CVE-2016-9811: GStreamer 1.10 release notes

GStreamer 1.10.0 was originally released on 1st November 2016. The latest bug-fix release in the 1.10 series is 1.10.5 and was released on 15 June 2017.

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 new features, bug fixes and other improvements.

See https://gstreamer.freedesktop.org/releases/1.10/ for the latest version of this document.

Last updated: Thursday 15 Jun 2017, 09:30 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 new features, bug fixes and other improvements.

Highlights

• Several convenience APIs have been added to make developers’ lives easier
• A new GstStream API provides applications a more meaningful view of the structure of streams, simplifying the process of dealing with media in complex container formats
• Experimental decodebin3 and playbin3 elements which bring a number of improvements which were hard to implement within decodebin and playbin
• A new parsebin element to automatically unpack and parse a stream, stopping just short of decoding
• Experimental new meson-based build system, bringing faster build and much better Windows support (including for building with Visual Studio)
• A new gst-docs module has been created, and we are in the process of moving our documentation to a markdown-based format for easier maintenance and updates
• A new gst-examples module has been created, which contains example GStreamer applications and is expected to grow with many more examples in the future
• Various OpenGL and OpenGL|ES-related fixes and improvements for greater efficiency on desktop and mobile platforms, and Vulkan support on Wayland was also added
• Extensive improvements to the VAAPI plugins for improved robustness and efficiency
• Lots of fixes and improvements across the board, spanning RTP/RTSP, V4L2, Bluetooth, audio conversion, echo cancellation, and more!

Major new features and changes****Noteworthy new API, features and other changes****Core API additions****Receive property change notifications via bus messages

New API was added to receive element property change notifications via bus messages. So far, applications had to connect a callback to an element’s notify::property-name signal via the GObject API, which was inconvenient for at least two reasons: one had to implement a signal callback function, and that callback function would usually be called from one of the streaming threads, so one had to marshal (send) any information gathered or pending requests to the main application thread which was tedious and error-prone.

Enter gst_element_add_property_notify_watch() and gst_element_add_property_deep_notify_watch() which will watch for changes of a property on the specified element, either only for this element or recursively for a whole bin or pipeline. Whenever such a property change happens, a GST_MESSAGE_PROPERTY_NOTIFY message will be posted on the pipeline bus with details of the element, the property and the new property value, all of which can be retrieved later from the message in the application via gst_message_parse_property_notify(). Unlike the GstBus watch functions, this API does not rely on a running GLib main loop.

The above can be used to be notified asynchronously of caps changes in the pipeline, or volume changes on an audio sink element, for example.

GstBin “deep” element-added and element-removed signals

GstBin has gained "deep-element-added" and "deep-element-removed" signals which makes it easier for applications and higher-level plugins to track when elements are added or removed from a complex pipeline with multiple sub-bins.

playbin makes use of this to implement the new "element-setup" signal which can be used to configure elements as they are added to playbin, just like the existing "source-setup" signal which can be used to configure the source element created.

Error messages can contain additional structured details

It is often useful to provide additional, structured information in error, warning or info messages for applications (or higher-level elements) to make intelligent decisions based on them. To allow this, error, warning and info messages now have API for adding arbitrary additional information to them using a GstStructure: GST_ELEMENT_ERROR_WITH_DETAILS and corresponding API for the other message types.

This is now used e.g. by the new GST_ELEMENT_FLOW_ERROR API to include the actual flow error in the error message, and the souphttpsrc element to provide the HTTP status code, and the URL (if any) to which a redirection has happened.

Redirect messages have official API now

Sometimes, elements need to redirect the current stream URL and tell the application to proceed with this new URL, possibly using a different protocol too (thus changing the pipeline configuration). Until now, this was informally implemented using ELEMENT messages on the bus.

Now this has been formalized in the form of a new GST_MESSAGE_REDIRECT message. A new redirect message can be created using gst_message_new_redirect(). If needed, multiple redirect locations can be specified by calling gst_message_add_redirect_entry() to add further redirect entries, all with metadata, so the application can decide which is most suitable (e.g. depending on the bitrate tags).

New pad linking convenience functions that automatically create ghost pads

New pad linking convenience functions were added: gst_pad_link_maybe_ghosting() and gst_pad_link_maybe_ghosting_full() which were previously internal to GStreamer have now been exposed for general use.

The existing pad link functions will refuse to link pads or elements at different levels in the pipeline hierarchy, requiring the developer to create ghost pads where necessary. These new utility functions will automatically create ghostpads as needed when linking pads at different levels of the hierarchy (e.g. from an element inside a bin to one that’s at the same level in the hierarchy as the bin, or in another bin).

Miscellaneous

Pad probes: IDLE and BLOCK probes now work slightly differently in pull mode, so that push and pull mode have opposite scenarios for idle and blocking probes. In push mode, it will block with some data type and IDLE won’t have any data. In pull mode, it will block before getting a buffer and will be IDLE once some data has been obtained. (commit, bug)

gst_parse_launch_full() can now be made to return a GstBin instead of a top-level pipeline by passing the new GST_PARSE_FLAG_PLACE_IN_BIN flag.

The default GStreamer debug log handler can now be removed before calling gst_init(), so that it will never get installed and won’t be active during initialization.

A new STREAM_GROUP_DONE event was added. In some ways it works similar to the EOS event in that it can be used to unblock downstream elements which may be waiting for further data, such as for example input-selector. Unlike EOS, further data flow may happen after the STREAM_GROUP_DONE event though (and without the need to flush the pipeline). This is used to unblock input-selector when switching between streams in adaptive streaming scenarios (e.g. HLS).

The gst-launch-1.0 command line tool will now print unescaped caps in verbose mode (enabled by the -v switch).

gst_element_call_async() has been added as convenience API for plugin developers. It is useful for one-shot operations that need to be done from a thread other than the current streaming thread. It is backed by a thread-pool that is shared by all elements.

