platform/upstream/gstreamer.git
8 years agomultifilesrc: set position as offset from start-index
Prashant Gotarne [Thu, 16 Jul 2015 04:18:46 +0000 (09:48 +0530)]
multifilesrc: set position as offset from start-index

query position in GST_FORMAT_BUFFER returns
offset from start-index rather than index.

https://bugzilla.gnome.org/show_bug.cgi?id=752462

8 years agotests: add unit test for JPEG-2000 rtp payloader leak
Tim-Philipp Müller [Fri, 27 May 2016 11:49:32 +0000 (12:49 +0100)]
tests: add unit test for JPEG-2000 rtp payloader leak

https://bugzilla.gnome.org/show_bug.cgi?id=766870

8 years agortpj2kpay: Fix buffer memory leak
Pierre Lamot [Wed, 25 May 2016 15:11:13 +0000 (17:11 +0200)]
rtpj2kpay: Fix buffer memory leak

Input buffer memory was not unmapped

https://bugzilla.gnome.org/show_bug.cgi?id=766870

8 years agov4l2object: fix caps leak
Guillaume Desmottes [Wed, 18 May 2016 09:12:15 +0000 (12:12 +0300)]
v4l2object: fix caps leak

gst_v4l2_object_probe_caps() was taking an extra ref on the returned
caps for no reason.

https://bugzilla.gnome.org/show_bug.cgi?id=766610

8 years agovideocrop mark crop properties as mutable in playing state
Tim-Philipp Müller [Sun, 22 May 2016 19:14:18 +0000 (20:14 +0100)]
videocrop mark crop properties as mutable in playing state

8 years agosouphttpsrc: fix buffer leak when flushing
Guillaume Desmottes [Fri, 20 May 2016 13:47:35 +0000 (16:47 +0300)]
souphttpsrc: fix buffer leak when flushing

When early returning in gst_soup_http_src_read_buffer() because the
element is FLUSHING, we need to unmap and unref the buffer which was just created.

https://bugzilla.gnome.org/show_bug.cgi?id=766718

8 years agoqtdemux: Set seek event seqnum on all SEGMENT events
Sebastian Dröge [Fri, 20 May 2016 08:15:44 +0000 (11:15 +0300)]
qtdemux: Set seek event seqnum on all SEGMENT events

Some were forgotten.

See https://bugzilla.gnome.org/show_bug.cgi?id=765935

8 years agoavidemux: Pass through seek event seqnums in all SEGMENT/EOS events and SEGMENT_DONE...
Sebastian Dröge [Fri, 20 May 2016 08:12:44 +0000 (11:12 +0300)]
avidemux: Pass through seek event seqnums in all SEGMENT/EOS events and SEGMENT_DONE messages/events

See https://bugzilla.gnome.org/show_bug.cgi?id=765935

8 years agomatroskademux: Set seek event seqnum in EOS and SEGMENT_DONE messages/events
Sebastian Dröge [Fri, 20 May 2016 07:56:52 +0000 (10:56 +0300)]
matroskademux: Set seek event seqnum in EOS and SEGMENT_DONE messages/events

Also actually store the seqnum in pull mode seeks.

See https://bugzilla.gnome.org/show_bug.cgi?id=765935

8 years agodeinterlace: fix caps leak
Guillaume Desmottes [Tue, 17 May 2016 10:40:38 +0000 (13:40 +0300)]
deinterlace: fix caps leak

The caps returned by gst_pad_get_current_caps() was never unreffed when
not early returning.

Fix a leak with the elements/deinterlace test.

https://bugzilla.gnome.org/show_bug.cgi?id=766558

8 years agortpsession: don't act on suspicious BYE RTCP
Mikhail Fludkov [Mon, 25 Jan 2016 15:25:51 +0000 (16:25 +0100)]
rtpsession: don't act on suspicious BYE RTCP

Some endpoints (like Tandberg E20) can send BYE packet containing our
internal SSRC. I this case we would detect SSRC collision and get rid
of the source at some point. But because we are still sending packets
with that SSRC the source will be recreated immediately.
This brand new internal source will not have some variables incorrectly
set in its state. For example 'seqnum-base` and `clock-rate` values will be
-1.
The fix is not to act on BYE RTCP if it contains internal or unknown
SSRC.

https://bugzilla.gnome.org/show_bug.cgi?id=762219

8 years agortpsession: Add test for locking of the stats signal
Mikhail Fludkov [Sun, 15 Nov 2015 13:54:28 +0000 (14:54 +0100)]
rtpsession: Add test for locking of the stats signal

Keeping the lock while emitting the stats signal introduces potential
deadlock in those situations when the signal callback wants the access
to rtpsession's properties which also requre the lock.

https://bugzilla.gnome.org/show_bug.cgi?id=762216

8 years agomatroskademux: don't hold object lock whilst pushing out headers
Seungha Yang [Thu, 19 May 2016 06:36:57 +0000 (15:36 +0900)]
matroskademux: don't hold object lock whilst pushing out headers

matroskademux would take the GST_OBJECT_LOCK in
- gst_matroska_demux_push_codec_data_all()
- gst_matroska_demux_query()

Some parse element such as FLAC checks upstream seekability, and
there is some use cases that matroska-demux is linked to a parse element
(e.g.,FLAC format) without intermediate elements (e.g., queue).
In this case, matroska-demux never returns from _push_codec_data_all()
because the parser can return only after it receives the response to
the upstream query, but that's not going to happen because it's
deadlocked.

