- buildRelayPayload now uses config.domain for local users instead of
empty string, so the relay receiver can resolve the user
- Fixed canonical_pair_id backfill: SQLite NULL || ':' || x = NULL,
so use COALESCE(home_user_id, id) instead of concatenation
Existing 1-on-1 DM channels created before federation had canonical_pair_id = NULL,
causing the relay to create duplicate channels on receipt of a message. On server
startup, migrations now: (1) compute and set canonical_pair_id for all existing
1-on-1 channels using home_user_id (falling back to regular id for local users),
and (2) merge any duplicate channels that share the same canonical_pair_id by
moving all messages, read states, and members into the oldest channel and deleting
the stale duplicate. Both steps are non-fatal and idempotent.
Federation relay was incorrectly defaulting to disabled, requiring
admins to discover a buried settings toggle. Since federation itself
is opt-in (you connect instances manually), relay should be on by
default. Also migrates existing instances from 0 → 1.
Adds a post-deploy step that removes Docker images and build cache
older than 24 hours. Prevents the gradual disk bloat from repeated
docker compose --build invocations (was 71GB on the Pi).
Task 13: Hook server-to-server peering into connectToRemote (non-fatal)
and add federation API namespace to the client (initiate, peers, revoke).
Task 14: Show a transparency notice in the DM welcome header when the
other user is on a remote instance, informing that messages are stored
on both home instances and are not end-to-end encrypted.
Task 15: Add Federation section to the instance settings General panel
with DM relay toggle, TTL config, and a live peer list with revoke
buttons. Also extends InstanceAdminSettings type and the server settings
route to support federationRelayEnabled / federationRelayTtlDays.
Add POST /api/federation/sync for peer instances to catch up on missed
DM mutations after downtime. Uses HMAC-SHA256 authentication (same as
relay), determines shared channels between peers via raw SQL join on
dm_members/users, queries the federation_mutation_log with timestamp
pagination, and returns FederationRelayEvent payloads for each mutation
type (create, update, delete, reaction_add, reaction_remove).
Add federation routes module with four endpoints:
- POST /api/federation/peer/initiate — admin-only, starts peering handshake
with a remote instance (validates origin, generates HMAC secret, sends
server-to-server accept request with 10s timeout)
- POST /api/federation/peer/accept — server-to-server first contact,
rate-limited to 10 req/min per IP, creates/activates peer record
- GET /api/federation/peers — admin-only, lists all peers (hmacSecret excluded)
- DELETE /api/federation/peers/:id — admin-only, revokes peer and cleans outbox
Adds column migrations (dm_channels.canonical_pair_id, dm_messages.source_*,
attachments.source_url, instance_settings.federation_relay_*), CREATE TABLE
for federation_peers/outbox/file_queue/mutation_log, partial/filtered indexes,
and an idempotent backfill that seeds the mutation log from existing DM messages.
Also makes migrateDmMessagesReplyToFk and migrateAttachmentsDmMessageFk
column-aware so table recreation preserves any columns added by earlier
migrations (prevents dropping federation columns on upgrade).
The dist < 150 check caused the ResizeObserver and media load
listener to miss large layout shifts (e.g., GIF loading adding
400+ px of height). The isAtBottomRef gate already determines
whether to auto-scroll — the pixel threshold is redundant.
Remove isAtBottom state from scroll effect deps to break the
re-triggering feedback loop. Use isAtBottomRef (ref) instead.
Gate ResizeObserver and media load effects behind the ref so
they only auto-scroll after the initial snap completes.
Images now size naturally within max constraints instead of being
letterboxed into a 4/3 bg-surface-input container. Fixes dark bars
on GIFs and images whose aspect ratio doesn't match 4/3.
PNG conversion strips GIF animation. GIFs are now detected by URL
pattern (.gif extension or Tenor/Klipy CDN) and copied as URL text
instead, so pasting back into chat re-renders the animated GIF.
Add data-embed-thumbnail marker to VideoEmbed, GenericEmbed, and
RichEmbed thumbnail containers. Update detection logic to skip
these alongside avatars. Also remove redundant filename derivation
in Save Image menu item.
Detect right-clicked images in handleContextMenu and prepend
Save Image, Copy Image, and Open Original to the custom menu.
Avatars are excluded via the data-avatar semantic marker.
Implements image save/download (blob fetch + anchor trigger, fallback to window.open)
and clipboard copy (PNG write via ClipboardItem, fallback to URL text copy) with
toast feedback. Also adds ClipboardItem polyfill and Response.blob() content-type
fix to the jsdom test setup so the clipboard test suite runs correctly.
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.
Lower minimum query length from 2 to 1 character so single-letter
searches return results.
Fix dedup to prefer native profiles (homeUserId=null) over replicated
ones. Previously the first-seen result won, which was usually the
local replicated profile (no instance badge, namespaced username).
Now when a native profile is found on the remote instance, it replaces
the replicated copy — showing the clean username with the instance badge.