Various race conditions have been fixed around the GstPoll API used by e.g. GstBus and GstBufferPool. Some of these manifested themselves primarily on Windows.

GstAdapter can now keep track of discontinuities signalled via the DISCONT buffer flag, and has gained new API to track PTS, DTS and offset at the last discont. This is useful for plugins implementing advanced trick mode scenarios.

GstTestClock gained a new "clock-type" property.

GstStream API for stream announcement and stream selection

New stream listing and stream selection API: new API has been added to provide high-level abstractions for streams (GstStream) and collections of streams (GstStreamCollections).

Stream listing

A GstStream contains all the information pertinent to a stream, such as stream id, caps, tags, flags and stream type(s); it can represent a single elementary stream (e.g. audio, video, subtitles, etc.) or a container stream. This will depend on the context. In a decodebin3/playbin3 one it will typically be elementary streams that can be selected and unselected.

A GstStreamCollection represents a group of streams and is used to announce or publish all available streams. A GstStreamCollection is immutable - once created it won’t change. If the available streams change, e.g. because a new stream appeared or some streams disappeared, a new stream collection will be published. This new stream collection may contain streams from the previous collection if those streams persist, or completely new ones. Stream collections do not yet list all theoretically available streams, e.g. other available DVD angles or alternative resolutions/bitrate of the same stream in case of adaptive streaming.

New events and messages have been added to notify or update other elements and the application about which streams are currently available and/or selected. This way, we can easily and seamlessly let the application know whenever the available streams change, as happens frequently with digital television streams for example. The new system is also more flexible. For example, it is now also possible for the application to select multiple streams of the same type (e.g. in a transcoding/transmuxing scenario).

A STREAM_COLLECTION message is posted on the bus to inform the parent bin (e.g. playbin3, decodebin3) and/or the application about what streams are available, so you no longer have to hunt for this information at different places. The available information includes number of streams of each type, caps, tags etc. Bins and/or the application can intercept the message synchronously to select and deselect streams before any data is produced - for the case where elements such as the demuxers support the new stream API, not necessarily in the parsebin compatibility fallback case.

Similarly, there is also a STREAM_COLLECTION event to inform downstream elements of the available streams. This event can be used by elements to aggregate streams from multiple inputs into one single collection.

The STREAM_START event was extended so that it can also contain a GstStream object with all information about the current stream, see gst_event_set_stream() and gst_event_parse_stream(). gst_pad_get_stream() is a new utility function that can be used to look up the GstStream from the STREAM_START sticky event on a pad.

Stream selection

Once the available streams have been published, streams can be selected via their stream ID using the new SELECT_STREAMS event, which can be created with gst_event_new_select_streams(). The new API supports selecting multiple streams per stream type. In the future, we may also implement explicit deselection of streams that will never be used, so elements can skip these and never expose them or output data for them in the first place.

The application is then notified of the currently selected streams via the new STREAMS_SELECTED message on the pipeline bus, containing both the current stream collection as well as the selected streams. This might be posted in response to the application sending a SELECT_STREAMS event or when decodebin3 or playbin3 decide on the streams to be initially selected without application input.

Further reading

See further below for some notes on the new elements supporting this new stream API, namely: decodebin3, playbin3 and parsebin.

More information about the new API and the new elements can also be found here:

• GStreamer stream selection design docs
• Edward Hervey’s talk “The new streams API: Design and usage” (slides)
• Edward Hervey’s talk “Decodebin3: Dealing with modern playback use cases” (slides)

Audio conversion and resampling API

The audio conversion library received a completely new and rewritten audio resampler, complementing the audio conversion routines moved into the audio library in the previous release. Integrating the resampler with the other audio conversion library allows us to implement generic conversion much more efficiently, as format conversion and resampling can now be done in the same processing loop instead of having to do it in separate steps (our element implementations do not make use of this yet though).

The new audio resampler library is a combination of some of the best features of other samplers such as ffmpeg, speex and SRC. It natively supports S16, S32, F32 and F64 formats and uses optimized x86 and neon assembly for most of its processing. It also has support for dynamically changing sample rates by incrementally updating the filter tables using linear or cubic interpolation. According to some benchmarks, it’s one of the fastest and most accurate resamplers around.

The audioresample plugin has been ported to the new audio library functions to make use of the new resampler.

Support for SMPTE timecodes

Support for SMPTE timecodes was added to the GStreamer video library. This comes with an abstraction for timecodes, GstVideoTimeCode and a GstMeta that can be placed on video buffers for carrying the timecode information for each frame. Additionally there is various API for making handling of timecodes easy and to do various calculations with them.

A new plugin called timecode was added, that contains an element called timecodestamper for putting the timecode meta on video frames based on counting the frames and another element called timecodewait that drops all video (and audio) until a specific timecode is reached.

Additionally support was added to the Decklink plugin for including the timecode information when sending video out or capturing it via SDI, the qtmux element is able to write timecode information into the MOV container, and the timeoverlay element can overlay timecodes on top of the video.

More information can be found in the talk about timecodes at the GStreamer Conference 2016.

GStreamer OpenMAX IL plugin

The last gst-omx release, 1.2.0, was in July 2014. It was about time to get a new one out with all the improvements that have happened in the meantime. From now on, we will try to release gst-omx together with all other modules.

This release features a lot of bugfixes, improved support for the Raspberry Pi and in general improved support for zerocopy rendering via EGL and a few minor new features.

At this point, gst-omx is known to work best on the Raspberry Pi platform but it is also known to work on various other platforms. Unfortunately, we are not including configurations for any other platforms, so if you happen to use gst-omx: please send us patches with your configuration and code changes!