Elements must not hold the object lock whilst pushing out events
or data.

https://bugzilla.gnome.org/show_bug.cgi?id=766645

8 years agosouphttpclientsink: Set sent_buffers and streamheader_buffers to NULL after freeing
Sebastian Dröge [Thu, 19 May 2016 09:43:01 +0000 (12:43 +0300)]
souphttpclientsink: Set sent_buffers and streamheader_buffers to NULL after freeing

Otherwise we might use an already freed list later and crash or worse.

8 years agoudpsrc: fix Since version for new "loop" property
Tim-Philipp Müller [Wed, 18 May 2016 17:32:57 +0000 (18:32 +0100)]
udpsrc: fix Since version for new "loop" property

8 years agortpdec: fix clock leak
Guillaume Desmottes [Mon, 16 May 2016 14:18:37 +0000 (16:18 +0200)]
rtpdec: fix clock leak

gst_system_clock_obtain() returns a new ref.

https://bugzilla.gnome.org/show_bug.cgi?id=766521

8 years agoudpsrc: add doc blurb with since marker for new "loop" property
Tim-Philipp Müller [Tue, 17 May 2016 04:33:35 +0000 (05:33 +0100)]
udpsrc: add doc blurb with since marker for new "loop" property

8 years agoavimux: add support for png
Dimitrios Katsaros [Fri, 13 Nov 2015 14:52:35 +0000 (15:52 +0100)]
avimux: add support for png

https://bugzilla.gnome.org/show_bug.cgi?id=758059

8 years agosplitmuxsrc: Connect to demux signals before activating
Jan Schmidt [Sun, 15 May 2016 12:07:14 +0000 (22:07 +1000)]
splitmuxsrc: Connect to demux signals before activating

Fix a race in splitmuxsrc by properly connecting to the
demuxer signals we're interested in *before* setting it running.

8 years agodocs: Update for git master
Sebastian Dröge [Sun, 15 May 2016 10:31:37 +0000 (13:31 +0300)]
docs: Update for git master

8 years agortpmp4gpay: Don't produce timestamps based on byte count
Olivier Crête [Sun, 15 May 2016 10:16:23 +0000 (12:16 +0200)]
rtpmp4gpay: Don't produce timestamps based on byte count

The GST_BUFFER_OFFSET of output buffers returned to GstRtpBasePayload
should reflect the number of "samples" in the unit of the RTP clock in this
buffer. If this is not true, then it shouldn't be set.

https://bugzilla.gnome.org/show_bug.cgi?id=761943

8 years agomatroska-mux: Fix strcmp usage
Edward Hervey [Sun, 15 May 2016 10:24:03 +0000 (12:24 +0200)]
matroska-mux: Fix strcmp usage

Just use g_strcmp0 which can handle NULL entries

8 years agosouphttpsrc: Use audio/x-unaligned-raw instead of audio/x-raw for L16 data
Carlos Rafael Giani [Fri, 4 Mar 2016 09:14:00 +0000 (10:14 +0100)]
souphttpsrc: Use audio/x-unaligned-raw instead of audio/x-raw for L16 data

Directly setting audio/x-raw caps leads to problems when the delivered
data blocks do not align properly at sample boundaries (for example, a
data block with 391 bytes). So, instead, set audio/x-unaligned-raw to
let a parser be autoplugged.

https://bugzilla.gnome.org/show_bug.cgi?id=689460

8 years agoqtdemux: Parsing elst box based on version
Seungha Yang [Thu, 12 May 2016 02:52:09 +0000 (11:52 +0900)]
qtdemux: Parsing elst box based on version

segment_duration and media_time should be parsed based on version
of elst box. Specification defines that an elst box with version 1
has uint64 and int64 values for segment_duration and media_time,
respectively.

https://bugzilla.gnome.org/show_bug.cgi?id=766301

8 years agosouphttpsrc: check if request was cancelled when sending message
Thiago Santos [Sat, 14 May 2016 15:57:41 +0000 (12:57 -0300)]
souphttpsrc: check if request was cancelled when sending message

It might be that the request was aborted by the application and
we can return immediatelly

8 years agosouphttpsrc: proxy resolver is on by default
Thiago Santos [Sat, 14 May 2016 15:43:54 +0000 (12:43 -0300)]
souphttpsrc: proxy resolver is on by default

Remove from the session creation parameters

8 years agosoup: update build to warn about newer deprecated functions
Thiago Santos [Sat, 14 May 2016 15:15:48 +0000 (12:15 -0300)]
soup: update build to warn about newer deprecated functions

We already depend on 2.48

8 years agosouphttpsrc: reduce reading latency by using non-blocking read
Thiago Santos [Sat, 14 May 2016 14:09:33 +0000 (11:09 -0300)]
souphttpsrc: reduce reading latency by using non-blocking read

Non-blocking read will return the amount of data available without
blocking to wait for the full requested size.

