There is a race condition where a BedJet unit previously had its BLE
"notify" flag enabled, and it continues to broadcast these notify
packets even after the ESP32 (and BLEClient) goes away, such as during
a crash or unplugging power.
BLEClient::setup_priority=AFTER_BLUETOOTH, while
BedJetHub::setup_priority=AFTER_WIFI. When the ESP32 starts back up
again, BLEClient::setup() happens first and will start receiving the
BLE notify packets almost immediately. Since we register the BLEClient
child from codegen, BedJetHub is registered as a child already by this
point, so BLEClient dispatches the notify status packet (and other gatt
events) to the BedJetHub handler, even though BedJetHub::setup() has
not been called yet.
We initialize BedJetHub::codec_ in setup(), so if BLEClient starts
dispatching gatt events before setup() is called, then codec_ will not
be initialized yet. This causes BedJetHub's gatt notify handler to call
`this->codec_->decode_notify()` on an uninitialized null pointer. Since
invoking a method does not have to dereference the pointer, that method
invocation is allowed; but later trying to access memory on that
instance results in a StoreProhibited panic.
Changing the BedJetHub's setup_priority to BLUETOOTH causes it to be
setup before BLEClient, so that by the time BLEClient starts to receive
BLE packets, BedJetHub is ready to receive them.
* Add push to talk voice assistant
* Refactor most code into voice_assistant
* Make voice_assistant the component and remove push_to_talk (can be done in yaml)
* Fix component setup
* Always AF_INET to match serverside
* Fix microphone and media player co-existence
* Format
* Update codeowners
* Update test file
* Fix endifs
* nullptr not NULL
* clang-tidy
* Format
* fixup: Add VA event data
* Generate proto
* Parse and log events
* Add default to switch
* Fix
* Add mic/va to test5
* limit configuration to 64 binary sensors to match code limitation
* remove superfluous debug logging
* improve state publishing logic eliminating NAN being sent on boot in certain cases
* adjust type for bitmask shift to match mask variable type
* Synchronise Device Classes from Home Assistant
* Remove count do the `DEVICE_CLASSES` list is also updated
* Format file
---------
Co-authored-by: esphomebot <esphome@nabucasa.com>
Co-authored-by: Jesse Hills <3060199+jesserockz@users.noreply.github.com>
* Move EntityBase Object Id from memory to flash.
* Sprinkler use common `setup_entity` method.
* Remove `EntityBase` from Sprinkler.
* Support for entity names set to None
* change so gh PR picks up commit.
---------
Co-authored-by: Your Name <you@example.com>
Co-authored-by: Jesse Hills <3060199+jesserockz@users.noreply.github.com>
* feat: Add support to unsubscribe from BLE advertisements
* Fix tests & clang
---------
Co-authored-by: Jesse Hills <3060199+jesserockz@users.noreply.github.com>
Animation resizing in RGB24 format is causing an error "Image cannot be resized to a bigger size". Other image types do not show the issue, and the only difference is the "image.thumbnail" call.
Removed the call and tested; the animation is shown with the desired size.
* Add ability to clear the gatt cache
With BlueZ we can fully clear the cache when something goes wrong with the services, however since this is also a cache on the ESP32 we need to be able to clear the on device cache as well for the proxies since if something goes wrong with the service resolution it can cache the bad resolution on NVS forever.
Our current client implementation is limited to clearing the memory cache in Home Assistant 89355e0879/homeassistant/components/esphome/bluetooth/client.py (L512)
related issue https://github.com/esphome/issues/issues/4156https://github.com/esphome/aioesphomeapi/pull/410
* naming
* lint
* lint
* naming
* naming
* naming
* 88 now that 87 is taken
* make const
* Update esphome/components/api/api_frame_helper.cpp
Animation resizing in RGB24 format is causing an error "Image cannot be resized to a bigger size". Other image types do not show the issue, and the only difference is the "image.thumbnail" call.
Removed the call and tested; the animation is shown with the desired size.