New Elements****decodebin3, playbin3, parsebin (experimental)

This release features new decoding and playback elements as experimental technology previews: decodebin3 and playbin3 will soon supersede the existing decodebin and playbin elements. We skipped the number 2 because it was already used back in the 0.10 days, which might cause confusion. Experimental technology preview means that everything should work fine already, but we can’t guarantee there won’t be minor behavioural changes in the next cycle. In any case, please test and report any problems back.

Before we go into detail about what these new elements improve, let’s look at the new parsebin element. It works similarly to decodebin and decodebin3, only that it stops one step short and does not plug any actual decoder elements. It will only plug parsers, tag readers, demuxers and depayloaders. Also note that parsebin does not contain any queueing element.

decodebin3's internal architecture is slightly different from the existing decodebin element and fixes many long-standing issues with our decoding engine. For one, data is now fed into the internal multiqueue element after it has been parsed and timestamped, which means that the multiqueue element now has more knowledge and is able to calculate the interleaving of the various streams, thus minimizing memory requirements and doing away with magic values for buffering limits that were conceived when videos were 240p or 360p. Anyone who has tried to play back 4k video streams with decodebin2 will have noticed the limitations of that approach. The improved timestamp tracking also enables multiqueue to keep streams of the same type (audio, video) aligned better, making sure switching between streams of the same type is very fast.

Another major improvement in decodebin3 is that it will no longer decode streams that are not being used. With the old decodebin and playbin, when there were 8 audio streams we would always decode all 8 streams even if 7 were not actually used. This caused a lot of CPU overhead, which was particularly problematic on embedded devices. When switching between streams decodebin3 will try hard to re-use existing decoders. This is useful when switching between multiple streams of the same type if they are encoded in the same format.

Re-using decoders is also useful when the available streams change on the fly, as might happen with radio streams (chained Oggs), digital television broadcasts, when adaptive streaming streams change bitrate, or when switching gaplessly to the next title. In order to guarantee a seamless transition, the old decodebin2 would plug a second decoder for the new stream while finishing up the old stream. With decodebin3, this is no longer needed - at least not when the new and old format are the same. This will be particularly useful on embedded systems where it is often not possible to run multiple decoders at the same time, or when tearing down and setting up decoders is fairly expensive.

decodebin3 also allows for multiple input streams, not just a single one. This will be useful, in the future, for gapless playback, or for feeding multiple external subtitle streams to decodebin/playbin.

playbin3 uses decodebin3 internally, and will supercede playbin. It was decided that it would be too risky to make the old playbin use the new decodebin3 in a backwards-compatible way. The new architecture makes it awkward, if not impossible, to maintain perfect backwards compatibility in some aspects, hence playbin3 was born, and developers can migrate to the new element and new API at their own pace.

All of these new elements make use of the new GstStream API for listing and selecting streams, as described above. parsebin provides backwards compatibility for demuxers and parsers which do not advertise their streams using the new API yet (which is most).

The new elements are not entirely feature-complete yet: playbin3 does not support so-called decodersinks yet where the data is not decoded inside GStreamer but passed directly for decoding to the sink. decodebin3 is missing the various autoplug-* signals to influence which decoders get autoplugged in which order. We’re looking to add back this functionality, but it will probably be in a different way, with a single unified signal and using GstStream perhaps.

For more information on these new elements, check out Edward Hervey’s talk decodebin3 - dealing with modern playback use cases

LV2 ported from 0.10 and switched from slv2 to lilv2

The LV2 wrapper plugin has been ported to 1.0 and moved from using the deprecated slv2 library to its replacement liblv2. We support sources and filter elements. lv2 is short for Linux Audio Developer’s Simple Plugin API (LADSPA) version 2 and is an open standard for audio plugins which includes support for audio synthesis (generation), digital signal processing of digital audio, and MIDI. The new lv2 plugin supersedes the existing LADSPA plugin.

WebRTC DSP Plugin for echo-cancellation, gain control and noise suppression

A set of new elements (webrtcdsp, webrtcechoprobe) based on the WebRTC DSP software stack can now be used to improve your audio voice communication pipelines. They support echo cancellation, gain control, noise suppression and more. For more details you may read Nicolas’ blog post.

Fraunhofer FDK AAC encoder and decoder

New encoder and decoder elements wrapping the Fraunhofer FDK AAC library have been added (fdkaacdec, fdkaacdec). The Fraunhofer FDK AAC encoder is generally considered to be a very high-quality AAC encoder, but unfortunately it comes under a non-free license with the option to obtain a paid, commercial license.

Noteworthy element features and additions****Major RTP and RTSP improvements

• The RTSP server and source element, as well as the RTP jitterbuffer now support remote clock synchronization according to [RFC7273][https://tools.ietf.org/html/rfc7273].
• Support for application and profile specific RTCP packets was added.
• The H265/HEVC payloader/depayloader is again in sync with the final RFC.
• Seeking stability of the RTSP source and server was improved a lot and runs stably now, even when doing scrub-seeking.
• The RTSP server received various major bugfixes, including for regressions that caused the IP/port address pool to not be considered, or NAT hole punching to not work anymore. [Bugzilla #766612][https://bugzilla.gnome.org/show_bug.cgi?id=766612]
• Various other bugfixes that improve the stability of RTP and RTSP, including many new unit / integration tests.

Improvements to splitmuxsrc and splitmuxsink

• The splitmux element received reliability and error handling improvements, removing at least one deadlock case. splitmuxsrc now stops cleanly at the end of the segment when handling seeks with a stop time. We fixed a bug with large amounts of downstream buffering causing incorrect out-of-sequence playback.

• splitmuxsrc now has a "format-location" signal to directly specify the list of files to play from.

• splitmuxsink can now optionally send force-keyunit events to upstream elements to allow splitting files more accurately instead of having to wait for upstream to provide a new keyframe by itself.

