This commit adds a patch from libav resolving issues with h263p encoder.
see https://bugzilla.libav.org/show_bug.cgi?id=912

Change-Id: I95f5e81063af01a57be47b70fadaf54c3393eb52
Tuleap: #138

Corruption was seen on transfert. Hangup of a call was done while
the call was destroyed. Use of shared_ptr resolved it.

Change-Id: I11b2b91e07bc1da77f05302fa08926193923af60
Tuleap: #122

- TCHAR conversion to std::string wasn't working

Change-Id: I8fa9a634f50a58515a0ba49be99dd166a0a537b3
Tuleap: #133

Eloi Bail authored Nov 24, 2015 and Guillaume Roguez committed Nov 24, 2015

Attended transfers led to an assertion in pjsip. According to high-level
librairie pjsua, unused in Ring, and specially function pjsua_call_on_incoming,
it looks like we should not call pjsip_inv_answer when replace header is
present. This function actually leads to the pjsip assertion.
pjsip_inv_initial_answer should be rather used.

NOTIFY is still missing.

Change-Id: I2e4a09533bd33dd32b2cc61a7d9737dc1fe79738
Tuleap: #122

Current size was defined before the LibAV jitter buffer
was introduced. The buffer size defined was already very
conservative: 32000 samples means 2/3 of a second (an eternity).

* reduce buffer size to 16000 samples (333 ms at 48kHz)
* reduces audio latency of up to 333 ms
* should slightly improve echo cancelling

Tuleap: #120
Change-Id: I0c3d4f83d18f4fdede7e89cd9a36327af716c1ca

8.36 is not hosted anymore. Bump to latest 8.38.

Tuleap: #124
Change-Id: Ib40e60979d2ac1909cf8810a45a7dc9a0bf013cc

Eloi Bail authored Nov 19, 2015 and Guillaume Roguez committed Nov 19, 2015

Padding needs sometimes to be done for scaling purpose when a change of video
source is performed. Padding calcul is wrong and leads to memory bad writting
access.

Tuleap: #87
Change-Id: Ied282941077c1e268529457d39cdcc403a741aaf

Eloi Bail authored Apr 28, 2015 and Guillaume Roguez committed Nov 18, 2015

H263 version 1 allows a limited range of resolutions outdated compared
to recent camera capabilities. It was thus nearly impossible
to use it on Ring.

This patchset forces encoding of H263+ which supports all resolutions
multiple of 4 from 4x4 to 2048x1152.
On Libav we see video artifacts if multithread is enabled
on encoder side. A workaround deals with forcing h263 encoder
to monothread.

See https://bugzilla.libav.org/show_bug.cgi?id=912

Tuleap: #138
Change-Id: I7121e182fa13139d28db010e2ed0de573c616de7

IceTransport::isXXX() API are not thread-safe, not clear in which
state the transport is and not easy to manipulate (order of state).

This patch tries to solve that by:
- Procted API by a mutex
- Remove uneeded isComplete() API
- Ensure that each API are ordered (one at true implies true
  on all lower states: INITIALIZED -> STARTED -> RUNNING).
  All return false on Fail state.
- Change usage accordingly

Tuleap: #107
Change-Id: I17211e54322d70bbfe18c28f06cf9967b9ef93d2

Adrien Béraud authored Nov 11, 2015 and Guillaume Roguez committed Nov 16, 2015

Ant 1.9.5 is not hosted anymore, which prevents Android to compile.
Bump to 1.9.6.

Tuleap: #62
Change-Id: Ib20a0b6ea40d229ddfd09e7850b4606d9f28f508

Eloi Bail authored Oct 28, 2015 and Guillaume Roguez committed Nov 13, 2015

Dynamic quality change was done by increase/decreasing encoder bitrate.
It led to video artefact issues, specialy at low bitrates.
In addition, it does not make sence to use same bitrate with different
resolutions and fps. With high resolutions and fps, maximal bitrate
need to be higher.

It is possible for H264 and VP8 to set video quality using CRF.
This parameter solves video artefact issues and choose a different
output bitrate depending on input resolution and fps.

