Fixes a critical bug where Android devices (acting as BLE centrals) disconnecting
from Pi GATT servers (acting as peripherals) never triggered cleanup, causing stale
peer entries to accumulate until the 7-peer limit was reached and blocked all new
connections.
## Root Cause
- When centrals disconnected from peripheral mode, no cleanup occurred
- `BLEGATTServer._handle_central_disconnected()` method didn't exist
- `on_central_disconnected` callback was never wired to driver
- No D-Bus signal monitoring for device disconnections
- Stale entries remained in `_peers` dict until daemon restart
## Implementation (TDD Approach)
**New Methods:**
- `LinuxBluetoothDriver._handle_peripheral_disconnected()` (line 852)
- Removes peer from `_peers` dictionary
- Notifies on_device_disconnected callback
- Triggers full cleanup chain in BLEInterface
- `BluezeroGATTServer._handle_central_disconnected()` (line 1945)
- Removes from `connected_centrals` dictionary
- Logs connection duration
- Invokes driver callback
- `BluezeroGATTServer._monitor_device_disconnections()` (line 1645)
- Monitors D-Bus PropertiesChanged signals
- Detects when Connected property becomes False
- Runs in separate daemon thread
- Automatically triggers cleanup on disconnect
**Callback Wiring:** (line 1558)
`on_central_disconnected = driver._handle_peripheral_disconnected`
## Testing
- Created comprehensive test suite (9 tests, all passing)
- `tests/test_peripheral_disconnect_cleanup.py`:
- Callback wiring verification
- Peer dictionary cleanup
- D-Bus signal handling simulation
- Edge cases (multiple disconnects, race conditions, shutdown)
- Reproduces real-world bug from production logs
- No regressions in existing tests (test_bluez_state_cleanup.py passes)
## Current Status
✅ Core cleanup logic implemented and tested
✅ Deployed to 4 production devices (10.0.0.80, .242, .39, .246)
⚠️ D-Bus monitoring thread needs debugging (not logging yet)
**Known Issue:** D-Bus signal subscription may need alternative approach.
See PERIPHERAL_DISCONNECT_FIX_SUMMARY.md for troubleshooting steps.
**Fallback Option:** Timeout-based polling can be implemented if D-Bus proves difficult.
Reference: Production logs showed device 4A:87:8C:C7:E3:F3 repeatedly blocked
by "max peers (7) reached" due to uncleaned peripheral disconnections.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Fixed race condition where started_event fires before peripheral.publish()
fully exports GATT services to D-Bus, causing "Reticulum service not found"
errors when central devices connect immediately after server startup.
Root cause:
- started_event.set() called on line 1665
- peripheral_obj.publish() called on line 1669 (exports to D-Bus)
- 50-200ms gap where server thinks it's ready but services aren't on D-Bus yet
- Central connects during gap -> "service not found" error
Fix:
- Added _verify_services_on_dbus() method to poll D-Bus adapter introspection
- Polls every 200ms with 5-second timeout after started_event fires
- Only returns from start() after confirming services are exported
- Graceful degradation: warns on timeout but doesn't fail startup
Impact:
- Eliminates "service not found" errors during server startup
- Ensures services are actually available before accepting connections
- Typical verification time: 100-300ms
- No runtime performance impact (only affects startup)
Files changed:
- src/RNS/Interfaces/linux_bluetooth_driver.py: Add D-Bus polling
- tests/test_gatt_server_readiness.py: Add test coverage
- BLE_PROTOCOL_v2.2.md: Document initialization race fix
- CHANGELOG.md: Record fix details
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
ConnectDevice() D-Bus method returns an object path (signature 'o') which
should be treated as success, not error. Previously, the return value was
not captured or logged, causing confusion when error messages like
"br-connection-profile-unavailable" appeared (which is expected for LE-only
connections).
