Mohamed Chibani authored 3 years ago and Sébastien Blin committed 3 years ago

Currently, when making an outgoing call to a Jami account with
multiple registered devices, a new ice media instance is created
for each subcall. This is not necessary, because all ice instances
will be similar until the peer answers (and provide its candidates)
But when the answer is received from the peer, all ice instances are
destroyed except the instance of the successful subcall, which will
be used to negotiate the media path.
Instead, only one instance will be created for the parent call, and
shared amongst the attached subcalls.

Gitlab: #619

Change-Id: I001c27f69d21b3ea640b189aea401e43c3f6bdef

Mohamed Chibani authored 3 years ago and Sébastien Blin committed 3 years ago

This patch removes the temporary ICE shared pointer

Gitlab: #619

Change-Id: Icacac9df1102327d4d1a0f0d67dfa457016048cd

Aline Gondim Santos authored 3 years ago and Adrien Béraud committed 3 years ago

Make sure we unload plugin before trying to load it.

Change-Id: I87ac7b36378c0e70ecd101b68c8fcb8b7ea7fde8

Amin Bandali authored 3 years ago and Sébastien Blin committed 3 years ago

Change-Id: I8b64659932709aa5642ef88be4bd6707812f7148

When requesting destruction of ice session and stream
transport instances from PJNATH, some operation will be
done asynchronously through scheduled timers. If not
properly handled, the timer heap will still contain
timer entries when the timer heap and the IO queueu are
destroyed.
Now, the timer heap is given more time to flush the
remaining timer entries before the IO queue and the
timer heap are destructed.
The timer heap may still contain timer entries if
PJNATH does not process all the timers within the
givent max time (currently set to 3s).

Gitlab: #637

Change-Id: I5ef2fe9d824e8b57191a51fac8f9e53e0e626fcd

Change-Id: Id5bd0b744e5b6775dd08b31b39a1ddea01c2be0c

Pixel 4 XL, 720p:
before: 6.69 ms (14,9 fps)
after: 5.92 ms (16,9 fps)

Change-Id: I8a2bdbed7388efbf95b066ace2f8032684a87a2e

Change-Id: Iada2486cffef5be26eac5618b8e62349ca3b01cf

Change-Id: I1c3d2943f3347fa9214456a2e180b0836bd6420f

Mohamed Chibani authored 3 years ago and Sébastien Blin committed 3 years ago

When an instance of JamiAccount class is destroyed, an attempt
is made  in the destructor to hangup all existing calls by
calling the hangupCalls() method. However, many methods used in
the process (such as terminateSipSession) uses a weak_ptr to
get the account instance, which is not accessible anymore.
The call to hangupCalls() is now called before the JamiAccount
destructor is called.

Gitlab: #627

Change-Id: I26e603a88c5074414d88e59d005d95a952c51d5e

Change-Id: I8c27143000c8e02d2245e0329ad87f5cf846a73f

using issuer->getId() is preferred, as the certificate can be
incomplete.

Change-Id: Ibf4129977d12b4867ce73f170e5c9fad04ad3952

Change-Id: I729e15f31fd1f7e9c0a2319966e50ed3ad86ec56

Change-Id: I7b7b67008bd1b3df9ec6d17c831acca753ba5662

Change-Id: I7bf01643731a1e9901b165dd0d54356b9febb8a4

Maxim Cournoyer authored 3 years ago and Sébastien Blin committed 3 years ago

* contrib/src/main.mak (GTKDOCIZE): Set and export variable to true.

Change-Id: I1901bdac034aa1e9b837f628f942f09b14160b52

This patch introduces many changes that prevent various crashes
and assertion failures.

In PJNATH source code:
 - Avoid duplication when adding host candidates.
 - Prtoect ICE context (lock) in TCP callbacks
 - Correctly set local and remote addres in STUN request

In jami source code:
 - Check that the address is valid before using PJSIP
 helpers

Gitlab: #617

Change-Id: I317d748db8d1f93dda6d5f510a2fae909a04704c

Also set agent logging to console by default.

Change-Id: I4993886a149eb4f60e1d61bad7ed18956350599f

Change-Id: Ica9005c67b0aff391166356aff0beb502b8e3f42

This avoid to use the certificate store when using an identified
channel.