The downside is that now it souphttpsrc needs to have a waiting
mechanism in case there is no data available yet to avoid busy
looping arond the inputstream.

8 years agortpsession: Take the lock already when reading the other stats, not just for the...
Sebastian Dröge [Sun, 15 May 2016 09:30:50 +0000 (12:30 +0300)]
rtpsession: Take the lock already when reading the other stats, not just for the hash table

https://bugzilla.gnome.org/show_bug.cgi?id=766025

8 years agomatroska: use math-compat.h for NAN define
Tim-Philipp Müller [Sat, 14 May 2016 16:04:57 +0000 (17:04 +0100)]
matroska: use math-compat.h for NAN define

8 years agosplitmuxsink: Use GstBin async-handling instead of our own.
Jan Schmidt [Sat, 14 May 2016 13:39:22 +0000 (23:39 +1000)]
splitmuxsink: Use GstBin async-handling instead of our own.

Set the async-handling property on GstBin to let it manage
async-handling instead of the local handling from the previous
commit. Works because of #174a5e in core

8 years agosouphttpsrc: refactor to use Soup's sync API
Thiago Santos [Fri, 13 May 2016 13:17:33 +0000 (10:17 -0300)]
souphttpsrc: refactor to use Soup's sync API

Replace the async API with the sync API to remove all the extra mainloop
and context handling. Currently it blocks reading until 'blocksize'
bytes are available but that can be improved by using:

https://developer.gnome.org/gio/unstable/GPollableInputStream.html#g-pollable-input-stream-read-nonblocking

https://bugzilla.gnome.org/show_bug.cgi?id=693911

8 years agotests: souphttpsrc: replace deprecated API
Thiago Santos [Sat, 14 May 2016 07:50:36 +0000 (04:50 -0300)]
tests: souphttpsrc: replace deprecated API

Avoid using soup_server_run_async and old get_port() APIs,
replace with me soup_server_listen and get the port through the
URIs list returned from the server.

8 years agojitterbuffer: Upgrade debug message to error
Olivier Crête [Sat, 14 May 2016 10:34:10 +0000 (12:34 +0200)]
jitterbuffer: Upgrade debug message to error

It causes the entire pipeline to fail, it should be easier to find.

8 years agosplitmuxsink: Hide internal async state changes.
Jan Schmidt [Sat, 14 May 2016 08:32:52 +0000 (18:32 +1000)]
splitmuxsink: Hide internal async state changes.

When switching fragments, hide the async-start/async-done
messages from the parent bin, as otherwise we sometimes (very rarely)
hang in PAUSED instead of returning / continuing to PLAYING
state.

8 years agosplitmuxsink: Remove stray carriage-return from debug
Jan Schmidt [Fri, 13 May 2016 11:20:28 +0000 (21:20 +1000)]
splitmuxsink: Remove stray carriage-return from debug

8 years agortp: Ship gstrtpj2kcommon.h file to fix distcheck
Sebastian Dröge [Fri, 13 May 2016 13:43:21 +0000 (16:43 +0300)]
rtp: Ship gstrtpj2kcommon.h file to fix distcheck

8 years agoavimux: Do not write index and header if idx is NULL
Jesper Larsen [Thu, 30 Apr 2015 12:43:04 +0000 (14:43 +0200)]
avimux: Do not write index and header if idx is NULL

Fixes criticals with e.g.
videotestsrc num-buffers=1 ! identity drop-probability=1.0 ! avimux ! fakesink

https://bugzilla.gnome.org/show_bug.cgi?id=748700

8 years agortpj2kpay: manage T tile invalidation bit correctly, update tile id in header correctly.
Aaron Boxer [Thu, 12 May 2016 12:43:39 +0000 (08:43 -0400)]
rtpj2kpay: manage T tile invalidation bit correctly, update tile id in header correctly.

1. according to RFC, T bit is only set when either the RTP packet only contains the J2K main header, or the packet contains tile parts from multiple tiles. This is now being managed correctly in the code. The second scenario cannot happen with our payloader, since tile headers are always placed in their own RTP packet, and so a packet cannot contain tile parts from multiple tiles.
However, I have added code to track if multiple tile parts are included in a single RTP packet, in case in the future we want to put header and data in same packet.

