feat(hid): Unifying receiver support

[cserby]

Summary

OpenLogi previously only enumerated Logi Bolt receivers (PID 0xC548). Users with a Unifying receiver (PID 0xC52B / 0xC532) saw no devices in the GUI or CLI.

This PR adds full Unifying receiver support:

• Protocol layer (openlogi-hidpp): implement the missing HID++ 1.0 methods on UnifyingReceiver — count_pairings, trigger_device_arrival, get_device_pairing_information, device-connection event listener (0x41 sub_id), DeviceKind enum (shifted vs Bolt at values ≥5).
• Transport filter (openlogi-hid): the hid-logitech-dj kernel driver creates a virtual hidraw node per paired Unifying device; these matched the HID++ filter and caused 5-second probe timeouts every enumeration tick. Added is_receiver_child_node() (Linux) that checks the sysfs path and filters them before probing.
• Inventory (openlogi-hid): split probe_one to handle Receiver::Unifying separately. Added probe_unifying_receiver / probe_unifying_slot that build PairedDevice entries from device-arrival events. Added UNIFYING_SLOT_PROBE (3.5 s) per-slot budget so a slow wireless HID++ 2.0 round-trip cannot consume the shared PROBE_BUDGET.
• Routing (openlogi-hid): add DeviceRoute::Unifying variant; add DeviceRoute::device_route_for() constructor that picks the right variant from the receiver's product ID; consolidate BOLT_PIDS / UNIFYING_PIDS as the single source of truth (previously duplicated in three places).
• App stack (openlogi-agent-core, openlogi-gui, openlogi-cli): update all routing call sites, fix "Bolt receiver" label, add "Unifying receiver" label, surface devices that lack HID++ 2.0 model info in the carousel (wpid-based config key), hide the mouse-model Buttons tab for keyboards without an asset.

Verified on hardware

• Logitech Unifying receiver (C52B) — K540/K545 keyboard (slot 1) and M510 mouse (slot 2) both appear in the carousel
• Logitech MX Anywhere 3 over Bluetooth — still appears correctly
• Device tab shows "Unifying receiver" (not "Bolt receiver")
• K540 keyboard shows only the Device tab (no spurious mouse Buttons panel)
• M510 mouse shows Buttons + Device tabs

Known limitations

• Offline Unifying devices (paired but sleeping) are not surfaced — the 0xB5/0x5N pairing-info register returns InvalidValue; the sub-register format differs from Bolt and needs further investigation.
• K540 model info is absent on the first few enumeration ticks — the K540 takes ~3 s for the HID++ 2.0 version ping (slow wireless round-trip); the 3.5 s per-slot budget covers it, but feature enumeration may still miss on very slow hardware.
• The CacheKey::Bolt sentinel [0,0,0,slot] is reused for Unifying slot caching (no unit_id available without working pairing-info register); documented in the code.

Test plan

• cargo test --workspace passes (18 new unit tests across 5 crates)
• cargo clippy --workspace -- -D warnings clean
• Run ./target/debug/openlogi-agent + ./target/debug/openlogi-gui — all three devices appear in the carousel
• Device tab on Unifying devices shows "Unifying receiver"
• K540 shows only Device tab; M510 shows Buttons + Device tabs

🤖 Generated with Claude Code

[greptile-apps]

Greptile Summary

This PR adds full Logi Unifying receiver support (PIDs 0xC52B/0xC532) across the whole stack, from the HID++ 1.0 protocol layer up to the GUI carousel and settings window. It also resolves a long-standing Linux issue where the hid-logitech-dj kernel driver's virtual per-device hidraw nodes caused 5-second probe timeouts on every enumeration tick.

• Protocol layer (openlogi-hidpp): UnifyingReceiver gains count_pairings, trigger_device_arrival, get_device_pairing_information, and a 0x41 sub_id event listener; the BoltReceiver is simultaneously refactored to the Arc<ListenerDropGuard> pattern that fixes the Clone/Drop listener-handle bug identified in the previous review.
• Transport/inventory (openlogi-hid): is_receiver_child_node() filters out virtual DJ child nodes; probe_unifying_receiver/probe_unifying_slot probe online slots concurrently with a 3.5 s per-slot budget; CacheKey::UnifyingSlot uses the full receiver serial to prevent cross-receiver cache collisions (addressed from the prior review); the per-slot timeout fallback now correctly returns cached data rather than an empty default.
• App stack (openlogi-agent-core, openlogi-gui): routing consolidated into DeviceRoute::device_route_for(); Buttons tab suppressed for keyboards lacking an asset; Linux systemd user-unit autostart reconcile added.

