Large video was being updated through scheduleLargeVideoUpdate even when
the large video container was hidden via CSS. This occurred in multiple
layout modes: tile view, stage filmstrip (with 2+ participants), and
etherpad editing. These updates caused expensive operations including
setting video streams, managing track listeners, updating avatars, and
running show/hide animations - all wasted CPU cycles since the container
wasn't visible.
The fix introduces a centralized shouldHideLargeVideo() function that checks
all cases where the large video container is hidden. This function is used in
selectParticipantInLargeVideo() to guard to not update the participant id.
A state listener monitors transitions from hidden to visible states and ensures
the large video participant id is properly updated when the container becomes
visible again and set to undefined when large video is hidden.
This improves performance by eliminating unnecessary video element manipulation
and handler execution across all layout modes where large video is not displayed.
In order to use gDM in Electron the flow is somewhat reversed. It starts
from the Electron main process, so we need an API in the external_api
that can trigger the builtin picker. The picker is still necessary.
The proposal never passed stage 1 and was last updated 4 years ago,
which signals it won't make it into the language: https://github.com/tc39/proposal-export-default-from
The alternative is just a couple of characters longer.
It was possible that join can be executed before conference.init have even started or we haven't reached the point ot create the initialGUMPromise. This was causing the following issues:
- users stuck on the prejoin screen
- participants join 2+ times in the call (we have been creating more than 1 local participants from a single page).
* feat(shared-video): Shows confirmation dialog before playing video.
* feat(shared-video/native): created ShareVideoConfirmDialog and unified actions
* squash: Simplifies state and fixes stop and then start scenario.
* squash: Use constants everywhere.
* squash: Use helper function.
* squash: Ignore any command with not matching video URL.
---------
Co-authored-by: Calin-Teodor <calin.chitu@8x8.com>
Many of the events are not used at all or used only on one place. For the rest of them the listeners were added 2 times on promoted visitors and not cleaned at all.
In the case where we switch from jvb to p2p when we need to switch the p2p and jvb track, they will be with the same source name. In order to add the streaming status listener we need to check if the isP2P flag is different. Without this check we won't have the correct stream status listener for the track. Normally the Thumbnail and ConnectionIndicator components will update the streaming status the same way and this may mask the problem. But if for some reason the update from the Thumbnail and ConnectionIndicator components don't happen this may lead to showing the avatar instead of the video because of the old track inactive streaming status.
Events such as "mouse-move", "mouse-leave" and "face-landmark-detected"
reach this code and pollute the logs. It's probably worth investigating
why this is the case and fixing it if necessary, but for now just remove
the log message.
* ref(transcriptions): refactor transcriptions api
* ref(transcriptions): refactor usage of translation label
Extend IFrame API to allow adding a transcriber in the room without the subtitles needing to be visible.
Allow transcription chunk messages to be passed through the IFrame API if a transcriber is present.
Clean-up transcription messages sent through the IFrame API to not include timeout field and possible conflicting states (stable / unstable /final)
* fix linting
* code review: extend api message to match webhook format