2. Old code would set the tile id to zero for all J2K packets. This is now set correctly to the appropriate tile id.

https://bugzilla.gnome.org/show_bug.cgi?id=745187

8 years agortpj2kpay: manage fragmented headers correctly
Aaron Boxer [Thu, 12 May 2016 12:41:51 +0000 (08:41 -0400)]
rtpj2kpay: manage fragmented headers correctly

J2K main header framentation across multiple RTP packets is now handled correctly

https://bugzilla.gnome.org/show_bug.cgi?id=745187

8 years agortpj2k: move common code to shared header, code clean up
Aaron Boxer [Wed, 11 May 2016 19:04:26 +0000 (15:04 -0400)]
rtpj2k: move common code to shared header, code clean up

https://bugzilla.gnome.org/show_bug.cgi?id=745187

8 years agortpj2k: update documentation
Aaron Boxer [Wed, 11 May 2016 19:01:32 +0000 (15:01 -0400)]
rtpj2k: update documentation

https://bugzilla.gnome.org/show_bug.cgi?id=745187

8 years agoauparse: Fix sticky event misordering warning
Patricia Muscalu [Thu, 12 May 2016 12:43:43 +0000 (14:43 +0200)]
auparse: Fix sticky event misordering warning

Make sure that src pad has caps before sending segment event.

https://bugzilla.gnome.org/show_bug.cgi?id=766359

8 years agortpsession: Don't notify about stats property changes while taking the session lock
Sebastian Dröge [Wed, 11 May 2016 06:28:13 +0000 (09:28 +0300)]
rtpsession: Don't notify about stats property changes while taking the session lock

The signal handlers might want to actually get the value of the stats
property, which would take the session lock again and deadlock.

This was introduced by 2e960e70750a0cb7e1117d0c09d08597866a29ee.

https://bugzilla.gnome.org/show_bug.cgi?id=766025

8 years agoqtdemux: improve edts segment handling after seeks in push mode
Thiago Santos [Tue, 3 May 2016 16:59:54 +0000 (13:59 -0300)]
qtdemux: improve edts segment handling after seeks in push mode

Properly handle edts segments for push-based operation seeking.
We only support edts that a single segment that has media at the end,
being preceeded by any number of gap segments.

This also allows the qt segment rate to be respected after seeks

https://bugzilla.gnome.org/show_bug.cgi?id=765669

8 years agoqtdemux: properly activate segment with rate != 1.0
Thiago Santos [Tue, 3 May 2016 13:41:06 +0000 (10:41 -0300)]
qtdemux: properly activate segment with rate != 1.0

Also use the qt rate to identify the position within a qt segment
to properly translate playback time to qt media time

https://bugzilla.gnome.org/show_bug.cgi?id=765669

8 years agortpjitterbuffer: Fix stall when receiving already lost packet
Havard Graff [Tue, 3 May 2016 09:45:01 +0000 (11:45 +0200)]
rtpjitterbuffer: Fix stall when receiving already lost packet

When a packet arrives that has already been considered lost as part of a
large gap the "lost timer" for this will be cancelled. If the remaining
packets of this large gap never arrives, there will be missing entries
in the queue and the loop function will keep waiting for these packets
to arrive and never push another packet, effectively stalling the
pipeline.

The proposed fix conciders parts of a large gap definitely lost (since
they are calculated from latency) and ignores the late arrivals.

In practice the issue is rare since large gaps are scheduled immediately,
and for the stall to happen the late arrival needs to be processed
before this times out.

https://bugzilla.gnome.org/show_bug.cgi?id=765933

8 years agortpsession: Take session lock when creating stats
Miguel París Díaz [Thu, 5 May 2016 12:18:21 +0000 (14:18 +0200)]
rtpsession: Take session lock when creating stats

The access to the session hash table must happen while the session lock is
taken, otherwise another thread might modify the hash table while we're
creating the stats.

https://bugzilla.gnome.org/show_bug.cgi?id=766025

8 years agoqtdemux: update segment when new duration is found
Thiago Santos [Wed, 4 May 2016 00:17:01 +0000 (21:17 -0300)]
qtdemux: update segment when new duration is found

Otherwise the old segment will have a shorter stop time and would
cause the stream to end too early.

8 years agoqtdemux: dismember activate_segment into 2 parts
Thiago Santos [Wed, 4 May 2016 14:37:29 +0000 (11:37 -0300)]
qtdemux: dismember activate_segment into 2 parts

One that updates and push a new segment, the other will move the
stream to the new segment starting position

8 years agodv: Use correct pixel-aspect-ratio values
Sebastian Dröge [Wed, 4 May 2016 06:30:27 +0000 (09:30 +0300)]
dv: Use correct pixel-aspect-ratio values

The previous ones resulted in odd display aspect ratios and were different
from the ones used by e.g. ffmpeg. The new ones now result in display aspect
ratios of 4:3 and 16:9.

https://bugzilla.gnome.org/show_bug.cgi?id=765946

8 years agotests: add splitmuxsrc test for new "format-location" signal
George Kiagiadakis [Mon, 9 Nov 2015 16:55:09 +0000 (17:55 +0100)]
tests: add splitmuxsrc test for new "format-location" signal

https://bugzilla.gnome.org/show_bug.cgi?id=753625

8 years agosplitmuxsrc: add a format-location signal that allows bypassing the location property
George Kiagiadakis [Mon, 9 Nov 2015 16:51:12 +0000 (17:51 +0100)]
splitmuxsrc: add a format-location signal that allows bypassing the location property

This signal allows a user to directly return a sorted list of
files to be joined, so that they don't have to follow the
filename pattern that the "location" property expects.

https://bugzilla.gnome.org/show_bug.cgi?id=753625

8 years agosplitmuxsink: Fix deadlock case when source reaches EOS
Xavier Claessens [Wed, 4 May 2016 15:15:20 +0000 (11:15 -0400)]
splitmuxsink: Fix deadlock case when source reaches EOS

https://bugzilla.gnome.org/show_bug.cgi?id=765072

8 years agowavparse: simplify and correct header scanning
Stefan Sauer [Wed, 4 May 2016 05:59:27 +0000 (22:59 -0700)]
wavparse: simplify and correct header scanning

The wav spec tells that 'fmt' (and 'bext' if present) must come before 'data'.
There is no requirement for 'fmt' to be first. We already had a list of chunks
to skip, but it is easier to just skip any chunk while seeking for 'fmt'.