Changes:
- Capture object path returned by call_connect_device()
- Log object path for debugging visibility
- Document that object path indicates successful LE connection initiation
- Clarify that BR/EDR profile unavailable is expected for BLE-only connections
Impact:
- Eliminates confusion from "profile unavailable" error messages
- Confirms LE connection was successfully initiated
- Improved debugging visibility through object path logging
Files changed:
- src/RNS/Interfaces/linux_bluetooth_driver.py: Capture and log object path
- tests/test_breddr_fallback_prevention.py: Add test coverage
- BLE_PROTOCOL_v2.2.md: Document object path behavior
- CHANGELOG.md: Record fix details
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Implements comprehensive BlueZ device state cleanup after connection failures
to prevent persistent "Operation already in progress" errors. This addresses
the issue where BlueZ maintains stale connection state after timeouts or
failures, preventing successful reconnection even after blacklist periods expire.
BlueZ State Cleanup Implementation:
- **Explicit client disconnect**: Call client.disconnect() in timeout and failure
exception handlers to release BlueZ resources
- **D-Bus device removal**: New _remove_bluez_device() method removes stale device
objects via BlueZ RemoveDevice() API
- **Post-blacklist cleanup**: Trigger BlueZ cleanup when peer is blacklisted after
reaching max_connection_failures (7 failures)
Impact:
- Enables successful reconnection after temporary connection failures
- Fixes persistent errors across blacklist periods
- Prevents BlueZ from maintaining corrupted connection state
- Particularly important for Android devices with MAC address rotation
Implementation Details:
- linux_bluetooth_driver.py:786-830: New _remove_bluez_device() method
- linux_bluetooth_driver.py:1029-1044: Timeout cleanup (disconnect + removal)
- linux_bluetooth_driver.py:1051-1066: Failure cleanup (disconnect + removal)
- BLEInterface.py:1270-1285: Post-blacklist cleanup hook
- tests/test_bluez_state_cleanup.py: 10 new tests (all passing)
Documentation Updates:
- BLE_PROTOCOL_v2.2.md: New troubleshooting section for persistent InProgress errors
- CLAUDE.md: Added to recent fixes list
- CHANGELOG.md: Comprehensive fix description
Related Issues:
- Addresses "Operation already in progress" errors persisting after connection timeouts
- Fixes reconnection failures after peer blacklisting
- Prevents BlueZ state machine corruption from abandoned BleakClient instances
Testing:
- All 10 new unit tests pass
- Cleanup methods properly handle missing devices and D-Bus unavailability
- Integration testing on Raspberry Pi pending
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Removed test_refactor_suite.py as it is completely superseded by the
comprehensive test suite:
Reasons for removal:
- Broken: Import errors, cannot run
- Incomplete: Contains TODO comments, no actual assertions
- Overlapped: Functionality covered by test_multi_device_simulation.py
- Inferior: 1 broken test vs 20 passing comprehensive tests
- Wrong approach: Tried to run real BLE instances instead of using mocks
- Already excluded: Ignored in CI via --ignore flag
The multi_device_simulation test suite provides superior coverage:
- MockBLEComponents (5 tests)
- SimulatedBLENode (3 tests)
- TwoDeviceSimulator (6 tests)
- IntegrationScenarios (4 tests)
- Performance (2 tests)
This was leftover scaffolding from the driver abstraction refactor.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Updated tests to reflect the new driver-based architecture where GATT
server and connection management are handled by the driver layer instead
of directly in BLEInterface.
Changes:
- test_integration.py: Updated to check for driver callbacks instead of
old GATT server methods (_data_received_callback vs on_data_received)
- test_integration.py: Added test for driver abstraction layer
- test_prioritization.py: Updated to check for driver.connect() instead
of removed _connect_to_peer() method
All 106 tests now pass (excluding test_refactor_suite.py which has
import issues and appears to be obsolete).
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Only install libffi-dev on armhf (32-bit ARM) systems where cffi needs
to compile from source. x86_64 and arm64 have pre-built cffi wheels
available, so they don't need the development headers.
Changes:
- install.sh: Detect architecture and conditionally add libffi-dev for armhf
- test_installer.sh: Show libffi-dev in output only for armhf systems
- test.yml: Update ARM CI summary to reflect conditional dependency
This reduces unnecessary dependencies on x86_64 and arm64 systems while
maintaining full compatibility with 32-bit Raspberry Pi devices.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Add libffi-dev to system dependencies for Debian/Ubuntu/Raspberry Pi OS
to provide FFI headers needed when cffi compiles from source on ARM
platforms. This fixes ARM 32-bit and 64-bit installation failures.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
The test script was failing because it changed directories during execution
(to /tmp/test-config/interfaces) and then tried to use a relative path
to navigate back to the repo root, which failed.