OpenGL/GLES improvements****iOS and macOS (OS/X)

• We now create OpenGL|ES 3.x contexts on iOS by default with a fallback to OpenGL|ES 2.x if that fails.
• Various zerocopy decoding fixes and enhancements with the encoding/decoding/capturing elements.
• libdispatch is now used on all Apple platforms instead of GMainLoop, removing the expensive poll()/pthread_*() overhead.

New API

• GstGLFramebuffer - for wrapping OpenGL frame buffer objects. It provides facilities for attaching GstGLMemory objects to the necessary attachment points, binding and unbinding and running a user-supplied function with the framebuffer bound.
• GstGLRenderbuffer (a GstGLBaseMemory subclass) - for wrapping OpenGL render buffer objects that are typically used for depth/stencil buffers or for color buffers where we don’t care about the output.
• GstGLMemoryEGL (a GstGLMemory subclass) - for combining EGLImages with a GL texture that replaces GstEGLImageMemory bringing the improvements made to the other GstGLMemory implementations. This fixes a performance regression in zerocopy decoding on the Raspberry Pi when used with an updated gst-omx.

Miscellaneous improvements

• gltestsrc is now usable on devices/platforms with OpenGL 3.x and OpenGL|ES and has completed or gained support for new patterns in line with the existing ones in videotestsrc.
• gldeinterlace is now available on devices/platforms with OpenGL|ES implementations.
• The dispmanx backend (used on the Raspberry Pi) now supports the gst_video_overlay_set_window_handle() and gst_video_overlay_set_render_rectangle() functions.
• The gltransformation element now correctly transforms mouse coordinates (in window space) to stream coordinates for both perspective and orthographic projections.
• The gltransformation element now detects if the GstVideoAffineTransformationMeta is supported downstream and will efficiently pass its transformation downstream. This is a performance improvement as it results in less processing being required.
• The wayland implementation now uses the multi-threaded safe event-loop API allowing correct usage in applications that call wayland functions from multiple threads.
• Support for native 90 degree rotations and horizontal/vertical flips in glimagesink.

Vulkan

• The Vulkan elements now work under Wayland and have received numerous bugfixes.

QML elements

• qmlglsink video sink now works on more platforms, notably, Windows, Wayland, and Qt’s eglfs (for embedded devices with an OpenGL implementation) including the Raspberry Pi.
• New element qmlglsrc to record a QML scene into a GStreamer pipeline.

KMS video sink

• New element kmssink to render video using Direct Rendering Manager (DRM) and Kernel Mode Setting (KMS) subsystems in the Linux kernel. It is oriented to be used mostly in embedded systems.

Wayland video sink

• waylandsink now supports the wl_viewporter extension allowing video scaling and cropping to be delegated to the Wayland compositor. This extension is also been made optional, so that it can also work on current compositors that don’t support it. It also now has support for the video meta, allowing zero-copy operations in more cases.

DVB improvements

• dvbsrc now has better delivery-system autodetection and several new parameter sanity-checks to improve its resilience to configuration omissions and errors. Superfluous polling continues to be trimmed down, and the debugging output has been made more consistent and precise. Additionally, the channel-configuration parser now supports the new dvbv5 format, enabling dvbbasebin to automatically playback content transmitted on delivery systems that previously required manual description, like ISDB-T.

DASH, HLS and adaptivedemux

• HLS now has support for Alternate Rendition audio and video tracks. Full support for Alternate Rendition subtitle tracks will be in an upcoming release.
• DASH received support for keyframe-only trick modes if the GST_SEEK_FLAG_TRICKMODE_KEY_UNITS flag is given when seeking. It will only download keyframes then, which should help with high-speed playback. Changes to skip over multiple frames based on bandwidth and other metrics will be added in the near future.
• Lots of reliability fixes around seek handling and bitrate switching.

Bluetooth improvements

• The avdtpsrc element now supports metadata such as track title, artist name, and more, which devices can send via AVRCP. These are published as tags on the pipeline.
• The a2dpsink element received some love and was cleaned up so that it actually works after the initial GStreamer 1.0 port.

GStreamer VAAPI

• All the decoders have been split, one plugin feature per codec. So far, the available ones, depending on the driver, are: vaapimpeg2dec, vaapih264dec, vaapih265dec, vaapivc1dec, vaapivp8dec, vaapivp9dec and vaapijpegdec (which already was split).
• Improvements when mapping VA surfaces into memory. It now differentiates between negotiation caps and allocations caps, since the allocation memory for surfaces may be bigger than one that is going to be mapped.
• vaapih265enc now supports constant bitrate mode (CBR).
• Since several VA drivers are unmaintained, we decide to keep a whitelist with the va drivers we actually test, which is mostly the i915 and to a lesser degree gallium from the mesa project. Exporting the environment variable GST_VAAPI_ALL_DRIVERS disables the whitelist.
• Plugin features are registered at run-time, according to their support by the loaded VA driver. So only the decoders and encoder supported by the system are registered. Since the driver can change, some dependencies are tracked to invalidate the GStreamer registry and reload the plugin.
• dmabuf importation from upstream has been improved, gaining performance.
• vaapipostproc now can negotiate buffer transformations via caps.
• Decoders now can do I-frame only reverse playback. This decodes I-frames only because the surface pool is smaller than the required by the GOP to show all the frames.
• The upload of frames onto native GL textures has been optimized too, keeping a cache of the internal structures for the offered textures by the sink.