Confidence Score: 5/5

Safe to merge; all three issues flagged in the previous review round are addressed in this version, and the new code is well-tested with 18 new unit tests.

All previous blocking findings (Clone/Drop listener handle, cache key collisions, timeout fallback ignoring the cache) have been resolved with the Arc pattern, CacheKey::UnifyingSlot using the full serial, and the corrected cached.map_or_else fallback. The two remaining notes are both minor style/performance observations with no impact on correctness.

crates/openlogi-gui/src/platform/permissions.rs — the Linux permission probe is called from a render closure; caching the result would prevent repeated filesystem I/O during UI redraws.

Important Files Changed

Filename	Overview
crates/openlogi-hidpp/src/receiver/unifying.rs	New UnifyingReceiver implementation with HID++ 1.0 register access; Clone issue from previous review resolved via Arc
crates/openlogi-hidpp/src/receiver/mod.rs	BoltReceiver refactored to Arc pattern, fixing the Clone/Drop listener handle bug from prior review; Unifying variant added to Receiver enum
crates/openlogi-hid/src/inventory.rs	probe_unifying_receiver / probe_unifying_slot added; CacheKey::UnifyingSlot uses full receiver UID to avoid cross-receiver collisions; timeout fallback correctly returns cached data
crates/openlogi-hid/src/route.rs	DeviceRoute::Unifying variant added; device_route_for() consolidates routing logic into a single canonical constructor; BOLT_PIDS/UNIFYING_PIDS now the single source of truth
crates/openlogi-hid/src/transport.rs	is_receiver_child_node() / is_receiver_child_sysfs_path() added to filter out hid-logitech-dj virtual child nodes on Linux; logic tested and correctly identifies terminal vs. parent sysfs components
crates/openlogi-agent/src/launch_agent.rs	Linux systemd user-unit reconcile added; escape_systemd_exec handles %, $, spaces and double-quotes; idempotent read-compare-write logic with best-effort systemctl calls
crates/openlogi-gui/src/platform/permissions.rs	Linux input-device access probe added (uinput + hidraw); classify() logic is well-tested; probe_logitech_hidraw reads /dev on every call which is invoked from a render closure
crates/openlogi-gui/src/windows/settings.rs	Permissions page split into macOS/Linux variants; permission_field function is #[cfg(target_os = "macos")]-gated but contains a redundant inner cfg block
crates/openlogi-gui/src/app.rs	Buttons tab now gated on can_show_mouse_model to suppress mouse-silhouette UI for keyboards without an asset; Unifying receiver label added to route_label()
crates/openlogi-agent-core/src/device_order.rs	DeviceRoute::Unifying matched alongside Bolt in DeviceStableId::from_parts; intentional shared stable-ID variant documented and unit-tested

Sequence Diagram

sequenceDiagram
    participant Enum as Enumerator
    participant Trans as transport
    participant Chan as HidppChannel
    participant UR as UnifyingReceiver
    participant Slot as probe_unifying_slot

    Enum->>Trans: enumerate_hidpp_devices()
    Trans-->>Enum: [hidraw nodes] (child nodes filtered)
    Enum->>Chan: open_hidpp_channel(dev)
    Chan-->>Enum: channel
    Enum->>UR: receiver::detect() → Receiver::Unifying
    Enum->>UR: get_unique_id()
    UR-->>Enum: serial (receiver_uid)
    Enum->>UR: count_pairings()
    UR-->>Enum: count
    Enum->>UR: listen() → rx
    Enum->>UR: trigger_device_arrival()
    UR-->>Enum: DeviceConnection events (drain 1.5s)
    loop per online slot (concurrent, 3.5s budget each)
        Enum->>Slot: probe_unifying_slot(channel, event, receiver_uid)
        Slot->>Chan: Device::new + enumerate_features
        Chan-->>Slot: ProbedFeatures
        Slot-->>Enum: (PairedDevice, CacheOutcome)
    end
    Enum-->>Enum: "DeviceInventory { receiver: Unifying, paired: [...] }"

Loading

_{Reviews (9): Last reviewed commit: "fix(hid): fall back to PID when Unifying..." | Re-trigger Greptile}