Fix: Save the absolute path to repo root at the beginning and reuse it
when needed instead of calculating it relative to the current directory.
Fixes the error: 'cd: tests/..: No such file or directory'
Fixes#3
BlueZ experimental mode is required for proper BLE connectivity. Without it,
BlueZ attempts Classic Bluetooth (BR/EDR) connections instead of BLE (LE)
connections, causing connection errors like "br-connection-profile-unavailable"
and immediate disconnections after pairing.
Changes:
- install.sh: Automatically enables BlueZ experimental mode during installation
- Detects BlueZ version (requires >= 5.49)
- Creates systemd override to add -E flag to bluetoothd
- Checks if already enabled to avoid duplicate configuration
- Shows strong warning if user skips with --skip-experimental flag
- Added --skip-experimental flag to opt-out (not recommended)
- Updated help text to document new flag
- tests/test_installer.sh: Added tests for experimental mode configuration
- README.md: Documented BlueZ experimental mode in installation sections
- Added to automated installation description
- Added as required step in manual installation
- Added troubleshooting section for BR/EDR connection errors
- examples/config_example.toml: Added troubleshooting entry for BR/EDR errors
The installer now:
1. Detects BlueZ version >= 5.49 (required for experimental mode)
2. Checks if already enabled (graceful skip)
3. Enables experimental mode by default unless --skip-experimental is used
4. Shows prominent warning if skipped (may cause BLE to break)
5. Handles edge cases (no systemd, old BlueZ, container environments)
This addresses the root cause reported in issue #3 where devices were
connecting then immediately disconnecting with BR/EDR profile errors.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Arch Linux has PyGObject 3.54.5 in python-gobject package, but bluezero
requires PyGObject <3.52.0, causing pip to fail when trying to replace
the system version.
Solution: Don't install python-gobject system package on Arch. Let pip
compile the compatible PyGObject version (3.50.2) instead.
Changes:
- install.sh: Remove python-gobject from Arch pacman install
- install.sh: Add explanatory warning about PyGObject compilation
- tests/test_installer.sh: Don't check for python-gobject on Arch
- tests/test_installer.sh: Add comment explaining why it's skipped
- tests/test_installer.sh: Update summary for Arch (PyGObject compiled)
- README.md: Remove python-gobject from Arch instructions
- README.md: Explain version incompatibility and compilation requirement
Result:
- Debian/Ubuntu: All system packages, zero compilation (~1 min)
- Arch Linux: System packages + PyGObject compilation (~2-3 min)
Trade-off accepted: Arch users get longer install time in exchange for
compatibility with bluezero's PyGObject version requirement.
Fixes: error: uninstall-no-record-file (PyGObject 3.54.5 conflict)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Arch Linux has unique pip/system package integration where pip doesn't
recognize system python-gobject as satisfying PyGObject dependency,
causing bluezero to try compiling PyGObject from source.
