The 100-500 kbit NVENC default squeezes every IDR of a high-resolution desktop
into a blur that sharpens over the GOP: at 3840x2160 hevc_nvenc CQ18 / 30M the
IDR frame lost 29-31 dB Y-PSNR against the frame before it, and the following
P-frames spent 15-18 Mbit/s re-sharpening a static desktop.
NVENC now takes one second of the bitrate cap within 500-8000 kbit, the rule
libx264 already uses (#8). UU_NVENC_BUFFER_MS still overrides it, so a small
value brings back the low-latency buffer. VAAPI is unchanged.
* feat(portal): follow the viewer's mouse mode with the embedded cursor
The portal stream hides the compositor cursor. A viewer that drives the pointer
with relative motion (UU's controlled-end mouse, also the Mac trackpad mode)
has no absolute position to draw its own pointer at, and the game's cursor
never reaches the picture, so the mouse can be moved and clicked but not seen.
Request cursor_mode=embedded while relative motion arrives, and keep it hidden
while absolute motion arrives: there the viewer draws its own pointer and an
embedded copy would trail it. The portal reads cursor_mode only when the
session is created, so switch by rebuilding the session with the saved consent,
the same path the privacy screen uses. A run has to last five events and 200 ms
before it counts, so a burst at connect or an alternating controller does not
rebuild; a failed rebuild backs off for five seconds.
The rebuilt session is checked against the monitor the previous one captured,
so a restored consent cannot silently move the share. The engine keeps the mode
for later opens (recovery, privacy, monitor switches) and skips the switch when
an older libuu-portal.so lacks uu_portal_set_cursor_mode.
UU_PORTAL_CURSOR_MODE=embedded sets the startup default.
* fix(portal): keep the cursor rebuild off locks and held input
Do not rebuild while the desktop is locked, and cancel an open that is already
waiting when it locks: GNOME closes the portal session on lock, so the request
would wait for consent that cannot be answered while holding `recovery`
against the post-unlock recovery.
Keys and buttons held through the portal input path are dropped when the
session is replaced, so wait for their release. Monitor selection takes the
same `recovery` lock, so it cannot install a stale source over the cursor
rebuild or the other way round, and the rebuilt session is revalidated against
the source it started from.
* fix(portal): carry the clipboard across cursor rebuilds
The engine keeps the viewer's text only in the session that provided it, and
the host re-sends it only for a new viewer update. A replacement session
therefore had no clipboard ownership, so changing the mouse mode made
previously synchronized text unpasteable until the viewer sent it again.
Offer the text to the replacement before the old session closes, but only
while that session still owns the selection: a local copy that has taken the
clipboard since is left alone. Touch consent upgrades replace the handle too,
so they take the same `recovery` lock as recovery, monitor selection and the
cursor rebuild and cannot install a handle opened under an older cursor mode.
* fix(portal): serialize clipboard writes with the cursor swap
The clipboard bridge calls clipboard_set from its own thread. Installing the
replacement handle and then carrying the old text over left a window where an
update could land on the new session and be overwritten by the older text,
with no pending update left to restore it.
Take a clipboard write lock before the handle is installed and hold it until
the carry-over finishes; clipboard_set takes the same lock. A write that
starts before the swap is included in the text carried over, a write that
starts after waits and wins.
* fix(portal): cancel a waiting rebuild when the desktop ends
The desktop ends its share when the last controller leaves (end_idle), but an
open that is still waiting for consent is not the installed handle, so it
survived that shutdown and installed a live share afterwards.
Bump an end generation in Session::end(), snapshot it before the open, and
cancel the request when it moves. The generation is checked again under the
handle lock before installing, so an end that races the swap either aborts
the rebuild or ends the handle it just installed.
* fix(portal): keep session swaps consistent with the viewer's state
Addresses the ownership race reported on the previous commit and the paths
that share its failure mode.
uu_portal_clipboard_set publishes the new text before SetSelection so a
transfer during the call serves it, restores the old text when the call fails,
and ignores a stale revision so a newer owner wins. The engine reports the
cursor mode it actually requested and the modes the backend advertised; the
host follows that instead of assuming the request was applied and stops
retrying where an embedded cursor is unavailable. The end generation is
sampled before the rebuild queues for `recovery`, a rebuild refuses to start
while the session is no longer alive, and cancellation now covers virtual
output opens.