V4L2 changes

• More pixels formats are now supported
• Decoder is now using G_SELECTION instead of the deprecated G_CROP
• Decoder now uses the STOP command to handle EOS
• Transform element can now scale the pixel aspect ratio
• Colorimetry support has been improved even more
• We now support the OUTPUT_OVERLAY type of video node in v4l2sink

Miscellaneous

• multiqueue's input pads gained a new "group-id" property which can be used to group input streams. Typically one will assign different id numbers to audio, video and subtitle streams for example. This way multiqueue can make sure streams of the same type advance in lockstep if some of the streams are unlinked and the "sync-by-running-time" property is set. This is used in decodebin3/playbin3 to implement almost-instantaneous stream switching. The grouping is required because different downstream paths (audio, video, etc.) may have different buffering/latency etc. so might be consuming data from multiqueue with a slightly different phase, and if we track different stream groups separately we minimize stream switching delays and buffering inside the multiqueue.
• alsasrc now supports ALSA drivers without a position for each channel, this is common in some professional or industrial hardware.
• libvpx based decoders (vp8dec and vp9dec) now create multiple threads on computers with multiple CPUs automatically.
• rfbsrc - used for capturing from a VNC server - has seen a lot of debugging. It now supports the latest version of the RFB protocol and uses GIO everywhere.
• tsdemux can now read ATSC E-AC-3 streams.
• New GstVideoDirection video orientation interface for rotating, flipping and mirroring video in 90° steps. It is implemented by the videoflip and glvideoflip elements currently.
• It is now possible to give appsrc a duration in time, and there is now a non-blocking try-pull API for appsink that returns NULL if nothing is available right now.
• x264enc has support now for chroma-site and colorimetry settings
• A new JPEG2000 parser element was added, and the JPEG2000 caps were cleaned up and gained more information needed in combination with RTP and various container formats.
• Reverse playback support for videorate and deinterlace was implemented
• Various improvements everywhere for reverse playback and KEY_UNITS trick mode
• New cleaned up rawaudioparse and rawvideoparse elements that replace the old audioparse and videoparse elements. There are compatibility element factories registered with the old names to allow existing code to continue to work.
• The Decklink plugin gained support for 10 bit video SMPTE timecodes, and generally got many bugfixes for various issues.
• New API in GstPlayer for setting the multiview mode for stereoscopic video, setting an HTTP/RTSP user agent and a time offset between audio and video. In addition to that, there were various bugfixes and the new gst-examples module contains Android, iOS, GTK+ and Qt example applications.
• GstBin has new API for suppressing various GstElement or GstObject flags that would otherwise be affected by added/removed child elements. This new API allows GstBin subclasses to handle for themselves if they should be considered a sink or source element, for example.
• The subparse element can handle WebVTT streams now.
• A new sdpsrc element was added that can read an SDP from a file, or get it as a string as property and then sets up an RTP pipeline accordingly.

Plugin moves

No plugins were moved this cycle. We’ll make up for it next cycle, promise!

Rewritten memory leak tracer

GStreamer has had basic functionality to trace allocation and freeing of both mini-objects (buffers, events, caps, etc.) and objects in the form of the internal GstAllocTrace tracing system. This API was never exposed in the 1.x API series though. When requested, this would dump a list of objects and mini-objects at exit time which had still not been freed at that point, enabled with an environment variable. This subsystem has now been removed in favour of a new implementation based on the recently-added tracing framework.

Tracing hooks have been added to trace the creation and destruction of GstObjects and mini-objects, and a new tracer plugin has been written using those new hooks to track which objects are still live and which are not. If GStreamer has been compiled against the libunwind library, the new leaks tracer will remember where objects were allocated from as well. By default the leaks tracer will simply output a warning if leaks have been detected on gst_deinit().

If the GST_LEAKS_TRACER_SIG environment variable is set, the leaks tracer will also handle the following UNIX signals:

• SIGUSR1: log alive objects
• SIGUSR2: create a checkpoint and print a list of objects created and destroyed since the previous checkpoint.

Unfortunately this will not work on Windows due to no signals, however.

If the GST_LEAKS_TRACER_STACK_TRACE environment variable is set, the leaks tracer will also log the creation stack trace of leaked objects. This may significantly increase memory consumption however.

New MAY_BE_LEAKED flags have been added to GstObject and GstMiniObject, so that objects and mini-objects that are likely to stay around forever can be flagged and blacklisted from the leak output.

To give the new leak tracer a spin, simply call any GStreamer application such as gst-launch-1.0 or gst-play-1.0 like this:

GST_TRACERS=leaks gst-launch-1.0 videotestsrc num-buffers=10 ! fakesink

If there are any leaks, a warning will be raised at the end.

It is also possible to trace only certain types of objects or mini-objects:

GST_TRACERS="leaks(GstEvent,GstMessage)" gst-launch-1.0 videotestsrc num-buffers=10 ! fakesink

This dedicated leaks tracer is much much faster than valgrind since all code is executed natively instead of being instrumented. This makes it very suitable for use on slow machines or embedded devices. It is however limited to certain types of leaks and won’t catch memory leaks when the allocation has been made via plain old malloc() or g_malloc() or other means. It will also not trace non-GstObject GObjects.

The goal is to enable leak tracing on GStreamer’s Continuous-Integration and testing system, both for the regular unit tests (make check) and media tests (gst-validate), so that accidental leaks in common code paths can be detected and fixed quickly.

For more information about the new tracer, check out Guillaume Desmottes’s “Tracking Memory Leaks” talk or his blog post about the topic.

GES and NLE changes

• Clip priorities are now handled by the layers, and the GESTimelineElement priority property is now deprecated and unused
• Enhanced (de)interlacing support to always use the deinterlace element and expose needed properties to users
• Allow reusing clips children after removing the clip from a layer
• We are now testing many more rendering formats in the gst-validate test suite, and failures have been fixed.
• Also many bugs have been fixed in this cycle!

GStreamer validate changes

This cycle has been focused on making GstValidate more than just a validating tool, but also a tool to help developers debug their GStreamer issues. When reporting issues, we try to gather as much information as possible and expose it to end users in a useful way. For an example of such enhancements, check out Thibault Saunier’s blog post about the new Not Negotiated Error reporting mechanism.