Tuleap: #80
Change-Id: I07d37175dfc335b5fc70cf54798f82eafe921ea9

Adrien Béraud authored Nov 04, 2015 and Guillaume Roguez committed Nov 12, 2015

This patch removes annoying build warnings on non casted unions.

Tuleap: #13
Change-Id: I9ec70dc559dea5973539b655b4d7a24ab22351ef

Make sure to not include OpenSSL when GnuTLS is used.

Tuleap: #77
Change-Id: I6f3ee4b80471d2562c6c5ef1ebb17cdc0002760a

Edric Milaret authored Nov 11, 2015 and Eloi Bail committed Nov 12, 2015

- dshow provide min/max value that don't always work. No pattern
have been observed as to why...
- ffmpeg chooses the highest value it can when we don't provide it.

Change-Id: I8d8b0993ded6d474f3ad9c741df7fd14fb61021b
Tuleap: #132

See http://trac.pjsip.org/repos/milestone/release-2.4.5 and similar for
detailed changelog.

* Bump from 2.4 to 2.4.5
* Mostly bug fixes
* Should fix #77
* Removes the need for patch "mainstream_fix_1844"

Tuleap: #77
Change-Id: Ib457203082fd7326d05bc8910a2aa2567974a4b8

Tuleap: #13
Change-Id: Ic948160e96a0f5e232871863407fa920292318ac

Let user choose the RING_REVISION and RING_DIRTY_REPO values.
This also permit packaging to build from non-git sources tree
with a fancy and valid revision string.

Just set previous environment variable before to run
the make command.

Change-Id: I24d71f985decc2187f09ab21eb72e588ede7fb7a
Tuleap: #71

Eloi Bail authored Nov 05, 2015 and Guillaume Roguez committed Nov 06, 2015

AVPacket are not freed after use. The solution deals with
allocating AVPacket on the stack and ensuring to free it
calling av_packet_unref.
This fix has also widly deployed on Video for having a unique
method other the code.

Change-Id: I049b29effb4e6748ee2dd12b4c52536c99fef498
Tuleap: #91

* use previous gettext version 0.18.3
* should fix build on Ubuntu Trusty 14.04
* keep gettext 0.19 on windows for mingw compilation

Tuleap: #62
Change-Id: I39f3d5b4938f7b2926f730a3b7822c19e05d9d1e

* bump from 3.4.2 to 3.4.6
* mostly fixes (including security)

Tuleap: #62
Change-Id: Ibe0e44e91f5cc7d9bb102aa9af9fed03c65f5754

This patch fixes a memory leak in SipsIceTransport class.
gnutls_priority_init() allocates memory, no gnutls_priority_deinit()
is called to free it.
This last has been added in SipsIceTransport destructor.
This patch also fixes code rules error on priority_cache_ naming
(this is the variable initialized by gnutls_priority_init).

Change-Id: I32a8a7347d4af96886a0def2d9a6d77fe0fad7c7
Tuleap: #92

Issue: #77219
Change-Id: I5bf71251212343e71f5e04081615a5890f7aa345

bump:
* cmake to 3.2.2
* libtool to 2.4.6
* automake to 1.15
* m4 to 1.4.17
* tar to 1.28
* xz to 5.2.2
* gettext to 0.19.6

Tuleap: #62
Change-Id: I7fa1f021ab1c8425d05416dc652aad060b2c2536

Adrien Béraud authored Oct 26, 2015 and Guillaume Roguez committed Oct 28, 2015

* add "active" runtime property to accounts
* accounts can only be used when active
* "active" state is not persistans
* add "active" state to volatile properties map
* add setAccountActive API
* "active" is the default state (backward compatibility)

The active state can be used to manage runtime states
defining if an account should be used (e.g. on Wifi/3G etc.)

Tuleap: #9
Tuleap: #42
Change-Id: Ie3b3aaadea6e43237fc22f6a1f96de8cbfe2bd95

Alexandre Lision authored Oct 22, 2015 and Guillaume Roguez committed Oct 27, 2015

missing override keyword to tag overridden virtual methods.