This fixes reading files generated by ProTools.

8 years agoosxaudio: Support audio device provider on osx
Hyunjun Ko [Sat, 30 Apr 2016 13:15:13 +0000 (22:15 +0900)]
osxaudio: Support audio device provider on osx

https://bugzilla.gnome.org/show_bug.cgi?id=753265

8 years agoavimux: set audio header rate according to calculated bps in stop_file
Mark Nauwelaerts [Sun, 1 May 2016 13:09:27 +0000 (15:09 +0200)]
avimux: set audio header rate according to calculated bps in stop_file

... now that set_fields is no longer called there by
e538608b3f90539003de21c1db238f3c9b946e30

8 years agoqtdemux: Store the segment sequence number in the EOS events and SEGMENT_DONE events...
Sebastian Dröge [Fri, 29 Apr 2016 12:04:11 +0000 (15:04 +0300)]
qtdemux: Store the segment sequence number in the EOS events and SEGMENT_DONE events/message

Also instead of storing it per stream, store it globally in the demuxer. It's
the same for each stream anyway.

https://bugzilla.gnome.org/show_bug.cgi?id=765806

8 years agoudpsrc: Always bind to ANY when address is a multicast address and not only on Windows
Sebastian Dröge [Mon, 11 Apr 2016 07:54:38 +0000 (10:54 +0300)]
udpsrc: Always bind to ANY when address is a multicast address and not only on Windows

For IPv6 addresses, binding to a multicast group does not work on Linux
either. Always bind to ANY and then later join the multicast group.

https://bugzilla.gnome.org/show_bug.cgi?id=764679

8 years agoximageutil: shouldn't implement transform if don't support it
Song Bing [Tue, 26 Apr 2016 09:01:49 +0000 (17:01 +0800)]
ximageutil: shouldn't implement transform if don't support it

shouldn't implement transform if don't support it. Or gst_buffer_copy_into()
will print ERROR log.

https://bugzilla.gnome.org/show_bug.cgi?id=765583

8 years agoqtmux: Allow MPEG-1 Layer 1 and 2 in addition to 3 in MP4
Sebastian Dröge [Thu, 28 Apr 2016 13:24:52 +0000 (16:24 +0300)]
qtmux: Allow MPEG-1 Layer 1 and 2 in addition to 3 in MP4

Via the MPEG-4 Part 3 spec we can support the other layers too.
Also correct the samples per frame calculation for MP3 if it's MPEG-2 or
MPEG-2.5.

https://bugzilla.gnome.org/show_bug.cgi?id=765725

8 years agortspsrc: Update caps for TCP whenever they change
Sebastian Dröge [Wed, 27 Apr 2016 17:46:34 +0000 (20:46 +0300)]
rtspsrc: Update caps for TCP whenever they change

We only changed them for UDP so far, which caused the wrong seqnum-base and
other information to be passed to rtpjitterbuffer/etc when seeking. This
usually wasn't that much of a problem as the code there is robust enough, but
every now and then it causes us to drop up to 32756 packets before we
continue doing anything meaningful.

https://bugzilla.gnome.org/show_bug.cgi?id=765689

8 years agortpjitterbuffer: Ensure to not take caps with the wrong pt for getting the clock...
Sebastian Dröge [Wed, 27 Apr 2016 17:33:38 +0000 (20:33 +0300)]
rtpjitterbuffer: Ensure to not take caps with the wrong pt for getting the clock-rate

Especially the caps on the pad might be out of date, and the new caps would be
provided for the current pt via the request-pt-map signal.

https://bugzilla.gnome.org/show_bug.cgi?id=765689

8 years agortspsrc: Don't propagate spurious state change returns from internal elements further
Sebastian Dröge [Wed, 27 Apr 2016 15:27:17 +0000 (18:27 +0300)]
rtspsrc: Don't propagate spurious state change returns from internal elements further

We handle them inside rtspsrc and override them in all other cases anyway, so
do the same for "internal" state changes like PAUSED->PAUSED and
PLAYING->PLAYING.

This keeps unexpected NO_PREROLL to confuse state changes in GstBin.

See also https://bugzilla.gnome.org/show_bug.cgi?id=760532

https://bugzilla.gnome.org/show_bug.cgi?id=765689

8 years agoavimux: Don't override maximum audio chunk size with the scale again just before...
Sebastian Dröge [Wed, 27 Apr 2016 11:09:03 +0000 (14:09 +0300)]
avimux: Don't override maximum audio chunk size with the scale again just before writing it

set_fields() should only be called in the beginning, otherwise we will never
remember the maximum audio chunk size and write a wrong block align... which
then causes wrong timestamps and other problems.