Playbin3 support has been added so we can run validate tests with playbin3 instead of playbin.

We are now able to properly communicate between gst-validate-launcher and launched subprocesses with actual IPC between them. That has enabled the test launcher to handle failing tests specifying the exact expected issue(s).

gst-libav changes

gst-libav uses the recently released ffmpeg 3.2 now, which brings a lot of improvements and bugfixes from the ffmpeg team in addition to various new codec mappings on the GStreamer side and quite a few bugfixes to the GStreamer integration to make it more robust.

Build and Dependencies****Experimental support for Meson as build system****Overview

We have have added support for building GStreamer using the Meson build system. This is currently experimental, but should work fine at least on Linux using the gcc or clang toolchains and on Windows using the MingW or MSVC toolchains.

Autotools remains the primary build system for the time being, but we hope to someday replace it and will steadily work towards that goal.

More information about the background and implications of all this and where we’re hoping to go in future with this can be found in Tim’s mail to the gstreamer-devel mailing list.

For more information on Meson check out these videos and also the Meson talk at the GStreamer Conference.

Immediate benefits for Linux users are faster builds and rebuilds. At the time of writing the Meson build of GStreamer is used by default in GNOME’s jhbuild system.

The Meson build currently still lacks many of the fine-grained configuration options to enable/disable specific plugins. These will be added back in due course.

Note: The meson build files are not distributed in the source tarballs, you will need to get GStreamer from git if you want try it out.

Windows Visual Studio toolchain support

Windows users might appreciate being able to build GStreamer using the MSVC toolchain, which is not possible using autotools. This means that it will be possible to debug GStreamer and applications in Visual Studio, for example. We require VS2015 or newer for this at the moment.

There are two ways to build GStreamer using the MSVC toolchain:

1. Using the MSVC command-line tools (cl.exe etc.) via Meson’s “ninja” backend.
2. Letting Meson’s “vs2015” backend generate Visual Studio project files that can be opened in Visual Studio and compiled from there.

This is currently only for adventurous souls though. All the bits are in place, but support for all of this has not been merged into GStreamer’s cerbero build tool yet at the time of writing. This will hopefully happen in the next cycle, but for now this means that those wishing to compile GStreamer with MSVC will have to get their hands dirty.

There are also no binary SDK builds using the MSVC toolchain yet.

For more information on GStreamer builds using Meson and the Windows toolchain check out Nirbheek Chauhan’s blog post "Building and developing GStreamer using Visual Studio".

Dependencies****gstreamer

libunwind was added as an optional dependency. It is used only for debugging and tracing purposes.

The opencv plugin in gst-plugins-bad can now be built against OpenCV version 3.1, previously only 2.3-2.5 were supported.

gst-plugins-ugly

• mpeg2dec now requires at least libmpeg2 0.5.1 (from 2008).

gst-plugins-bad

• gltransformation now requires at least graphene 1.4.0.

• lv2 now plugin requires at least lilv 0.16 instead of slv2.

Packaging notes

Packagers please note that the gst/gstconfig.h public header file in the GStreamer core library moved back from being an architecture dependent include to being architecture independent, and thus it is no longer installed into $(libdir)/gstreamer-1.0/include/gst but into the normal include directory where it lives happily ever after with all the other public header files. The reason for this is that we now check whether the target supports unaligned memory access based on predefined compiler macros at compile time instead of checking it at configure time.

Platform-specific improvements****Android****New universal binaries for all supported ABIs

We now provide a “universal” tarball to allow building apps against all the architectures currently supported (x86, x86-64, armeabi, armeabi-v7a, armeabi-v8a). This is needed for building with recent versions of the Android NDK which defaults to building against all supported ABIs. Use the Android player example as a reference for the required changes.

Miscellaneous

• New ahssrc element that allows reading the hardware sensors, e.g. compass or accelerometer.

macOS (OS/X) and iOS

• Support for querying available devices on OS/X via the GstDeviceProvider API was added.
• It is now possible to create OpenGL|ES 3.x contexts on iOS and use them in combination with the VideoToolbox based decoder element.
• many OpenGL/GLES improvements, see OpenGL section above

Windows

• gstconfig.h: Always use dllexport/import on Windows with MSVC
• Miscellaneous fixes to make libs and plugins compile with the MVSC toolchain
• MSVC toolchain support (see Meson section above for more details)

New Modules for Documentation, Examples, Meson Build

Three new git modules have been added recently:

gst-docs

This is a new module where we will maintain documentation in the markdown format.

It contains the former gstreamer.com SDK tutorials which have kindly been made available by Fluendo under a Creative Commons license. The tutorials have been reviewed and updated for GStreamer 1.x and will be available as part of the official GStreamer documentation going forward. The old gstreamer.com site will then be shut down with redirects pointing to the updated tutorials.

Some of the existing docbook XML-formatted documentation from the GStreamer core module such as the Application Development Manual and the Plugin Writer’s Guide have been converted to markdown as well and will be maintained in the gst-docs module in future. They will be removed from the GStreamer core module in the next cycle.

This is just the beginning. Our goal is to provide a more cohesive documentation experience for our users going forward, and easier to create and maintain documentation for developers. There is a lot more work to do, get in touch if you want to help out.

If you encounter any problems or spot any omissions or outdated content in the new documentation, please file a bug in bugzilla to let us know.

We will probably release gst-docs as a separate tarball for distributions to package in the next cycle.

gst-examples

A new module has been added for examples. It does not contain much yet, currently it only contains a small http-launch utility that serves a pipeline over http as well as various GstPlayer playback frontends for Android, iOS, Gtk+ and Qt.

More examples will be added over time. The examples in this repository should be more useful and more substantial than most of the examples we ship as part of our other modules, and also written in a way that makes them good example code. If you have ideas for examples, let us know.