Portal-bound keys and buttons are tracked separately, so an input held through
the portal defers the swap on every routing path, and the check is repeated
after the open. Every handle replacement (privacy screen, monitor switch,
touch consent, recovery after lock or stream loss) carries the viewer's
clipboard under the clipboard write lock, and the rebuilt session is checked
against the captured geometry.
* fix(portal): stop the swap bookkeeping from outliving a deferred rebuild
open_inner publishes the replacement's cursor mode before the swap. When held
input defers the rebuild, that session is never installed, so the host belief
described the discarded handle and later motion looked like it already matched.
Restore the belief from the installed session on every abort path.
The extra portal releases in release_held now stop at the first failure like
the sweep above; with a dead portal bus each remaining input waited out the
three-second timeout while the input and desktop-owner locks were held.
* fix(portal): invalidate clipboard ownership when a session ends
A closed session stops receiving owner changes, so a cached clipboard_owns
survived into the recovery paths and could republish old viewer text over a
newer local copy. Clear the claim in uu_portal_end and the Closed handler, and
let uu_portal_clipboard_copy refuse a source that is no longer alive.
Mark the embedded cursor unavailable when the engine did not apply it: with
AvailableCursorModes unreadable the session opens hidden, and retrying on every
motion run rebuilt it every 200 ms without changing the mode.
* fix(portal): retire held portal input when the devices are replaced
A key or button held through the native portal path is recorded when its
injection succeeds. If the session dies first, the key-up fails and the record
survives; recovery then installs new devices while the tracker still says an
input is held, so pointer-mode detection keeps deferring.
Keep the tracker in module state and clear it whenever the portal replaces or
ends its devices: the rebuild, both recovery paths and Session::end.
* fix(portal): scope held-input retirement to the retired handle
A press delivered to the replacement after the swap could be recorded while
the old handle was still being retired, and the unconditional reset then
deleted it. Hold the recording lock from before the swap until the reset
finishes, so only inputs sent to the retired devices are dropped.
A portal touchscreen contact was not tracked at all: the swap retired the
active contact while TouchDrag stayed down, so the following move reached a
session that never saw the press. Track contact down/up and include it in the
deferral state.
* fix(portal): serialize touch injection with handle replacement
inject_touch wrote the contact state outside the recording lock the handle
swap holds, so a contact delivered to the replacement between the swap and the
reset could be cleared, and the following phase-1 would fail on new devices
that never saw the press. Take the same lock across the injection and the
state update.
* fix(portal): decide the swap under the held-input lock
The final held-input check ran before install_replacement() took the serial
lock, so an input could be injected and recorded against the old handle while
the install waited for the clipboard lock. The install then retired the
records and swapped, destroying a contact the new devices never saw. Repeat
the check under the serial lock that recording uses, and defer the rebuild
when it hits.
* fix(portal): hold the input lock across native injection, refresh privacy geometry
inject_button and inject_key injected through the portal and only took the
recording lock afterwards, so a swap could take the lock, see no held input
and replace the devices that had just received the press. Take the serial lock
across the injection and its record, as the touch path does; the record
helpers no longer take it themselves.
A cursor rebuild that replaces the privacy virtual output invalidates the
cached pointer region: recompute it from the settled layout so absolute input
keeps landing inside the captured share.
The iOS viewer sends PrivateScreenSettingRequest{enable=0, privacy_type=3}
when the user turns the display off: its enable bit is the picture-mode
switch and the explicit type carries the intent. The host treated enable=0
as off, so the screen was restored instead of turned off.
Normalize that combination for iOS controllers only. Android 4.43.0 was
verified with the field-1 semantics, so keep Android and desktop viewers
unchanged. Record the platform after a desktop session is created, update
it on capture_desktop promotion, and cancel a pending UU_PRIVACY_AUTO even
when the switch is refused.
* perf(portal): wait for frames without polling
* fix(portal): use monotonic clock for condition variable wait deadline
* fix(portal): compute wait_for_first_frame deadline with monotonic clock
* annotation: answer viewer draw RPCs and draw strokes on an X11 overlay
The official viewer's watch-mode annotation (批注) sends RpcRequest 27
draw_req, which this host never answered, so the viewer reported
批注开启失败 (9998). The schema is taken from the Mac 4.42.0
UURemoteService SwiftProtobuf metadata. Replies go on TEXT because a
reply on CONTROL times out.
