[W5TSU]

Closes #75, closes #46, closes #47, closes #80, closes #42.

All four open issues, plus the broken v26.10.6 x86_64 AppImage.

v26.10.6 shipped without an x86_64 AppImage

HackRfDiscovery.cpp used BOARD_ID_UNDETECTED, which Ubuntu 22.04's libhackrf 2021.03.1 (the x86_64 AppImage's build host) calls BOARD_ID_INVALID. Only the tag build uses that header, so only the release broke, and release signing timed out waiting for the missing file. It now uses the number, 0xFF. A new ctest, libhackrf_api_floor, fails when src/ uses a libhackrf name the 2021.03.1 header doesn't declare. This is the second break of this kind (#67).

#75: the demo radio sent a frame after "disconnected"

RadioModel disconnects the simulator's vended connection directly, and SimBackend split that into two queued slots: re-emit disconnected(), then close the gate. A spectrum row could land between them. One handler now closes the gate and then announces the disconnect, the core of upstream aethersdr#6088.

#47: ASan lane

• Fail-safe unkey held until the next tick (Icom). sendUserCommand() read the clock once and queueWrite() stamped the request with a later reading, so the pump skipped the PTT-off as not yet due whenever a millisecond boundary fell between the two. This failed 2 of 40 local runs. It now pumps with a fresh reading. A clock seam reproduces the race on every run: the new check failed before the fix, and 40 of 40 runs pass after it.
• Out of disk. The build had failed since 2026-10-03, with every test linking its own static copy of the core. A new AETHER_SHARED_CORE option (sanitizer CI only) builds the core as one shared library, from upstream ci(sanitizers): link the ASan lane against a shared core so it fits on disk. Principle VIII. aethersdr/AetherSDR#6136. A failed Configure or Build now reaches the tracking issue instead of skipping the whole report chain.

#46: TSan lane

weather_radar_loading_test timed out every week, with a 30 s limit against ~15 s plain and 26 s under ASan. In instrumented builds every TIMEOUT is now ×4 (upstream aethersdr#6208); ordinary builds keep their limits.

#80: TSan on Qt 6.12

The TSan image moved from Qt 6.8.3 to 6.12.0, and four tests (tci_automation, tci_server_review, hl2_tci_signaling, whats_new_dialog) now fail on two reports wholly inside Qt: a lock-order inversion in its TLS backend init, and ~QHttpThreadDelegate racing a main-thread reply delete. Upstream suppressed exactly these, anchored on the Qt frames, in aethersdr#6173; tsan.supp takes those entries verbatim.

#42: HackRF backend, the last two items

• Unplug mid-session. HackRfWorker::streamStopped was declared and handled but never emitted. An unplugged HackRF froze the spectrum, stayed "connected", and refused a reconnect. The backend's 50 ms tick now runs checkStreamHealth(): if libhackrf says the stream ended, or no transfer arrives for 2 s, the backend reports "HackRF stream stopped unexpectedly" and disconnects, in that order (rule 6).
• Windows. scripts/setup/setup-hackrf.ps1 builds libhackrf 2026.01.3 with MSVC against vcpkg's libusb and pthreads, as Great Scott Gadgets' guide does. The installer and portable ZIP ship hackrf.dll, libusb-1.0.dll and pthreadVC3.dll, and REQUIRE_HACKRF=ON gates both the installer and the per-PR Windows build. The firmware asks Windows for WinUSB, so there's no Zadig step.
• Multi-slice receive, the one other v1 goal in HackRF backend — highly experimental (RX+TX, satellite FM/CW) #42's design, isn't in this PR. It moves to HackRF: multi-slice receive across the wide capture #79.

Testing

• Whole tree with AETHER_SHARED_CORE=ON: builds, and 512/512 ctest pass.

• A -fsanitize=thread configure scales weather_radar_loading_test to 120 s; a plain configure leaves it at 30 s.

• Every HackRF source compiles against the libhackrf 2021.03.1 header.

• Unplug, on hardware (HackRF Pro). tools/hackrf_unplug_test.py (new, receive only) resets the board with libhackrf's reset request, so it drops off USB and re-enumerates. It passed 9/9:

  • disconnected in 0.3–0.4 s;
  • still running;
  • reconnected and receiving again.

  On the old build the app never disconnected and refused to reconnect.

• Red then green: sim_backend_test (2 checks), icom_identity_test (1), hackrf_backend_test (2 ordering checks).

• Windows: CI only, not yet run against a HackRF on Windows. The Windows Installer and AppImage workflows are dispatched on this branch before merge.

🤖 Generated with Claude Code