No decision has been made yet if this module will be released and/or packaged. It probably makes sense to do so though.

gst-build

gst-build is a new meta module to build GStreamer using the new Meson build system. This module is not required to build GStreamer with Meson, it is merely for convenience and aims to provide a development setup similar to the existing gst-uninstalled setup.

gst-build makes use of Meson’s subproject feature and sets up the various GStreamer modules as subprojects, so they can all be updated and built in parallel.

This module is still very new and highly experimental. It should work at least on Linux and Windows (OS/X needs some build fixes). Let us know of any issues you encounter by popping into the #gstreamer IRC channel or by filing a bug.

This module will probably not be released or packaged (does not really make sense).

Contributors

Aaron Boxer, Aleix Conchillo Flaqué, Alessandro Decina, Alexandru Băluț, Alex Ashley, Alex-P. Natsios, Alistair Buxton, Allen Zhang, Andreas Naumann, Andrew Eikum, Andy Devar, Anthony G. Basile, Arjen Veenhuizen, Arnaud Vrac, Artem Martynovich, Arun Raghavan, Aurélien Zanelli, Barun Kumar Singh, Bernhard Miller, Brad Lackey, Branko Subasic, Carlos Garcia Campos, Carlos Rafael Giani, Christoffer Stengren, Daiki Ueno, Damian Ziobro, Danilo Cesar Lemes de Paula, David Buchmann, Dimitrios Katsaros, Duncan Palmer, Edward Hervey, Emmanuel Poitier, Enrico Jorns, Enrique Ocaña González, Fabrice Bellet, Florian Zwoch, Florin Apostol, Francisco Velazquez, Frédéric Bertolus, Fredrik Fornwall, Gaurav Gupta, George Kiagiadakis, Georg Lippitsch, Göran Jönsson, Graham Leggett, Gregoire Gentil, Guillaume Desmottes, Gwang Yoon Hwang, Haakon Sporsheim, Haihua Hu, Havard Graff, Heinrich Fink, Hoonhee Lee, Hyunjun Ko, Iain Lane, Ian, Ian Jamison, Jagyum Koo, Jake Foytik, Jakub Adam, Jan Alexander Steffens (heftig), Jan Schmidt, Javier Martinez Canillas, Jerome Laheurte, Jesper Larsen, Jie Jiang, Jihae Yi, Jimmy Ohn, Jinwoo Ahn, Joakim Johansson, Joan Pau Beltran, Jonas Holmberg, Jonathan Matthew, Jonathan Roy, Josep Torra, Julien Isorce, Jun Ji, Jürgen Slowack, Justin Kim, Kazunori Kobayashi, Kieran Bingham, Kipp Cannon, Koop Mast, Kouhei Sutou, Kseniia, Kyle Schwarz, Kyungyong Kim, Linus Svensson, Luis de Bethencourt, Marcin Kolny, Marcin Lewandowski, Marianna Smidth Buschle, Mario Sanchez Prada, Mark Combellack, Mark Nauwelaerts, Martin Kelly, Matej Knopp, Mathieu Duponchelle, Mats Lindestam, Matthew Gruenke, Matthew Waters, Michael Olbrich, Michal Lazo, Miguel París Díaz, Mikhail Fludkov, Minjae Kim, Mohan R, Munez, Nicola Murino, Nicolas Dufresne, Nicolas Huet, Nikita Bobkov, Nirbheek Chauhan, Olivier Crête, Paolo Pettinato, Patricia Muscalu, Paulo Neves, Peng Liu, Peter Seiderer, Philippe Normand, Philippe Renon, Philipp Zabel, Pierre Lamot, Piotr Drąg, Prashant Gotarne, Raffaele Rossi, Ray Strode, Reynaldo H. Verdejo Pinochet, Santiago Carot-Nemesio, Scott D Phillips, Sebastian Dröge, Sebastian Rasmussen, Sergei Saveliev, Sergey Borovkov, Sergey Mamonov, Sergio Torres Soldado, Seungha Yang, sezero, Song Bing, Sreerenj Balachandran, Stefan Sauer, Stephen, Steven Hoving, Stian Selnes, Thiago Santos, Thibault Saunier, Thijs Vermeir, Thomas Bluemel, Thomas Jones, Thomas Klausner, Thomas Scheuermann, Tim-Philipp Müller, Ting-Wei Lan, Tom Schoonjans, Ursula Maplehurst, Vanessa Chipirras Navalon, Víctor Manuel Jáquez Leal, Vincent Penquerc’h, Vineeth TM, Vivia Nikolaidou, Vootele Vesterblom, Wang Xin-yu (王昕宇), William Manley, Wim Taymans, Wonchul Lee, Xabier Rodriguez Calvar, Xavier Claessens, xlazom00, Yann Jouanin, Zaheer Abbas Merali

… and many others who have contributed bug reports, translations, sent suggestions or helped testing.

Bugs fixed in 1.10

More than 750 bugs have been fixed during the development of 1.10.

This list does not include issues that have been cherry-picked into the stable 1.8 branch and fixed there as well, all fixes that ended up in the 1.8 branch are also included in 1.10.

This list also does not include issues that have been fixed without a bug report in bugzilla, so the actual number of fixes is much higher.

Stable 1.10 branch

After the 1.10.0 release there will be several 1.10.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 1.10.x bug-fix releases will be made from the git 1.10 branch, which is a stable branch.

1.10.0

1.10.0 was released on 1st November 2016.

1.10.1

The first 1.10 bug-fix release (1.10.1) was released on 17 November 2016. This release only contains bugfixes and it should be safe to update from 1.10.x.

Major bugfixes in 1.10.1

• Security-relevant bugfix in the vmnc decoder (CVE-2016-9445, CVE-2016-9446)
• Various bugfixes to playbin3/decodebin3
• Fix error at the end of playing any WAV file
• Fix usability of androidmedia plugin if the camera or sensor API is not available, but codecs are
• Handle redirections on PLAY, and missing control attribute in the RTSP source
• Various OpenGL related bugfixes
• … and many, many more!

