheek Chauhan, Ognyan Tonchev, okuoku, Oleksandr
Kvl,Olivier Crête, Ondřej Hruška, Pablo Marcos Oltra, Patricia Muscalu,
Peter Seiderer, Peter Workman, Philippe Normand, Philippe Renon, Philipp
Zabel, Pieter Willem Jordaan, Piotr Drąg, Ralf Sippl, Randy Li, Rasmus
Thomsen, Ratchanan Srirattanamet, Raul Tambre, Ray Tiley, Richard
Kreckel, Rico Tzschichholz, R Kh, Robert Rosengren, Robert Tiemann,
Roman Shpuntov, Roman Sivriver, Ruben Gonzalez, Rubén Gonzalez,
rubenrua, Ryan Huang, Sam Gigliotti, Santiago Carot-Nemesio, Saunier
Thibault, Scott Kanowitz, Sebastian Dröge, Sebastiano Barrera, Seppo
Yli-Olli, Sergey Nazaryev, Seungha Yang, Shinya Saito, Silvio
Lazzeretti, Simon Arnling Bååth, Siwon Kang, sohwan.park, Song Bing,
Soohyun Lee, Srimanta Panda, Stefano Buora, Stefan Sauer, Stéphane
Cerveau, Stian Selnes, Sumaid Syed, Swayamjeet, Thiago Santos, Thibault
Saunier, Thomas Bluemel, Thomas Coldrick, Thor Andreassen, Tim-Philipp
Müller, Ting-Wei Lan, Tobias Ronge, trilene, Tulio Beloqui, U. Artie
Eoff, VaL Doroshchuk, Varunkumar Allagadapa, Vedang Patel, Veerabadhran
G, Víctor Manuel Jáquez Leal, Vivek R, Vivia Nikolaidou, Wangfei, Wang
Zhanjun, Wim Taymans, Wonchul Lee, Xabier Rodriguez Calvar, Xavier
Claessens, Xidorn Quan, Xu Guangxin, Yan Wang, Yatin Maan, Yeongjin
Jeong, yychao, Zebediah Figura, Zeeshan Ali, Zeid Bekli, Zhiyuan Sraf,
Zoltán Imets,
… and many others who have contributed bug reports, translations, sent
suggestions or helped testing.
Stable 1.18 branch
After the 1.18.0 release there will be several 1.18.x bug-fix releases
which will contain bug fixes which have been deemed suitable for a
stable branch, but no new features or intrusive changes will be added to
a bug-fix release usually. The 1.18.x bug-fix releases will be made from
the git 1.18 branch, which will be a stable branch.
1.18.0
1.18.0 was released on 7 September 2020.
Schedule for 1.20
Our next major feature release will be 1.20, and 1.19 will be the
unstable development version leading up to the stable 1.20 release. The
development of 1.19/1.20 will happen in the git master branch.
The plan for the 1.20 development cycle is yet to be confirmed, but it
is now expected that feature freeze will take place some time in January
2021, with the first 1.20 stable release around February/March 2021.
1.20 will be backwards-compatible to the stable 1.18, 1.16, 1.14, 1.12,
1.10, 1.8, 1.6, 1.4, 1.2 and 1.0 release series.
------------------------------------------------------------------------
These release notes have been prepared by Tim-Philipp Müller with
contributions from Mathieu Duponchelle, Matthew Waters, Nirbheek
Chauhan, Sebastian Dröge, Thibault Saunier, and Víctor Manuel Jáquez
Leal.
License: CC BY-SA 4.0