Replicated profiles on remote instances have different local IDs but
share the same homeUserId as the native profile. Use homeUserId ?? id
as the dedup key so the same person only appears once. Home instance
is queried first, so the native profile (with clean username) wins.
Replace old "Add Friend" form tests (placeholder/button/inline messages) with
new tests for the Find People panel: search input, Direct Add row, toast feedback,
and searchUsers debounce. Also fix ancillary mocks (discoverStore, authStore,
activityStore, ConnectInstanceModal, findExistingDmForUser) required by the
rewritten component.
Replace the two separate inputs (direct add + discover search) with a
single unified search field. When the user types, results come from
socialStore.searchUsers() across all connected instances. When empty,
the discover grid is shown as before. A user@instance pattern surfaces
a Direct Add action row for sending friend requests to federated users.
Parent FriendsPage no longer owns add-friend state — it all lives inside
AddFriendTab now, keeping the component self-contained.
When a WS connection receives its ready event, push cached myActivities
to that instance so federated users see activities immediately without
waiting for the next debounce cycle.
Switch pushActivities and setShowActivity from wsSend (home-only) to
wsSendAll (all instances). Add myActivities cache for re-push on late
connections.
Inserts a new tier 3 between DM member lookup and identity-only fallback.
When the current space's member list doesn't contain the voice participant
(user navigated to a different space/instance), the hook now falls back
to the cached User object from ParticipantInfo. This preserves avatar,
avatarColor, displayName, and homeUserId for federated users in PiP.
The isLocal branch was falling through to cachedUser.id when
homeUserId was null, diverging from the null-for-home-instance
contract used by the memberMatch and carry-forward branches.
Avatar.tsx already falls back to user.id internally.
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.
InstanceNotConnectedError and InstanceDisconnectedError now open the
ConnectInstanceModal instead of being swallowed silently. After the user
connects/reconnects, the friend request is retried automatically and a
success toast confirms both the connection and the sent request. All
other friend action errors (remove, cancel, accept, decline) now surface
as warning toasts instead of being silently discarded.
stopScreenShare() resets hwOverdrive: false, which clobbers the
user's intent when switching VP9 → H.264 HW mid-stream. Save
and restore the flag around the restart.
Add hwOverdrive: false to resetSession(), leaveVoice(), and
handleForceDisconnect() store resets. Add deactivateHwOverdrive()
to disconnect callback and cleanup effect in useLiveKit.
Ensures "resets when stream ends" contract holds for edge cases
like server shutdown, identity collision, and logout.
- buildScreenShareOptions reads hwOverdrive from voiceStore instead of config.codec; both VP9 and H.264 paths now always get a VP8 SIMULCAST backup
- startScreenShare activates SDP override before WebRTC negotiation and deactivates on failure
- stopScreenShare and handleScreenShareUnpublished both deactivate the override and reset hwOverdrive: false
- scheduleEncoderDetection checks encoderImplementation in WebRTC stats after 4s and shows a warning toast if OpenH264 (software) is detected
VP9 is now always the default codec, so the persisted codec field is
removed. hwOverdrive is a transient boolean (excluded from partialize)
that drives hardware H.264 on demand without surviving page reload.
Persist version bumped 12→13 with migration to strip stale codec field.
AcceleratedVideoEncoder only enables VA-API encoding on Linux.
Windows/macOS hardware H.264 encoding in Chromium's WebRTC is not
reliably controllable — VP9 via libvpx remains the recommended codec.
LiveKit uses screenShareEncoding (not videoEncoding) for screen share
tracks. Without it, both primary and backup fall back to the default
ScreenSharePresets.h1080fps15 (15fps). Primary was masked by our
overdrive mechanism, but backup had no overdrive and was stuck at 15fps.
- 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.
Calling e.preventDefault() on touchstart inside [data-context-menu]
elements prevented the browser from synthesizing click events
(touchstart → touchend → click chain). This broke tapping on spaces,
channels, DMs, and every other surface with data-context-menu.
Text selection prevention is already handled by CSS user-select: none
on [data-context-menu] elements — the JS preventDefault was redundant
and destructive. Reverted touchstart listener back to passive.
- Gaming mode defaults to H.264 (NVENC hardware encoding, zero CPU)
- Text mode defaults to VP9 (better compression, CPU is free)
- Changing mode auto-updates codec to smart default
- User can manually override codec via UI pills
- Changing codec mid-stream auto-restarts the stream (~1s interruption)
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.
- Gaming mode uses H.264 primary (hardware NVENC encoding, zero CPU
impact on games). Text mode uses VP9 primary with H.264 backup
and SIMULCAST policy for Safari compatibility.
- Fix native mode starting at 30fps by decoupling frameRate constraint
from resolution constraint in both screenShare.ts overdrive and
useLiveKit.ts updateActiveTracks.
- Filter paused backup codec tracks in Connection Info stats so dead
0kbps entries don't show alongside active codec tracks.
MobileVoiceFullScreen and MobileVoiceMiniBar were missing voice_leave
(and dm_call_end) WS messages in their disconnect handlers, leaving
the server showing stale voice state. Now matches VoiceControls pattern.
Jannis Braun