@@ -74,6 +74,35 @@ release.
7474- The generated sample-data README now annotates each IQ file as a
7575 * (noise-floor reference)* or * (burst captured)* .
7676- ` .gitignore ` : ` tests/fm_testdata/ ` (deterministic, regenerable output).
77+ - ` tests/test_interrupt_clean.py ` : the clean-interrupt contract as a tested
78+ guarantee on BOTH platforms, replacing the "untested on Windows" caveat in
79+ ` core.py ` . The stub child now records WHICH signal it caught, so the tests
80+ distinguish the clean interrupt (SIGINT on POSIX; SIGBREAK from
81+ CTRL_BREAK_EVENT on Windows, traversing the ` .bat ` launcher layer like the
82+ real tools) from the ` terminate() ` escalation -- previously a broken
83+ CTRL_BREAK path could hide behind a working escalation. Also asserted:
84+ output written by the child's interrupt handler is drained into ` stop() ` 's
85+ result (the no-truncated-capture contract), and a child that ignores the
86+ interrupt is still reaped and reported unclean. The Windows CI leg proves
87+ the Windows path on the next push. (plan:#4 , Phase 3a)
88+ - OS dead-man for handle-mode children (closes the long-standing
89+ ` TODO(os-deadman) ` in ` core.py ` ): the atexit backstop never runs on
90+ SIGKILL / TerminateProcess, so a hard-killed parent could leave a
91+ transmitter on the air. Now the OS itself ends the child when the parent
92+ dies: Linux children set ` PR_SET_PDEATHSIG ` to SIGINT in preexec (the
93+ CLEAN interrupt -- the dead-man is the flush path, with a ` getppid() `
94+ check closing the fork-window race), and Windows children are assigned to
95+ a Job Object with ` KILL_ON_JOB_CLOSE ` (terminating the whole
96+ ` .bat ` -> python tree; the job handle lives on the ` _Process ` ). macOS has
97+ no equivalent primitive; the atexit backstop remains the net there,
98+ documented. Gated exactly like the atexit registry: TX always, RX unless
99+ ` backstop_rx=False ` . ` tests/test_deadman.py ` proves it with a real
100+ SIGKILL of an intermediate parent on POSIX -- the child dies AND its
101+ interrupt handler ran -- plus the opt-out gate; the Windows Job Object
102+ leg proves on the next CI push. (plan:#4 , Phase 3b)
103+ - ` hackrfpy.__version__ ` , resolved from installed package metadata
104+ (` importlib.metadata ` ), with a ` 0.0.0+unknown ` fallback for uninstalled
105+ checkouts. (plan:#7 )
77106- ` examples/fm_demod_to_wav.py ` : the missing last mile -- demodulate a
78107 captured broadcast-FM IQ file to an audible mono 16-bit WAV using only
79108 numpy and the stdlib. Channelize (staged windowed-sinc decimation to
@@ -174,16 +203,35 @@ release.
174203 below 1e-9 (a few hundred taps at 200 kHz); a 2 s reference file demodulates
175204 in ~ 3 s end to end, with output identical to within 1 LSB. Stage timings are
176205 now printed so a future stall is self-locating.
177-
178- <!--
179- Roadmap for this branch (append entries above as each lands):
180- - Added: type annotations across the public API; mypy promoted to a CI gate
181- (makes the shipped py.typed marker accurate).
182- - CI: --cov-fail-under=85; confirm the Windows leg exercises the CTRL_BREAK
183- lifecycle path.
184- - Docs: note that a single HackRF instance is not safe to share across
185- threads (deferred pending broader testing).
186- -->
206+ - ` sweep() ` / ` sweep_to_file() ` silently truncated the TOP of the requested
207+ band: both MHz edges were floored, so ` sweep(433.9e6, 434.1e6) ` ran
208+ ` 433:434 ` and never covered 434.0-434.1 MHz even while warning about the
209+ snap. Edges now snap OUTWARD (floor the low edge, ceil the high edge) so
210+ the swept range always contains the requested band; the warning text says
211+ so. (plan:#2 )
212+ - ` monitor_frequencies() ` reported the MEAN dB of the whole covering sweep
213+ segment as the power "at" a frequency, diluting a narrowband carrier
214+ toward the noise floor (a -20 dB carrier in one bin of a 10-bin segment
215+ read as -74). It now reads the bin covering the frequency (max of that
216+ bin +/-1 for tuning slop). Behavior change for ` examples/channel_monitor.py `
217+ and any monitor consumer: readings of narrowband signals rise to their
218+ true level. (plan:#3 )
219+ - Handle-mode output was not visible to the parent until process exit unless
220+ the child flooded: the drain thread read pipes with ` BufferedReader.read(N) ` ,
221+ which blocks until N bytes (64 KB) accumulate, so small periodic writes --
222+ e.g. hackrf_transfer's ~ 60-byte-per-second stats lines -- sat invisible for
223+ what would be ~ 18 minutes of real capture. Now ` read1() ` : bytes appear in
224+ the drain as soon as the child writes them. Found by the new clean-interrupt
225+ tests; also cut the test suite's wall time roughly in half by removing the
226+ same latency from every stubbed lifecycle test. (Phase 3a finding)
227+ - ` _Process.stop() ` 's escalation fired ` terminate() ` and returned without
228+ reaping: a child that ignored the interrupt could outlive ` stop() ` , and
229+ ` result() ` reported ` returncode None ` . The escalation is now a bounded,
230+ reaped ladder (interrupt -> terminate -> kill), so ` stop() ` always returns
231+ with the child dead and a real exit status. (Phase 3a finding)
232+ - ` from_device() ` guarded its parsed-info invariant with a bare ` assert ` ,
233+ which ` python -O ` strips, letting raw text flow onward; it now raises
234+ ` HackRFDeviceError ` . (plan:#9 )
187235
188236## [ 1.0.0] - 2026-06-16
189237
0 commit comments