This reverts commit 0b9dd0e6.

Change-Id: I3d42ca9673de395f8f6a36c7708e0654b638ec73

Change-Id: I5a7764f810d39ea3f691a821a9669d549700517c

Change-Id: I7d266d9d82d76715f8f88807e4944821a04aee20

This is a bit a dirty fix, but the ICE transport can use several
sockets or even an IPv4 socket on a side and an IPv6 on the other
side. But, if it's the case, the header sent contains an IPv6
address and sip_transport will return PJ_ENOTSUITABLE because
TLS6 != TLS4.

In this patch, we just ignore the V6 extension.

Change-Id: I3ded796dd6ee95a08ec54656dc2ad8107f0e62e3
Gitlab: #134

This mainly solve the "TLS decryption failed" error happening when
auto-answer is enabled. Because we only used only one buffer for
rx datas, if two sockets answered at the same time, the buffer was
mixing both answer into one packet.

Now, each active sock has a independant rx_buffer to separate
incoming packets.

Change-Id: I95320a10804b00a0f2558a1a26a1f219b377ddbf
Gitlab: #138

Add media-sdk contrib folder. Includes fetch_and_patch.bat and
SHA512SUMS files.

Modify ffmpeg contrib build to link with libmfx and enable qsv
encoders/decoders.

Change-Id: I50d4e479d4619922f98377d3588424331d12aec2

Change-Id: I23bc4b136f6c546d58ba5771d3a9336d90fff5a8

Avoid needless wakeups by waiting for 10 s instead of 100 ms.
Thread is always woken up when necessary anyway.

Change-Id: I75078a56c13a649aa81c224a2fa20defd4297b78

When setting the state directly, state change callbacks are not called.

State is now always set by the TLS handler thread, which will call state
callbacks.

Change-Id: Iffd03a2b20a3bf0527aa404914728ed1d889288b

Change-Id: I5d6c119143c6eec64724226b3a2d5ebe279e0406

Change-Id: Ic1381e5f00b386f9af801e138c86d0cee1af6abb

- now, audio only call can be allowed to enable the previewing in the setting
- in LRC, call start startpreview

Change-Id: I671841ee663fdf50790d772ce6298d1d5bd88900

Change-Id: I8a219f22bf6dc119f2c89c38b75b0093f0622d7b

Change-Id: I4e05d780481c340e288fcbb56898f30fb40bf90c

1 rtcp packet every 5 rtp packets instead of a percentage of byte sent

Change-Id: I6baf4672840ae80031b03e3bc995221aa92f0191

When using nvenc and the media sender gets restarted, the peer's srtp
decryption breaks and only gets set up correctly again after a pause and
play.

This issue was traced back to the RTP sequence numbers. Skipping a few
numbers before reinitializing the sender works.

Change-Id: Ic93f1368fc7eb01b3c57341f31d3832c9f2f2c1f

Also removes log message.

Change-Id: I5fc4dff440641f604f7f522d8319523715f4ef89

Change-Id: Ib925c5d60c1d4700e8ba2e06d2469daff2350072

reduce drop threshold and diminution rate

Change-Id: I651e3f730b638c1e94fca3882ef69f2f9e0e7732

220 kbit/s to 600 kbit/s

Change-Id: Ib6a1fb9c84f5da7a8fd2b83d113621a131d48f00

Don't choke the network connection with multiple keyframes in
succession. Only 1 keyframe should be necessary.

Change-Id: I6c1bcc4c3973fd6cb7eb58a9928e823bc0c5fcf7

avformat_find_stream_info takes a few seconds to compute the
real/average framerate if it is not in the metadata. We don't
need this, so we can safely skip it.

Change-Id: I134b6f478e4275c3588788c475cdfb630f3f76b5

Fixes recording.

Change-Id: I56042a37d86dcdc2a514a9319a71e12f4ee148ed

Change bitrate on the fly for supported codec, restart encoder only for other
(~10ms) instead of restart all pipeline

Change-Id: I1faad6e3cdc83f846d2279d9b8b997f1656f8596

If multiple transfers occurs at the same time, responses can come
with another order. So, we must check the id to get the correct
response.

Change-Id: Ica5d2f78fee6cfd385accc8cce6fff462dd4c4f8

Change-Id: Ie257ba783b6dbb1a7b2fd95771d73275a5c264bf

Change-Id: I37302afd2b8b7f18b92c213ba505f85120e72b02

Nvidia's hardware type is named cuda, but its hardware encoders are
suffixed with "_nvenc" instead of "_cuda".

Change-Id: I2e55297b4254570249421d60dd48259971281f89

Change-Id: I6da65c9b61ea0def790710cb55120c0dab632fda

Because the candidate is destroyed if turn_alloc fails, the
candidate should be committed before turn_alloc and turn_create.

Change-Id: I84f7a206ef42c2a21049d0bea135f5f60c92c0ad

Gitlab: #132
Change-Id: I138813de976c14321d59c9d423c1467b7f0fba5f

Hardware decoders output nv12, but Jami's software encoders want
yuv420p. If using a hardware decoder, make sure the input frames are
yuv420p. If using a hardware encoder, keep the frames as nv12.

Change-Id: I5b20238786104da4ddf25bee061899be18fa0d59

Change-Id: I7910b8a0b08910716fe0fba2a72f5caa776ed2f8
Reviewed-by: Philippe Gorley <philippe.gorley@savoirfairelinux.com>

Change-Id: I5ad8f7293fda713e91342a16de463750697c83e6

Change-Id: I85c88cfd1f2d5712fc271ad33da9f4b1b424bcfb

Change-Id: I73923252daa1c59a9278b8b90cbf6153f022296e

Change-Id: Ie05bd2b1ecdd89ca1134e9fc4dde91b5845435d3

Change-Id: I053ca22980267a8e2fd04e73e9a0c1e4a1064248

The UpnpDownloadXmlDoc is now called using shared_future. The api
function is a blocking call and there were instances where,
depending on the router and wifi speeds, the function might fail
or hang. By using std::shared_future, if such a thing arises, the
program won't block.

Change-Id: Ibca8d926e8980ebd4519f0e6bbd6ce2125396159

Nvdec shows a lot of artifacts. Disable until fixed.

Change-Id: I8df7e6072005c1bc0c19f820cec437c877c6e63a