Commit Graph

17 Commits

Author SHA1 Message Date
Ken Sanislo d159081893
device: Fix operator precedence bug and end-of-configuration timing in device.changed() (#3173)
* Fix WIRELESS_DEVICE_STATUS reconfiguration: push settings and ack with proper cookie

When a device sends WIRELESS_DEVICE_STATUS with config_needed=1, the host
should re-push settings and acknowledge via ConfigChange SetComplete. The
existing code relied on the push flag in changed(), but operator precedence
caused push=True to be ignored for devices that have the WIRELESS_DEVICE_STATUS
feature — the exact devices that send this notification.

Handle the reconfiguration entirely in the notification handler: explicitly
push settings and ack, rather than trying to overload the changed() condition.

Also replace hardcoded cookie 0x11 with proper GetCookie/SetComplete protocol:
read the device's current configuration cookie and echo it back, per the
ConfigChange (0x0020) specification.

* WIRELESS_DEVICE_STATUS reconfig: gate apply on ConfigChange cookie

Add a cookie-comparison gate so devices that emit multiple reconfig
notifications (PRO X 2 sends two on power-on) don't get their settings
re-applied for every one. The gate also benefits any path that calls
apply_settings_if_needed — devices that retain config through
power-save now skip the redundant apply.

New flow:

- Device.apply_settings_if_needed() reads ConfigChange (0x0020) cookie
  via GetCookie. If the live value matches the cookie we stored after
  the last successful sync (persister key `_config_cookie`), the
  device hasn't drifted and we skip apply_all_settings entirely. On a
  real apply, SetComplete echoes the cookie back and we persist it.
  Devices without CONFIG_CHANGE bypass the gate (always apply).

- device.changed(active=True) calls apply_settings_if_needed in place
  of the old apply_all_settings + signal_configuration_complete pair.

- WIRELESS_DEVICE_STATUS reconfig handler drops the redundant explicit
  apply + ack (changed() already handles it on transition to active)
  and instead calls apply_settings_if_needed to cover the case of
  follow-up reconfig notifications on an already-active device — the
  cookie gate makes the second/third call essentially free.

Protocol cleanup:

- set_configuration_complete no longer auto-increments the cookie
  before echoing. The device owns the cookie and bumps it when its
  config drifts; the host's job is to confirm which value it synced
  with, not to advance the value. Host-side increment introduced a
  needless race with device-side bumps.

- signal_configuration_complete gains an optional cookie= arg so the
  caller can pass a value it already read (saves a redundant GetCookie
  round-trip from inside the gated apply path).

Tests:

- get_configuration_cookie returns the two-byte cookie from fn 0.
- set_configuration_complete echoes a provided cookie unchanged.
- set_configuration_complete with cookie=None reads the live cookie
  and echoes it (no increment).
2026-05-13 11:30:26 -04:00
MattHag bd00cc97ad
Estimate accurate battery level for some rechargable devices (#2745)
* battery: Extract battery level estimation into function

Test battery level estimation with sharp edges based on predefined
steps. Rename variable for clarity and add type hints.

Related #2744

* battery: Interpolate battery level for some rechargeable devices in percent

Estimate remaining battery based on measured battery voltage. Use linear
interpolation to achieve a smooth line instead of 10 percent jumps.

Fixes #2744
2025-01-02 10:58:07 -05:00
MattHag 5f5c7cdcce Fixes on top of refactoring 2025-01-01 10:46:04 -05:00
MattHag 1afcfe4b57
refactor: use IntEnum for firmware and cidgroup constances
* Refactor: test_named_ints_flag_names

Shorten test and clarify behavior using binary numbers.

* Introduce plain flag_names function

This replicates the NamedInts functionality as plain function.

* Refactor FeatureFlag to use IntFlag

Replace NamedInts implementation with IntFlag enum and plain flag_names
function.

Related #2273

* Refactor FirmwareKind to use IntEnum

- Move general FirmwareKind to common module.
- Replace NamedInts implementation with IntEnum.
- Harden related HIDPP 1.0 get_firmware test.

Related #2273

* Refactor CID_GROUP, CID_GROUP_BIT to use IntEnum

Related #2273
2024-10-23 16:25:35 -04:00
MattHag 0cd9c0c9b5 Refactor: Introduce Feature enum
Convert Feature NamedInts to SupportedFeature integer enum.

Related #2273
2024-10-14 07:28:09 -04:00
MattHag 741c0861c2 Clarify that fake hidpp is used
This module shouldn't be necessary on the long run. Remove pieces from
it whenever possible.
2024-10-08 14:35:16 -04:00
Matthias Hagmann c9dc232951 Refactor: Use dataclasses and enums
Replace unnecessary NamedInts in favour of default data types.
Simplify interfaces by reducing possible input from strings to members
of an enum.
2024-05-22 21:14:41 -04:00
Peter F. Patel-Schneider e64eec18e9 tests: extend testing of hidpp20 2024-04-13 18:38:44 -04:00
Peter F. Patel-Schneider c70e8b54bf tests: remove unused code 2024-03-27 18:02:55 -04:00
Peter F. Patel-Schneider 12f3f2e856 tests: adjust imports to always import installed version 2024-03-27 18:02:55 -04:00
Peter F. Patel-Schneider 88ac4c9f89 tests: use new test methods in test_hidpp20_simple 2024-03-23 10:11:30 -04:00
Peter F. Patel-Schneider 6164317a64 tests: improve hidpp20 coverage 2024-03-16 16:20:39 -04:00
Peter F. Patel-Schneider d76eed85f6 device: fix bug in setting configuration cookie due to bad documentation 2024-03-14 17:06:17 -04:00
Peter F. Patel-Schneider d5bdf2b0f5 tests: complete testing of common 2024-03-13 16:08:16 -04:00
Peter F. Patel-Schneider 4632c46e30 tests: expand coverage of hidpp20 2024-03-13 16:08:16 -04:00
Matthias Hagmann 5b09ace1f5 ruff: Apply single line import format
# Usage
pre-commit run --all-files

Related #2295
2024-03-13 15:41:21 -04:00
Peter F. Patel-Schneider 15ed26887b tests: tests for a few simple hidpp20 functions 2024-03-05 15:27:07 -05:00