Neither had a test before the move, so the moved lines showed up as
uncovered. The client initialisation is tested through a fake client
and mocked stores: which store it syncs into with and without
IndexedDB, with and without the worker, the fallback to memory when
the IndexedDB store cannot start, clearing the stores only when not
restoring, carrying on without crypto, and counting as ready only
after a sync from the network rather than the one restored from the
store. The session restore is tested for no stored session, a
restored one, one the homeserver no longer knows, and any other
failure.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The package carried the standalone app's IndexedDB store worker: 11 MB
with a copy of the Matrix SDK and its crypto wasm, referenced by an
absolute path no host could serve, and never used by the component,
which takes the host's client. The bundler emits the worker file as
soon as it loads the `?worker` import, used or not, and the import sat
in the Matrix utilities the component reaches for the room URLs.
The client initialisation moves into a module of its own, with the
worker import, which only the standalone login, registration and
session restore paths import. ClientProvider takes the session loader
through a `loadClient` prop instead of importing it itself, since even
a dynamic import keeps a module in the package: the standalone app
passes a lazy `initSPA` from App.tsx, the component passes nothing,
and given neither a client nor a loader no one is signed in.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The runtime path was `./` plus the asset's path from the package root,
which is right for every chunk the build emits today, all of which sit
at that root, and wrong for one emitted under a directory. Vite names
the referencing chunk, so the path is now computed relative to it.
The sound loader kept a branch for sounds inlined as data URLs, with a
comment saying the component build produces them. No build does any
more, and no host would inline a sound of this size, so the branch and
its tests go.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
- The effects state's timeline through preparing, settling and attached,
and that a failed trial build withholds the effects for the session while
keeping the choice.
- The footer offers nothing once the pipeline fails to build, and passes
on the wait as the state's single wait field.
- The call leaves the camera out of the join while the pipeline is
prepared, publishes it once that is known, and turns on a camera turned
off before the processor was chosen; the lobby opens no camera meanwhile.
Both view models are driven by one state subject.
- The transformer's trial build answers whether a segmenter can be made.
- End to end, a browser whose pipeline cannot be built sees the effects
withheld.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
- A segmenter that can't be built, with the model unreachable or a GPU
its delegate refuses, took the camera down with the pipeline: the SDK
fails the track it was attaching to, and the lobby muted camera and
microphone, on every load while the choice was kept.
- One is now built on trial when an effect is first wanted, before the
pipeline is attached to anything, on the CPU where the GPU refuses.
Where it can't be built, the options are disabled with the existing
notice for the rest of the session. The choice is kept: a browser
update may fix what failed.
- The view model takes that verdict from the pipeline's state rather
than
asking the SDK, so both say the same, and it no longer builds a WebGL
context of its own to ask.
- Until the trial is over, the lobby opens no camera and the call makes
no
camera track, so neither starts without the effect.
Rebased onto main, where the Publisher is gone and the call view model
publishes the camera through the SDK:
- CallViewModel takes videoProcessorPreparing$. While it is true, the
camera's mute handler waits before making a track, and the join leaves
the camera out of its requests (initialPublishRequests); the view
model
publishes it once it is known whether the pipeline builds. The
Publisher's test became CallViewModel tests.
Cherry-picked from spec/background-effects/12-unbuildable-pipeline onto
toger5/background-effect/camera-preview, where the camera menu's preview
takes the camera from the footer's caller rather than from a track
synced
on the background-effects state. The footer factories take the
pipeline's
state again, as this commit needs its verdict and not only the settling
flag; the synced camera and the preview's wrapper type are not brought
along.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The component module was 18 MB, of which 15 MB was MediaPipe's runtime
as a base64 data URL, with the segmentation model, every sound and the
background pictures inlined the same way: a library build inlines every
asset unless its import says `no-inline`. Every host downloaded all of
it before the first call, whether or not a background effect was ever
switched on, and a host with a Content Security Policy without
`script-src data:` could not run effects at all, since MediaPipe
injects its loader as a script tag.
The imports now say `no-inline`, and the build renders each reference
as `new URL("./assets/…", import.meta.url)`, which webpack 5, Vite and
Rollup follow as an asset of the host's own; the `./` is spelled out
because a bare `assets/…` is read as a package name. The module is
2.3 MB. The runtime is fetched when an effect is first switched on and
cached on its own. MediaPipe's files come through the package's
`exports` rather than a path into node_modules.
MediaPipe's JavaScript (160 kB) stays in the module: LiveKit's track
processors import it statically, so a dynamic import moves nothing.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
- A cross on the tile's top-right corner removes an added background,
critical-coloured at rest and named by its tooltip, as design draws it.
The one in force offers none, since taking it away would leave nothing
chosen; the view model refuses it too.