Strokes are drawn on a click-through override-redirect ARGB window,
which needs a compositing manager. The window exists only while
annotation is on and is removed when the desktop session ends.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* annotation: follow root resizes and scope resets to the owning peer
Addresses the Codex review on #174.
- A RandR resize (on X11 the viewer's resolution choice does this) left
the overlay at its old size. The overlay now watches root
ConfigureNotify and is rebuilt at the new geometry. Strokes keep the
viewer's points and are laid out again.
- A delayed reset from a replaced desktop peer could remove the
annotation of the viewer that took over. Requests and resets now carry
the peer (its PeerActions address), and only the owner's reset,
toggle-off or clear applies.
The ignored overlay smoke test covers both cases. It always checks the
owner takeover, and checks the resize when UU_ANNOTATION_SMOKE_RESIZE is
set on a disposable display.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(privacy): match recorded input devices after a reboot
The recovery record stores KWin's `eventN` sysNames, but udev renumbers
them between boots. A record replayed after a reboot (privacy was on at
shutdown) then set `enabled` on a device path that no longer exists; the
restore failed, kept the record, and every later switch failed at the
same step. The device that had been blocked could also stay disabled.
Records now keep name/product/vendor next to the sysName, and a restore
matches the current device by that identity, falling back to the exact
sysName for older string-only records. A device that no longer exists is
skipped with a log line instead of failing the restore, so the record,
the guard and the outputs are cleaned up; a device that exists but cannot
be enabled still fails and keeps the record for the next attempt.
Verified by replaying records with the release build on KDE Plasma 6.8: a
stale `event14` for a mouse now on `event12` is re-enabled and the record
removed; a record for a missing device logs the skip and restores
normally. `cargo test --features wayland,portal`: 330 passed, 10 ignored.
* fix(privacy): restore every device sharing an identity
- enable every current device whose identity matches a record entry,
deduplicated by sysName: two record entries can describe one device, and
one identity can cover several (identical keyboards, or a receiver
exposing several event nodes under the same name)
- a failed property read during restore is an error instead of a missing
device, so the record and the guard stay for the next attempt; blocking
keeps the lenient read
- the gone-device warning logs a count only, matching the log-privacy rule
* fix(privacy): re-enable readable devices before reporting incomplete enumeration
A failed property read during restore no longer aborts enumeration before
set_sys runs: the readable matching devices are re-enabled first, then the
incomplete read is returned as an error so the record and the guard stay
for another attempt. Blocking still skips unreadable devices.
* input: park the pointer worker until pointer or gesture work exists
A fixed 4 ms tick kept ~250 wakeups/s (clock_gettime storms, worker
steal spins, pow-heavy flush math) burning ~3% CPU on an idle host with
no viewer connected. The worker now parks on a Notify permit until a
pointer/scroll/button event or a live touch gesture needs the 4 ms
cadence; notify_one permit semantics make the park check race-free.
Sub-notch scroll residue cannot inject alone, so it parks too.
* fix(input): preserve pointer wake ordering and motion cadence
Hold the serial lock before notifying button transitions so a parked worker cannot miss a newly pending gesture. Keep the 4 ms coalescing cadence until an empty tick and reset the interval after each parked phase.
Add regression coverage for button wake ordering, 1 ms motion streams, and idle cadence reset using Tokio test-only clock control.
---------
Co-authored-by: ferstar <[email protected]>
* fix(privacy): keep the cover window out of the task manager
The cover is a plain Wayland toplevel and GTK's skip-taskbar hint is X11
only, so KWin listed it: the task manager showed an unidentified window
(class uuremote-server, caption "uu-remote-linux privacy") with the icon
theme's generic "wayland" icon, a yellow W in Papirus.
Mark the window through KWin scripting after the cover reports ready. The
same one-shot script mechanism already moves windows to the virtual output.
Failure only logs; the cover itself is unaffected.
Verified on KDE Plasma 6.8 Wayland: setting skipTaskbar on the running
cover removes the task manager entry immediately, and the entry returns
when a new cover window is created without the fix. `cargo test --features
wayland,portal`: 320 passed, 10 ignored.