8 years agoavimux: Actually store the largest audio chunk size for the VBR case of MP2/MP3
Sebastian Dröge [Wed, 27 Apr 2016 10:53:00 +0000 (13:53 +0300)]
avimux: Actually store the largest audio chunk size for the VBR case of MP2/MP3

3ea338ce271e1f6a96d2ed49d4472b091f6f8b7e changed avimux to do that, but it
never actually kept track of the max audio chunk for MP3 and MP2. These are
knowing the hdr.scale only after parsing the frames instead of at setcaps
time.

8 years agomultiudpsink: Allow setting "socket-v6" without setting "socket" too
Mats Lindestam [Mon, 25 Apr 2016 13:03:14 +0000 (15:03 +0200)]
multiudpsink: Allow setting "socket-v6" without setting "socket" too

https://bugzilla.gnome.org/show_bug.cgi?id=764897

8 years agovpxenc: Properly handle frames with too low duration
Mario Sanchez Prada [Fri, 22 Apr 2016 14:02:16 +0000 (15:02 +0100)]
vpxenc: Properly handle frames with too low duration

When a frame's duration is too low, calling gst_util_uint64_scale()
to scale its value can result into it being truncated to zero, which
will cause the vpx encoder to return an VPX_CODEC_INVALID_PARAM error
when trying to encode.

To prevent this from happening, we simply ignore the duration when
encoding if it becomes zero after scaling, logging a warning message.

https://bugzilla.gnome.org/show_bug.cgi?id=765391

8 years agodeinterlace: fix description of linear interlacing method
Tim-Philipp Müller [Fri, 22 Apr 2016 14:48:08 +0000 (15:48 +0100)]
deinterlace: fix description of linear interlacing method

8 years agoflv: Handle the case where we do not get any CollectData in handle_buffer
Thibault Saunier [Thu, 21 Apr 2016 17:08:19 +0000 (14:08 -0300)]
flv: Handle the case where we do not get any CollectData in handle_buffer

https://bugzilla.gnome.org/show_bug.cgi?id=765320

8 years agoqtdemux: Do not use unreliable framerate
Seungha Yang [Mon, 11 Apr 2016 13:41:20 +0000 (22:41 +0900)]
qtdemux: Do not use unreliable framerate

timescale/1 is unreliable value for framerate. Due to downstream
element usually use framerate generated by qtdemux, let it be omitted
until the framerate can be reliably calculated.

https://bugzilla.gnome.org/show_bug.cgi?id=764733

8 years agoRevert "qtdemux: expose streams with first moof for fragmented format"
Sebastian Dröge [Thu, 21 Apr 2016 09:53:33 +0000 (12:53 +0300)]
Revert "qtdemux: expose streams with first moof for fragmented format"

This reverts commit d8bb6687ea251570c331038279a43d448167d6ad.

https://bugzilla.gnome.org/show_bug.cgi?id=764733

8 years agoqtdemux: support seeking of CENC encrypted streams
Alex Ashley [Tue, 9 Feb 2016 17:17:09 +0000 (17:17 +0000)]
qtdemux: support seeking of CENC encrypted streams

When playing a stream that has been protected by DASH CENC, playback
will fail if a seek is performed. Qtdemux produces the error "stream
is protected using cenc, but no cenc protection system information
has been found" and playback stops.

The problem is that gst_qtdemux_reset() gets called as part of the
FLUSH during a seek. This function frees the protection_system_ids
array. When gst_qtdemux_configure_protected_caps() is called after the
seek has completed, the protection_system_ids array is empty and
qtdemux is unable to create the correct output caps for the protected
stream.

This commit changes it to only free the protection_system_ids on
hard resets.

https://bugzilla.gnome.org/show_bug.cgi?id=761787

8 years agoudpsrc: add "retrieve-sender-address" property
Tim-Philipp Müller [Mon, 18 Apr 2016 13:33:10 +0000 (14:33 +0100)]
udpsrc: add "retrieve-sender-address" property

This allows disabling of sender address retrieval, which might
be useful in certain scenarios, like when the socket is connected,
or the sender address is not of interest (e.g. when receiving an
MPEG-TS stream). Disabling sender address retrieval in those
cases can have minor performance advantages.

https://bugzilla.gnome.org/show_bug.cgi?id=563323

8 years agov4l2: Change warning handling to break infinite message loop
Dimitrios Katsaros [Thu, 26 Nov 2015 12:15:06 +0000 (13:15 +0100)]
v4l2: Change warning handling to break infinite message loop

v4l2src can cause an "infinite message loop" when a base control exposed as a
property is not provided by the device. In these cases, if in the warning message
handling for the bus, the GST_DEBUG_BIN_TO_DOT_FILE* category of functions are used,
the src lookup causes a new warning to be posted on the bus, causing a loop.

