- 25 Feb, 2021 2 commits
-
-
Kateryna Kostiuk authored
Change-Id: I6cc29ad13e50cd5c58976bdc99e5deee68c3720c
-
Mohamed Chibani authored
Gitlab: #380 Change-Id: I3886c4c3b13710464f25ac779482e8e3726bda5c
-
- 24 Feb, 2021 3 commits
-
-
Kateryna Kostiuk authored
When frame decoding is fast, av_read_frame called before new data available. It will cause high CPU usage. This patch adds wait time when no data is available. Change-Id: I675092e2497f969976f797cc1bbb6b466b9dbcb5
-
Adrien Béraud authored
Change-Id: Idb36ee1a0ac04c577ae454a855280d9bd63f3245
-
Mohamed Chibani authored
Gitlab: #382 Change-Id: I9b6adba40262f2feb4bbc8445c98b52d0297a194
-
- 22 Feb, 2021 3 commits
-
-
Sébastien Blin authored
Change-Id: I6f92c33c8b0087b58e7d692bde4d3421906fb61a
-
Ming Rui Zhang authored
Gitlab: #346 Change-Id: Ia7b37abf31a14e8648f61a2415687d2400aa9937
-
Adrien Béraud authored
Change-Id: I2a59248df58ad5e019d3ded0b97e4db95c5520cb
-
- 19 Feb, 2021 3 commits
-
-
Ming Rui Zhang authored
daemon cmake ZERO_CHECK reference from ring-daemon vcxproj file will cause the build to fail for client qt sln on windows when using the newest Visual Studio Change-Id: I0488f65a636b28ab1ba9a2add4550d5ce9a52ca1
-
Sébastien Blin authored
4463ed9e introduced a regression and was checking for 70000+err instead of 120000+err. Connection refused is 120111 (PJ_ERRNO_START_SYS + 111) and not 70111 (PJ_ERRNO_START_STATUS + 111) leading to incorrect error detection and some negotiation failed. This patch also introduces some of the logs I used to debug this issue that can be useful if a similar scenario come back. Change-Id: I2a0e23f07f07ae81db14daea5d257976463c1133 GitLab: #438
-
Adrien Béraud authored
Change-Id: I6ccffb6ed28255922db2ca752503c42269b982e2
-
- 18 Feb, 2021 5 commits
-
-
Aline Gondim Santos authored
GitLab: #395 Change-Id: I8347ffe1a37ccf63d30e028e36e7597226d4ae66
-
Pierre Lespagnol authored
Gitlab: #236 Change-Id: I43cb644abc730a5ce9487bfd6cccee454eeab5d1
-
Maxim Cournoyer authored
Commit 8e2b334d normalized how the onnx sources are fetched; compared to earlier code, it strips the .git metadata. While usually a nice thing to do (as the .git metadata is not fixed and thus a checksum for the resulting tarball archive cannot be computed), in this case it breaks the build, which relies on it. The solution is to add a 4th argument to the 'download_git' procedure, that allows preserving the .git repository metadata. * contrib/src/main.mak (download_git): Add a new argument and document it. * contrib/src/onnx/rules.mak ($(TARBALLS)/onnxruntime-$(ONNX_VERSION).tar.xz): Use it. Reported-by:
Aline <aline.gondimsantos@savoirfairelinux.com> Change-Id: Ia5a5dad2a8dd2c292b8b8d8b8e5f63361a9484ef
-
Pierre Lespagnol authored
Otherwise the resolution is not updated and the remaining call keep the conference resolution. Gitlab: #389 Change-Id: Ib7a3880fceb1d1c0eafa3bac8aa0830c29c437d7
-
Aline Gondim Santos authored
- add jami preference to list non standard installation paths GitLab: #396 GitLab: #397 Change-Id: I65de9e4e9e345e866df9720918d080636541bbdc
-
- 17 Feb, 2021 2 commits
-
-
Sébastien Blin authored
the map should be protected to avoid erasing requests during the shutdown and in the callback Change-Id: I003b868239209c7ccd4f22274cbc2c949fefba2a GitLab: #440
-
Mohamed Chibani authored
Currently, a released mapping is not deleted but rather set as available for future use. However, in some circumstances, the socket bound to the mapped port may take some time to be completed closed and ready for reuse. Which means if the mapping is reused shortly after it was released, the socket binding may fail on the mapped port. To prevent this race, a relased mapping is always deleted. Note that in the current implementation, new mapping will be automatically requested if needed. Gitlab: #413 Change-Id: I9e6dce912aa14e29050669940b2846038b0a591c
-
- 16 Feb, 2021 1 commit
-
-
Amin Bandali authored
Change-Id: Ic30140e02be588ef36e6315235a33ebc11b6e32d
-
- 15 Feb, 2021 3 commits
-
-
Mohamed Chibani authored
This patch introduces the following modifcations: 1- The search for IGD and all requests will be performed only if the local address is valid. 2- A new search retry mechanism was introduced (similar to nat-pmp) to improve resilience against transient failures, in particular during a connectivity change. The search is perfomed a pre-defined number of times before it is definitely set as failed. 3- The device UID was used to identify the IGD. However, some devices may have more than one IGD with the same device ID, which caused some of IGDs to wrongly ignored. This is problematic if the first IGD with the same device ID that was added does not pass the validation or does not respond successfully to mapping requests. Now, we use both the deviceId and the URL as identifier for the discoverd IGDs. 4- Mapping address management was simplified. 5- ICE will only considere mappings that have valid local address. Gitlab: #420 Change-Id: I7eed337e9c1a190f6c42ebb1381957660f604410
-
Mohamed Chibani authored
Change-Id: Ia723407cc2cdb4793e0d8b11c0d625b655005d45 Gitlab: #420
-