* fix(privacy): match every cover window, not just the first caption
Picture mode creates one cover per physical monitor with the same title,
and KWin appends " <2>" and similar suffixes to duplicate captions, so the
exact match skipped every cover after the first on a multi-monitor host.
Match the title as a prefix instead.
* fix(privacy): report the cover ready only after its windows are mapped
The child wrote "ready" before gtk_main(), so the host marked skip-taskbar
while KWin may not have listed the cover yet; a cover that mapped later
kept the task manager entry. Pump the pending GTK events to map the windows
before the handshake, and run the mark script a second time to catch a
compositor-side delay.
* feat(privacy): auto-enable the privacy screen on desktop connect
`UU_PRIVACY_AUTO=picture` / `screen-off` (or 2 / 3) enables the privacy
screen when the first desktop controller connects. Off by default. The
switch waits for the portal capture session and retries while it is
missing; the viewer's own RPC 7 cancels a pending auto-enable for that
session.
The privacy state poller that pushes SystemMetrics and SystemStateChange
now starts for desktop peers too (it only ran for a tool peer promoted to
desktop), and it pushes once on its first tick. A mode the viewer did not
request, such as an auto-enable, therefore reaches its menu, and the
state message also goes to TEXT, which viewers read for privacy replies.
Verified on KDE Plasma Wayland with an iOS viewer: auto picture and auto
screen-off both engage, the menu reflects the mode after the push, and a
disconnect restores the physical outputs. `cargo test --features
wayland,portal`: 320 passed, 10 ignored.
* fix(privacy): let a viewer request win against auto-enable races
Two paths could still override an explicit choice:
- the override was checked before acquiring STATE, so an RPC 7 applied in
that window could disable privacy and the auto thread could turn it back
on; the check now happens inside the same lock that enters the mode
- `session_started` captured the override counter as its baseline, so a
request that arrived after CONTROL opened but before the session counter
started was treated as unrelated and auto-enable ran anyway; the counter
is now a flag set by any RPC 7 and cleared only when the last desktop
session ends
docs/privacy-screen.md now says remote-assistance (guest) sessions are
covered too, and that supported hosts push the current privacy state once
on every desktop connect.
The WebSocket transport already reduces errors to a category or HTTP status
without the token or endpoint, so log it instead of 'details omitted' to tell
a refused upgrade from a network failure (#166).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A CF_UNICODETEXT name made the Windows viewer register a custom
clipboard format instead of using id 13, so Linux text never reached
the standard Unicode format. Fixes#151.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* feat(audio): capture and encode Opus in process
* feat(portal): encode VAAPI from leased DMA-BUF frames
* fix(input): route Portal controls and committed ASCII through uinput
* docs(portal): shorten setup notes and colocate hardware acceptance guidance
* fix(input): preserve GNOME IME guard for committed uinput text
* build: link pulse/opus after the static C archive (--as-needed)
SDL viewer forwards one mapped controller (gpd_connect/gpd_update/gpd_disconnect) and receives two-motor gpd_vibration over the existing CONTROL channel. Default off (UU_VIEWER_NATIVE_GAMEPAD=1).
Replayed as touch, a phone drag scrolled in browsers but still selected
text in VTE terminals (Ptyxis, GNOME Terminal), as on a real touchscreen.
By default on the portal backend the host now decides per press: moving
past 12 px within 450 ms scrolls smoothly (NotifyPointerAxis or EIS
scroll, content follows the finger, `finish` on release so clients may
coast), holding still selects with button 1 down, and a short press
clicks. This works in terminals too and needs no touchscreen consent.
UU_TOUCH_INPUT=touch keeps the touchscreen replay (with its one-time
consent upgrade); UU_TOUCH_INPUT=0 keeps plain mouse drags.
* portal: mobile touch-mode drags become touch, consent asks once for it
In touch mode the Android viewer sends a finger as a left-button drag at
absolute positions (observed: press, mouse_move_absolute stream, release),
which on GNOME selects text instead of scrolling. SelectDevices now also
asks for the touchscreen type where AvailableDeviceTypes offers it, and a
drag from an Android/iOS controller (controller_platform 2/3) that follows
absolute motion is replayed as one finger through NotifyTouch* or EIS
touch, so the desktop scrolls and taps as on a touch screen. Relative
motion (the trackpad mode) and every other viewer keep mouse buttons.
