The card claimed a camera's ONVIF audio backchannel in two places that
had
nothing to do with a call: the capability probe at camera init
(`µphone`),
and a pre-armed `sendonly` audio transceiver on every live WebRTC offer.
Merely
looking at a dashboard occupied the camera's speaker line. Outbound
audio now
travels on its own audio-only WebRTC connection, opened when a call is
answered
and closed when it ends.
- The backchannel is claimed only for the duration of a call. Idle
viewing
claims nothing.
- Two-way audio now works in `mse`, `mp4` and `mjpeg` modes (note: the
outbound
audio still traverses WebRTC).
- No renegotiation and no video blink at call start or end.
- A call that cannot carry audio now reports it and ends, instead of
showing a
live microphone that goes nowhere.
- `live.microphone.always_connected` is now purely about the browser
microphone
permission prompt.
- Call setup measured at 66ms (LAN) and ~260ms (cellular) for ICE and
DTLS, plus
~300ms for `go2rtc` to open an RTSP backchannel.
Verified against a live Frigate + `go2rtc` instance, and by unit tests
at 100%
coverage.
- Closes#2691
- Closes#2039
- Closes#2178
Ref #2299 -- the probe no longer opens a backchannel, but it still runs
per
camera on every load and reconnect, and still dials the camera on the
direct-`go2rtc` path. Caching remains to be done.
Ref AlexxIT/go2rtc#1860 -- once a call has opened a backchannel,
`go2rtc` keeps
that media set up on the camera's RTSP session for the life of the
producer.
Diagnoses #2678
- Closes: #2679
BREAKING CHANGE: `selected` and `unselected` are removed from
`live.microphone.auto_unmute` / `auto_mute`. A camera change ends any
call and the microphone only carries audio during a call, so neither
were useful.
The microphone is connected when a call needs it and released the moment
that
call ends, so the browser stops reporting it as in use at hangup rather
than
`disconnect_seconds` later.
Existing configurations are migrated automatically by the visual editor.
- Closes: #2681
BREAKING CHANGE: `live.microphone.disconnect_seconds` is removed. The
microphone is released when a call ends, so there is no idle countdown
to
configure. Use `live.microphone.always_connected` to hold it open
instead.
BREAKING CHANGE: The `microphone_connect` and `microphone_disconnect`
actions
are removed. The card owns the microphone lifecycle; `microphone_mute`
and
`microphone_unmute` remain.
BREAKING CHANGE: `call` is removed from `live.microphone.auto_mute`,
whose
default is now `[]`. The microphone is muted when a call ends regardless
of
this option.
BREAKING CHANGE: `microphone_unmute` has no effect outside a call.
Nothing
carries the audio at any other time, so the request is ignored rather
than
opening the microphone.
This is a high risk change: every byte the card ships is emitted by a
different bundler, minified by a different minifier, and every
stylesheet is compiled by a different sass. The intent is that the
card's behaviour is unchanged.
**Significant development win**: Build time drops from 24s to 1.1s 🎉
Also adds a test suite ('dist') that runs against the built bundle.
Mounts the card from `dist/` the way Home Assistant does, by URL with a
HACS tag, and asserts the entry stays a facade, that no chunk imports
it, that the browser is served the file unmodified, and that the card
starts up and fetches a language chunk lazily.
Runs in Chromium, Firefox and WebKit, so a change of bundler or minifier
is checked against every engine.
The `initialized` state (used in conditions/triggers) was written once
and never cleared, so it meant "has this card ever been initialized"
while everything reading it took it as "is this card usable now". Home
Assistant takes a card off the page and puts it back whenever its
dashboard tab is left and returned to, so the card initialized again
while the state claimed it was initialized throughout: `trigger:
initialized` fired once per card rather than once per startup, and
automations were dropped in between.
The card lifecycle is now an explicit state machine (`SessionManager`),
the only writer of `initialized`, which separates a card that is
starting up from one initializing part of itself again while it runs. A
new `ever` parameter (conditions/triggers) selects the old latched
behaviour.
`remote_control` uses that parameter to keep its two camera priorities
correct under repeated starts. With `camera_priority: entity` the card
now re-reads the entity every time it starts, so a camera selected while
the card was away is picked up on return. With `camera_priority: card`
the card writes the entity on its first start only, unchanged, since
repeating that write would overwrite a camera the user had selected.
Closes: #2642
BREAKING CHANGE: `condition: initialized` is now `false` whenever the
card is not usable, and `trigger: initialized` fires each time the card
starts up rather than only the first time. Set `ever: true` on either to
keep the previous behaviour.
- Closes#2556
- Closes#2450
**Key intended features:**
- go2rtc compatible
- 100% test coverage to significantly improve ability to test, maintain
and work around browser weirdnesses (e.g. Safari).
- Written from the ground up in the style of the rest of the project.
**To use:**
- Change `live_provider` from `go2rtc` to `go2rtc-experimental`.
Closes#2531.
## Summary
The full `nunjucks` templating engine (~226KB) plus `ha-nunjucks`
(~46KB) — roughly **272KB, ~13% of the eager entry chunk** — was
statically imported and downloaded by every card on initial load, even
though templates only apply when a config value contains a `{{ … }}` /
`{% … %}` delimiter. Most cards use no templates and never need the
engine.
This defers the engine behind a dynamic `import('ha-nunjucks/dist')` so
it ships in a separate, on-demand chunk instead of the eager
`card-*.js`.
## Approach
The render path (`TemplateRenderer.renderRecursively`) is **kept
synchronous** — it is called from many synchronous hot paths
(condition/trigger evaluators, picture-elements rendering, actions,
folder matchers), and making it async would be a large, high-risk
refactor of the evaluation core.
Instead:
- **New `src/card-controller/templates/engine.ts`** — a module-level
singleton lazy loader (`loadTemplateEngine()` / `getTemplateEngine()`)
shared across all `TemplateRenderer` instances, plus a
`containsTemplate()` delimiter helper.
- **Delimiter gating** — strings without a delimiter never touch the
engine (the overwhelming majority of renders).
- **Pre-warm at config time** — because every template string originates
in the config, a new mandatory `TEMPLATE_ENGINE` initialization aspect
loads the engine before first render whenever the config contains a
delimiter. This **guarantees no raw `{{ … }}` flash**: content/condition
rendering is blocked until the engine is present for template-using
cards. Cards without templates never load it.
## Result
- nunjucks + ha-nunjucks move out of the eager chunk into a separate
chunk fetched only when a card actually uses templates.
- No change to the synchronous public render API; condition/trigger
evaluation core untouched.
## Tests
- New `engine.ts` loader coverage (concurrent load, cached reuse,
not-loaded fallback).
- The 11 existing test files that render real (delimiter-bearing)
templates declare their dependency explicitly via
`beforeAll(loadTemplateEngine)` — no global/implicit setup hook.
- Full suite green (4764 tests), lint and ts-prune clean, per-file 100%
coverage maintained for the affected directories.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
https://claude.ai/code/session_016ykWemdkZrgywvC71pHc6c
---
_Generated by [Claude
Code](https://claude.ai/code/session_016ykWemdkZrgywvC71pHc6c)_
Dermot Duffy