Loading
fix: BlueZ discovery drops its UUID filter, so rig phones are observed; linux-hw probe (private#770)
What
This MR has two changes, both from the first rig check of linux-hw against
the three rig phones on Glance (2026-11-02, this laptop's hci0).
Fix: BlueZ discovery drops its UUID filter.
- With
SetDiscoveryFilter { UUIDs: [service], Transport: le }, the probe observed nothing in 90 s.bluetoothctlwith the same UUID filter also reported no device. - Without the UUID filter, the probe observed 213 advertisements in 30 s, 171 of them from one phone advertising the service.
- The filter keeps
Transport: le.Devices::observealready keeps only devices that advertise the service, anddevices_without_the_service_are_not_observedstill covers that. - Why BlueZ's UUID filter drops these phones is not established.
v0.97.0does not contain this fix. GTK scans through this code too.
Tool: linux-hw probe (cargo run --example probe --). The modes are
report | scan | connect <address> | advertise | dial.
- It sends the commands an app would and prints the events.
dialscans, connects to the first phone it sees and reads the exchange payload, as Core's central does.- Plan Phase 2 "Tools", ADR-077 item 9.
Rig results
| Check | Result |
|---|---|
| Capability report | has_ble, has_camera, audio: Full; no NFC (no PC/SC reader) |
| Discovery, before the fix | 0 observations in 90 s |
Discovery, after the fix (probe built from the fix commit) |
122 observations in 20 s from two phones |
Outbound link (dial) |
Connected to "Vauchi-Diag", 14 characteristics, MTU 517, BleConnected (Outbound), read the 32-byte exchange payload, then the same value as a notification after StartNotify |
Peripheral (advertise, 45 s) |
No connection, no error. Inconclusive: whether a phone dials depends on Core's token comparison |
Evidence
- The
testcommit failed before the fix: every scan setUUIDs [service]where none was expected. - After the fix, the 10
bluez_discoverytests pass, and so dojust check linux-hwand the pre-push suite (83 tests).