Solution: Install base-devel on Arch to provide build tools (gcc, make, meson)
Changes:
- install.sh: Add base-devel to Arch system dependencies
- install.sh: Add note explaining why build tools needed on Arch
- install.sh: Use --needed flag to skip already installed packages
- README.md: Document base-devel requirement for Arch users
- README.md: Explain Arch vs Debian/Ubuntu compilation differences
- tests/test_installer.sh: Expect build tools on Arch (verify base-devel installed)
- tests/test_installer.sh: Update summary to reflect Arch compilation
Rationale:
- AUR python-bluezero is outdated (v0.9.0 vs pip v0.9.1)
- AUR package has 0 votes (rarely used by community)
- base-devel commonly installed on Arch systems anyway
- Keeps latest bluezero version
- Simpler than full AUR integration
Impact:
- Debian/Ubuntu: No compilation (< 1 min install)
- Arch Linux: Some compilation (~3 min install)
- Still faster than compiling everything on Debian
Fixes Arch Linux CI failure: "Unknown compiler(s): gcc not found"
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Fix two issues preventing installer tests from passing:
1. Arch Linux: Sync package database before installing packages
- Fresh Arch containers have no package database (core, extra)
- Added pacman -Sy before pacman -S in both basic prereqs and system deps
- Error was: "warning: database file for 'core' does not exist"
- Applied to both root and non-root installation paths
2. Debian/Ubuntu: Fix package check pattern for architecture suffixes
- dpkg shows packages as "python3-cairo:amd64" not "python3-cairo "
- Changed grep pattern from "^ii $pkg " to "^ii $pkg"
- Now matches packages with or without :amd64/:arm64 suffixes
- Error was: "FAIL: python3-cairo not installed" (even though it was)
Changes:
- install.sh lines 132-134, 233-234: Add pacman -Sy sync before install
- tests/test_installer.sh line 41: Fix dpkg grep pattern
This allows all 5 OS versions to pass:
- Debian 12 (Bookworm)
- Debian Trixie (testing)
- Ubuntu 22.04 LTS
- Ubuntu 24.04 LTS
- Arch Linux (rolling) [NEW]
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Add comprehensive Arch Linux testing to installer-test job.
Changes to .github/workflows/test.yml:
- Add archlinux:latest to test matrix (5 OS versions tested now)
- Set continue-on-error for Arch (rolling release can expose bleeding-edge issues)
- Arch tests run in parallel with Debian/Ubuntu tests
Changes to tests/test_installer.sh:
- Refactored to be OS-agnostic (supports Debian/Ubuntu AND Arch Linux)
- Added OS type detection (apt-get vs pacman)
- Added check_package() helper function (uses dpkg or pacman based on OS)
- Conditional Debian environment setup (DEBIAN_FRONTEND only for Debian/Ubuntu)
- OS-specific package name verification:
- Debian/Ubuntu: python3-gi, python3-dbus, python3-cairo, bluez
- Arch Linux: python-gobject, python-dbus, python-cairo, bluez, bluez-utils
- OS-specific build tool checks (dpkg -l vs pacman -Q)
- Updated summary output to show correct packages per OS
install.sh changes:
- NONE - Arch Linux support already complete and correct!
CI Matrix now tests:
- Debian 12 (Bookworm - current stable)
- Debian Trixie (testing - next release) [non-blocking]
- Ubuntu 22.04 LTS (Jammy)
- Ubuntu 24.04 LTS (Noble)
- Arch Linux (rolling release) [non-blocking] [NEW]
Benefits:
- Validates install.sh Arch support works in practice
- Tests with newer BlueZ/Python versions (rolling release)
- Forward compatibility testing
- Broader Linux distribution coverage
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
bleak doesn't always expose __version__ attribute, causing test failures.
Changed to just verify the module can be imported successfully.
Fixes: AttributeError: module 'bleak' has no attribute '__version__'
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Major architectural improvement: install.sh now handles all prerequisites,
eliminating duplicate logic and making CI test exactly what users run.