A consent saved before this restores keyboard and pointer only. The first
mobile drag then asks the host once per run in the background (GNOME's
share dialog, also visible on the phone; the consent timeout closes it)
and swaps the new session in only when it grants the touchscreen for the
same monitor and size. The saved consent is never removed beforehand.
UU_TOUCH_INPUT=0 keeps mouse drags and never asks.
* input: type mobile IME text as portal keysyms on Latin GNOME layouts
Each mobile text_input was set as the clipboard and pasted with
Shift+Insert, which VTE terminals bind to the PRIMARY selection: text
typed from the phone into a terminal pasted whatever was last selected
(observed: a selected line of box-drawing characters for every letter).
When GNOME's active input source (first of mru-sources) is a Latin xkb
layout, printable ASCII, newline and tab now go through
NotifyKeyboardKeysym, where Mutter picks the key and Shift level of the
active layout. Other text, IBus engines, KDE and EIS sessions keep the
clipboard paste, which no longer replaces the clipboard for plain typing.
* portal test: accept libei before 1.3 rejecting a touch outside one of two regions
libei 1.2 (Ubuntu 24.04, CI) accepts a touch only inside every region of the
device; 1.3.0 checks any region. The harness device also carries a region for
another screen, so the older library rejects the touch with a client-bug
diagnostic. A real one-stream session has one region.
The host starts every session at the official defaults (Auto, 60 fps,
4:2:0), so a choice made for a slow link was lost on reconnect. Save them
in viewer.json with the other preferences and send the ones that differ
once CONTROL answers the echo.
* viewer: decode at the host's native size and filter window scaling
Every frame was scaled into a fixed 1920x1080 canvas and SDL then stretched it
to the window with nearest-neighbour sampling, so remote text looked uneven and
blurry. Decode at the native size (read from ffmpeg's verbose scaler line),
restart the decoder when the host resolution changes, and use linear render
scaling. UU_VIEWER_SIZE keeps the fixed canvas, now resampled with Lanczos.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* Add Chinese/English UI language
All GTK windows (viewer launcher, tool windows, host main window and QR login)
and viewer messages are available in Chinese and English. The language comes
from UU_LANG, the choice saved in Settings -> General, then the locale.
README: document the fork's changes and language selection.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* Host main window: add the language selector
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* viewer: control center, network status, loading screen, device previews
In-window chrome drawn on the SDL canvas with a fontconfig font (ab_glyph):
- loading animation with the connection stage instead of a black window
- control center (Ctrl+Alt+M or the top tab): quality, fps, true color,
resolution, fit/1:1, volume, auto lock and host mute (SimpleAction 19,
14/18/16), system shortcut grab, file transfer, watch mode, full screen,
disconnect
- network status panel (Ctrl+Alt+G): time, route, fps, bitrate, RTT, loss
- launcher device previews from the last remote picture (local 0600 cache)
- viewer sign-in keeps the local host user service running, so this
computer joins the account's device list; the settings switch controls it
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* viewer: compact cascading control center
Narrow category list with an options card beside the selected row, sized to
its content; opaque cards with a hairline border and soft shadow, so the
remote picture no longer shows through.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* viewer: keep the control center responsive; resync on a key frame
The menu was rasterized on every mouse event just to hit-test it; hit testing
now uses the layout alone and the drawn menu is cached until what it shows
changes. A stalled UI overflowed the access-unit queue, and the reset decoder
then smeared P-frames into blocks; after a reset it now waits for an IDR/IRAP
while requesting key frames. The tab icon is drawn, since fonts without U+2630
showed a box.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* viewer: measured RTT, no crash on closing tool windows, account card
webrtc-ice never measured the round trip, so the network panel always read
'RTT --'. Binding responses now record it on their pair, and the selected
pair is pinged when the sample is older than the keepalive interval, since
a streaming pair never idles (vendor/webrtc-ice/PATCHES.md).
On Wayland, libdecor's GTK plugin drew SDL's title bar with GTK3 from SDL's
thread while the file window ran gtk_main on its own; closing that window
crashed in GTK's style code. GTK builds now offer libdecor only the cairo
plugin unless LIBDECOR_PLUGIN_DIR is set.
