Fixes a typo in the weather forecast characteristic writer that causes
Gadgetbridge to send the current weather twice rather than sending the
forecast.
Zepp OS 3 firmware upgrades are big (200MB+). Gadgetbridge was crashing,
since the entire firmware file would be pulled into memory.
This commit unifies all the logic for Zepp OS firmware handling.
However, since the needed refactor was big, this commit duplicates some
of the code from Huami classes, namely:
- ZeppOsFirmwareUpdateOperation clones UpdateFirmwareOperation2020
- ZeppOsFwInstallHandler clones AbstractMiBandFWInstallHandler
This avoids changes to older device logic and introducing regressions.
Lost functionality:
- Repackaging firmwares as UIHH (does not seem to be needed, and was not
used anyway). Code can be recovered from this commit if needed in the
future
- Whitelisted firmwares by checksum (we do not have a lot of them at
this point anyway)
Other misc changes:
- Rename ZipFile to GBZipFile not to clash with the java class
Tested by updating the Amazfit GTR 4 to Zepp OS 3.
Initial support for sony wi-sp600n reported working . minor difference with the inspiring wf-sp800n
Reviewed-on: https://codeberg.org/Freeyourgadget/Gadgetbridge/pulls/3522
Co-authored-by: Denis Pitzalis <denics@free.fr>
Co-committed-by: Denis Pitzalis <denics@free.fr>
If the device connection state is updated from two threads simultaneously
(as in, from the main looper and from the thread that handles
BluetoothDevice.connectGatt), a second update may get overridden by the
first update if the broadcasts are handled out-of-order by the
LocalBroadcastManager.
By updating the device state through a handler on the main looper, the
broadcasts are sent in order as they are processed from the looper's
queue.
This may be a potential solve for issue #3524.
- Some devices send a random int 0, which would prevent sleep stage
parsing
- Some devices send the details as a file of type summary, but same
structure
- It is still not stable for all devices
Thanks to @opcode for the parsing logic
Not all communication was moved to services, and some might not be
respecting the encryption flag sent during initialization implemented
in 3a2b02df2. Some services are encrypted or not across different
watches - see #3308.
This allows sleep stage detection to work by parsing
some of the data sent in SleepDetails.
It's still missing parsing the summary contained inside SleepDetails.
and decoding the large amount of other mostly unknown data.
Disconnect devices in all states except GBDevice.State.NOT_CONNECTED.
This should fix devices getting stuck in GBDevice.State.CONNECTING state
when Bluetooth is switched off.
* Add general_reconnectonlytoconnected setting.
* Replace last_device_address shared prefs string with
last_device_addresses shared prefs string set.
Bluetooth address of a device is added to last_device_addresses when
connecting to the device.
Bluetooth address of a device is removed from last_device_addresses only
when deleting the device or explicitly disconnecting from the device
(e.g. by selecting "Disconnect" in the device tile menu).
* Adjust ExternalPebbleJSActivity to better support multiple connected
devices.
Math.multiplyExact is only available on SDK24 onward. We do not expect
an overflow to occur here, since the hashLength is fixed and the outputLength
is small (32 on all current calls).
Explicitely disable the request MTU feature from the NRF DFU library to ensure that it'll send 20 bytes packets (instead of 253B, which will crash InfiniTime).
Changes:
- Support locking (keep visible and let hands display time)
- Support merge navigation instruction
- Support wrist flick gesture to move hands
- Support GB-configurable foreground and vibration behaviour
After find phone is stopped, watch sends a `Command` message back
without `system` set, and because of missing hasSystem() check, we'd
wrongly go to condition that sets findPhoneEvent.event to START.
Reports in #3455 and #2955 mention that everything works, except HR,
which crashes the devices (they do not have an HR sensor), so we just
disable HR support.
Gadgetbridge can be configured to perform an action when a
Huami device is taken off or the user was detected to fall asleep or
wake up. This functionality was specific to Huami devices, but this
changeset moves this upstream to the AbstractDeviceSupport class in
combination with new GBDeviceEvents.
Now that the ADS has centralized support for this functionality, the
same logic can be used for other devices. In this case, an
implementation is added for supported Xiaomi devices.
Works:
- notifications
- call notification
- set time
- setting brightness
- setting 24h/12h format
- sending weather
Note
- this is implemented using using the classic bluetooth serial protocol, the
device can do BLE, but I don't know how to use it, as I did not have the
offical app to sniff.
- The information about the protocol comes from here
https://github.com/jfroehlich/node-p1x3lramen/blob/main/source/devices/pixoo.js
TODO:
- Enable beep? Possible? I heard it beep once at least when switching it on
- Getting out of factory mode? Why does it always play animations even when I
switch to the clock?
- Implement switching modes (can be done with the button)
- Implement sending own images and animations
- Firmware update?
- ...
Because the reporting of battery state is inconsistent between different
models, the device's battery state was not correctly processed in GB.
For at least the firmware on the Xiaomi Watch S1 Active, the charger
state is broadcast through a separate message from the message
containing the battery level. Even though the battery level was
requested by GB upon receiving this broadcast, the charger state got
discarded as it was expected to also be included in the result of the
subsequent request.
This patch changes the name of the `Charger` message to `DeviceState`
and includes more fields that may be presented by some device models.
Furthemore, the broadcast is cached so that the charger state can be
processed from this cache instead of the battery level response message.
Exceptions raised while handling actions in the device's support class
may result in the DeviceCommunicationService crashing as a whole and not
being started again until the user forces GB to make a connection by
pressing the device from the list.
This change is made, because Xiaomi devices make use of proto2 messages,
where optional fields cannot be set to null values as that will make it
throw NPEs.
This will prevent uncaught exceptions that occur during the
parsing of activity data from breaking up the activity fetching
chain and causing the task to never get completed.
The activity parser may return null pointers in case the version
of the data structure is not supported. Not checking for null here may
result in the activity fetching task may never complete and cause
further communication with the device to grind to a halt.
We should use NFKD instead of NFD since we are flattening to US-ASCII
afterwards anyway. This allows various Unicode characters which would
end up as a question mark to be represented by their compatibility
decomposition. This applies to e.g. ligatures (e.g. U+FB01 LATIN SMALL
LIGATURE FI will now be replaced with plain fi instead of a question
mark), and also the U+00A0 NO-BREAK SPACE [NBSP] to be replaced by
a normal space instead of a question mark.
+Add Czech fancy quotes to the Czech transliterator
+Add a unit test for Multitransliterator