This patch changes the warning for these controls so they are not posted on the bus.

https://bugzilla.gnome.org/show_bug.cgi?id=758703

8 years agospitmuxsink: Avoid creating small file at EOS
Xavier Claessens [Fri, 15 Apr 2016 14:44:02 +0000 (10:44 -0400)]
spitmuxsink: Avoid creating small file at EOS

When EOS is reached, the current file get closed and the last
GOP in the mq was written in a new file.

https://bugzilla.gnome.org/show_bug.cgi?id=765072

8 years agoscaletempo: S16 uses S32 temporary buffers, float/double their own type
Sebastian Dröge [Fri, 15 Apr 2016 16:59:15 +0000 (19:59 +0300)]
scaletempo: S16 uses S32 temporary buffers, float/double their own type

Make sure to allocate not only a S16 buffer for S16 but a twice as big one to
hold S32.

https://bugzilla.gnome.org/show_bug.cgi?id=765116

8 years agoRevert "pulsesink: uncork if needed upon commit"
Jan Schmidt [Fri, 15 Apr 2016 16:17:26 +0000 (02:17 +1000)]
Revert "pulsesink: uncork if needed upon commit"

This reverts commit 0dd46accf6d282ff07065852bd91c85c78af3394.

With some audiosinks, starting the ringbuffer on the first commit
causes audio glitches at startup by starting to output segments
from the ringbuffer before it has been filled / fully prerolled. This
doesn't usually happen with pulsesink because we map the pulseaudio
ringbuffer directly, but we should keep things consistent with
other sinks with regards to startup latency, plus it gives more
headway to avoid glitching, should the initial 2nd segment take
more than 10ms to generate.

https://bugzilla.gnome.org/show_bug.cgi?id=657076

8 years agortspsrc: add srtp rollover counters from mikey crypto sessions
Aleix Conchillo Flaqué [Fri, 15 Apr 2016 07:46:56 +0000 (00:46 -0700)]
rtspsrc: add srtp rollover counters from mikey crypto sessions

The server can send multiple crypto sessions, one for each SSRC with its
own rollover counter. We parse this information and pass it to the SRTP
decoder via the "request-key" signal.

https://bugzilla.gnome.org/show_bug.cgi?id=730540

8 years agortpjitterbuffer: Fix debug output when resyncing
Jan Schmidt [Fri, 15 Apr 2016 14:35:07 +0000 (14:35 +0000)]
rtpjitterbuffer: Fix debug output when resyncing

Don't output the pointer value of the time() function as a timestamp
by using the correct variable.

Fixes build on Raspberry Pi 3.

8 years agosouphttpclientsink: If no proxy is set by properties, use the default libsoup proxy...
Sebastian Dröge [Fri, 15 Apr 2016 08:36:36 +0000 (11:36 +0300)]
souphttpclientsink: If no proxy is set by properties, use the default libsoup proxy resolver

That is, use whatever system settings there might exist. This is the same
behaviour we use in the HTTP source.

8 years agoAutomatic update of common submodule
Julien Isorce [Thu, 14 Apr 2016 09:01:28 +0000 (10:01 +0100)]
Automatic update of common submodule

From 6f2d209 to ac2f647

8 years agosplitmuxsink: Add max_files_number property
Damian Ziobro [Wed, 13 Apr 2016 17:45:07 +0000 (18:45 +0100)]
splitmuxsink: Add max_files_number property

https://bugzilla.gnome.org/show_bug.cgi?id=744612

8 years agovideomixer: drop reference to videomixer 2
Reynaldo H. Verdejo Pinochet [Wed, 13 Apr 2016 17:57:03 +0000 (10:57 -0700)]
videomixer: drop reference to videomixer 2

Fix a small grammar mistake on "overlayed" while at it.

8 years agoximage: Initialize all fields in the meta explicitly
Sebastian Dröge [Wed, 13 Apr 2016 06:57:16 +0000 (09:57 +0300)]
ximage: Initialize all fields in the meta explicitly

The meta is not allocated with all fields initialized to zeroes.

https://bugzilla.gnome.org/show_bug.cgi?id=764902

8 years agortpmux: Forward sticky events on buffer lists too, not only on buffers
Paolo Pettinato [Tue, 12 Apr 2016 09:41:00 +0000 (09:41 +0000)]
rtpmux: Forward sticky events on buffer lists too, not only on buffers

https://bugzilla.gnome.org/show_bug.cgi?id=764933

8 years agodeinterlace: Drain the field history if the caps are changing
Sebastian Dröge [Tue, 12 Apr 2016 12:01:28 +0000 (15:01 +0300)]
deinterlace: Drain the field history if the caps are changing

Otherwise we will use fields from the old caps with everything set up for the
new caps, causing crashes and worse.

Also don't do anything if the same caps are set twice.

8 years agodeinterlace: Instead of confusing crashes later, just error out immediately if mappin...
Sebastian Dröge [Tue, 12 Apr 2016 12:00:31 +0000 (15:00 +0300)]
deinterlace: Instead of confusing crashes later, just error out immediately if mapping a video frame fails

This probably still crashes but at least we get some hint about what goes
wrong instead of random behaviour later.