Tuleap: #69
Change-Id: I27d10cc49835e0e42858912cdfa11ef3e8761c04

* bump MsgPack to stable release 1.2.0
* bump OpenDHT to latest master supporting MsgPack 1.2.0
* adapt RingAccount to latest OpenDHT API

Issue: #81946
Change-Id: I331626120ca5e1c4c0d6f9a2dcf69e8d2934a636

Adrien Béraud authored Oct 15, 2015 and Guillaume Roguez committed Oct 23, 2015

Tuleap: #3
Change-Id: If43ac48729f4f55ba7fa02faa548b69d7f41dc78

Adrien Béraud authored Oct 13, 2015 and Guillaume Roguez committed Oct 22, 2015

Tuleap: #13
Change-Id: I8bc2d5e5b73bfb9461aeaea13ec1923567c64e7d

Calling this API causes a race condition with RtpSession::start().
The former needs the latter to be run first, this was not enforced,
causing a crash by nullptr access on socketPair_ member.
This patch checks for a non-nullptr socketPair_ to procceed the restart.

Issue: #81987
Change-Id: I20886fe70398f018621aade1c7bac3ff15ef9055

Call::getCallId() method is accessed in a nullptr pointer context.
This patch fix it by using the user given callId.

Issue: #82045
Change-Id: I8700aeaa173706909cf303be6d1efff5a4d1fc44

During a conference, video display blink on a full black image
at each video restart (due to adaptative bitrate algorithm).
This patch fixes this effect due to an unvoluntary mixer's sink
start/stop.

Issue: #82044
Change-Id: I181051de7230189d0a951f65f02b5673f202ffb5

Alexandre Lision authored Oct 19, 2015 and Guillaume Roguez committed Oct 19, 2015

YUV420P was allocating one large buffer on Linux and 3 on OSX.
This was causing read overflows. We use YUYV422 which is packed (always one buf)

Issue: #81116
Change-Id: I76456f72c6f155d181aa533aa62cb70b92f6eff1

This method doesn't shift correctly in case of YUYV422 output format.
This patch simplify the code and force the x-shift to be correct.

Issue: #81116
Change-Id: I049d61f51a944476f5a1c3743ecb905308becc83

This method wasn't able to treat some non-planar YUV pixel formats,
like YUYV422, giving a green background.
This patchset fixes that by giving a true black background when format
is non-planar YUV.

Note: this method continue to not working for non-YUV format,
as before, but it's not checked.

Issue: #81116
Change-Id: I55ec947d1474fc55dee861510623b8e91ef7f467

Some of internal stats of a frame could be re-used by ffmpeg,
but the frame doesn't have been reset for that.
This endings into potential multiple crashes.

This patch forces frame reset before re-use.

Issue: #81689
Change-Id: I9fa1c7e0d7a19d4abc8e44cb8978fa9379d51abf

Video scaler now follows sink lifecycle

Issue: #81116
Change-Id: Ibb053991b265131e1f63e1b2a169821e1d0f3bae

Adrien Béraud authored Oct 09, 2015 and Guillaume Roguez committed Oct 16, 2015

Prefixing ring IDs allows to recognize them easily
and to link them to Ring from external tools.

Tuleap: #14
Change-Id: I4ef64fbd2973dec5a8c1151da3ab3370d5330370

Eloi Bail authored Oct 08, 2015 and Guillaume Roguez committed Oct 15, 2015

If video_rtp_session is restarted due to bitrate changement
we need to setup again conferencepipeline. Otherwise videomixer sink
would be never attached and video never send.

Issue: #81547
Change-Id: I6de180afab5a7dd320fe3b31a1d61735106d76d1

Adrien Béraud authored Jul 24, 2015 and Guillaume Roguez committed Oct 14, 2015

Experimental

Issue: #81125
Change-Id: Id0e579fa0fd70ef152c7e4074ed713730b542923

This patchset prevents client from reading a frame that is being written
by the daemon using a push/pull mechanism.

**WARNING** This breaks the direct rendering API

Issue: #81542
Change-Id: I55d4ee9b986ac682903b326fd86c32d8f5f68825
Signed-off-by: Guillaume Roguez <guillaume.roguez@savoirfairelinux.com>