requestFullscreen() on the voice container puts only its descendants in the
browser's top layer; overlays portaled to document.body were rendered outside
that layer and stayed invisible — most visibly the right-click context menu on
stream tiles and voice user panels.
Add usePortalContainer() hook returning document.fullscreenElement ?? document.body
and re-rendering on fullscreenchange. Migrate every overlay reachable during a
call: ContextMenuRenderer (desktop, submenu, mobile sheet), Tooltip,
ConfirmDialog, ConnectionInfoPopover, ScreenShareSettingsPopover, and
ScreenSharePicker (which previously rendered inline at App root).
Wire broadcastStreamWatch (LiveKit data-channel ping) to all three
explicit user-action sites in StreamTile: "Stop Watching" context-menu
item, "Watch Stream" context-menu item, and the handleWatch button
callback. Broadcast is intentionally confined to click sites only —
not wired in voiceStore actions — to prevent noisy stream_user_left
sounds on automatic teardown paths in useLiveKit.ts.
- Voice-bar and mobile camera buttons now use the canonical handler
(fixes mobile no-op and voice-bar wrong-preset bugs)
- Remove dead useLiveKit.toggleCamera
- Add _intentionalCameraOff flag with mark/consume helpers
The identity flip during disconnect: updateParticipants resolves
homeUserId → localSnowflake when activeDmCall is set, but reverts
to raw homeUserId when activeDmCall is cleared (before LiveKit
disconnect completes). SoundController sees the snowflake "leave"
and the homeUserId "join" — two phantom events for the same person.
Previous fix only checked homeUserId OR id. Now checks BOTH via
isSelf(id) which matches against a Set of {id, homeUserId}. This
recognizes the user as "self" regardless of which identity format
the participant currently has.
Root cause: SoundController compared LiveKit participant p.userId
(which is homeUserId from the home instance) against currentUser.id
(local snowflake on the current instance). For federated users these
are different IDs, so the controller thought the user's own presence
was a stranger — playing user_join/user_leave for self.
Fix: use homeUserId || id for the self-check. This matches the
LiveKit identity format used in federated calls.
Added justDisconnected guard to the participant sound loop. When
the user hangs up, isLiveKitConnected transitions to false — but
in a separate or same subscription tick, the participants list
also empties. Without the guard, SoundController plays user_leave
for every departed participant AND the disconnect sound simultaneously.
Now: if justDisconnected is true, the entire participant loop is
skipped. Only the disconnect sound plays.
Bug A: handleAccept relied on dm_call_accepted server response to set
activeDmCall. But connectFn's async AudioContext resume yields to the
event loop, dm_call_accepted arrives during the yield, finds
isLiveKitConnected=false (connectFn just reset it), and skips
setActiveDmCall. The acceptor connects to LiveKit but the UI never
shows the call. Fix: set activeDmCall and clear incomingCall
directly in the click handler.
Bug B: ready handler no longer sets activeDmCall for active calls.
On refresh/restart the client has no LiveKit connection — showing
"Connecting..." with no connection is broken. The call exists on
the server but this client session is disconnected.
Four fixes addressing the full state management problem:
1. Passive ready handler: no longer auto-connects to LiveKit on
page refresh. Prevents identity conflicts when the same user
has multiple sessions fighting for one LiveKit identity slot.
The user must re-accept to join; state is shown but not acted on.
2. SoundController sync guard: incomingCallLoading/outgoingCallLoading
refs prevent multiple playSound calls during async audio load.
If call is cancelled while sound loads, stops it immediately on
completion. Eliminates the "5 ringtones at once" bug.
3. Host dm_call_accepted broadcasts now include federatedCallId so
all clients (including remote instances) can match the event.
4. Removed all diagnostic console.log statements.
Enable federated DM calls to route accept/reject/end through the correct
WebSocket connection using callOrigin, and include federatedCallId in all
dm_call payloads for server-side FederatedCallEntry lookup.