8 years agoqtdemux: check stream is available in PIFF parser
Luis de Bethencourt [Tue, 12 Apr 2016 10:38:51 +0000 (11:38 +0100)]
qtdemux: check stream is available in PIFF parser

qtdemux->streams is an array, it will never evaluate to true when comparing
to NULL. Instead we want to check the number of streams to make sure the
stream is available.

https://bugzilla.gnome.org/show_bug.cgi?id=753614
CID 1358389

8 years agoRevert "qtdemux: redundant check in PIFF parser"
Luis de Bethencourt [Tue, 12 Apr 2016 10:37:36 +0000 (11:37 +0100)]
Revert "qtdemux: redundant check in PIFF parser"

This reverts commit 41e10524f3babdd92aac8c8c9d5b9cdf184c2d4e.

8 years agoqtdemux: redundant check in PIFF parser
Luis de Bethencourt [Tue, 12 Apr 2016 10:05:50 +0000 (11:05 +0100)]
qtdemux: redundant check in PIFF parser

qtdemux->streams is an array of size GST_QTDEMUX_MAX_STREAMS, it will never
evaluate to true when comparing to NULL.

https://bugzilla.gnome.org/show_bug.cgi?id=753614
CID 1358389

8 years agov4l2: avoid leaking GValues
Wim Taymans [Tue, 12 Apr 2016 09:56:08 +0000 (11:56 +0200)]
v4l2: avoid leaking GValues

unset the GValue if we don't use it any more to avoid leaks.

8 years agortpjitterbuffer: Fix rtp_jitter_buffer_get_ts_diff() fill level calculation
Sebastian Dröge [Tue, 12 Apr 2016 07:15:39 +0000 (10:15 +0300)]
rtpjitterbuffer: Fix rtp_jitter_buffer_get_ts_diff() fill level calculation

The head of the queue is the oldest packet (as in lowest seqnum), the tail is
the newest packet. To calculate the fill level, we should calculate tail-head
while considering wraparounds. Not the other way around.

Other code is already doing this in the correct order.

https://bugzilla.gnome.org/show_bug.cgi?id=764889

8 years agortpmanager: It's GST_LIBS, not GST_LIBS_LIBS
Sebastian Dröge [Mon, 11 Apr 2016 07:44:56 +0000 (10:44 +0300)]
rtpmanager: It's GST_LIBS, not GST_LIBS_LIBS

8 years agoqtdemux: Fix parsing segment duration of empty edit list box
Seungha Yang [Sun, 10 Apr 2016 23:33:17 +0000 (08:33 +0900)]
qtdemux: Fix parsing segment duration of empty edit list box

For empty edit list, segment-duration in edit list box should not be
used for segment event.

https://bugzilla.gnome.org/show_bug.cgi?id=764870

8 years agomatroskamux: make timecodescale configurable
Nicola Murino [Fri, 8 Apr 2016 11:05:57 +0000 (13:05 +0200)]
matroskamux: make timecodescale configurable

In some use cases the default timecodescale will produce blocks with the same timestamp

https://bugzilla.gnome.org/show_bug.cgi?id=764769

8 years agojiterbuffer: Move assertion to the right location
Edward Hervey [Thu, 7 Apr 2016 11:01:52 +0000 (13:01 +0200)]
jiterbuffer: Move assertion to the right location

We shouldn't have "late" lost timers at that point

8 years agojitterbuffer: Speed up lost timeout handling
Edward Hervey [Wed, 2 Mar 2016 13:25:24 +0000 (14:25 +0100)]
jitterbuffer: Speed up lost timeout handling

When downstream blocks, "lost" timers are created to notify the
outgoing thread that packets are lost.

The problem is that for high packet-rate streams, we might end up with
a big list of lost timeouts (had a use-case with ~1000...).

The problem isn't so much the amount of lost timeouts to handle, but
rather the way they were handled. All timers would first be iterated,
then the one selected would be handled ... to re-iterate the list again.

All of this is being done while the jbuf lock is taken, which in some use-cases
would return in holding that lock for 10s... blocking any buffers from
being accepted in input... which would then arrive late ... which would
create plenty of lost timers ... which would cause the same issue.

In order to avoid that situation, handle the lost timers immediately when
iterating the list of pending timers. This modifies the complexity from
a quadratic to a linear complexity.

https://bugzilla.gnome.org/show_bug.cgi?id=762988

8 years agojitterbuffer: Don't create lost events if we don't need them
Edward Hervey [Wed, 2 Mar 2016 13:23:01 +0000 (14:23 +0100)]
jitterbuffer: Don't create lost events if we don't need them

When "do-lost" is set to FALSE we don't use/send the lost events.
In that case, don't create them to start with :)

https://bugzilla.gnome.org/show_bug.cgi?id=762988

8 years agojitterbuffer: Add tracing of lock usage
Edward Hervey [Wed, 2 Mar 2016 12:57:07 +0000 (13:57 +0100)]
jitterbuffer: Add tracing of lock usage

Helps with debugging lock usage

https://bugzilla.gnome.org/show_bug.cgi?id=762988