Change-Id: I7b9c5c6d896ee9e14ff1f3f456a63fd2452d887b

Adrien Béraud authored 3 years ago and Sébastien Blin committed 3 years ago

Change-Id: I82413cac3557666de6c3407f01a9f0dc93cea7a1

This gives the possibility to know what channel are initiated
by the daemon via connectDevice()

Change-Id: I6dc42a4765cdd2987dc675ef455d119cbd31406d

shutdownConnections is only called when the account is disabled to
avoid to call it if the account's config is changed.
However, when stopping the daemon, doUnregister is called if the
account is enabled, and this should also call shutdownConnections()
to avoid any lock.

Change-Id: I35fba66b1b7f86360fa29cbeec78697f2b7c3328

Andreas Traczyk authored 3 years ago and Sébastien Blin committed 3 years ago

Change-Id: I9244f02a7c48da930cd34b1e20c6fe8d10c048b1

Change-Id: Ia220b318c6995a97157b5d3f122ae58d71d5ceb1

If somehow we have obtained a convInfo file with invalid conversations
(reproducible during a post link account sync), remove them when
loading the conversations.

Gitlab: #528
Change-Id: I45f759f470e58f567e70abafaca21dcce728183f

Change-Id: I750ee4f9b1f174d0f6aacab77293506c633a00f4

Change-Id: Ife1e1a111bbda8d8b10fc967502f713d7f09de58

Change-Id: I48cf424123ef137dc1c6fee0ab68c94ab6ac4c56

* contrib/src/main.mak (TARBALLS_DEFAULT): New variable.
(TARBALLS): Define in terms of the above.
(FLOCK, FLOCK_PREFIX): New variables.
(download): Use $(FLOCK_PREFIX), which may be empty when flock is not
available and a default TARBALLS location is used.

Change-Id: I95ca0ec0cdd893fc50f3924619cc87fe8cd6a5b7

Change-Id: Iedf63f2c4a1cf0c3a4fdc78a5ccb452576c522f6

Andreas Traczyk authored 3 years ago and Adrien Béraud committed 3 years ago

Replaces an atomic flag with a mutex preventing async calls to
switchInput from just failing or possibly causing a crash if a
resource already in progress.

jami-client-qt#522

Change-Id: I5f726306a2d54b1dcaa396a463d221a6efca13d8

This homogeneize build between autotools and CMake. jami_contact.h
is a source.

Change-Id: I5bf7a46c858dbfd6ccc4b8c0d053b3b834eebf13

This reverts commit 5e453368.

Reason for revert: Cause crash due to re-ordering the checklist while some checks are in progress

The resort of is causing inconstitency between the checks. The current uses
the list index as the ID for the STUN request. When the STUN response
(for this request) is received, the ID is used as is to retrieve the
check in the check list.
If the list is re-ordered (when receiving an new incoming check), before
the STUN response is received and processed, the ID of STUN request will
point to a different check on reception of the STUN response.

This a temporary revert until a more robust solution to the ordering
issue is found.

Change-Id: I912a9861118461ff7f6296f61c53531fa2b9384e

Change-Id: I5a30a4f5b7c7f2b74e5457973fdf42849fbc604c
GitLab: https://git.jami.net/savoirfairelinux/ring-project/-/issues/1282

SipTransportBroker::shutdown triggers pjsip_transport_shutdown
that can call transportStateChanged synchronously. In this case,
transportMapMutex_ is locked, causing a deadlock on a single thread.
We don't care about the transport's map at this point as we are
destroying the transportbroker, so we can just ignore the state
changed here.

Change-Id: I180ff1807fe87d1c48d72851f17e94ae2f892823

The handleEvent should finish. If after 3 secs nothing it's still
running, it should be stopped anyway. Add a log to detect this case.

Change-Id: Icc64f9450d5a8aecfcb6a5aa2e6a13882f4d0e8b

With this patch, contact left is removed in one to one. So, if
a contact left, they still receive message notification and they
will generate conversation requests and retrieve the old history.
Moreover, if they remove and re-add a contact, they will receives
a confirmation for the previous conversation and retrieve the old
history.
This avoid duplicates and read-only conversation.

GitLab: https://git.jami.net/savoirfairelinux/ring-project/-/issues/1282
Change-Id: Id820482e82e23334632e76bf4920255efb45f4cb

Olivier Dion authored 3 years ago and Sébastien Blin committed 3 years ago

Change-Id: I8d5eaf4c90cb425e8f96eaac25e727b6a6aae24f

Do not emit CallSignal::StateChange while locked. Because if the
client uses any method that lock callMutex_ it will be a deadlock.

Change-Id: I49aa7cb4ee60b23ef069751a7c1abbf4c9fcec95