Settings > General gets an Account card with the signed-in user and Sign out.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* README: keep the language section, without the fork notes
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* viewer: English server errors in the English UI; stable type_ipc tests
UU servers answer in Chinese only. The English UI now translates the known
messages (2002 target already controlled, assistance refusals) and shows the
code for unknown ones instead of Chinese text.
type_ipc tests shared one temp directory per process and read
XDG_RUNTIME_DIR while another test pointed it at a planted directory; each
call now gets its own directory and socket tests hold the environment lock.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* viewer launcher: switching the language restarts the launcher
The choice was saved but took effect only on the next start, so the window
seemed to ignore it. The launcher now re-executes itself in place; session
windows are separate processes and stay connected.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* viewer: show pictures as soon as decoded; coalesce mouse motion
The raw-stream parser closed an access unit only when the next began, and
frames were collected only when another one arrived, so a single remote
change (a typed letter) waited for the next one. An access unit delimiter
now follows each unit and the loop drains finished frames between units.
Motions within one 8 ms poll collapse to the newest position (click order
kept), so a drag no longer queues behind itself on a relayed path.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* viewer launcher: hosting is opt-in
Signing in to the viewer no longer installs or enables uuremote-native.service.
Only the Settings switch enables or disables it, and the unit's enabled state is
the saved choice, so turning it off survives a restart. A new sign-in restarts the
host only when the user turned it on and it is running.
PATCHES.md: the RTT keepalive also runs on the host; checked with the official
Android client controlling a Linux host.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* viewer: drop SDL event types sdl2 0.37 does not define
sdl2 0.37 transmutes every event type into SDL_EventType. sdl2-compat on SDL3
delivers types SDL 2 does not define (0x207), which is undefined behaviour and
aborts with debug checks (clipboard_requests_use_existing_window on Fedora 44).
Poll raw events and skip those types before the crate converts them.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* Keep the package version at 0.2.39
The maintainer sets the version at release time.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* portal: switch monitors on GNOME, precise touch, stop sharing when idle
GNOME grants one monitor per portal consent and refuses multi-select, so
each Mutter monitor gets its own saved consent (gnome@<connector>). The
monitors are advertised as screens; a switch opens that monitor's session
(silently once consented, GNOME asks on the host the first time) and swaps
it into the shared PortalSession, whose frame size follows the monitor.
Absolute motion and clicks go through the portal session: a uinput tablet
spans every monitor, and clicks on the separate uinput queue could land
before the compositor moved the pointer.
An idle watch ends the portal session (screen-share indicator off) once no
desktop controller is left for two checks; the next one reopens it with the
saved consent.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* portal: stop screen sharing as soon as the last controller leaves
The idle watch stays as the fallback for controllers that vanish without
a release event.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* portal test: accept the runtime-engine message in portal builds
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
The official iPhone viewer sends RpcRequest case 15 SendVideoTrackReq (no
fields) right after CONTROL opens. Without it a second session only joins the
stream already encoded for another viewer: no SPS/PPS/IDR, so it stays black.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Upstream webrtc-ice only gathers UDP relays. Forced-relay sessions also offer
turns:<ip>:<port>?transport=tcp, the only media path on hosts whose UDP cannot
reach the TURN servers but whose TCP goes through an HTTP proxy. Vendor
webrtc-ice 0.10.1 with a TCP/TLS dialer (optional HTTP CONNECT) that frames
STUN/ChannelData for the unchanged turn client. Certificates are chain-verified;
a name mismatch is accepted only for IP-literal TURN URLs. Enabled only with
UU_USE_PROXY=1; default UDP gathering is unchanged.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A session that joins a host already streaming to another viewer can receive a
clean stream with no IDR in flight, so the loss-driven PLI never fired. Ask
every 30 access units until something is presented, and warn after 300.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Route host and viewer WebSocket signaling through the shared reqwest client so UU_USE_PROXY=1 applies to signaling as well (HTTP/HTTPS proxies only, no direct fallback, no redirects, strict upgrade validation). ICE/WebRTC media is not proxied.
Co-Authored-By: Claude Opus 5.5 <[email protected]>