Refactor useGridLayout from RefObject to callback-ref API so the
ResizeObserver reattaches when the grid container remounts after
focus-mode toggle. Fixes tiles rendering at stale fullscreen dimensions
on 4:3 monitors.
Also clamp focus-mode participant strip height (max-h-[20vh] min-h-[80px])
to prevent it from crowding the focused stream on short viewports.
Extends ParticipantInfo with cachedUser: User | null. In updateParticipants,
snapshots the previous participant list and carries forward the cached User
object when spaceStore member lookup fails (user navigated to a different
space). Local user falls back to authStore. homeUserId is now derived from
cachedUser when the direct member lookup misses. Updates EMPTY_PARTICIPANT
in PictureInPicture.tsx.
- Backup codec changed from H.264 (OpenH264, slow) to VP8 (libvpx,
fast) — same encoder family as VP9, lower dual-encode overhead
- Backup capped at 30fps with proportional bitrate to keep CPU
overhead low during SIMULCAST dual-encoding
- H.264 pill visually dimmed with amber warning when selected
- Connection Info now shows encoder impl (e.g. libvpx, OpenH264,
ExternalEncoder) on send-side video tracks alongside quality
limitation reason
- Amber warning "VP9 is recommended" when H.264 is selected
VP9 empirically outperforms H.264 in gaming (Chrome may not use
NVENC for screen share H.264). Codec pills now match other pill
styles — simple labels, no description text. Mode change no longer
auto-switches codec; user's codec choice is independent.
VP9 is default — better quality per bit with H.264 SIMULCAST backup
for Safari. H.264 option available for zero-CPU hardware encoding.
Codec choice is independent of gaming/text mode, which still controls
contentHint and degradationPreference.
Replace the useEffect-based auto-connect pattern in AppLayout with
direct connect/disconnect calls from user gesture contexts. This is
required for iOS (AudioContext.resume + getUserMedia must happen in
a gesture handler) and aligns with tightening autoplay policies on
desktop browsers.
Architecture:
- voiceStore gains connectFn/disconnectFn refs, registered by AppLayout
from the single useLiveKit() instance.
- All voice join paths (ChannelSidebar, MobileSpacesScreen, MainContent,
voice_moved WS handler) pass connectFn to joinVoiceChannel().
- All disconnect paths (VoiceControls, voiceActions, MobileVoiceFullScreen,
MobileVoiceMiniBar, dm_call_ended/rejected WS handlers, ready handler)
call disconnectFn() directly.
- dm_call_accepted WS handler calls connectFn() to initiate the DM call
LiveKit connection.
- The 55-line auto-connect useEffect and lastAttemptedRef are removed.
Use channelToSpaceMap.get(channelId) for permission checks instead of
currentSpaceId from the navigation store. On mobile, UI navigation
decouples from the active voice session.
Move mute/deafen/camera/screen-share/disconnect logic out of VoiceControlBar into a shared voiceActions.ts utility so the same handlers can be called from both UI buttons and the upcoming keybind dispatcher.
- Extract VolumeSliderItem to voiceMenuItems.tsx (was duplicated in
VoiceUser.tsx and VoiceChannel.tsx)
- Add e.stopPropagation() to all imperative context menu handlers
for consistency with useContextMenu hook behavior
After a server restart, navigating to a previously-visited channel showed
messages at a stale middle position instead of scrolling to the bottom.
The in-memory message cache survived the reconnect, so loadMessages()
bailed (cache hit) and the scroll-to-bottom logic never fired.
Now the ready handler clears the messages and hasMore maps for all
channels belonging to the reconnecting origin (including DMs for home).
The currently open channel is force-reloaded immediately; other channels
get fresh-fetched on next visit, triggering proper scroll-to-bottom.
Also fixes voice channel settings gear icon placement to match text
channels (flex-1 pushes icon to right edge).
The channel sidebar voice user list was maintained by a separate
voiceUsers Map (fed by WS events + fragile hydration code) that diverged
from reality after server restarts — users shown in wrong channels,
duplicated across channels. The VoiceGrid was always correct because it
reads LiveKit participants directly.