- The cross sits beside the item rather than in it, which would take the
press first and choose the tile, and isn't a control of its own, so the
menu keeps one keyboard model: Delete or Backspace on the tile does the
same, and the tile says so with aria-keyshortcuts.
- Removed from IndexedDB and its picture released. End to end on
Chromium and Firefox, a background removed by Delete is still gone
after a reload.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- Five of one's own at most, so the grid is whole: at the limit the add
tile goes, and it comes back once one is removed. The store holds the
limit too.
- An added image is refused if it isn't an image, is animated, or can't
be decoded; reduced to 1920px on its longest edge; and laid on an opaque
ground, since a transparent pixel would be a hole in the background.
- Animation is read from the file, as a still and an animated WebP share
a type. Chromium has no selected track until the decoder's tracks are
ready; without waiting for them an animated GIF got through.
- A refusal says why, in place of the notice, and is scrolled into view.
It can be dismissed, is gone once the menu closes, and the next shows
even if it reads the same. What was in force stays.
- End to end on Chromium and Firefox, from what the device then keeps: an
animated GIF is refused with its message, a 4000x3000 image is kept at
1920x1440, and a half transparent one is kept opaque. All three fail
with the preparation bypassed.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- An add tile after the pictures opens the file picker, and the image
chosen joins the backgrounds on offer without being put on; it is in
force once chosen, like a shipped one. The tile is a command rather
than a choice, so a menu item beside the radios, and it isn't offered
where effects can't run.
- Kept in IndexedDB on this device and never sent anywhere, behind a
small interface, so the unit tests use an in-memory store; the
IndexedDB store itself is tested against fake-indexeddb, a new dev
dependency.
- The picker takes the focus, which the menu reads as a click elsewhere,
so the menu holds itself open until a file is chosen or the picker is
cancelled.
- A chosen image the device no longer keeps, as after its storage is
cleared, falls back to no effect once the device has said what it
keeps.
- Switching is checked between any two kinds of effect. End to end on
Chromium and Firefox: an added image shows in the preview once chosen;
once it is in force, no request carries the file or the copy kept,
which a deliberate upload fails; chosen, it is in force again after a
reload, gone once storage is cleared, and can be added again.
- Kept as given here; what can be added is checked in the commit after.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- A remembered effect this browser can't run is cleared rather than kept,
as the menu already showed no effect in force there.
- A camera turned off and on again with an effect chosen gets the same
pipeline as its track arrives, neither rebuilt nor switched again.
- End to end, on Chromium and Firefox: the effect chosen before a call is
still chosen, and worn, in the next call after leaving; and a second
tab still shuts the first down with the existing "Opened in another
tab" error while an effect runs.
- Switching camera and sharing a screen are checked by hand: they need
two cameras and a screen, which the fake devices don't give yet.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- The desktop rule from #4059 goes: now that the menu says where effects
run slowly, using them there is the user's choice. A phone on the fast
route held 99% of its frame rate and was refused; a desktop on the slow
one held 73% and was allowed.
- Asks @livekit/track-processors directly; the wrapper module goes.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- The first effect chosen in a session shows a spinner on its own tile,
in place of the tick, until a frame carrying it is drawn. Later changes
show nothing, having nothing to prepare.
- Ended by the frame, not the promises, which resolve while the segmenter
is still being built. The transformer reports the first frame that went
through the segmenter; one passed through untouched doesn't count.
- The pipeline publishes the wait with its state, and both footer view
models take it as an input.
- The lobby tests' provider mock gains the observable hook, built once: a
fresh observable per render rebuilt the footer on every render.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- Where the browser has no MediaStreamTrackProcessor, effects draw each
frame through a canvas: slower, and the first build stalls the page.
They stay offered and choosable there, with a notice under the grid
that says so and describes the group.
- Decided in the view model, beside whether effects can run at all, so
the lobby and the call give the same answer.
- The notice stays within the menu's width however long the sentence,
which the story checks.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- The camera menu's blur toggle becomes a Background effects section
under the cameras: no effect, blur and the two shipped backgrounds as
one radio choice of tiles, three across. It scrolls with the device
list, so it stays reachable in a short call, and choosing keeps the
menu open. The lobby gets it too, as both render the same footer.
- The footer's view model gives the options, the notice and the actions;
the footer only names them.
- Each tile is Compound's MenuItem as a menuitemradio, styled as a tile.
Compound makes it an item of the dropdown, so the arrow keys walk the
tiles with the device rows, and takes its tap in the phone drawer. A
tile that can't be chosen has no select handler, so the keyboard passes
over it.
- Where the browser can't run effects the tiles are shown, disabled and
explained by a notice the group is described by. No effect stays
choosable and shows as in force. The notice is in the scrolling list,
so a short call doesn't push the menu out of it.