Mohamed Chibani authored
A callback from PUPNP lib may be received during the shutdown of the daemon. To prevent this race, we lock the ctrl point and we ignore the callback if the client is registered. Change-Id: Ie6a3cb37fc189a271673e0935a5f03abce1ffc73 Gitlab: #335
-
- 14 Feb, 2021 3 commits
-
-
Maxim Cournoyer authored
Some systems do not ship a cc executable. Change-Id: Idb8ee8a5da0433bd96fb75dfcf1bd2d0aa6355e1
-
Maxim Cournoyer authored
This fixes the spurious "sh: call: command not found" error that would be printed on any 'make' invocation. The source of the error was pinpointed using strace. Change-Id: Iea866e23a5b98bba0d108cf5f7b1723052fe81a7
-
Maxim Cournoyer authored
* contrib/src/main.mak (git_download): Document procedure. * contrib/src/onnx/rules.mak (ONNX_URL): Use it. Change-Id: Ia1759e8d10c79b1109481895dc13166440709228
-
- 13 Feb, 2021 1 commit
-
-
Vladimir Stoiakin authored
Change-Id: Ib33a80885bebfbd5c9180de44c91f6a97c2e21b0
-
- 10 Feb, 2021 3 commits
-
-
Sébastien Blin authored
Before this patch, connectDevice() triggered a new ICE negotiation until no TLS socket exists. This means, if when starting a communication with a Device by triggering two connectDevice() like a SIP channel and a data-transfer, two requests will be send. Now, connectDevice will store new connectDevice() requests into pendingCbs to only do one request at a time. Note that the other side can ask a new request via the DHT, in this case, behavior will not change, we accept all requests if authorized and this must not interact with the pendingCbs_ Change-Id: I42a12458c364174e08531f568826da5b8348e08a
-
Sébastien Blin authored
Previous answerTo's logic was weird. hasPublicIp() should be unnecessary here and we should send remote candidates when gathered (before ice->start). Change-Id: I3dc9466a1979878e5557fc7dd05cee6049935523
-
Pierre Lespagnol authored
Change-Id: Ia8e96b9bf414b215a9fee01474754778a8c24295
-
- 09 Feb, 2021 1 commit
-
-
Adrien Béraud authored
Change-Id: I7b87bad47c66bda0d66140b59315ea45a43a4839
-
- 08 Feb, 2021 1 commit
-
-
Philippe Gorley authored
Change-Id: I42375710410f3ceca06bd546a66bc3258c60d5fb
-
- 05 Feb, 2021 5 commits
-
-
Pierre Lespagnol authored
The ice transport destruction can take a while, move this in the io pool to avoid blocking the main thread. This patch fix the incorrect orientation for android devices after conferences. Change-Id: I7ce467628aa265c2e8afc140d3bb4980beebd2ae
-
Mohamed Chibani authored
The external address of the IGD might not be always routable if the router is not directly connected to the internet. So we need to always add the published address returned by the DHT even if UPNP is enabled and actif. Failing to do so, and if the peer use only relay candidates, the TURN will reject all incoming connections from our public address because it were not in the candidates list and thus not authorized by TURN. The automatic reallocation of failing mappings was also improved. Change-Id: I5fb1424d5dc61b990595c229be647099b017eba1 Gitlab: #335
-
Sébastien Blin authored
Change-Id: I9d11246efa4c86534d2e8940510df3165e1119f6 GitLab: #426
-
Sébastien Blin authored
First, on_connect_complete dropped the connection errors but it must be handled for restarting failed check if necessary. Then, a problematic case was detected when the controlling agent got only a few active checks to do, but a passive will succeed. In the previous version the controlling was failing as soon as all active checks were checked. Now, if all active are failing the controlling agent will let some time for passive agents and then will decide to set the negotiation as failed. Change-Id: Ica728f55083a34496b031c859ed73b576e1eef26 GitLab: #411
-
Mohamed Chibani authored
Calls to DeletePortMapping method are made synchronous which may cause lags. This might be particulary critical if the IGD does not respond fast enough since the calls are made on the shared main thread. Change-Id: I57366a70ffeb294e6277c4e8d946b339a8277309 Gitlab: #420
-
- 04 Feb, 2021 1 commit
-
-
Sébastien Blin authored
answering and joining a participant can take quite some time, avoid to lock main thread for that. Change-Id: I4a2982ba8c35cdc7861ff5db01cffd0678777bea
-
- 03 Feb, 2021 2 commits
-
-
Sébastien Blin authored
Change-Id: I5d6f9d8fe3b44a5bbe2045cc3d48f11a81a2512c
-
Pierre Lespagnol authored
Reduce the bandwidth and CPU/GPU usage Change-Id: Icfc19b7d0232e01bf4d78a99649daeee3295fdbd
-
- 02 Feb, 2021 1 commit
-
-
Mohamed Chibani authored
In the previous implementation, a failed mapping was not removed if it's beeing used. The responsability was left to the owner to release it when done with it. This requires that all the owners implement the mapping callback and properly release the mapping. This might be somehow cumbersome and prune to error. Now, a failed mapping is automatically removed, and no further action is required from the owner. The mapping's owner will still receive a notification callback (if registered) when the state of the mapping changes to failed (or to any other state). Change-Id: I014ba3906b836efce3173fb22fbe298a4023db7e Gitlab: #335
-