TikTokTTS v1.2.5

v1.2.5

• Fix: Selecting the (loading...) placeholder in the device dropdown (shown when no media players are available yet) no longer crashes the select.select_option service call with an IndexError
• Fix: Calling tiktoktts.set_random_voices after the last config entry was unloaded (e.g. the entry is disabled) no longer crashes with a KeyError; the pool is still validated and persisted
• Refactor: The shared select/text/button entities are now forwarded by a single owning config entry instead of every entry, deleting the per-platform singleton guards and per-entry bookkeeping flags. Re-homing the shared entities to a surviving entry (when the owner is unloaded while another entry remains) now runs as a tracked Home Assistant task that is skipped during shutdown and no-ops when the owner was merely reloaded - replacing a raw call_soon hook that could race HA shutdown
• Refactor: The language and voice dropdowns no longer hold direct references to each other. They coordinate through dispatcher signals plus one hass.data key, which removes the startup readiness event, its 10-second timeout, and the entity object previously stashed in hass.data for the service handler. Behaviour (restore on restart, default-voice preservation, random-voice handling) is unchanged and covered by tests
• Perf: Direct API mode now remembers the endpoint that last served a request and tries it first, so when the configured endpoint is down, a multi-chunk message no longer re-probes it (with full retries) for every chunk
• Perf: Direct API mode now has an overall per-message deadline (120 s) so a long message during an outage cannot stall a tts.speak call for many minutes while every chunk walks the whole endpoint fallback list
• Perf: The device dropdown's media-player discovery now filters state events inside Home Assistant's event dispatch (event_filter), so routine attribute updates from active players never wake the integration at all
• Perf: Static assets (card JS, icon) are now served with cache headers; the card URL already carries a ?v=<version> cache-buster so upgrades still load the new card immediately
• Security: The TikTok session_id field in the setup and options forms is now a masked password input instead of plain text
• Fix: Connection-test responses in the config flow are now properly released via context managers, and unexpected connection-test failures are logged with a traceback instead of being silently reported as "unknown error"
• Meta: manifest.json now declares integration_type: service and iot_class: cloud_push (matching core TTS integrations)
• Tests: 133 tests (up from 118); new coverage for the two crash fixes above, endpoint memoization, the per-message deadline, device selection and restore-wait behaviour, Speak-button guard paths, the options-save reload cycle, and config-flow unknown-error branches. The frontend tests now read the version from manifest.json instead of hardcoding it, so version bumps no longer break the suite
• CI: Added ruff check and ruff format --check to the test workflow; the codebase is now Ruff-formatted

v1.2.4

• Fix: Dashboard card no longer crashes with the name "tiktoktts-card" has already been used with this registry after upgrading. Older versions registered the card as a Lovelace dashboard resource; the current version registers it via the frontend module API, so on upgrade both could load the card and define the element twice. The card now guards against double-registration. If you upgraded, you can also remove the leftover /tiktoktts/tiktoktts-card.js entry under Settings -> Dashboards -> Resources.