Now VoiceChannel.tsx derives its user list from LiveKit participants for
the connected channel (single source of truth) and only falls back to
server-provided voiceUsers for channels the user is not connected to.
Removed all hydration band-aids that tried to sync the two systems:
- useLiveKit ParticipantDisconnected → removeVoiceUser
- useLiveKit ConnectionStateChanged → addVoiceUser hydration loop
- useWebSocket ready handler → dynamic import LiveKit hydration
Also includes: voice channel settings gear icon on hover, persist
per-user volume/mute prefs across sessions, default screen share
audio off on Electron (no system audio capture support).
Electron requires setDisplayMediaRequestHandler for getDisplayMedia() to
work — removing it broke screen sharing entirely. Restored the handler
with useSystemPicker: true, which on macOS 15+ uses the native system
picker (honoring restrictOwnAudio), while Windows/Linux fall back to
the custom picker with the shareAudio toggle for echo control.
The custom screen share picker intercepted getDisplayMedia() and created
a raw loopback stream, bypassing Chromium's constraint pipeline entirely.
restrictOwnAudio was silently discarded. Removing the handler lets
Chromium 144's native getDisplayMedia run end-to-end with restrictOwnAudio
applied, eliminating the audio feedback loop in the desktop app.
Screen sharing with audio captured the app's own voice playback, causing
users to hear themselves echoed back. Fixed via two layers:
- Add restrictOwnAudio constraint (Chrome 141+/Chromium 144) to exclude
the app's own audio from system audio capture
- Add shareAudio toggle so users can disable system audio entirely
- Remove outdated macOS audio block (now supported via ScreenCaptureKit)
- Upgrade Electron 33→40 (Chromium 130→144) so restrictOwnAudio works
natively in the desktop app
- Add NSAudioCaptureUsageDescription for macOS 14.2+ audio capture
- Add GTK 3 fallback for Linux GNOME compatibility (Electron 36+)
- Custom screen share picker for Electron (ScreenSharePicker.tsx) with
Aether Drift design: glass-bubble footer, adaptive grid, pill tabs,
border-based selection (avoids overflow clipping), hover brightness
- Instance URL picker so Electron connects to any Backspace server
- System audio loopback on Windows/Linux via desktopCapturer callback
- macOS: video-only callback (OS blocks system audio capture)
- IPC bridge for screen source enumeration and selection
- Purge stale service worker caches on Electron launch
Replace hard-edged ring ripples with blurred radial gradient orbs, add
subtle glass refraction shimmer, and use gentler glow/breathing curves
for a calmer incoming call experience.
- SSRF protection: DNS resolution + private IP blocking on metadata fetcher
- Upload security: CSP/X-Frame-Options headers, SVG forced download, nosniff
- Auth hardening: JWT secret min length, password min 8 chars, token revocation via password_changed_at
- Attachment ownership verification before linking to messages
- Message length limit (4000 chars) enforced on client and server
- Asset URL validation on avatar/banner updates
- Federation instance validation (domain regex, origin scheme, length limits)
- DB indexes on all FK columns for query performance
- Migrations: nullable moderator columns, dm_messages reply_to FK constraint
- File cleanup on avatar/banner replacement and space deletion
- Fastify trustProxy, AbortController on fetches, typing map size cap
Track last speaker via ref so avatar fallback doesn't snap to array-order
first remote when everyone stops talking. Parse :stream suffix from
focusedParticipantId so clicking a stream tile in the grid pins that
screen share to PiP. Remember last focus via ref so PiP content persists
when grid focus is cleared.
Three changes to bring output volume closer to native apps:
- Insert masterBoost GainNode (+3dB) before the compressor/limiter
- Raise default system sound volume from 0.5 to 0.8
- Add configurable Sound Effects Volume slider (0–200%) in Voice settings
Voice channels already support video/screen share, so the separate video
type was redundant. Adds migration to convert existing video channels.
Also adds border-border-soft to CreateChannel input fields for visibility.
Jannis Braun