For a full list of bugfixes see Bugzilla. Note that this is not the full list of changes. For the full list of changes please refer to the GIT logs or ChangeLogs of the particular modules.

1.10.2

The second 1.10 bug-fix release (1.10.2) was released on 29 November 2016. This release only contains bugfixes and it should be safe to update from 1.10.x.

Major bugfixes in 1.10.2

• Security-relevant bugfix in the FLI/FLX/FLC decoder (CVE-2016-9634, CVE-2016-9635, CVE-2016-9636)
• Various fixes for crashes, assertions and other failures on fuzzed input files. Among others, thanks to Hanno Böck for testing and reporting (CVE-2016-9807, CVE-2016-9808, CVE-2016-9809, CVE-2016-9810, CVE-2016-9811, CVE-2016-9812, CVE-2016-9813).
• SAVP/SAVPF profile in gst-rtsp-server works for live streams again, and the correct MIKEY policy message is generated
• Further OpenGL related bugfixes
• gst-libav was updated to ffmpeg 3.2.1
• … and many, many more!

For a full list of bugfixes see Bugzilla. Note that this is not the full list of changes. For the full list of changes please refer to the GIT logs or ChangeLogs of the particular modules.

1.10.3

The third 1.10 bug-fix release (1.10.3) was released on 30 January 2017. This release only contains bugfixes and it should be safe to update from 1.10.x.

Major bugfixes in 1.10.3

• Various fixes for crashes, assertions, deadlocks and memory leaks on fuzzed input files and in other situations (CVE-2016-10198, CVE-2016-10199, CVE-2017-5837, CVE-2017-5838, CVE-2017-5839, CVE-2017-5840, CVE-2017-5841, CVE-2017-5842, CVE-2017-5843, CVE-2017-5844, CVE-2017-5845, CVE-2017-5846)
• Regression fixes for souphttpsrc with redirection tracking and retrying
• Regression fix for gst-rtsp-server not handling TCP-only medias anymore
• Various other bugfixes the RTP/RTSP codebase
• Fix for gmodule on 64 bit Android, it was never working before there
• vp8enc works again on 32 bit Windows
• Fixes to Opus PLC handling in the decoder
• Fix for stream corruption in multihandlesink when removing clients
• gst-libav was updated to ffmpeg 3.2.2
• … and many, many more!

For a full list of bugfixes see Bugzilla. Note that this is not the full list of changes. For the full list of changes please refer to the GIT logs or ChangeLogs of the particular modules.

1.10.4

The fourth 1.10 bug-fix release (1.10.4) was released on 23 February 2017. This release only contains bugfixes and it should be safe to update from 1.10.x.

Major bugfixes in 1.10.4

• Various fixes for crashes, assertions, deadlocks and memory leaks on fuzzed input files and in other situations (CVE-2017-5847, CVE-2017-5848)
• More regression fixes for souphttpsrc redirection tracking
• Regression fix for gmodule on 32 bit Android, which was introduced as part of the 64 bit Android fix in 1.10.3 and broke the androidmedia plugin
• Various bugfixes for regressions and other problems in the V4L2 plugin
• Fix for 5.1, 6.1 and 7.1 channel layouts for Vorbis
• Fixes for timestamp generation of Android video encoder element
• gst-libav was updated to ffmpeg 3.2.4, fixing a couple of CVEs
• … and many, many more!

For a full list of bugfixes see Bugzilla. Note that this is not the full list of changes. For the full list of changes please refer to the GIT logs or ChangeLogs of the particular modules.

1.10.5

The fifth 1.10 bug-fix release (1.10.5) was released on 15 June 2017. This release only contains bugfixes and it should be safe to update from 1.10.x.

This is most likely the last release in the stable 1.10 release series.

Major bugfixes in 1.10.5

• Various fixes for crashes, assertions, deadlocks and memory leaks
• Fix for regression in rtpjitterbuffer that caused PTS to be reset to 0 on
• Fix detection of some HLS streams
• Various fixes to the MOV/MP4 muxer and demuxer gaps in the input
• gst-libav was updated to ffmpeg 3.2.5, fixing a couple of CVEs
• … and many, many more!

For a full list of bugfixes see Bugzilla. Note that this is not the full list of changes and also does not include bug reports that were closed after the 1.12.0 release. For the full list of changes please refer to the GIT logs or ChangeLogs of the particular modules.

Known Issues

• iOS builds with iOS 6 SDK and old C++ STL. You need to select iOS 6 instead of 7 or 8 in your projects settings to be able to link applications. Bug #766366
• Code signing for Apple platforms has some problems currently, requiring manual work to get your application signed. Bug #771860
• Building applications with Android NDK r13 on Windows does not work. Other platforms and earlier/later versions of the NDK are not affected. Bug #772842

Schedule for 1.12

Our next major feature release will be 1.12, and 1.11 will be the unstable development version leading up to the stable 1.12 release. The development of 1.11/1.12 will happen in the git master branch.

The plan for the 1.12 development cycle is yet to be confirmed, but it is expected that feature freeze will be around early/mid-February, followed by several 1.11 pre-releases and the new 1.12 stable release in April.

1.12 will be backwards-compatible to the stable 1.10, 1.8, 1.6, 1.4, 1.2 and 1.0 release series.

These release notes have been prepared by Olivier Crête, Sebastian Dröge, Nicolas Dufresne, Edward Hervey, Víctor Manuel Jáquez Leal, Tim-Philipp Müller, Reynaldo H. Verdejo Pinochet, Arun Raghavan, Thibault Saunier, Jan Schmidt, Wim Taymans, Matthew Waters

License: CC BY-SA 4.0