- Section headings are raised, or the positioned tiles scroll over the
one they sit under; raised within the list only, so the menu's frame
stays over them. Tiles clip with overflow: clip; with hidden they are
scroll containers, which focus scrolls into view without their border.
- End to end on Chromium and Firefox: a shipped background chosen in the
lobby shows in the preview, and None after Blur draws nothing, counted
in WebGL draws (Blur: about 1400 a second on Chromium, 2300 on
Firefox).
- The Quick Audio Menu's test of the camera menu keeps its name and now
finds blur among the effects rather than as a toggle.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- A catalogue of what can be chosen: no effect, blur, and two shipped
backgrounds, the app's own gradients flattened onto its canvas colour.
Their ids are what a saved choice stores, so they are named for the
pictures.
- One setting for the choice, typed as the ids an effect can have and
defaulting to blur for anyone who had it on. A stored value we no
longer ship, or one that isn't even a string, reads as no effect: it
comes from storage another build or a hand may have written.
- The pipeline follows it, switching in place to a picture as readily as
to blur, and the camera menu's blur toggle reads and writes it.
- So does Settings' blur checkbox: a picture reads as blur off, and
checking it replaces the picture with blur. The old on/off setting is
only read, to carry blur over. First tests for the Settings modal,
which fail with the checkbox on the old setting.
- The transformer's own init loads a picture chosen before it was built,
as the base init does and it had not: the first effect always is, and
the picture was drawn black.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
@livekit/track-processors skips a frame on
`isDisabled ?? backgroundDisabled`, and `isDisabled` is never
undefined, so a disabled effect that keeps its blur radius is still
segmented and drawn on every frame. Still the case in 0.8.2. The
override stays until that is fixed upstream.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The SDK now builds a processor chosen while the camera was off when the
camera turns on, so the call view model no longer unpublishes the
camera and publishes a new one. TrackSupply.replace, the record of
which camera took which processor, and followProcessor's applied
callback go with it.
Answers robintown's review on #4282 ("what's wrong with attaching
processors to already published camera tracks?").
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The controls and the pipeline each asked the SDK on their own whether
background processing is supported, so one could offer what the other
refused. One function answers both, and it asks the browser, kept to
desktops until the menu says what the effects cost elsewhere.
Ported from spec/background-effects/2-pipeline (#4282): f78e9c05 and
87638325, the tests on the lobby and call footers as they are.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
A camera track made without the processor lets the room see the camera
until the processor is built. So the processor travels in the publish
request: the lobby opens the camera with it, the join publishes with it,
the room's capture defaults follow the requests made since the join, and
a camera turned off before the processor was chosen is replaced on turning
it on rather than unmuted, so that the new track starts with it.
Ported from spec/background-effects/2-pipeline (#4282): the publish half
of a91dab8f. The replacement lives in the mute switch's track supply,
which the call view model shares with the lobby.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Blur runs through one pipeline, shared by the pre-join preview and the
call: attached the first time an effect is wanted and switched in place
after, never detached. The logic is BackgroundEffects in src/state;
BackgroundEffectsContext only builds it, from the app's view model, so
the standalone app and the component hand the same instance to every
camera. Disabled drops the blur radius, so no effect costs no
segmentation. The SDK gains fromLivekitProcessor beside the adapter,
now toLivekitProcessor.
Ported from spec/background-effects/2-pipeline (#4282): a91dab8f with
its review tidy-ups 445e010c, bbc73d07, 51729bcf, f52d2b79, 813f1e6f
and 584bbaa0. Left out as already solved here: OneStepPipeline (the SDK
runs a processor's steps in turn) and the LiveKit-track sync helpers
(the lobby's camera is an SDK track, synced by followProcessor).
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
join
The lobby captured its own LiveKit tracks and the join captured again:
two permission prompts, a grey tile while the camera restarted, and the
background pipeline built twice. The lobby now opens the devices on the
SDK's `LocalMedia`, follows the mute switches as they change (they start
off until the devices are enumerated), the device choice and the
processor through controls shared with the call view model, and the call
joins with that media, so the tile carries the very track the lobby
previewed.
`useBehavior` subscribes without `useSyncExternalStore`, which left a
view without a re-render for an emission React had been told about,
about one time in three, when its store had switched. The LiveKit
preview hook and the processor sync helpers go, with
`@livekit/components-react`.
Background effects run under the dev server: the node polyfill no longer
rewrites MediaPipe's WASM loader. Taken from fkwp's 2fad4e08 on
spec/background-effects/2-pipeline (#4282).
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Storybook's react-docgen plugin parses .ts files with Babel's jsx plugin
enabled, so the angle-bracket assertion `<EncryptionSystem>useMemo(...)`
reads as an opening JSX tag and the Storybook build fails with
"Unexpected token". Passing the type as useMemo's generic is equivalent
and parses either way.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>