## Changes to install.sh:
**1. Added pip_install() helper function (lines 37-49)**
- Detects pip version capabilities
- Uses --break-system-packages flag on pip 23.0+ (Debian 12+, Ubuntu 24.04+)
- Falls back to no flag on pip 22.x (Ubuntu 22.04)
- Single source of truth for all pip operations
- Fixes compatibility across all OS versions
**2. Added basic system package installation (lines 91-128)**
- Checks and installs: python3, python3-pip, git, sudo
- Supports both Debian/Ubuntu (apt-get) and Arch (pacman)
- Only installs missing packages (idempotent)
**3. Changed Reticulum check to auto-install (lines 171-190)**
- Previously: exited with error if Reticulum not found
- Now: automatically installs Reticulum using pip_install()
- Verifies installation succeeded
- Falls back to manual instructions if auto-install fails
**4. Updated all pip install commands to use helper (lines 242, 251)**
- Consistent --break-system-packages handling
- Works on Ubuntu 22.04, Debian 12, Trixie, Ubuntu 24.04
**5. Updated header comment**
- Reflects that script is now self-contained
- Documents all responsibilities
## Changes to tests/test_installer.sh:
**Simplified from 127 lines to 126 lines, but more importantly:**
**Removed (no longer needed):**
- Manual apt-get install of base packages
- Manual pip install of Reticulum
- Duplicate pip compatibility logic
**Kept:**
- Non-interactive environment setup
- Verification tests
- BLE interface import test
**Added:**
- Reticulum verification check
- Updated summary to reflect self-contained nature
## Benefits:
1. ✅ **Single source of truth** - No duplicate pip logic
2. ✅ **CI tests real workflow** - Exactly what users run
3. ✅ **Better user experience** - One command does everything
4. ✅ **Cross-version compatibility** - Works on all OS/pip versions
5. ✅ **Easier maintenance** - Changes in one place
6. ✅ **Self-contained** - install.sh has zero external dependencies
## Testing:
Works across all CI matrix OS versions:
- Ubuntu 22.04 (pip 22.0.2 - no --break-system-packages)
- Debian 12 (pip 23.0+ - requires --break-system-packages)
- Debian Trixie (pip 23.0+ - requires --break-system-packages)
- Ubuntu 24.04 (pip 24.0+ - supports --break-system-packages)
Fixes#4🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Fix two issues preventing Ubuntu 22.04 CI from passing:
1. Remove --break-system-packages flag from pip install
- Ubuntu 22.04 has pip 22.0.2 (flag added in pip 22.3)
- Container runs as root, so no permission issues
- Flag not needed for compatibility
2. Fix /workspace absolute path to use relative path
- GitHub Actions containers use different workspace structure
- Changed to: cd "$(dirname "$0")/.." to navigate from tests/ to repo root
- More portable across CI environments
These changes make the test script compatible with:
- Ubuntu 22.04 (pip 22.0.2)
- Ubuntu 24.04 (pip 24.0+)
- Debian 12 and Trixie
- Any environment where script location may vary
Fixes Ubuntu 22.04 installer-test CI failure.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Ubuntu 24.04 CI was hanging on tzdata interactive timezone prompt.
Changes:
- tests/test_installer.sh: Set DEBIAN_FRONTEND=noninteractive, pre-configure timezone
- tests/test_installer.sh: Add apt-get options to suppress prompts
- .github/workflows/test.yml: Set environment variables in installer-test container
- install.sh: Auto-detect CI environment and enable non-interactive mode
This follows Debian/Ubuntu best practices for containerized environments and
prevents interactive prompts from blocking CI runs.
Fixes Ubuntu 24.04 installer-test CI failure.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Strategy 1: Use pre-compiled system packages instead of building from source
- install.sh: Add python3-gi, python3-cairo to apt-get install
- install.sh: Add python-gobject, python-cairo to pacman install
- install.sh: Install only bleak and bluezero via pip (skip compiled packages)
- README.md: Update dependency instructions with system packages
- README.md: Add explanation of why system packages are preferred
Strategy 2: Add CI integration test for fresh Debian/Ubuntu systems
- tests/test_installer.sh: New integration test script
- .github/workflows/test.yml: Add installer-test job with matrix for Debian 12, Ubuntu 22.04, 24.04
- Tests reproduce real user experience and catch missing dependencies
Benefits:
- Zero compilation time (seconds vs minutes)
- No build tools needed (meson, cmake)
- No dev headers needed (libglib2.0-dev, libcairo2-dev, etc.)
- Faster installation on resource-constrained devices (Raspberry Pi)
- Prevents future dependency documentation issues
Fixes#4🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
The BLE interface now dynamically resolves the interface directory
by checking RNS.Reticulum.configdir when loaded via exec() by Reticulum.
This allows users to specify custom config directories using the
--config flag without encountering import errors.
Changes:
- Update BLEInterface.py to use RNS.Reticulum.configdir when available
- Add fallback to default ~/.reticulum/interfaces for backward compatibility
- Add comprehensive test coverage for config directory resolution
- Update documentation to mention custom config directory support
Fixes#2🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
