commit ca4e8fe7a26c2851baddc8b099916400afdaef7f Author: Mooneer Salem Date: Sat Oct 31 22:21:30 2026 -0700 Add changelog entry for PR #1614. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit cc806b11ecc864e152610103dc629235678be18a Author: Mooneer Salem Date: Sat Oct 31 22:21:19 2026 -0700 RADE: fill in the high band of the features for 8 kHz mics. Turns on freedv-backend's RADEFeatureExpander when the TX mic's sample rate is 8 kHz or below, so the far end doesn't hear a muffled voice. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit e4b475160655b3fd32596a58c19eac550a2a2343 Author: Mooneer Salem Date: Sun Nov 1 00:57:25 2026 +0000 latest user manual PDF commit b5acb98d9b56017f24e946d15be312b562ca2eb1 Merge: 5427b2a7 710af8ce Author: Mooneer Salem Date: Sat Oct 31 17:32:26 2026 -0700 Merge branch 'v3.0-dev' into ms-rade-v2 commit 710af8ce701330aa8a2f1995911ef1a58832a360 Merge: d6bc4162 4afdc48a Author: Mooneer Salem Date: Sat Oct 31 17:31:44 2026 -0700 Merge pull request #1612 from freedv/test-reporter-focus-selection Test that the Reporter list gaining the focus doesn't select a row commit 4afdc48aba7e87af028b6ae96174896cd2dab1eb Merge: 1cb1f8db 56da90b9 Author: Mooneer Salem Date: Sat Oct 31 17:30:59 2026 -0700 Merge branch 'master' into test-reporter-focus-selection commit 56da90b9b19dfda981fa446935bf83393852a84a Author: Mooneer Salem Date: Sun Nov 1 00:22:19 2026 +0000 latest user manual PDF commit 1e062689dad3645941202b6e157c74cb9b88d1df Author: Mooneer Salem Date: Sat Oct 31 17:18:57 2026 -0700 User manual: Update reference to Experimental Features section. (#1613) * User manual: Update reference to Experimental Features section. * Add PR #1613 to changelog. commit a5b614f65352c2deb892a4d4ebd8c2193a1edb9b Author: Mooneer Salem Date: Sat Oct 31 17:18:25 2026 -0700 Add PR #1613 to changelog. commit 1332ea66e3b9dc3973a8a5b1db799ea081b89706 Author: Mooneer Salem Date: Sat Oct 31 17:17:02 2026 -0700 User manual: Update reference to Experimental Features section. commit 102aa6e48318ec668c2b96c1de669a83ed70dc98 Author: Mooneer Salem Date: Sat Oct 31 17:11:38 2026 -0700 FreeDV Reporter: don't select the row under the mouse on focus without a click; begin 2.4.2 development (#1611) * FreeDV Reporter: only select the row under the mouse on focus if a button is down. OnSetFocus() selected whatever row the mouse was over whenever the list gained the focus, including when that came from the window being brought to the front or a tooltip closing rather than a click. On Linux the selection guard doesn't stop this either, as hovering over a row already allows selections. This caused an occasional gui_reporter_row_actions failure on v3.0-dev ("Selected stations after hovering messages is ['K1ABC']"). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN * Add PR #1611 to changelog. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN * Re-enable dev version tag and six-month build expiry now that 2.4.1 is out. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN * Increment version to 2.4.2 to begin development. Also builds against freedv-backend's main branch again instead of the v1.0.1 release, as was done when 2.4.1 development began (#1473). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN --------- Co-authored-by: Claude Opus 5.5 commit f500c198a37028907046790a02c2dd822debeff4 Author: Mooneer Salem Date: Sat Oct 31 14:33:11 2026 -0700 Increment version to 2.4.2 to begin development. Also builds against freedv-backend's main branch again instead of the v1.0.1 release, as was done when 2.4.1 development began (#1473). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit 5013d1615f5f81123e40b6add14722a3e65493cc Author: Mooneer Salem Date: Sat Oct 31 14:31:02 2026 -0700 Re-enable dev version tag and six-month build expiry now that 2.4.1 is out. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit 9a0412b51ac2ee815f1c60172900a7e0bc05b181 Author: Mooneer Salem Date: Sat Oct 31 14:31:02 2026 -0700 Add PR #1611 to changelog. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit 1cb1f8db9afef78fa739804e757c78d7c725b4ee Author: Mooneer Salem Date: Sat Oct 31 14:26:25 2026 -0700 Test that the Reporter list gaining the focus doesn't select a row. Gives the list the focus while the pointer rests over a row, as bringing the window to the front or a tip window closing can, and checks that nothing ends up selected (#1611). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit f045506c9f6aa3c36861452778ea1aa9bba3a17a Author: Mooneer Salem Date: Sat Oct 31 12:00:14 2026 -0700 FreeDV Reporter: only select the row under the mouse on focus if a button is down. OnSetFocus() selected whatever row the mouse was over whenever the list gained the focus, including when that came from the window being brought to the front or a tooltip closing rather than a click. On Linux the selection guard doesn't stop this either, as hovering over a row already allows selections. This caused an occasional gui_reporter_row_actions failure on v3.0-dev ("Selected stations after hovering messages is ['K1ABC']"). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN (cherry picked from commit 6d2c9e5fc1b78c7e8c1fc1716281ba61d037955e) commit 6d2c9e5fc1b78c7e8c1fc1716281ba61d037955e Author: Mooneer Salem Date: Sat Oct 31 12:00:14 2026 -0700 FreeDV Reporter: only select the row under the mouse on focus if a button is down. OnSetFocus() selected whatever row the mouse was over whenever the list gained the focus, including when that came from the window being brought to the front or a tooltip closing rather than a click. On Linux the selection guard doesn't stop this either, as hovering over a row already allows selections. This caused an occasional gui_reporter_row_actions failure on v3.0-dev ("Selected stations after hovering messages is ['K1ABC']"). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit d6bc416230667301f384b6f43fb2f7d36b3f5948 Author: Mooneer Salem Date: Fri Oct 30 22:58:53 2026 -0700 Restore v3.0-dev's dev version tag and build expiry. The master merge brought over the 2.4.1 release settings. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit 727fec3a399e57e344c52d0f70783577696b9818 Author: Mooneer Salem Date: Fri Oct 30 22:58:52 2026 -0700 Fix v3.0-dev's low sample rate messages and tests after the master merge. The merge took master's validateSoundCardSetup() wholesale, which dropped v3.0-dev's translation wrappers and brought back "Tools->Audio Config..." in place of "Tools->Settings". The GUI tests also still expected microphone/speaker devices below 16 kHz to be refused. - Restore _() around the messages and point to Tools->Settings. - test_low_sample_rate: below 8 kHz is refused for every device; with everything at 8 kHz FreeDV starts with one Low Sample Rate warning that lists only the microphone and speaker, and doesn't repeat it on the next Start. - test_partly_configured: the 8 kHz mic/speaker cases now start with the warning. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit 5427b2a789c91bd662f805c7d56dbb9b94f7a827 Merge: 4a3d37ba 957795c3 Author: Mooneer Salem Date: Fri Oct 30 18:12:46 2026 -0700 Merge branch 'v3.0-dev' into ms-rade-v2 commit 957795c377ba129dfa449919ef67538646cac87d Merge: bb71eedc 321db15f Author: Mooneer Salem Date: Fri Oct 30 17:21:08 2026 -0700 Merge branch 'master' into v3.0-dev commit 321db15f631db102bb4c3f112148a600b130b68b Author: Mooneer Salem Date: Fri Oct 30 16:36:05 2026 -0700 Add freedv-backend bug fixes to the 2.4.1 changelog (#1608) * Add freedv-backend bug fixes to the 2.4.1 changelog. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN * Add freedv-backend PR #78 fixes to the 2.4.1 changelog. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN * Prepare for 2.4.1 release. * Update freedv-backend to point to newly tagged release. * Warn instead of refusing to start when a mic/speaker device is below 16 kHz. FreeDV resamples every device to the rate each step needs, so an 8 kHz mic or speaker (e.g. a Bluetooth headset using HFP's narrowband codec) still works; it just loses audio above 4 kHz. Devices below 8 kHz are still rejected. The warning lists the affected devices and is shown at most once per session; each Start also logs it. Fixes #1610. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN * Avoid listing duplicate device names for the Low Sample Rate warning. --------- Co-authored-by: Claude Opus 5.5 commit 03117b8c5267d8d08416efb2f86551c32f51fd0d Author: Mooneer Salem Date: Fri Oct 30 16:35:07 2026 -0700 [skip ci] Avoid listing duplicate device names for the Low Sample Rate warning. commit cbee286898e27067ac397c90e5c25623834540bd Author: Mooneer Salem Date: Fri Oct 30 16:01:59 2026 -0700 Warn instead of refusing to start when a mic/speaker device is below 16 kHz. FreeDV resamples every device to the rate each step needs, so an 8 kHz mic or speaker (e.g. a Bluetooth headset using HFP's narrowband codec) still works; it just loses audio above 4 kHz. Devices below 8 kHz are still rejected. The warning lists the affected devices and is shown at most once per session; each Start also logs it. Fixes #1610. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit 16947b8d50f3b48bdb78274916478c7e4c580297 Author: Mooneer Salem Date: Fri Oct 30 12:19:22 2026 -0700 Update freedv-backend to point to newly tagged release. commit 77882e493f6218a656e4251b0efd64ac1c6d2a45 Author: Mooneer Salem Date: Fri Oct 30 12:16:28 2026 -0700 Prepare for 2.4.1 release. commit 41a15e60036415b935368b5e3b37c198f56dc893 Author: Mooneer Salem Date: Fri Oct 30 12:18:26 2026 -0700 Add freedv-backend PR #78 fixes to the 2.4.1 changelog. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit 9f66009c70d3236dd327ce279fd8f7f2550dd584 Author: Mooneer Salem Date: Wed Oct 28 13:03:11 2026 -0700 Add freedv-backend bug fixes to the 2.4.1 changelog. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit 4a71ce089abd25b99c22ec736479cc9dfe47a31f Author: Mooneer Salem Date: Tue Oct 27 22:50:41 2026 -0700 Warm up the RADE decoder and FARGAN as well as the encoder The warmup only ran RADE on silence, which never syncs, so the decoder and FARGAN first ran on the first synced frame. On macOS CI's app-folder test that frame took ~1 s every time (40253103674, 40411090564, 40540593975, 40608808684), sometimes enough for the RX loss test to fail. Run a short over through a throwaway encoder/receiver until the decoder produces features, then synthesize once with a throwaway FARGAN. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit 4a3d37ba95ebf3f18c375b4966ff78e269e3f0fb Merge: 9cfb3a12 bb71eedc Author: Mooneer Salem Date: Tue Oct 27 17:14:33 2026 -0700 Merge remote-tracking branch 'origin/v3.0-dev' into ms-rade-v2 commit bb71eedca7c696681388ed5b100d3ee2e3fb346e Merge: 657516e0 5834e2f4 Author: Mooneer Salem Date: Tue Oct 27 17:14:30 2026 -0700 Merge branch 'master' into v3.0-dev # Conflicts: # USER_MANUAL.html # USER_MANUAL.pdf commit 932b90e15f2ea7542fcdfe15254aad667400f4ca Author: Mooneer Salem Date: Wed Oct 28 00:13:12 2026 +0000 latest user manual PDF commit 657516e0d6b2cc784007424ec9710293419dd1e7 Author: Mooneer Salem Date: Tue Oct 27 17:11:27 2026 -0700 test_tone: listen until the test and VOX tones are heard (#1607) * test_tone: listen until the test and VOX tones are heard The 1500 Hz check and the VOX tone check each judged a single one-second capture, so one gap in what was played (seen on Windows CI, and the VOX check once on Linux) failed the test. Listen for up to 10 s, as the 700 Hz check already does, and report the last capture if the tone never shows up. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN * Add #1607 to the release notes Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN --------- Co-authored-by: Claude Opus 5.5 commit 5834e2f45feecab77acf6905882e502ee7641c1c Author: Mooneer Salem Date: Tue Oct 27 17:10:50 2026 -0700 Initialize FreeDV Reporter's selection flag; fix RecordStepTest flake (#1606) * Initialize FreeDV Reporter's selection flag; make RecordStepTest wait for the writer isSelectionPossible_ was never initialized, so on Linux whether a selection was honored before the pointer had been over a row depended on whatever was in memory (caught by UBSan in gui_reporter_numeric_filters). recordsSamplesInOrder queued bursts faster than real time with a fixed 5 ms sleep between them; on Windows runners the writer thread fell far enough behind for the 1 s FIFO to fill and drop blocks. Wait for each burst to be written instead. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN * Add release note for #1606 Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN --------- Co-authored-by: Claude Opus 5.5 commit 3114fc83271f16ffe6f3bc4cde2f359ac5300230 Author: Mooneer Salem Date: Tue Oct 27 17:09:38 2026 -0700 Add #1607 to the release notes Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit 9e9b423a72d3663001ba805c26ccd2ca4075c70a Author: Mooneer Salem Date: Tue Oct 27 17:09:27 2026 -0700 test_tone: listen until the test and VOX tones are heard The 1500 Hz check and the VOX tone check each judged a single one-second capture, so one gap in what was played (seen on Windows CI, and the VOX check once on Linux) failed the test. Listen for up to 10 s, as the 700 Hz check already does, and report the last capture if the tone never shows up. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit dee4b3507c333809729c77e1e3d156e4cf415484 Author: Mooneer Salem Date: Tue Oct 27 16:02:15 2026 -0700 Add release note for #1606 Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit 39733cb70a25c200cd4aff92f48971e7877b7b4e Author: Mooneer Salem Date: Tue Oct 27 15:59:21 2026 -0700 Initialize FreeDV Reporter's selection flag; make RecordStepTest wait for the writer isSelectionPossible_ was never initialized, so on Linux whether a selection was honored before the pointer had been over a row depended on whatever was in memory (caught by UBSan in gui_reporter_numeric_filters). recordsSamplesInOrder queued bursts faster than real time with a fixed 5 ms sleep between them; on Windows runners the writer thread fell far enough behind for the 1 s FIFO to fill and drop blocks. Wait for each burst to be written instead. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit 9cfb3a1288efbfcb3d7bf8a2e582675d998daeef Merge: d1921e0d 9734dd1f Author: Mooneer Salem Date: Tue Oct 27 09:24:05 2026 -0700 Merge remote-tracking branch 'origin/v3.0-dev' into ms-rade-v2 commit 9e7d24c24448bc3affdb652b2a1f1009790e7335 Author: Mooneer Salem Date: Tue Oct 27 16:04:43 2026 +0000 latest user manual PDF commit 9734dd1f67ebeb9d9a07d97cc4c94c70a3dae672 Author: Mooneer Salem Date: Tue Oct 27 15:51:03 2026 +0000 latest user manual PDF commit fca4edc24a8185851b559e3e655dcf05c175d458 Author: Mooneer Salem Date: Tue Oct 27 08:48:00 2026 -0700 Add an option to disable saving and restoring the frequency on start and stop (#1600) * Add an option to not restore the last used frequency on start Settings->Rig Control gets "Set the radio to the last used frequency on start" (/Hamlib/RestoreLastFrequency, on by default, which keeps the current behavior). Unchecked, pushing Start leaves the radio on its current frequency, for both Hamlib and OmniRig, and FreeDV shows that frequency once connected; the mode is still set if FreeDV is allowed to change it. The option is greyed out with "No frequency or mode changes" and while running. Also regenerates freedv.pot and merges it into pt_BR.po (which picks up the strings changed by #1532 too), and documents the option in the manual. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN * Add release note for #1600 Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN * Make the option control returning the radio to its frequency on stop The option added in the previous commit stopped FreeDV setting the radio to the last used frequency on start, but what it was meant to control is the other direction: both rig controllers save the radio's frequency and mode when they connect and set them back when FreeDV stops. Undo the start-time change and pass the option through to the controllers' restore-on-disconnect flag instead. The option is now "Return the radio to its previous frequency and mode on stop" (/Hamlib/RestoreFreqModeOnStop, on by default), and the rig control test checks the radio after Stop with it on and off. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN * Make the option cover both Start and Stop The option now controls both directions: "Save and restore the frequency on start and stop" (/Hamlib/SaveRestoreFrequency, on by default, which keeps the current behavior). Unchecked, Start no longer sets the radio to the frequency FreeDV last used (FreeDV shows the radio's current frequency instead), and Stop no longer returns the radio to the frequency and mode it had before Start. This brings back the Start-time change from the first commit alongside the Stop-time one. The rig control test checks both, with the option on and off. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN --------- Co-authored-by: Claude Opus 5.5 commit 507f94c1a320e814f6b76e66c6ec8ef7ec780499 Merge: af149670 8fc6ba70 Author: Mooneer Salem Date: Tue Oct 27 08:47:26 2026 -0700 Merge branch 'v3.0-dev' into restore-last-freq-option commit 8fc6ba70d7d0823aae32b0596d4ecd616f548c44 Merge: 55998549 8fac7565 Author: Mooneer Salem Date: Tue Oct 27 08:46:27 2026 -0700 Merge pull request #1602 from freedv/fix-rx-mute-latency Don't believe implausible audio latencies when going from TX to RX (#1601) commit 8fac756502819cf43d88f2b49c060ba00887a696 Merge: b3745a5e 296143dd Author: Mooneer Salem Date: Tue Oct 27 08:46:12 2026 -0700 Merge branch 'master' into fix-rx-mute-latency commit 296143dd6a94aa074a3018679dc4f594b9e7c5da Author: Mooneer Salem Date: Tue Oct 27 08:44:51 2026 -0700 Fix linting issue from previous PR. (#1605) commit b2af29485166e69257c4a4a0d053228e087d44ca Author: Mooneer Salem Date: Tue Oct 27 08:44:42 2026 -0700 Add PR #1605 to changelog. commit af149670309e88e05815af48a9f97a57e451e3b5 Merge: eca18c31 55998549 Author: Mooneer Salem Date: Tue Oct 27 08:40:38 2026 -0700 Merge branch 'v3.0-dev' into restore-last-freq-option commit 559985496ea46631f5efd5ab32e4e5ce9e454a29 Author: Mooneer Salem Date: Tue Oct 27 08:39:21 2026 -0700 Save the config on Stop from the UI thread (#1604) * Save the config on Stop from the UI thread The save added by #1532 ran on the thread doing the Stop and read settings (e.g. the reporting frequency) that the UI thread can change at the same time, which TSan reports in the rig control tests. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN * Add #1604 to the release notes Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN --------- Co-authored-by: Claude Opus 5.5 commit 45a1c38d202a466e14f4ff6e1ec8c8fcb034a4fa Author: Mooneer Salem Date: Tue Oct 27 08:37:59 2026 -0700 Fix linting issue from previous PR. commit b3745a5ecbeb6e504b08e3be42bb4cf70def8f6c Author: Mooneer Salem Date: Tue Oct 27 07:50:38 2026 -0700 Build the serial test shim with CMake too Both shims are now built by one CMake function into test/gui in the build folder, and found with freedvtest.shim_path() ( overrides where). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit 8571ece1a85fa69dab52301bc2a91596b9f64bb3 Author: Mooneer Salem Date: Tue Oct 27 07:48:15 2026 -0700 Build the PulseAudio test shim with CMake instead of from the test Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit 19e2604ab00ab3ece2962ba4186f65767ad23052 Merge: 09465acd dbbc7aec Author: Mooneer Salem Date: Tue Oct 27 07:45:39 2026 -0700 Merge remote-tracking branch 'origin/master' into fix-rx-mute-latency # Conflicts: # USER_MANUAL.pdf commit 349358c3cfbb9504942a99f4ac6a8a9c8ac1f6b5 Author: Mooneer Salem Date: Tue Oct 27 07:40:19 2026 -0700 Add #1604 to the release notes Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit 1cb29a72bbfccf16dc0e08e08b41eef2c82d4012 Author: Mooneer Salem Date: Tue Oct 27 07:40:08 2026 -0700 Save the config on Stop from the UI thread The save added by #1532 ran on the thread doing the Stop and read settings (e.g. the reporting frequency) that the UI thread can change at the same time, which TSan reports in the rig control tests. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit dbbc7aec418396f2a750a1e0bbfe69b5f270d5e1 Author: Mooneer Salem Date: Tue Oct 27 14:39:49 2026 +0000 latest user manual PDF commit 04a5228d2e2d2e06d6081b8d6a2e72048bdd4053 Author: Mooneer Salem Date: Tue Oct 27 07:37:30 2026 -0700 Don't believe implausible audio latencies when going from TX to RX (#1601) (#1603) * Don't believe implausible audio latencies when going from TX to RX (#1601) Some PipeWire versions report about 2055 seconds of latency for record streams. FreeDV resumes RX that long after the radio confirms PTT off, so RX stayed muted, and as the mute wasn't cleared on Start, Stop and Start didn't help either. Treat any latency over a second (or below zero) as a second (or zero), with a warning in the log. This applies to the output device's latency before dropping PTT too, which a bogus value would have held up in the same way. Also clear any leftover RX mute when starting. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN * Add release note for #1602/#1603 Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN * PR is a fix for a previous bugfix, update changelog accordingly. --------- Co-authored-by: Claude Opus 5.5 commit 0711f64a1c4c75c94d4b401772cbe0e95c179950 Author: Mooneer Salem Date: Tue Oct 27 01:06:42 2026 -0700 PR is a fix for a previous bugfix, update changelog accordingly. commit 796c032a5433c60e6b0f673522d64f91e22357db Author: Mooneer Salem Date: Tue Oct 27 00:59:58 2026 -0700 Add release note for #1602/#1603 Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit 09465acd953f449df76d6f112f2edb5f50ecb95e Author: Mooneer Salem Date: Tue Oct 27 00:59:56 2026 -0700 Add release note for #1602/#1603 Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit b978746b7a2be5fcc58b38f29e2454590c002c4b Author: Mooneer Salem Date: Tue Oct 27 00:56:53 2026 -0700 Don't believe implausible audio latencies when going from TX to RX (#1601) Some PipeWire versions report about 2055 seconds of latency for record streams. FreeDV resumes RX that long after the radio confirms PTT off, so RX stayed muted, and as the mute wasn't cleared on Start, Stop and Start didn't help either. Treat any latency over a second (or below zero) as a second (or zero), with a warning in the log. This applies to the output device's latency before dropping PTT too, which a bogus value would have held up in the same way. Also clear any leftover RX mute when starting. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit 8cbce93f48811a197cc2f17227d1dc2e4c299102 Author: Mooneer Salem Date: Tue Oct 27 00:56:53 2026 -0700 Test that RX resumes after TX when PulseAudio reports a bogus latency New rig control scenario, receive_after_transmit: with PTT through Hamlib (rigctld), FreeDV transmits, and the transmission is then played back to it as a reply, which it has to sync on; again after Stop and Start. On Linux, pulse_shim.c (loaded with LD_PRELOAD) makes PulseAudio report the 2055 s record latency from #1601; elsewhere the scenario checks RX after TX with the real latency. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit 9bef2c22ee56b03b9856c6277c2a6616df868cea Author: Mooneer Salem Date: Tue Oct 27 00:56:53 2026 -0700 Don't believe implausible audio latencies when going from TX to RX (#1601) Some PipeWire versions report about 2055 seconds of latency for record streams. FreeDV resumes RX that long after the radio confirms PTT off, so RX stayed muted, and as the mute wasn't cleared on Start, Stop and Start didn't help either. Treat any latency over a second (or below zero) as a second (or zero), with a warning in the log. This applies to the output device's latency before dropping PTT too, which a bogus value would have held up in the same way. Also clear any leftover RX mute when starting. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit eca18c31ad9566b771d15b349c5af33c7e53debf Author: Mooneer Salem Date: Mon Oct 26 21:25:50 2026 -0700 Make the option cover both Start and Stop The option now controls both directions: "Save and restore the frequency on start and stop" (/Hamlib/SaveRestoreFrequency, on by default, which keeps the current behavior). Unchecked, Start no longer sets the radio to the frequency FreeDV last used (FreeDV shows the radio's current frequency instead), and Stop no longer returns the radio to the frequency and mode it had before Start. This brings back the Start-time change from the first commit alongside the Stop-time one. The rig control test checks both, with the option on and off. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit 0adfe7bc2258e1a5eea13cc30be3b7aeb618bc2f Author: Mooneer Salem Date: Mon Oct 26 21:17:12 2026 -0700 Make the option control returning the radio to its frequency on stop The option added in the previous commit stopped FreeDV setting the radio to the last used frequency on start, but what it was meant to control is the other direction: both rig controllers save the radio's frequency and mode when they connect and set them back when FreeDV stops. Undo the start-time change and pass the option through to the controllers' restore-on-disconnect flag instead. The option is now "Return the radio to its previous frequency and mode on stop" (/Hamlib/RestoreFreqModeOnStop, on by default), and the rig control test checks the radio after Stop with it on and off. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit 592fdfaf7ae4bcecfabd9ab4928e9a74081ee170 Author: Mooneer Salem Date: Mon Oct 26 20:58:21 2026 -0700 Add release note for #1600 Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit 71aea344d18ee3dc6c2bfc3872e02b5aeb48ff03 Author: Mooneer Salem Date: Mon Oct 26 20:58:09 2026 -0700 Add an option to not restore the last used frequency on start Settings->Rig Control gets "Set the radio to the last used frequency on start" (/Hamlib/RestoreLastFrequency, on by default, which keeps the current behavior). Unchecked, pushing Start leaves the radio on its current frequency, for both Hamlib and OmniRig, and FreeDV shows that frequency once connected; the mode is still set if FreeDV is allowed to change it. The option is greyed out with "No frequency or mode changes" and while running. Also regenerates freedv.pot and merges it into pt_BR.po (which picks up the strings changed by #1532 too), and documents the option in the manual. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit 7ad7cc4fe69d5751efe8f183265211a1bad62295 Author: Mooneer Salem Date: Tue Oct 27 03:27:23 2026 +0000 latest user manual PDF commit 7e9a23735238503cc2fb1e6ff8475a8832a14157 Author: Mooneer Salem Date: Mon Oct 26 20:25:09 2026 -0700 Fix the wxWidgets 3.0 build of LevelMeterLed; retry the AudioDeviceCmdlets install (#1599) * Fix the wxWidgets 3.0 build of LevelMeterLed; retry the AudioDeviceCmdlets install wxWidgets 3.0 (Ubuntu 22.04) has no wxWindow::UseBackgroundColour(); use UseBgCol(), which 3.0 has and which newer versions keep (UseBackgroundColour() just calls it). The Windows CI's one-shot Install-Module of AudioDeviceCmdlets from the PowerShell Gallery failed in two jobs of the same run; retry it up to five times with a growing delay. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN * Add PR #1599 to changelog. --------- Co-authored-by: Claude Opus 5.5 commit befb5a3279e467ba6efb1f65ffdc89be7823241a Author: Mooneer Salem Date: Mon Oct 26 20:25:00 2026 -0700 Add PR #1599 to changelog. commit ad7b9ee6d8cd5fa713aaa025f635089badce36a0 Author: Mooneer Salem Date: Mon Oct 26 17:07:37 2026 -0700 Fix the wxWidgets 3.0 build of LevelMeterLed; retry the AudioDeviceCmdlets install wxWidgets 3.0 (Ubuntu 22.04) has no wxWindow::UseBackgroundColour(); use UseBgCol(), which 3.0 has and which newer versions keep (UseBackgroundColour() just calls it). The Windows CI's one-shot Install-Module of AudioDeviceCmdlets from the PowerShell Gallery failed in two jobs of the same run; retry it up to five times with a growing delay. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit ff257393ca1cee54e9b73f81722102cbc21e3ee0 Author: Mooneer Salem Date: Mon Oct 26 23:27:53 2026 +0000 latest user manual PDF commit d1921e0d416cd4c51949dc7dfd8815691bb9dfe0 Merge: 8337b627 cad62515 Author: Mooneer Salem Date: Mon Oct 26 15:13:45 2026 -0700 Merge branch 'v3.0-dev' into ms-rade-v2 commit a8dd52ad3715f313b13d4067a57c4d5821b4affe Merge: e50c4efc a2d5365f Author: Mooneer Salem Date: Mon Oct 26 13:20:04 2026 -0700 Merge branch 'master' into v3.0-dev commit a2d5365fbbb651be67959d47c351573f9374d5e0 Author: Mooneer Salem Date: Mon Oct 26 13:16:05 2026 -0700 Add missed attribution to changelog. (#1598) commit ff37c549e00a27c0e40d53de7b136b3c19d3bba8 Author: Mooneer Salem Date: Mon Oct 26 13:15:32 2026 -0700 Add missed attribution to changelog. commit 8337b6279c5f73855737949c59e22b35d10553a2 Merge: edda0fb4 e50c4efc Author: Mooneer Salem Date: Sun Oct 25 18:01:48 2026 -0700 [full ci] Merge branch 'v3.0-dev' into ms-rade-v2 # Conflicts: # .github/workflows/cmake-linux.yml # .github/workflows/cmake-macos.yml # .github/workflows/cmake-windows.yml # CMakeLists.txt # cmake/BuildFreeDVBackend.cmake # po/freedv.pot # src/gui/dialogs/dlg_options.cpp # src/main.cpp # test/gui/test_main_window.py commit f80b68be79d60149b0e947f60f509b07cb3bd157 Author: Mooneer Salem Date: Mon Oct 26 01:01:29 2026 +0000 latest user manual PDF commit e50c4efc5670ff9846286cee0f45a5063e1f6ca5 Merge: 71eb68ee 2d14abf6 Author: Mooneer Salem Date: Sun Oct 25 17:55:07 2026 -0700 Merge branch 'master' into v3.0-dev # Conflicts: # USER_MANUAL.pdf commit 2d14abf6918dbed59cc4b0688cfab29fa024a258 Author: Mooneer Salem Date: Sun Oct 25 17:53:31 2026 -0700 WASAPI: only request raw mode on devices that support it (#1594) * WASAPI: only request raw mode on devices that support it Requesting AUDCLNT_STREAMOPTIONS_RAW on an endpoint without raw mode support can make IAudioClient::Initialize() fail with E_INVALIDARG (0x80070057), reported as "Could not initialize AudioClient". Check PKEY_Devices_AudioDevice_RawProcessingSupported first and fall back to AUDCLNT_STREAMOPTIONS_NONE when it is absent or false. Co-Authored-By: Claude Opus 5.5 * Add PR #1594 to changelog. --------- Co-authored-by: Claude Opus 5.5 commit 71eb68eebdfd1b9fb2e064a3824ab1aa423c53a9 Author: Mooneer Salem Date: Sun Oct 25 17:53:09 2026 -0700 Make the Hamlib test's "VFO unavailable" fault flag atomic (#1596) setsFrequencyDespiteUnavailableVfoInformation sets and clears RadioFaults::getVfoUnavailable on the main thread while the fake get_vfo reads it on the controller's Hamlib thread, which may still be polling the radio. The fault counters beside it are already atomic; this flag was left a plain bool, so TSan reported a race on Linux. Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN Co-authored-by: Claude Opus 5.5 commit 0b0aabc42b4b97c45e692273c10ef566f2ea07f8 Author: Mooneer Salem Date: Sun Oct 25 15:43:31 2026 -0700 Make the Hamlib test's "VFO unavailable" fault flag atomic setsFrequencyDespiteUnavailableVfoInformation sets and clears RadioFaults::getVfoUnavailable on the main thread while the fake get_vfo reads it on the controller's Hamlib thread, which may still be polling the radio. The fault counters beside it are already atomic; this flag was left a plain bool, so TSan reported a race on Linux. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit 60398b91887152b7b1cb35cfb25e7b93ae646fbd Author: Mooneer Salem Date: Sun Oct 25 20:46:41 2026 +0000 latest user manual PDF commit 4a5e09e35b6301c2dcc26b1ee7a70f349a3a7ecc Author: Mooneer Salem Date: Sun Oct 25 13:46:05 2026 -0700 Add PR #1594 to changelog. commit 8bb9997bbe61ef35d564a6f018be0f2102e4be9b Author: Mooneer Salem Date: Sun Oct 25 13:43:12 2026 -0700 Fix two TSan-reported test races; show the notary log when macOS notarization fails (#1595) * Fix two data races in tests that TSan reports - PulseAudioDeviceTest read a running device's name without the mainloop lock that StreamMovedCallback_ holds while changing it on PulseAudio's thread. Read it under that lock. - logsHamlibDebugMessagesAtEachLevel turned ulog's quiet flag back on while the async logger's thread could be reading it. Wait for that thread to go idle first. TSan aborted the test on macOS, and since stderr was redirected at the time, the report never reached the log. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN * Show Apple's notary log when macOS notarization fails notarytool submit --wait exits successfully even when the submission is rejected, so the release target went on to staple and failed there with "Record not found", without saying why Apple rejected the build. Submit through a script that checks the result's status, prints the notary log when it isn't Accepted, and fails the build at that point. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN * Add PR #1595 to changelog. --------- Co-authored-by: Claude Opus 5.5 commit c34637031c80479d8aa19693d662babe4d8fd84e Author: Mooneer Salem Date: Sun Oct 25 13:42:59 2026 -0700 Add PR #1595 to changelog. commit 1f764ca9b40a626477d523153770f456028905db Author: Mooneer Salem Date: Sun Oct 25 19:17:28 2026 +0000 latest user manual PDF commit c7af9fe3bc15cba9d443165e7c9624ee82153945 Author: Mooneer Salem Date: Sun Oct 25 11:57:11 2026 -0700 Show Apple's notary log when macOS notarization fails notarytool submit --wait exits successfully even when the submission is rejected, so the release target went on to staple and failed there with "Record not found", without saying why Apple rejected the build. Submit through a script that checks the result's status, prints the notary log when it isn't Accepted, and fails the build at that point. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit 61d687683de26d015b8abde72a1cc7ce43cf659d Author: Mooneer Salem Date: Sun Oct 25 18:38:15 2026 +0000 latest user manual PDF commit 4ba5966d100c94d0a2030e74fc3f9ccaf1c3ebc8 Author: Mooneer Salem Date: Sun Oct 25 11:32:14 2026 -0700 AppImage: stop forcing GDK_BACKEND=x11 (#1593) * AppImage: stop forcing GDK_BACKEND=x11. linuxdeploy's GTK plugin adds an AppRun hook that exports GDK_BACKEND=x11, overwriting any value the user set. On Wayland desktops this runs the bundled GTK 3.24.33 (from the Ubuntu 22.04 build host) under XWayland, where a user on Ubuntu 26.04 saw plot panels flicker after pressing Start and the Settings dialog draw text over itself after switching tabs. Removing just that export from the extracted AppImage fixed both; a native build (system or bootstrapped wxWidgets) under X11 was fine, so the problem is the bundled GTK stack on X11. Delete the export from the plugin before linuxdeploy runs it so GTK chooses the backend itself (Wayland when available, else X11) and GDK_BACKEND is respected again, and fail the build if the hook still sets it. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Add PR #1593 to changelog. --------- Co-authored-by: Claude Opus 5.5 commit 33739308f4ffee93a6a8489eaf961a4f5cafe29d Author: Mooneer Salem Date: Sun Oct 25 11:31:54 2026 -0700 Add PR #1593 to changelog. commit c0096f07fa42e9da675183059d05afd225979970 Author: Mooneer Salem Date: Sun Oct 25 11:28:25 2026 -0700 Fix two data races in tests that TSan reports - PulseAudioDeviceTest read a running device's name without the mainloop lock that StreamMovedCallback_ holds while changing it on PulseAudio's thread. Read it under that lock. - logsHamlibDebugMessagesAtEachLevel turned ulog's quiet flag back on while the async logger's thread could be reading it. Wait for that thread to go idle first. TSan aborted the test on macOS, and since stderr was redirected at the time, the report never reached the log. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_019q8F5ZSDKDxXJoLJTFQ8wN commit 32d9691b6236a2c42a41e7c877b25f9fde74bdbe Author: Mooneer Salem Date: Sun Oct 25 10:13:09 2026 -0700 WASAPI: only request raw mode on devices that support it Requesting AUDCLNT_STREAMOPTIONS_RAW on an endpoint without raw mode support can make IAudioClient::Initialize() fail with E_INVALIDARG (0x80070057), reported as "Could not initialize AudioClient". Check PKEY_Devices_AudioDevice_RawProcessingSupported first and fall back to AUDCLNT_STREAMOPTIONS_NONE when it is absent or false. Co-Authored-By: Claude Opus 5.5 commit 2784368d04d358829a83fd5b7f2334a5a5436c01 Author: Mooneer Salem Date: Sun Oct 25 08:32:12 2026 -0700 AppImage: stop forcing GDK_BACKEND=x11. linuxdeploy's GTK plugin adds an AppRun hook that exports GDK_BACKEND=x11, overwriting any value the user set. On Wayland desktops this runs the bundled GTK 3.24.33 (from the Ubuntu 22.04 build host) under XWayland, where a user on Ubuntu 26.04 saw plot panels flicker after pressing Start and the Settings dialog draw text over itself after switching tabs. Removing just that export from the extracted AppImage fixed both; a native build (system or bootstrapped wxWidgets) under X11 was fine, so the problem is the bundled GTK stack on X11. Delete the export from the plugin before linuxdeploy runs it so GTK chooses the backend itself (Wayland when available, else X11) and GDK_BACKEND is respected again, and fail the build if the hook still sets it. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit b8d92621dcea6e6b20c377f650f0e92b8c4f671e Author: Mooneer Salem Date: Sun Oct 25 08:25:34 2026 -0700 Extend branch coverage of FreeDV to ~92% (#1588) * Recover branch coverage with extended reporter and settings GUI scenarios. - Rewrite ConfigurationDataElementTest around a DefineConfigUnitTest macro registered per element in src/config. - Add reporter scenarios/probes: filters and saved messages, status-message combo behaviour, row actions including message-column hovering and list mouse behaviour; frequency-list keyboard probes in the settings scenario. - Fix MsgListPopup PageUp selection when the popup list is open. CI-chain branch coverage: 89.6% -> 90.1% (12891/14311). * Extend rig and audio unit tests to fault paths; open serial ports without stealing the terminal * Cover PTT input port fault paths; open serial ports without stealing the terminal * Recover configuration store and element conversion branches in unit tests. * Recover plot fault, colour and translated-layout branches with new GUI scenarios. * Cover reporter rows, voice keyer cache, TxRxThread and clean shutdown paths * Cover kHz unit switching, none-output input test and RTS keying test * Extend setup wizard coverage: none/unplug audio paths, baud fallback and range filtering, toggle handlers, network-rig and partial WSJT-X imports * Extend GUI tests for ongui.cpp branches * Extend unit recording test with timed and feature-capture runs * Cover heard-list updates when a logged entry is heard again * Keep wxWidgets log messages' UTF-8 text alive while logging them. msg.ToUTF8() returns a temporary buffer, so the pointer taken from it was left dangling by the time it reached the log. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Don't assert when quitting with a message/frequency list open. Destroying a combo whose list is open makes ~wxComboCtrlBase() hide the popup, which asks MsgListPopup for its value. By then the wxComboCtrl part of the object is gone, so GetComboCtrl()'s cast asserts, and the assert dialog keeps FreeDV from exiting. Use the base class pointer. With that, and wxWidgets' log messages no longer shown in dialogs, the GUI tests no longer need to close the frequency list or give back the settings folder before quitting, so they now cover both. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Let three tests fail when they should. - test_ut_recording.sh checked tee's exit status rather than FreeDV's, so a FreeDV run that crashed partway through still passed. Its ctest pass expression was also just "PASS", which anything FreeDV printed could match, and its sync check looked at every run's output rather than the receive run's. - gui_rig_control_serial_ptt was never registered with ctest. - aMovedStreamCarriesOnWithTheNewDeviceName failed now and then because PipeWire sometimes accepts a move without carrying it out; ask again until the stream has moved, and fail if it never does. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Put back the GUI test checks the rewrites dropped. - column_filters: the menu's exact contents and order, which comparison is ticked, Clear Column Filter being enabled once there's a filter, filters on two columns at once, Cancel leaving the filter alone, and clearing one filter leaving the other. - status_message: editing a saved message after adding one has moved it down, rather than before. - audio: refreshing the TX Audio tab's lists refreshes the RX Audio tab's too, and starting with the transmit devices unset as well as saved as "none". Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Keep a typed status message when the saved list closes, and fix tests failing on macOS and Windows. FreeDV: opening the saved status messages and closing them again without picking one emptied the message box on Linux and Windows, since with nothing highlighted the list's value was "". It's now what's in the box. status_message expected the old behaviour; macOS already kept the text. Tests that only failed outside Linux: - selected_khz, frequency_in_khz and frequency_list_from_main_window expected kHz frequencies without the thousands separators that the region adds on macOS and Windows (e.g. 14,236.0); leave them out, as the other tests already do. - config_migration_failure is skipped where there's no migration (only Linux has one). - missing_port named a /dev port on Windows, where FreeDV only checks COM names, so Hamlib reported it instead; use a COM port there. - logsHamlibDebugMessagesAtEachLevel captured to /tmp, which Windows doesn't have; use the temporary folder, in binary mode so that no carriage returns are added. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Keep the plots' arrow key checks' double-click clear of the Spectrum's labels. On macOS the Spectrum is narrower (407 px against 451 on Linux) and its dB labels take up its first 13%, so a double-click at 12% across landed on them and tuned nowhere. The Down presses then started from the top and ran out at 2400 Hz. Double-click a quarter of the way across, as the wheel checks already do, and allow enough presses to get to the bottom from there. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Don't use a PulseAudio stream the sound server put on another device. Streams aren't connected with PA_STREAM_DONT_MOVE, so that they can be moved between devices later, but that also lets the sound server put a new one elsewhere when the device it's asked for isn't there: PipeWire plays to the default device instead of a sink that's missing (1.6.9), and on 1.6.2 recording from a speaker's name isn't refused either. Once a stream is ready, check that it's on the device asked for; if not, say so and disconnect it. aPlaybackDeviceWithNoSinkReportsTheRefusal covers the playback case, which kept playing on the default device before this. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Only log when a PulseAudio stream ends up on another device, as on master. Match #1591: leave the stream connected and don't report an audio error, just note in the log where the sound server put it. The playback test for a missing sink goes, as nothing's refused any more, and the recording test accepts what PipeWire 1.6.2 does (recording from another device instead of refusing) so long as the stream is somewhere and nothing else was reported. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Skip config_migration_failure where FreeDV is built without migration. Configuration files are only migrated with wxWidgets 3.3 or later, and the ubuntu-26.04 jobs build with the system's 3.2. Those jobs never got as far as the GUI tests before, as the unit tests failed first. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Stop the radio recording budget at zero instead of wrapping it. g_recFromRadioSamples is unsigned, so a write that ran past the end of the budget wrapped it around and the "<= 0" check never saw the end. Writes after the end also kept wrapping it, which is what failed radioRecordingStopsWhenItsBudgetIsUsedUp on macOS CI. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Don't assert on macOS when quitting with the frequency list open. Destroying the main window destroys the frequency box, and ~wxComboCtrlBase() then closes its open list. On macOS that works out where the mouse is relative to the window, which by then is no longer a wxTopLevelWindow, so it asserts ("TopLevel Window missing") and the assert dialog keeps FreeDV from exiting. Close the list at the start of ~MainFrame() instead. Also, on macOS a key the list doesn't use closes it (wxPopupFocusHandler::OnChar), so frequency_list_from_main_window now reopens the empty list for each key. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Update tests for the fixes merged from master. - Drop this branch's radioRecordingStopsWhenItsBudgetIsUsedUp, which #1592 brought in again (in a version that always ends the budget mid-write). - no_udp: Log QSO now stays off without UDP, even with callsigns heard (#1589). - receive_reports: a restart no longer lists the station twice (#1589). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Add PR #1588 to changelog. --------- Co-authored-by: Claude Opus 5.5 commit 42156edd50a6910fd293b523d1a15078355b00cc Author: Mooneer Salem Date: Sun Oct 25 08:25:23 2026 -0700 Add PR #1588 to changelog. commit 503c8c77023ebb78e5f19cdff3bc2f9fe1353ff0 Author: Mooneer Salem Date: Sun Oct 25 03:12:20 2026 +0000 latest user manual PDF commit 5de22f3d5c3c00664db889a0bafe2053921c55dc Author: Mooneer Salem Date: Sat Oct 24 20:11:02 2026 -0700 Update tests for the fixes merged from master. - Drop this branch's radioRecordingStopsWhenItsBudgetIsUsedUp, which #1592 brought in again (in a version that always ends the budget mid-write). - no_udp: Log QSO now stays off without UDP, even with callsigns heard (#1589). - receive_reports: a restart no longer lists the station twice (#1589). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit 339f575abb2fc642a31cc71294c1fc47615fa103 Merge: 9595f819 085783da Author: Mooneer Salem Date: Sat Oct 24 20:06:00 2026 -0700 Merge branch 'v3.0-dev' into ms-cov-extend commit 085783da8f34ba152ae76b6abe05e11b300f0909 Merge: 048d437c b679c400 Author: Mooneer Salem Date: Sat Oct 24 20:05:38 2026 -0700 Merge branch 'master' into v3.0-dev commit b679c4005e083814d154dfccb3db94e178855ee9 Author: Mooneer Salem Date: Sat Oct 24 19:58:53 2026 -0700 Stop the radio recording budget at zero instead of wrapping it (#1592) * Stop the radio recording budget at zero instead of wrapping it. g_recFromRadioSamples is unsigned, so a write that ran past the end of the budget wrapped it around, the "<= 0" check never saw the end and recording was never reported finished. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Add PR #1592 to changelog. --------- Co-authored-by: Claude Opus 5.5 commit 467165ac2c737079d7bda758e703d756bfc30f15 Author: Mooneer Salem Date: Sat Oct 24 19:58:12 2026 -0700 Add PR #1592 to changelog. commit c68cb98e66b8c4cbb2ea50083f173e18ba746137 Merge: 1e12a7a2 937b00d1 Author: Mooneer Salem Date: Sat Oct 24 19:57:31 2026 -0700 Merge branch 'master' into ms-record-budget-underflow commit 937b00d126c7c0152b4e58b1948bdff520f20982 Author: Mooneer Salem Date: Sat Oct 24 19:57:13 2026 -0700 Log when the sound server puts a PulseAudio stream on another device (#1591) * Don't use a PulseAudio stream the sound server put on another device. Streams aren't connected with PA_STREAM_DONT_MOVE, so that they can be moved between devices later, but that also lets the sound server put a new one elsewhere when the device it's asked for isn't there: PipeWire plays to the default device instead of a sink that's missing. Once a stream is ready, check that it's on the device asked for; if not, say so and disconnect it. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Only log when a PulseAudio stream ends up on another device. Leave the stream connected and don't report it as an audio error; just note in the log where the sound server put it. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Add PR #1591 to changelog. --------- Co-authored-by: Claude Opus 5.5 commit 0051e67cad9d98d7292060f2b53efc69a0cd226b Author: Mooneer Salem Date: Sat Oct 24 19:57:00 2026 -0700 Add PR #1591 to changelog. commit 47d5ca5422175601594749b1c4b00f38b710bcd6 Merge: a238dda8 73f17460 Author: Mooneer Salem Date: Sat Oct 24 19:56:21 2026 -0700 Merge branch 'master' into ms-pulse-wrong-device commit 73f17460c2346f49c2dfaa4501dcd3d4c77a53b4 Author: Mooneer Salem Date: Sat Oct 24 19:55:57 2026 -0700 [skip ci] Keep a typed FreeDV Reporter status message when the saved list closes (#1590) * Keep a typed status message when the saved list is closed without a pick. Opening FreeDV Reporter's saved status messages and closing the list again without choosing one emptied the message box on Linux and Windows: the combo takes the list's value as it closes, and with nothing highlighted that was "". Leave what's in the box alone instead. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Add PR #1590 to changelog. --------- Co-authored-by: Claude Opus 5.5 commit d45758dd4f82fe9ab0231318f53f7651dc06a555 Author: Mooneer Salem Date: Sat Oct 24 19:55:24 2026 -0700 Add PR #1590 to changelog. commit 688cbfd706ab254fd6436dc294c3228618742f91 Merge: 56ee996c 32c8f46f Author: Mooneer Salem Date: Sat Oct 24 19:54:28 2026 -0700 Merge branch 'master' into ms-status-message-keep-text commit 32c8f46f6dc86e8d7f4b48f456888931ec5c669e Author: Mooneer Salem Date: Sat Oct 24 19:54:15 2026 -0700 [skip ci] Fix duplicate stations heard after restart, and Log QSO dropping contacts without UDP (#1589) * Don't list stations heard more than once after a restart, or offer Log QSO without a logger. Stations heard: the CSV log gets a row every time a callsign is decoded, but on the air repeated decodes of the same station on the same frequency only update the top row's SNR. Restoring the list at startup inserted every row, so a station heard for a while came back as many identical entries. Collapse them the same way when restoring. Log QSO: on the air the button is only enabled when there's a UDP logger to send contacts to, but when stopped it was enabled whenever the list of stations heard wasn't empty. With UDP logging off, confirming the log entry then threw the contact away without a word. Require UDP logging when stopped as well, and recheck after Settings is closed. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Add PR #1589 to changelog. * Clean up function naming since Log QSO enablement doesn't depend on modem execution. --------- Co-authored-by: Claude Opus 5.5 commit ba0fa85ad693343ef0cd96c51b8e943312de3409 Author: Mooneer Salem Date: Sat Oct 24 19:53:16 2026 -0700 Clean up function naming since Log QSO enablement doesn't depend on modem execution. commit 326a43db3476dfff65c17f48b810839a038932bb Author: Mooneer Salem Date: Sat Oct 24 19:53:11 2026 -0700 Add PR #1589 to changelog. commit 9595f81991314758d12c60b96e912a910e0bb8fa Author: Mooneer Salem Date: Sat Oct 24 19:06:47 2026 -0700 Don't assert on macOS when quitting with the frequency list open. Destroying the main window destroys the frequency box, and ~wxComboCtrlBase() then closes its open list. On macOS that works out where the mouse is relative to the window, which by then is no longer a wxTopLevelWindow, so it asserts ("TopLevel Window missing") and the assert dialog keeps FreeDV from exiting. Close the list at the start of ~MainFrame() instead. Also, on macOS a key the list doesn't use closes it (wxPopupFocusHandler::OnChar), so frequency_list_from_main_window now reopens the empty list for each key. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit 1e12a7a281ae28e566f7a55f8e012e3da57ced7e Author: Mooneer Salem Date: Sat Oct 24 18:15:10 2026 -0700 Stop the radio recording budget at zero instead of wrapping it. g_recFromRadioSamples is unsigned, so a write that ran past the end of the budget wrapped it around, the "<= 0" check never saw the end and recording was never reported finished. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit ee9825c223ce5eef32416c0bcfa4cdc6eadb654e Author: Mooneer Salem Date: Sat Oct 24 17:52:50 2026 -0700 Stop the radio recording budget at zero instead of wrapping it. g_recFromRadioSamples is unsigned, so a write that ran past the end of the budget wrapped it around and the "<= 0" check never saw the end. Writes after the end also kept wrapping it, which is what failed radioRecordingStopsWhenItsBudgetIsUsedUp on macOS CI. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit 8eed136e9648226bd8b64a0769a433f3bfe87f47 Author: Mooneer Salem Date: Sat Oct 24 16:29:42 2026 -0700 Skip config_migration_failure where FreeDV is built without migration. Configuration files are only migrated with wxWidgets 3.3 or later, and the ubuntu-26.04 jobs build with the system's 3.2. Those jobs never got as far as the GUI tests before, as the unit tests failed first. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit fef8f1d33bb4fee8ae68712eb58a909befb14038 Author: Mooneer Salem Date: Sat Oct 24 15:40:31 2026 -0700 Only log when a PulseAudio stream ends up on another device, as on master. Match #1591: leave the stream connected and don't report an audio error, just note in the log where the sound server put it. The playback test for a missing sink goes, as nothing's refused any more, and the recording test accepts what PipeWire 1.6.2 does (recording from another device instead of refusing) so long as the stream is somewhere and nothing else was reported. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit a238dda8b9c39da9f076f4cac335fecd9741f62a Author: Mooneer Salem Date: Sat Oct 24 15:34:15 2026 -0700 Only log when a PulseAudio stream ends up on another device. Leave the stream connected and don't report it as an audio error; just note in the log where the sound server put it. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit ebf1b95ddf27a7e97243823d58a42fcdd4ed004e Author: Mooneer Salem Date: Sat Oct 24 15:00:58 2026 -0700 Don't use a PulseAudio stream the sound server put on another device. Streams aren't connected with PA_STREAM_DONT_MOVE, so that they can be moved between devices later, but that also lets the sound server put a new one elsewhere when the device it's asked for isn't there: PipeWire plays to the default device instead of a sink that's missing. Once a stream is ready, check that it's on the device asked for; if not, say so and disconnect it. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit 31a9a82ba56c0d286171abec9bc7f1f9699603df Author: Mooneer Salem Date: Sat Oct 24 15:00:41 2026 -0700 Don't use a PulseAudio stream the sound server put on another device. Streams aren't connected with PA_STREAM_DONT_MOVE, so that they can be moved between devices later, but that also lets the sound server put a new one elsewhere when the device it's asked for isn't there: PipeWire plays to the default device instead of a sink that's missing (1.6.9), and on 1.6.2 recording from a speaker's name isn't refused either. Once a stream is ready, check that it's on the device asked for; if not, say so and disconnect it. aPlaybackDeviceWithNoSinkReportsTheRefusal covers the playback case, which kept playing on the default device before this. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit 3272ad44d044571b4b3e87eff4d03fea5cf8e2a5 Author: Mooneer Salem Date: Sat Oct 24 13:35:58 2026 -0700 Keep the plots' arrow key checks' double-click clear of the Spectrum's labels. On macOS the Spectrum is narrower (407 px against 451 on Linux) and its dB labels take up its first 13%, so a double-click at 12% across landed on them and tuned nowhere. The Down presses then started from the top and ran out at 2400 Hz. Double-click a quarter of the way across, as the wheel checks already do, and allow enough presses to get to the bottom from there. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit 56ee996c2d4e9c3874bf9f7f89d3264622340d75 Author: Mooneer Salem Date: Sat Oct 24 13:07:35 2026 -0700 Keep a typed status message when the saved list is closed without a pick. Opening FreeDV Reporter's saved status messages and closing the list again without choosing one emptied the message box on Linux and Windows: the combo takes the list's value as it closes, and with nothing highlighted that was "". Leave what's in the box alone instead. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit d097f808931055a0b596c8173fc4e6d9981b3354 Author: Mooneer Salem Date: Sat Oct 24 13:07:19 2026 -0700 Keep a typed status message when the saved list closes, and fix tests failing on macOS and Windows. FreeDV: opening the saved status messages and closing them again without picking one emptied the message box on Linux and Windows, since with nothing highlighted the list's value was "". It's now what's in the box. status_message expected the old behaviour; macOS already kept the text. Tests that only failed outside Linux: - selected_khz, frequency_in_khz and frequency_list_from_main_window expected kHz frequencies without the thousands separators that the region adds on macOS and Windows (e.g. 14,236.0); leave them out, as the other tests already do. - config_migration_failure is skipped where there's no migration (only Linux has one). - missing_port named a /dev port on Windows, where FreeDV only checks COM names, so Hamlib reported it instead; use a COM port there. - logsHamlibDebugMessagesAtEachLevel captured to /tmp, which Windows doesn't have; use the temporary folder, in binary mode so that no carriage returns are added. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit 91f0d04da11ddaa45048c5089a541bc9bc437bd5 Author: Mooneer Salem Date: Sat Oct 24 11:21:52 2026 -0700 Put back the GUI test checks the rewrites dropped. - column_filters: the menu's exact contents and order, which comparison is ticked, Clear Column Filter being enabled once there's a filter, filters on two columns at once, Cancel leaving the filter alone, and clearing one filter leaving the other. - status_message: editing a saved message after adding one has moved it down, rather than before. - audio: refreshing the TX Audio tab's lists refreshes the RX Audio tab's too, and starting with the transmit devices unset as well as saved as "none". Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit ef6613b627c6ff3c3c4b9ae67fc71d1a59f224f6 Author: Mooneer Salem Date: Sat Oct 24 10:21:26 2026 -0700 Let three tests fail when they should. - test_ut_recording.sh checked tee's exit status rather than FreeDV's, so a FreeDV run that crashed partway through still passed. Its ctest pass expression was also just "PASS", which anything FreeDV printed could match, and its sync check looked at every run's output rather than the receive run's. - gui_rig_control_serial_ptt was never registered with ctest. - aMovedStreamCarriesOnWithTheNewDeviceName failed now and then because PipeWire sometimes accepts a move without carrying it out; ask again until the stream has moved, and fail if it never does. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit ca89de055145011b8664b3b411797df182c4b25f Author: Mooneer Salem Date: Sat Oct 24 09:42:33 2026 -0700 Don't assert when quitting with a message/frequency list open. Destroying a combo whose list is open makes ~wxComboCtrlBase() hide the popup, which asks MsgListPopup for its value. By then the wxComboCtrl part of the object is gone, so GetComboCtrl()'s cast asserts, and the assert dialog keeps FreeDV from exiting. Use the base class pointer. With that, and wxWidgets' log messages no longer shown in dialogs, the GUI tests no longer need to close the frequency list or give back the settings folder before quitting, so they now cover both. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit cfac6f308ba0aed2b77d4f0b0f8b72303e69ac9e Author: Mooneer Salem Date: Sat Oct 24 09:33:47 2026 -0700 Don't list stations heard more than once after a restart, or offer Log QSO without a logger. Stations heard: the CSV log gets a row every time a callsign is decoded, but on the air repeated decodes of the same station on the same frequency only update the top row's SNR. Restoring the list at startup inserted every row, so a station heard for a while came back as many identical entries. Collapse them the same way when restoring. Log QSO: on the air the button is only enabled when there's a UDP logger to send contacts to, but when stopped it was enabled whenever the list of stations heard wasn't empty. With UDP logging off, confirming the log entry then threw the contact away without a word. Require UDP logging when stopped as well, and recheck after Settings is closed. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit c73393948ca8a30fd2d03b4dd31ac099cfae3a6a Author: Mooneer Salem Date: Sat Oct 24 09:33:46 2026 -0700 Keep wxWidgets log messages' UTF-8 text alive while logging them. msg.ToUTF8() returns a temporary buffer, so the pointer taken from it was left dangling by the time it reached the log. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit f43b9a453d971a7aa5fb3b3dc3b846dc9c4e8a55 Merge: d25f47a5 048d437c Author: Mooneer Salem Date: Sat Oct 24 09:03:10 2026 -0700 Merge branch 'v3.0-dev' into ms-cov-extend commit d25f47a55d667df286757a561d3a33be92c1369e Author: Mooneer Salem Date: Sat Oct 24 06:10:56 2026 -0700 Cover heard-list updates when a logged entry is heard again commit faf47eaff49f850ef4205e2fb0f8da0078c9fff4 Author: Mooneer Salem Date: Sat Oct 24 06:10:01 2026 -0700 Extend unit recording test with timed and feature-capture runs commit ce91770a394990fa9bb2bcb7b4ce1e5963b8bdbd Author: Mooneer Salem Date: Sat Oct 24 06:01:29 2026 -0700 Extend GUI tests for ongui.cpp branches commit ac6c9d0c4cc22fe373f9dd6230bcb7f98907726e Author: Mooneer Salem Date: Sat Oct 24 04:47:34 2026 -0700 Extend setup wizard coverage: none/unplug audio paths, baud fallback and range filtering, toggle handlers, network-rig and partial WSJT-X imports commit de4d59a65668e9725b3d757492cd9c5043072a24 Author: Mooneer Salem Date: Sat Oct 24 04:15:23 2026 -0700 Cover kHz unit switching, none-output input test and RTS keying test commit 97c42dae5c47cf0d0c57e52452b78d1aff7268a1 Author: Mooneer Salem Date: Sat Oct 24 03:53:52 2026 -0700 Cover reporter rows, voice keyer cache, TxRxThread and clean shutdown paths commit 048d437c57899f1cf799efa2593dbe396be1ab91 Author: Mooneer Salem Date: Sat Oct 24 09:05:30 2026 +0000 latest user manual PDF commit 72078ea91a5cc45a7d22bd0e08973ae91e81f0ad Author: Mooneer Salem Date: Sat Oct 24 02:03:13 2026 -0700 Run a smaller set of tests on every pull request (#1585) * Run a smaller set of tests on every pull request. Pull requests got lint and packages only unless a commit said "[full ci]", and the full set takes hours, more so with several runs queued: the concurrency group fell back to the run ID on push events, so a newer push never cancelled an older run. - Build pull requests on pull_request (the merge result) and keep push for master and v*-dev only. Group runs by ref, and cancel superseded runs except on master and v*-dev. - Add a "pr" tier between nothing and everything: on Linux, ubuntu-26.04 plain and with ASan+UBSan, GUI tests on the plain one; on macOS, plain; on Windows, plain x86_64 unit and GUI tests and the audio tests against the plain installer. Plus lint and the plain packages, as now. - Replace "[full ci]" with a "full-ci" label on the pull request. "Run workflow" takes a tier and defaults to full. - Delete a closed pull request's refs/pull/N/merge caches as well as its branch's. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Don't git pull in the build scripts under CI. Pull request builds check out the merge commit, which isn't on a branch, so the pull failed. Elsewhere in CI it could also build a newer commit than the one being tested. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * EXPERIMENT: disable App Nap on the macOS test runner. To see whether it's why sleeps and timed waits run 2-3x long there (fullduplex_RADEV1, HamlibRigControllerTest, TxRxThreadTest), and how long a sleep takes on the runner to begin with. To be reverted. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Revert "EXPERIMENT: disable App Nap on the macOS test runner." App Nap made no difference: the runner's sleeps overshoot just as much with it disabled. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * EXPERIMENT: time a GUI test's phases on the macOS test runner. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Revert "EXPERIMENT: time a GUI test's phases on the macOS test runner." Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Start the test jobs without waiting for lint, and speed up GUI test waits. Waiting for lint held the Linux and macOS test jobs back by about seven minutes on every run. The GUI test harness now sends commands without TCP's coalescing delay and checks conditions every 20 ms rather than 100 ms, which matters most on the macOS runners, where sleeps run long. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Go back to the GUI test harness's old polling and socket settings. Polling every 20 ms with TCP_NODELAY didn't make the GUI tests measurably faster on any platform, and returning sooner from waits made test_tone start recording before the VOX tone was fully on the left channel. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * EXPERIMENT: check whether Gatekeeper slows FreeDV launches on the macOS test runner. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Revert "EXPERIMENT: check whether Gatekeeper slows FreeDV launches on the macOS test runner." This reverts commit da0aeebe. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * EXPERIMENT: time library loading at FreeDV launch on the macOS test runner. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Revert "EXPERIMENT: time library loading at FreeDV launch on the macOS test runner." This reverts commit 21c52c82. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * EXPERIMENT: record each FreeDV launch's timing in the GUI tests on CI. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Revert "EXPERIMENT: record each FreeDV launch's timing in the GUI tests on CI." This reverts commit b9395d9d. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Stop the GUI tests' FreeDV Reporter server without a 10 s wait. The server never closes its aiosqlite connection, whose non-daemon thread kept Python running after the server had shut down on SIGTERM, until ReporterServer gave up waiting and killed it 10 s later. That cost about 270 s per run on Linux and macOS (Windows terminates outright). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * EXPERIMENT: compare storage speeds on the macOS test runner, including a RAM disk. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Revert "EXPERIMENT: compare storage speeds on the macOS test runner, including a RAM disk." This reverts commit 104a95b4. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Speed up the GUI tests' FreeDV Reporter server and stations. Stopping the server no longer waits for whoever's connected (aiohttp waits up to a minute by default, so ReporterServer gave up and killed it after 10 s): FreeDV is always connected when a test restarts the server. Stations now wake as soon as the update they're waiting for arrives, rather than checking every 50 ms, and close their event loop when they disconnect. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Have connection_burst churn without pauses, for a set number of actions. A station hiding for a moment held everything else up while it was gone, which was about 70% of each 20 s round. It now comes back later while the churn carries on, and a round is $FREEDV_BURST_ACTIONS (default 250) actions, so what's done really is the same from run to run. The scenario takes about 32 s locally, down from about 121 s, while doing more per second to FreeDV. (Churning this fast found a race in the FreeDV Reporter server, since fixed there.) Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * EXPERIMENT: give rig control and Mac audio devices a responsive thread. Carries freedv-backend's side as a patch applied at fetch time. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Revert "EXPERIMENT: give rig control and Mac audio devices a responsive thread." This reverts commit 6c832b6e. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * EXPERIMENT: measure how much sleeps overrun on the Linux and macOS test runners. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Revert "EXPERIMENT: measure how much sleeps overrun on the Linux and macOS test runners." This reverts commit 3e6c87b2. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * EXPERIMENT: see whether keeping a CPU busy stops sleeps overrunning on the macOS runner. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Revert "EXPERIMENT: see whether keeping a CPU busy stops sleeps overrunning on the macOS runner." This reverts commit ec7dc16f. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Add PR #1585 to changelog. --------- Co-authored-by: Claude Opus 5.5 commit 296966b7776e16b3a0f8d26c34222e48aa021768 Author: Mooneer Salem Date: Sat Oct 24 02:03:00 2026 -0700 Add PR #1585 to changelog. commit a4e82107e28f1bf197ba122749d8fcc28d1e1d1c Author: Mooneer Salem Date: Sat Oct 24 01:54:49 2026 -0700 Revert "EXPERIMENT: see whether keeping a CPU busy stops sleeps overrunning on the macOS runner." This reverts commit ec7dc16f. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit ec7dc16fad09b106554bcfa211ada3c7958022fc Author: Mooneer Salem Date: Sat Oct 24 01:04:45 2026 -0700 EXPERIMENT: see whether keeping a CPU busy stops sleeps overrunning on the macOS runner. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit f189abd3bfb85edc42b6e70f88ef6128b0799555 Author: Mooneer Salem Date: Sat Oct 24 01:03:01 2026 -0700 Revert "EXPERIMENT: measure how much sleeps overrun on the Linux and macOS test runners." This reverts commit 3e6c87b2. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit 4e978f4715f9d6630ba8d817130d507b4d7ec108 Author: Mooneer Salem Date: Sat Oct 24 00:36:02 2026 -0700 Recover plot fault, colour and translated-layout branches with new GUI scenarios. commit 3e6c87b22e6161ae286f57d2df732939f24f672f Author: Mooneer Salem Date: Sat Oct 24 00:05:39 2026 -0700 EXPERIMENT: measure how much sleeps overrun on the Linux and macOS test runners. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit 68d6eeeb73437787a9cff15a1bb896358b840a4d Author: Mooneer Salem Date: Fri Oct 23 23:38:40 2026 -0700 Revert "EXPERIMENT: give rig control and Mac audio devices a responsive thread." This reverts commit 6c832b6e. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit 6c832b6e61a7a15fb35968e789b3c32f227eaf54 Author: Mooneer Salem Date: Fri Oct 23 22:52:24 2026 -0700 EXPERIMENT: give rig control and Mac audio devices a responsive thread. Carries freedv-backend's side as a patch applied at fetch time. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit 1d77f55b5a5f208005cbf33720a048503fb31ae8 Author: Mooneer Salem Date: Fri Oct 23 20:37:51 2026 -0700 Recover configuration store and element conversion branches in unit tests. commit 1a6a92f565457b2bef37a3e7c326bd2c7ff7ed91 Author: Mooneer Salem Date: Fri Oct 23 20:30:16 2026 -0700 Cover PTT input port fault paths; open serial ports without stealing the terminal commit 0783c515cd0fbdb5ecbe0d91566bf0c927c17400 Author: Mooneer Salem Date: Fri Oct 23 20:27:43 2026 -0700 Extend rig and audio unit tests to fault paths; open serial ports without stealing the terminal commit 5886bf79bf09a43eea0a0f7baa8bc6c17f62d986 Author: Mooneer Salem Date: Fri Oct 23 19:41:11 2026 -0700 Have connection_burst churn without pauses, for a set number of actions. A station hiding for a moment held everything else up while it was gone, which was about 70% of each 20 s round. It now comes back later while the churn carries on, and a round is $FREEDV_BURST_ACTIONS (default 250) actions, so what's done really is the same from run to run. The scenario takes about 32 s locally, down from about 121 s, while doing more per second to FreeDV. (Churning this fast found a race in the FreeDV Reporter server, since fixed there.) Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit dedfd6d1fba2457af26f3ea5a4882df0b4f25172 Author: Mooneer Salem Date: Fri Oct 23 19:25:52 2026 -0700 Speed up the GUI tests' FreeDV Reporter server and stations. Stopping the server no longer waits for whoever's connected (aiohttp waits up to a minute by default, so ReporterServer gave up and killed it after 10 s): FreeDV is always connected when a test restarts the server. Stations now wake as soon as the update they're waiting for arrives, rather than checking every 50 ms, and close their event loop when they disconnect. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit f85aa5225e4b3c0e2b02554fbf5e64b842373582 Author: Mooneer Salem Date: Fri Oct 23 15:59:16 2026 -0700 Revert "EXPERIMENT: compare storage speeds on the macOS test runner, including a RAM disk." This reverts commit 104a95b4. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit 104a95b42d1340a7a630a6909e529e57e8e23ad1 Author: Mooneer Salem Date: Fri Oct 23 15:26:19 2026 -0700 EXPERIMENT: compare storage speeds on the macOS test runner, including a RAM disk. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit da7bdd896d857ae7714159b147cd652937d510b5 Author: Mooneer Salem Date: Fri Oct 23 14:25:03 2026 -0700 Stop the GUI tests' FreeDV Reporter server without a 10 s wait. The server never closes its aiosqlite connection, whose non-daemon thread kept Python running after the server had shut down on SIGTERM, until ReporterServer gave up waiting and killed it 10 s later. That cost about 270 s per run on Linux and macOS (Windows terminates outright). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit f0d84a094013bf2ea181368f671091ce6ca11986 Author: Mooneer Salem Date: Fri Oct 23 14:02:33 2026 -0700 Revert "EXPERIMENT: record each FreeDV launch's timing in the GUI tests on CI." This reverts commit b9395d9d. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit b9395d9d0d3ad4a015a91e7ee7681d6584ac20f8 Author: Mooneer Salem Date: Fri Oct 23 12:43:06 2026 -0700 EXPERIMENT: record each FreeDV launch's timing in the GUI tests on CI. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit c7b4a7794fb24e224812f790404419dd8180004f Author: Mooneer Salem Date: Fri Oct 23 12:30:37 2026 -0700 Revert "EXPERIMENT: time library loading at FreeDV launch on the macOS test runner." This reverts commit 21c52c82. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit 21c52c82b75c0a160f3e5f4ef9dcd5b00d15e329 Author: Mooneer Salem Date: Fri Oct 23 12:21:45 2026 -0700 EXPERIMENT: time library loading at FreeDV launch on the macOS test runner. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit 2654354f159c26e68b8bc87455977ed1ad6e8e07 Author: Mooneer Salem Date: Fri Oct 23 08:06:09 2026 -0700 Revert "EXPERIMENT: check whether Gatekeeper slows FreeDV launches on the macOS test runner." This reverts commit da0aeebe. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit da0aeebe51984b39c7842c0dbfed012ce60ccb41 Author: Mooneer Salem Date: Fri Oct 23 07:54:56 2026 -0700 EXPERIMENT: check whether Gatekeeper slows FreeDV launches on the macOS test runner. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit 29eca08e9fbd1ac698cf097653050283a9873e91 Author: Mooneer Salem Date: Fri Oct 23 07:50:10 2026 -0700 Go back to the GUI test harness's old polling and socket settings. Polling every 20 ms with TCP_NODELAY didn't make the GUI tests measurably faster on any platform, and returning sooner from waits made test_tone start recording before the VOX tone was fully on the left channel. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit 5a4477062f0b0e8b2a305468f9feb4fffca74ec2 Author: Mooneer Salem Date: Fri Oct 23 07:47:10 2026 -0700 Recover branch coverage with extended reporter and settings GUI scenarios. - Rewrite ConfigurationDataElementTest around a DefineConfigUnitTest macro registered per element in src/config. - Add reporter scenarios/probes: filters and saved messages, status-message combo behaviour, row actions including message-column hovering and list mouse behaviour; frequency-list keyboard probes in the settings scenario. - Fix MsgListPopup PageUp selection when the popup list is open. CI-chain branch coverage: 89.6% -> 90.1% (12891/14311). commit bc325da43efbae881484d28eb0779a137a0bbac0 Merge: cd8d361e edb8d0fd Author: Mooneer Salem Date: Fri Oct 23 00:21:25 2026 -0700 Merge remote-tracking branch 'origin/v3.0-dev' into ms-ci-pr-tier Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit cd8d361ead6d1c9d86323b2192964ab3f8246062 Author: Mooneer Salem Date: Fri Oct 23 00:21:24 2026 -0700 Start the test jobs without waiting for lint, and speed up GUI test waits. Waiting for lint held the Linux and macOS test jobs back by about seven minutes on every run. The GUI test harness now sends commands without TCP's coalescing delay and checks conditions every 20 ms rather than 100 ms, which matters most on the macOS runners, where sleeps run long. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit edb8d0fda027e9038bfea3f37a1dcbaab0f763d7 Author: Mooneer Salem Date: Fri Oct 23 00:20:28 2026 -0700 Keep TxRxThreadTest's gaps between turns from adding up on slow timers. (#1587) The round trip through files sleeps 1 ms between turns, several thousand times. On the macOS CI runners a 1 ms sleep lasts about 10 ms, which made the test take 82 s there against 27 s on Linux. What a sleep overruns by is now taken off the turns that follow. Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ Co-authored-by: Claude Opus 5.5 commit 4202e29909679a117e6aefd59ddec00d349ecd54 Author: Mooneer Salem Date: Thu Oct 22 23:57:24 2026 -0700 Keep TxRxThreadTest's gaps between turns from adding up on slow timers. The round trip through files sleeps 1 ms between turns, several thousand times. On the macOS CI runners a 1 ms sleep lasts about 10 ms, which made the test take 82 s there against 27 s on Linux. What a sleep overruns by is now taken off the turns that follow. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit 4918eed3d7e97c4bf89656af4a7ab43020082bf1 Author: Mooneer Salem Date: Thu Oct 22 23:57:21 2026 -0700 Revert "EXPERIMENT: time a GUI test's phases on the macOS test runner." Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit 5fbb5b80dc5043bb50e097ddcba73fa8ee88d738 Author: Mooneer Salem Date: Thu Oct 22 23:30:04 2026 -0700 EXPERIMENT: time a GUI test's phases on the macOS test runner. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit ec85db195c42dd95cd0367081c87aa76e5843cf1 Merge: 38267751 f665e8d0 Author: Mooneer Salem Date: Thu Oct 22 22:24:03 2026 -0700 Merge remote-tracking branch 'origin/v3.0-dev' into ms-ci-pr-tier Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit f665e8d0bef260c30fef8bcc9363f473fcdc0190 Author: Mooneer Salem Date: Thu Oct 22 22:22:05 2026 -0700 Time test-mode waits by the clock instead of counting sleeps. (#1586) A 100 ms sleep can last 170 ms or more on the macOS CI runners, so counting them made fullduplex_RADEV1's 60 s receive run for over 100 s there, and stretched the 20 minute limit on the file playback loops. Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ Co-authored-by: Claude Opus 5.5 commit 9b74f0ca2bd658aff6cd744b0c3a7b73bef17418 Author: Mooneer Salem Date: Thu Oct 22 22:00:26 2026 -0700 Time test-mode waits by the clock instead of counting sleeps. A 100 ms sleep can last 170 ms or more on the macOS CI runners, so counting them made fullduplex_RADEV1's 60 s receive run for over 100 s there, and stretched the 20 minute limit on the file playback loops. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit 382677510c3e696b49af4d6db8868f15f16d1c2e Author: Mooneer Salem Date: Thu Oct 22 21:46:29 2026 -0700 Revert "EXPERIMENT: disable App Nap on the macOS test runner." App Nap made no difference: the runner's sleeps overshoot just as much with it disabled. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit 418c419a12208fca95bb0d5a7773f7d1a2cbb019 Author: Mooneer Salem Date: Fri Oct 23 04:09:32 2026 +0000 latest user manual PDF commit 5dca835e74ecfa7916b2f5a2d38e7ceab35a1215 Author: Mooneer Salem Date: Thu Oct 22 20:46:22 2026 -0700 EXPERIMENT: disable App Nap on the macOS test runner. To see whether it's why sleeps and timed waits run 2-3x long there (fullduplex_RADEV1, HamlibRigControllerTest, TxRxThreadTest), and how long a sleep takes on the runner to begin with. To be reverted. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit ba3e2aa9215f346b6bf49e3df8d339a7aa521ac8 Author: Mooneer Salem Date: Thu Oct 22 19:14:10 2026 -0700 Don't git pull in the build scripts under CI. Pull request builds check out the merge commit, which isn't on a branch, so the pull failed. Elsewhere in CI it could also build a newer commit than the one being tested. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit d8003cc61e24a7c2417f963ec8011feb38c53bc2 Author: Mooneer Salem Date: Thu Oct 22 18:58:45 2026 -0700 Run a smaller set of tests on every pull request. Pull requests got lint and packages only unless a commit said "[full ci]", and the full set takes hours, more so with several runs queued: the concurrency group fell back to the run ID on push events, so a newer push never cancelled an older run. - Build pull requests on pull_request (the merge result) and keep push for master and v*-dev only. Group runs by ref, and cancel superseded runs except on master and v*-dev. - Add a "pr" tier between nothing and everything: on Linux, ubuntu-26.04 plain and with ASan+UBSan, GUI tests on the plain one; on macOS, plain; on Windows, plain x86_64 unit and GUI tests and the audio tests against the plain installer. Plus lint and the plain packages, as now. - Replace "[full ci]" with a "full-ci" label on the pull request. "Run workflow" takes a tier and defaults to full. - Delete a closed pull request's refs/pull/N/merge caches as well as its branch's. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit 58c0613aa22a164ccb9c1294986ed802bdb3789b Merge: 04212802 781f37f7 Author: Mooneer Salem Date: Thu Oct 22 17:20:04 2026 -0700 Merge pull request #1584 from freedv/ms-agc-branch-update Update 3.0 backend branch to incorporate AGC changes. commit 781f37f7ca20403a578275093a04b101fc694484 Merge: 51d9847a 04212802 Author: Mooneer Salem Date: Thu Oct 22 14:13:21 2026 -0700 [full ci] Merge remote-tracking branch 'origin/v3.0-dev' into ms-agc-branch-update Picks up the ConfigMigrationTest fix from #1581. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit 51d9847a4e12c2e8bc331deedf7aaadf438c9466 Author: Mooneer Salem Date: Thu Oct 22 14:13:08 2026 -0700 Fix CI failures from the master merge. - Drop the conflict markers left in util.cpp. v3.0-dev moved these functions to the backend, so its side (removing them) is kept. - Remove m_schedule_restore = false from MainFrame's constructor; master's neighboring line came in with the OptionsDlg lazy-creation change, but that member doesn't exist in v3.0-dev. - Restore the brace closing UtilTest.cpp's anonymous namespace, which the merge dropped along with master's freq_shift_coh() tests. - automaticGainControlRaisesQuietMicrophone: use a 1000-peak tone. The new backend AGC treats the old 300-peak tone as silence and holds its gain. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit 6ef7c56ee2d6ae110d8e51d1a83d51eea7a42776 Author: Mooneer Salem Date: Thu Oct 22 13:20:33 2026 -0700 [full ci] Revert to base repo now that AGC code has merged. commit a1eb75e5bd2cd413eeb76286b5f05cf930192274 Merge: 9d5caf90 8d35c3d9 Author: Mooneer Salem Date: Thu Oct 22 13:19:57 2026 -0700 [full ci] Merge branch 'master' into ms-agc-branch-update commit 9d5caf902d5052ddf03d76d5e0d72a390cdaaddb Author: Mooneer Salem Date: Thu Oct 22 12:51:00 2026 -0700 Update 3.0 backend branch to incorporate AGC changes. commit 8d35c3d95f3410f4a25393972224870c8aef0684 Author: Mooneer Salem Date: Thu Oct 22 19:15:25 2026 +0000 latest user manual PDF commit cb16d654ebee82b4a3004d2db2caaf3ccefc9f49 Author: Mooneer Salem Date: Thu Oct 22 12:13:18 2026 -0700 Fix data races on RX frequency offset, reporter list, device latency and device stop (#1583) * Fix data races on RX frequency offset, reporter list and device latency. - The GUI read the RX frequency offset straight out of the modem state the RX thread was updating. The RX thread now publishes it to an atomic after each frame (RADE and legacy modes). - m_reporters was cleared and filled from the threads that start and stop FreeDV while the UI thread iterates it. All changes to it now happen on the UI thread. - getLatencyInMicroseconds() could run on the UI thread during the TX->RX wait while the device was being stopped. PortAudioDevice now guards deviceStream_ with a mutex, and PulseAudioDevice takes objLock_ before using mainloop_. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Add PR #1583 to changelog. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Make Mac and WASAPI audio device stop() safe to repeat. - MacAudioDevice: stop() ran its whole teardown unconditionally, so stopping a device that never started (or stopping twice) disposed an uninitialized or already-disposed audio unit and freed its buffers again. auHAL_/bufferList_ now start out null, stop() only stops the unit and removes listeners if it's running, clears pointers after freeing them, and gives up waiting for the unit to stop if its state can't be read. The destructor now always calls stopImpl_() so it also cleans up after a failed start(). - WASAPIAudioDevice: stop() closed semaphore_ and highResTimer_ while stopRealTimeWork() could still be waiting on them. They're now created once and only closed in the destructor; stop() just wakes any waiter. timeEndPeriod() is only called if start() called timeBeginPeriod(), since stop() also runs from the destructor. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ --------- Co-authored-by: Claude Opus 5.5 commit 1bafa5888390d2276037f771f1cb324764213f80 Author: Mooneer Salem Date: Thu Oct 22 09:45:57 2026 -0700 Make Mac and WASAPI audio device stop() safe to repeat. - MacAudioDevice: stop() ran its whole teardown unconditionally, so stopping a device that never started (or stopping twice) disposed an uninitialized or already-disposed audio unit and freed its buffers again. auHAL_/bufferList_ now start out null, stop() only stops the unit and removes listeners if it's running, clears pointers after freeing them, and gives up waiting for the unit to stop if its state can't be read. The destructor now always calls stopImpl_() so it also cleans up after a failed start(). - WASAPIAudioDevice: stop() closed semaphore_ and highResTimer_ while stopRealTimeWork() could still be waiting on them. They're now created once and only closed in the destructor; stop() just wakes any waiter. timeEndPeriod() is only called if start() called timeBeginPeriod(), since stop() also runs from the destructor. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit 26f84414953cbdf10c446e666e49fda4fc56fc82 Author: Mooneer Salem Date: Thu Oct 22 09:13:22 2026 -0700 Add PR #1583 to changelog. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit 703fc8ef7521918ea7aa137470e90070c741f595 Author: Mooneer Salem Date: Thu Oct 22 09:13:13 2026 -0700 Fix data races on RX frequency offset, reporter list and device latency. - The GUI read the RX frequency offset straight out of the modem state the RX thread was updating. The RX thread now publishes it to an atomic after each frame (RADE and legacy modes). - m_reporters was cleared and filled from the threads that start and stop FreeDV while the UI thread iterates it. All changes to it now happen on the UI thread. - getLatencyInMicroseconds() could run on the UI thread during the TX->RX wait while the device was being stopped. PortAudioDevice now guards deviceStream_ with a mutex, and PulseAudioDevice takes objLock_ before using mainloop_. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit 01fa099c74daeadcd29f570b9916cc240f4d9855 Author: Mooneer Salem Date: Thu Oct 22 15:26:48 2026 +0000 latest user manual PDF commit 071007686013b57a6b20efdb7537d85b34a94c3f Author: Mooneer Salem Date: Thu Oct 22 08:24:00 2026 -0700 OptionsDlg: Create only on user request. (#1579) * OptionsDlg: Create only on user request. * Add PR #1579 to changelog. commit 73c883953876abe9f11e0374da0383ab328ef651 Author: Mooneer Salem Date: Thu Oct 22 08:23:32 2026 -0700 Add PR #1579 to changelog. commit 04212802e4776adf9a1e6eaf6f97cce5ac019a6b Author: Mooneer Salem Date: Thu Oct 22 02:17:39 2026 +0000 latest user manual PDF commit 8d22e52ed2123c30cbb24f1e9665ead286f8709c Author: Mooneer Salem Date: Wed Oct 21 19:15:06 2026 -0700 Fix v3.0-dev CI failures from #1570 (#1581) * Fix v3.0-dev CI failures from #1570. - ConfigMigrationTest: expect the frequency list in ascending order on load, now that ReportingConfiguration sorts it. - Only save the frequency list's scroll position on wxWidgets 3.1+, as wxListBox::GetTopItem() doesn't exist in 3.0 (Ubuntu 22.04 builds). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ * Add PR #1581 to changelog. --------- Co-authored-by: Claude Opus 5.5 commit 3e9a4d3dc0542c6e75ec354ef49942522c939527 Author: Mooneer Salem Date: Wed Oct 21 19:13:21 2026 -0700 Add PR #1581 to changelog. commit 5216905b5f35491bfb4bc2beacd56acdf9dd7176 Author: Mooneer Salem Date: Wed Oct 21 18:42:54 2026 -0700 Fix v3.0-dev CI failures from #1570. - ConfigMigrationTest: expect the frequency list in ascending order on load, now that ReportingConfiguration sorts it. - Only save the frequency list's scroll position on wxWidgets 3.1+, as wxListBox::GetTopItem() doesn't exist in 3.0 (Ubuntu 22.04 builds). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01SYMA9QiGksM1Dox6TuavLZ commit 349a7b08e3f223b6aa3ff0d2d900ad97cd414037 Author: Mooneer Salem Date: Wed Oct 21 17:14:31 2026 -0700 [skip ci] Update .pot to provide valid contact information. (#1580) commit a92704e6500de467758493c8f5ac6a1d4e22ab06 Author: Mooneer Salem Date: Wed Oct 21 17:13:09 2026 -0700 [skip ci] Update .pot to provide valid contact information. commit e0e0eda5830b938bca1a17dcc6a5eede4c2febda Merge: fe7b97b5 faaa0b42 Author: Mooneer Salem Date: Wed Oct 21 17:00:47 2026 -0700 Merge branch 'v3.0-dev' into bcj-freq-list-edit commit fe7b97b5c6bc60d5bf1283c15cf4695bb2431ffb Author: Mooneer Salem Date: Wed Oct 21 17:00:44 2026 -0700 Add PR #1570 to changelog. commit 70261c064282f0a852162d36f316525b9a0ea892 Author: Mooneer Salem Date: Wed Oct 21 16:59:51 2026 -0700 Avoid magic numbers when setting up deselect timer. commit dd66cedcfc376fb4f06c10d9d8d0e3a079287689 Author: Mooneer Salem Date: Wed Oct 21 16:13:50 2026 -0700 OptionsDlg: Create only on user request. commit 08db5b942ecd2e8627c02603defe974687add019 Author: Mooneer Salem Date: Wed Oct 21 15:12:23 2026 +0000 latest user manual PDF commit fa21ad9cdb94de0ed4c62a1844927b428f57ea7d Author: Mooneer Salem Date: Wed Oct 21 08:09:44 2026 -0700 Resolve data races discovered by Qwen-Flash-Next. (#1578) * Resolve data races discovered by Qwen-Flash-Next. * Correct PR ID in changelog. * Drop session-transition gating and teardown marshalling from race fixes. The g_sessionTransitioning guards silently discarded user changes and rig frequency updates during start/stop, and the teardown changes guarded against a crash that can't occur since all OnTimer timers are already stopped on the UI thread before teardown. Keep only the atomic and GenericFIFO fixes, and condense the changelog entries accordingly. Co-Authored-By: Claude Opus 5.5 --------- Co-authored-by: Claude Opus 5.5 commit e8233d0a4ac8eeebdaf90cc99b8873848c41b133 Author: Mooneer Salem Date: Wed Oct 21 00:57:37 2026 -0700 Drop session-transition gating and teardown marshalling from race fixes. The g_sessionTransitioning guards silently discarded user changes and rig frequency updates during start/stop, and the teardown changes guarded against a crash that can't occur since all OnTimer timers are already stopped on the UI thread before teardown. Keep only the atomic and GenericFIFO fixes, and condense the changelog entries accordingly. Co-Authored-By: Claude Opus 5.5 commit 8cbb6c480c1869768f2d8ee46309dcc53b598fbf Author: Mooneer Salem Date: Wed Oct 21 00:43:24 2026 -0700 Correct PR ID in changelog. commit ac0ac054e42b35f979809c37bda95ead99439998 Merge: fa5f5b8d 169258a1 Author: Mooneer Salem Date: Wed Oct 21 00:42:15 2026 -0700 Merge branch 'master' into ms-data-race-fix commit fa5f5b8dae2f64b6a3cfb748c17d93baa86d9ce2 Author: Mooneer Salem Date: Wed Oct 21 00:41:59 2026 -0700 Resolve data races discovered by Qwen-Flash-Next. commit 169258a1afc22237e854b9d8fb0f9d3d657989f7 Author: Mooneer Salem Date: Wed Oct 21 07:05:43 2026 +0000 latest user manual PDF commit 628a58eeaeeb81705307a258004dd0d8e6ff165f Author: Mooneer Salem Date: Wed Oct 21 00:03:43 2026 -0700 Link cairo explicitly into fdv_gui_controls on Linux (#1577) * Link cairo explicitly into fdv_gui_controls on Linux plot_scalar.cpp calls cairo directly when HAS_CAIRO is defined, but the library was only ever linked transitively via GTK3_LIBRARIES on freedv. When gtk+-3.0.pc isn't found (or wx isn't detected as a GTK3 build), the final link fails with 'undefined reference to symbol cairo_destroy ... DSO missing from command line' (reported on openSUSE Tumbleweed). Co-Authored-By: Claude Opus 5.5 * Add PR #1577 to changelog. --------- Co-authored-by: Claude Opus 5.5 commit aa81cccebe71c6dd62b4d4700179f4aad6db33b5 Author: Mooneer Salem Date: Wed Oct 21 00:03:32 2026 -0700 Add PR #1577 to changelog. commit 2096ee081bdc3e7277138de467e81979e65e6d2f Author: Mooneer Salem Date: Tue Oct 20 19:54:11 2026 -0700 Link cairo explicitly into fdv_gui_controls on Linux plot_scalar.cpp calls cairo directly when HAS_CAIRO is defined, but the library was only ever linked transitively via GTK3_LIBRARIES on freedv. When gtk+-3.0.pc isn't found (or wx isn't detected as a GTK3 build), the final link fails with 'undefined reference to symbol cairo_destroy ... DSO missing from command line' (reported on openSUSE Tumbleweed). Co-Authored-By: Claude Opus 5.5 commit edda0fb4c78f7737a8bfd96a0e6ada39a923df22 Merge: 520334dd faaa0b42 Author: Mooneer Salem Date: Tue Oct 20 16:49:54 2026 -0700 [full ci] Merge branch 'v3.0-dev' into ms-rade-v2 commit faaa0b42457f5bffd53602faf90e3737da766fb7 Merge: 7f61930c db5bb00a Author: Mooneer Salem Date: Tue Oct 20 16:49:35 2026 -0700 Merge branch 'master' into v3.0-dev commit db5bb00aa0b4913530ea4fa380a9d74482c14ca7 Author: Mooneer Salem Date: Tue Oct 20 16:43:38 2026 -0700 Use new location for macOS audio driver. (#1576) commit cc32a92e90b2a60261316fcd5d23ca0264039944 Author: Mooneer Salem Date: Tue Oct 20 16:23:31 2026 -0700 Use new location for macOS audio driver. commit 603b3280a094527b2ed71f76b20ed183e76b0d82 Author: Mooneer Salem Date: Tue Oct 20 16:41:35 2026 -0700 Make MainFrame::m_RxRunning atomic (#1575) * Make MainFrame::m_RxRunning atomic. It's set by startRxStream()/stopRxStream() on the thread that starts and stops FreeDV (see OnTogBtnOnOff()) and read all over the UI thread, so reading it there was a data race (found by ThreadSanitizer in the macOS GUI tests on v3.0-dev). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_013G7dQDriDiWypo4B3J83gx * Add PR #1575 to changelog. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_013G7dQDriDiWypo4B3J83gx * Use acquire/release ordering for m_RxRunning. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_013G7dQDriDiWypo4B3J83gx --------- Co-authored-by: Claude Opus 5.5 commit 45b9d0d5700939ed2683b5d854db8b1978363a63 Author: Mooneer Salem Date: Tue Oct 20 16:23:31 2026 -0700 Use new location for macOS audio driver. commit c063cc9aca66f9da1c04fd234b5e727ebd8f49b3 Author: Mooneer Salem Date: Tue Oct 20 15:12:56 2026 -0700 Use acquire/release ordering for m_RxRunning. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_013G7dQDriDiWypo4B3J83gx commit 5ff0d010ac5d5a4633dca14da9649e6b350b23c0 Author: Mooneer Salem Date: Tue Oct 20 14:43:54 2026 -0700 Add PR #1575 to changelog. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_013G7dQDriDiWypo4B3J83gx commit c5f1872748c74c91e266f3cc4fee2b4452937b7a Author: Mooneer Salem Date: Tue Oct 20 14:43:42 2026 -0700 Make MainFrame::m_RxRunning atomic. It's set by startRxStream()/stopRxStream() on the thread that starts and stops FreeDV (see OnTogBtnOnOff()) and read all over the UI thread, so reading it there was a data race (found by ThreadSanitizer in the macOS GUI tests on v3.0-dev). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_013G7dQDriDiWypo4B3J83gx commit 255cd192dd96b837c7a398350f5394d926f6012a Author: Mooneer Salem Date: Tue Oct 20 21:41:38 2026 +0000 latest user manual PDF commit d0cc792c1b04eff4e91674d2ade72e47e36b9158 Author: Mooneer Salem Date: Tue Oct 20 14:39:00 2026 -0700 FreeDV Reporter: keep the milliseconds of the times the server sends (#1573) * FreeDV Reporter: keep the milliseconds of the times the server sends. The fraction of a second was removed so that ParseISOCombined() could parse the rest, and discarded, so stations that connected (or last transmitted or reported) within the same second compared equal. Ties in a sorted column then fell back to comparing the list's internal pointers, which made the order of e.g. two stations transmitting in the same mode random (as gui_reporter_sorted_updates on ms-rade-v2 found). The milliseconds are now added back once the rest is parsed. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_013G7dQDriDiWypo4B3J83gx * Add PR #1573 to changelog. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_013G7dQDriDiWypo4B3J83gx --------- Co-authored-by: Claude Opus 5.5 commit 3ac4d917d360439e336e36affab5a5b5080fbedc Author: Mooneer Salem Date: Tue Oct 20 12:01:16 2026 -0700 Add PR #1573 to changelog. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_013G7dQDriDiWypo4B3J83gx commit cc426d358ad1e66d95f084c6d57c05902226a6f8 Author: Mooneer Salem Date: Tue Oct 20 12:01:04 2026 -0700 FreeDV Reporter: keep the milliseconds of the times the server sends. The fraction of a second was removed so that ParseISOCombined() could parse the rest, and discarded, so stations that connected (or last transmitted or reported) within the same second compared equal. Ties in a sorted column then fell back to comparing the list's internal pointers, which made the order of e.g. two stations transmitting in the same mode random (as gui_reporter_sorted_updates on ms-rade-v2 found). The milliseconds are now added back once the rest is parsed. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_013G7dQDriDiWypo4B3J83gx commit 520334ddbd900f0772cebf7229b1779efad39b29 Merge: c8f9d67f 7f61930c Author: Mooneer Salem Date: Tue Oct 20 08:36:07 2026 -0700 [full ci] Merge branch 'v3.0-dev' into ms-rade-v2 Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_013G7dQDriDiWypo4B3J83gx commit 7f61930c1ac1f3afc6aa48c059caffcac3bfdbb5 Merge: 1a77c065 98896422 Author: Mooneer Salem Date: Tue Oct 20 08:29:21 2026 -0700 Merge branch 'master' into v3.0-dev commit 6f8e545ba5aaf43806b3d96e2b5618e1da926cb5 Author: Mooneer Salem Date: Tue Oct 20 15:18:03 2026 +0000 latest user manual PDF commit 988964228c410f9f3d9c53d4ff78f5085db7ccf5 Author: Mooneer Salem Date: Tue Oct 20 08:14:03 2026 -0700 rade_reporting_mpp: put the whole recording through the fading channel; initialize PlotPanel's sync state (#1572) * rade_reporting_mpp: put the whole recording through the fading channel. ch's MPP fading file holds 60 s of fading and ch stops when it runs out, so only the first 60 s of the ~100 s recording reached the receiver: about 4 of the 7 overs, rather than all of them, to decode the callsign from. When none of those 4 did, the test failed. ch is now given a fading file three times as long (the original's samples repeated), made when the test runs. The script also now checks ch's exit status, and fails without receiving anything if ch stopped part of the way through, rather than silently testing with what was left. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_013G7dQDriDiWypo4B3J83gx * Initialize PlotPanel's sync state. sync_ was only ever set by the waterfall's constructor and by setSync(), so the spectrum plot could paint its sync tuning line before its first timer tick from an uninitialized bool (found by UBSan in the macOS GUI tests: "load of value 96, which is not a valid value for type 'bool'" at plot_spectrum.cpp). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_013G7dQDriDiWypo4B3J83gx * Add PR #1572 to changelog. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_013G7dQDriDiWypo4B3J83gx --------- Co-authored-by: Claude Opus 5.5 commit bbd8bfd8d4ac244b85458fd86dbbe88eea8b37d8 Author: Mooneer Salem Date: Tue Oct 20 07:44:07 2026 -0700 Add PR #1572 to changelog. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_013G7dQDriDiWypo4B3J83gx commit f8a7be355feff1acce4d10cee8db4a7f018a6e96 Author: Mooneer Salem Date: Tue Oct 20 07:32:35 2026 -0700 Initialize PlotPanel's sync state. sync_ was only ever set by the waterfall's constructor and by setSync(), so the spectrum plot could paint its sync tuning line before its first timer tick from an uninitialized bool (found by UBSan in the macOS GUI tests: "load of value 96, which is not a valid value for type 'bool'" at plot_spectrum.cpp). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_013G7dQDriDiWypo4B3J83gx commit c8f9d67f9ab4227bafecd3dc46025c0da5594ab0 Author: Mooneer Salem Date: Tue Oct 20 07:39:18 2026 -0700 GUI tests: transmit speech, not a tone, while looking at the plots. With a tone from the microphone, RADE V2's receiver kept taking what FreeDV sent for the end of an over (~40 times in one run of plots:drawing), so sync came and went all the way through and the check of it after looking at each plot passed or failed by chance. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_013G7dQDriDiWypo4B3J83gx commit 824ccbc183296c1ba9de779e038f694f83aa1ce0 Author: Mooneer Salem Date: Tue Oct 20 00:39:08 2026 -0700 rade_reporting_mpp: put the whole recording through the fading channel. ch's MPP fading file holds 60 s of fading and ch stops when it runs out, so only the first 60 s of the ~100 s recording reached the receiver: about 4 of the 7 overs, rather than all of them, to decode the callsign from. When none of those 4 did, the test failed. ch is now given a fading file three times as long (the original's samples repeated), made when the test runs. The script also now checks ch's exit status, and fails without receiving anything if ch stopped part of the way through, rather than silently testing with what was left. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_013G7dQDriDiWypo4B3J83gx commit 84830a72ab1af3983ac6b71c6131d2a22b114be6 Author: Mooneer Salem Date: Tue Oct 20 06:03:10 2026 +0000 latest user manual PDF commit 17e046092219166653f0e366082f15955f3b07a2 Merge: 348fb8c6 1a77c065 Author: Mooneer Salem Date: Mon Oct 19 23:02:47 2026 -0700 [full ci] Merge branch 'v3.0-dev' into ms-rade-v2 Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_013G7dQDriDiWypo4B3J83gx commit 1a77c0651f420bc3015a87645e83458564323743 Merge: 5bfce276 2e33cfee Author: Mooneer Salem Date: Mon Oct 19 23:01:45 2026 -0700 Merge branch 'master' into v3.0-dev commit 2e33cfee0e6571ee0ad7b70bc94b19dd4a4b0eb0 Author: Mooneer Salem Date: Mon Oct 19 23:01:12 2026 -0700 Fix the voice keyer and other timed updates running twice as fast as intended (#1571) * Fix the voice keyer and other timed updates running twice as fast as intended. MainFrame::OnTimer() did its once-per-DT work (the voice keyer's state machine, the SNR display filter, the time since sync was lost and the RADE "--" reporting interval) for both ID_TIMER_UPDATE_OTHER and ID_TIMER_SNR, which both fire every DT, so all of these went twice as fast as they should: e.g. the voice keyer waited half its RX pause between calls, and stopped calling after 2.5 rather than 5 seconds of sync. ID_TIMER_SNR now only adds the next sample to the SNR plot. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_013G7dQDriDiWypo4B3J83gx * Add PR #1571 to changelog. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_013G7dQDriDiWypo4B3J83gx --------- Co-authored-by: Claude Opus 5.5 commit 4b007ad06e027d861f3e0fe730a6587acc49729d Author: Mooneer Salem Date: Mon Oct 19 22:03:21 2026 -0700 Add PR #1571 to changelog. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_013G7dQDriDiWypo4B3J83gx commit 8cd25903f59b2657ff35084da1dc0dbec5ad7df6 Author: Mooneer Salem Date: Mon Oct 19 22:00:17 2026 -0700 Fix the voice keyer and other timed updates running twice as fast as intended. MainFrame::OnTimer() did its once-per-DT work (the voice keyer's state machine, the SNR display filter, the time since sync was lost and the RADE "--" reporting interval) for both ID_TIMER_UPDATE_OTHER and ID_TIMER_SNR, which both fire every DT, so all of these went twice as fast as they should: e.g. the voice keyer waited half its RX pause between calls, and stopped calling after 2.5 rather than 5 seconds of sync. ID_TIMER_SNR now only adds the next sample to the SNR plot. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_013G7dQDriDiWypo4B3J83gx commit 348fb8c69a59a92813109b935de959b393c62cd4 Author: Mooneer Salem Date: Mon Oct 19 21:53:25 2026 -0700 GUI tests: watch for sync to be lost as soon as a full duplex transmission ends. Full duplex, FreeDV carries on transmitting once PTT is released, and RADE V2 finds that signal again within a second of losing sync at the end-of-over. The receive scenario only looked for the loss of sync after the callsign arrived, which with RADE V2 (sending it all the way through rather than at the end) is well after the receiver has found the signal again. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_013G7dQDriDiWypo4B3J83gx commit 576ed9395ec6c4e89cef5a9359fe3e109d44da88 Author: Mooneer Salem Date: Mon Oct 19 21:50:00 2026 -0700 GUI tests: send speech, not a tone, as the voice keyer's reply. A tone isn't something RADE V2 carries, and its receiver takes one for the end of an over again and again, so the reply never held sync for long enough to stop the voice keyer. The reply is now made from RADE's all.wav (FREEDVTEST_SPEECH, which the Windows GUI tests are given). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_013G7dQDriDiWypo4B3J83gx commit b8fbee01d75b4f4b91880b1b8d589a9870de8508 Merge: 8b2bed57 5bfce276 Author: Mooneer Salem Date: Mon Oct 19 19:04:06 2026 -0700 [full ci] Merge branch 'v3.0-dev' into ms-rade-v2 commit e69e7077dce9ec609c3ea202bd55001ac38f520b Author: Mooneer Salem Date: Mon Oct 19 23:09:59 2026 +0000 latest user manual PDF commit 5bfce276682159c10ccd97c82082646dca38c2f3 Author: Mooneer Salem Date: Mon Oct 19 16:03:26 2026 -0700 Update translations' bug report address to the new repo URL. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_013G7dQDriDiWypo4B3J83gx commit 0f47b27e547f59e22ced46fb64d55c589ec3ad63 Author: Mooneer Salem Date: Mon Oct 19 16:03:26 2026 -0700 Only fail GUI tests in sanitizer builds on sanitizer reports. As with the other tests: the GUI tests now go by freedv_set_test_pass_criteria(), and the test harness shows FreeDV's output whenever it has a sanitizer report in it (it otherwise only shows it when a test fails), so that ctest can see the report. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_013G7dQDriDiWypo4B3J83gx commit 60b84eec7af58ebab4270066fd178c78d38bb193 Merge: d0b5d381 7ecc1ab9 Author: Mooneer Salem Date: Mon Oct 19 16:00:11 2026 -0700 Merge branch 'master' into v3.0-dev Conflicts resolved: - CMakeLists.txt: v3.0-dev's test_rade_reporting.sh arguments, with master's freedv_set_test_pass_criteria(). - README.md, src/ongui.cpp: v3.0-dev's content (gettext packages, LaunchBrowser()) with master's new repo URL. - USER_MANUAL.html: master's wording, as in the merged USER_MANUAL.md. - USER_MANUAL.pdf: v3.0-dev's; it's regenerated from USER_MANUAL.md. Also updates the Help menu URLs that test/gui/test_main_window.py expects to the new repo URL. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_013G7dQDriDiWypo4B3J83gx commit 7ecc1ab9aa4ffcd0d7ec67820e8c02a12a59fcb8 Author: Mooneer Salem Date: Mon Oct 19 15:55:52 2026 -0700 Fix reported issue with TX not occurring. (#1568) * Fix reported issue with TX not occurring. * AI-proposed fix for failing tests. * Revert TxRxThread changes from #1567, implement minimum TX FIFO fill differently. * Revert main.cpp changes too. * Add PR #1568 to changelog. * Update tests, further unnecessary file cleanup. * Only fail sanitizer test runs on sanitizer reports; fix TxRxThreadTest file playback race. In sanitizer builds every ctest (including the unit tests) now passes unless a sanitizer reports something, crashes or times out; the tests' own checks only decide pass/fail in non-sanitizer builds. TxRxThreadTest's "files" round trips gave the file playback thread only 1 ms per turn to keep up with a loop running faster than real time, so slow or busy machines (e.g. sanitizer CI runners) saw spurious late underruns. Once a transmission is going, give that thread the time it would have had before the radio's sound card takes its block. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_013G7dQDriDiWypo4B3J83gx * Windows: only fail sanitizer unit test runs on sanitizer reports, crashes or timeouts. Matches what ctest now does for sanitizer builds on Linux and macOS: RunUnitTests.ps1 gains -SanitizerMode, which the UBSAN/ASAN unit-tests jobs pass, so a test whose own (timing-dependent) checks fail under sanitizer overhead no longer fails the job. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_013G7dQDriDiWypo4B3J83gx --------- Co-authored-by: Claude Opus 5.5 commit 1a9429ade3587be4238d49d35df3bb2835e8086e Author: Mooneer Salem Date: Mon Oct 19 13:31:41 2026 -0700 Windows: only fail sanitizer unit test runs on sanitizer reports, crashes or timeouts. Matches what ctest now does for sanitizer builds on Linux and macOS: RunUnitTests.ps1 gains -SanitizerMode, which the UBSAN/ASAN unit-tests jobs pass, so a test whose own (timing-dependent) checks fail under sanitizer overhead no longer fails the job. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_013G7dQDriDiWypo4B3J83gx commit 36fc1a817356c2b9f72dcc186453f601239e3216 Author: Mooneer Salem Date: Mon Oct 19 13:19:22 2026 -0700 Only fail sanitizer test runs on sanitizer reports; fix TxRxThreadTest file playback race. In sanitizer builds every ctest (including the unit tests) now passes unless a sanitizer reports something, crashes or times out; the tests' own checks only decide pass/fail in non-sanitizer builds. TxRxThreadTest's "files" round trips gave the file playback thread only 1 ms per turn to keep up with a loop running faster than real time, so slow or busy machines (e.g. sanitizer CI runners) saw spurious late underruns. Once a transmission is going, give that thread the time it would have had before the radio's sound card takes its block. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_013G7dQDriDiWypo4B3J83gx commit 1052493e2e636a76b9e19191f3a784dcbe82a334 Merge: 68c11766 764deda2 Author: Mooneer Salem Date: Mon Oct 19 09:08:56 2026 -0700 Merge branch 'master' into ms-mic-zero-audio commit 764deda2e516dac0929e31a5f3eb0d99ba0c965e Author: Mooneer Salem Date: Mon Oct 19 09:08:23 2026 -0700 Update documentation to reflect new repo URL. (#1569) * Update documentation to reflect new repo URL. * Remove no longer needed Python package script. * Add PR #1569 to changelog. commit c39fd723a8a4bd315250c9ac304af9ae75c7743e Author: Mooneer Salem Date: Mon Oct 19 08:27:31 2026 -0700 Add PR #1569 to changelog. commit c2340847208c1fe18ce4cc68bdaa018fea07c8b8 Author: Mooneer Salem Date: Sun Oct 18 23:53:53 2026 -0700 Remove no longer needed Python package script. commit 7310b305e1bbcd61ce971ff9650dd37abd73681d Author: Mooneer Salem Date: Sun Oct 18 23:50:15 2026 -0700 Update documentation to reflect new repo URL. commit 68c11766a43afe2896d74d4c9d1e4c392796f202 Author: Mooneer Salem Date: Sun Oct 18 22:43:44 2026 -0700 Update tests, further unnecessary file cleanup. commit 20178d08bb21356ab05fdf4bb985dc2d65dfddb0 Author: Mooneer Salem Date: Sun Oct 18 22:43:33 2026 -0700 Add PR #1568 to changelog. commit f2e262161e3f12e007e64c1630f2db4b0285151f Author: Mooneer Salem Date: Sun Oct 18 19:04:57 2026 -0700 Revert main.cpp changes too. commit 263b16fb228392265a8e0fb85139de5c3fcff7ac Author: Mooneer Salem Date: Sun Oct 18 10:10:59 2026 -0700 Revert TxRxThread changes from #1567, implement minimum TX FIFO fill differently. commit 3110893e139bcfbb628fc5ccc905b92fe16fca4b Author: Mooneer Salem Date: Sun Oct 18 07:06:00 2026 -0700 AI-proposed fix for failing tests. commit 18bad83a72bd7b26de0254f47e64c54a881736f0 Author: Mooneer Salem Date: Sat Oct 17 19:21:10 2026 -0700 Fix reported issue with TX not occurring. commit 8b2bed57506df187ba850ca75bbd42da53ffdf36 Author: Mooneer Salem Date: Sat Oct 17 10:05:35 2026 -0700 Point test_log_qso's reporter test at the renamed hostname setting It was still setting /Reporting/FreeDV/Hostname, so freedvtest filled in Hostname3.0 with its no-server default and the fake reporter's station never appeared. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 4df8cb4f5e6edc684fe4ca4d6bdd3774ac98293b Author: Mooneer Salem Date: Sat Oct 17 00:35:04 2026 -0700 [full ci] RADEV1 -> RADEV2 in tests. commit 316ed1f078fc0500e51ed39b1c84a2f57d26e612 Author: Mooneer Salem Date: Fri Oct 16 23:20:03 2026 -0700 [full ci] Config for GUI tests needs to change temporarily too. commit 298483bc541e69738f033f39200d324a5deb2e66 Author: Mooneer Salem Date: Fri Oct 16 22:33:15 2026 -0700 [full ci] Actually fix syntax error this time. commit 9719353c8b700c8b731495273a1d07c7b6e863db Author: Mooneer Salem Date: Fri Oct 16 22:08:46 2026 -0700 [full ci] Fix accidental syntax error. commit 41c1052db926cce8d799b7fba25429e28fa38281 Author: Mooneer Salem Date: Fri Oct 16 21:47:06 2026 -0700 [full ci] Allow tests to override FreeDV Reporter hostname. commit 572f297970ce692149b84fe14dd40b324bde4d98 Merge: f9430966 a78a4252 Author: Mooneer Salem Date: Fri Oct 16 21:37:54 2026 -0700 [full ci] Merge branch 'master' into ms-rade-v2 commit f94309660ad62a010559ecce97adcc931634523a Author: Mooneer Salem Date: Fri Oct 16 21:20:33 2026 -0700 [full ci] Revert "Fake change." This reverts commit f0f8d1b7b8bf8d7bfbaaaec49569f1852d35031e. commit f0f8d1b7b8bf8d7bfbaaaec49569f1852d35031e Author: Mooneer Salem Date: Fri Oct 16 21:20:29 2026 -0700 Fake change. commit e15d3264cc372847ac76329889169251e508615c Author: Mooneer Salem Date: Fri Oct 16 20:43:16 2026 -0700 Fix FreeDVInterfaceTest. commit 643e06049017381f11ee1c45433c53db298ead63 Merge: b7d2dee7 d0b5d381 Author: Mooneer Salem Date: Fri Oct 16 19:02:13 2026 -0700 [full ci] Merge branch 'v3.0-dev' into ms-rade-v2 Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit b7d2dee701dec76497eb1a42da1f4d73f0acabb7 Author: Mooneer Salem Date: Fri Oct 16 19:02:13 2026 -0700 Judge TxRxThreadTest's round trips against RADE V2's software-only loss As rade_loss does on this branch: leave the first 100 and last 300 frames received out of the loss (loss.py's --clip_start and --clip_end), and compare it with RADE V2 on its own with the same speech, no threads or sound cards (rade_tx_wav/rade_rx_wav --v2, done in-process). It has to be within 10% of that either way, rather than under 0.15, which was for V1. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 0178a506056204e01d1dca7d7bd1188644994087 Author: Mooneer Salem Date: Fri Oct 16 19:02:13 2026 -0700 Fix the v3.0-dev merge: unused TX FIFO target, rade-loss-baseline needs, backend branch The merge kept v3.0-dev's wait for the rest of a microphone frame but left the branch's TX_OUTFIFO_TARGET_MS and nsam_outfifo_target behind, unused, which broke every build under -Werror. It also dropped rade-loss-baseline from the needs of the Linux and macOS jobs that read its outputs. And TxRxThreadTest from v3.0-dev needs SpeechSample.h, which freedv-backend's ms-radev2-build-fix branch doesn't have, so build against radev2-dev. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit d0b5d381233a1ee45db574d23e0a5d21f795987e Author: Mooneer Salem Date: Sat Oct 17 00:04:47 2026 +0000 latest user manual PDF commit c2e6089920cc6b7544fce5d469aaa7dd77dd48e9 Merge: afd36a96 944900d9 Author: Mooneer Salem Date: Fri Oct 16 17:04:33 2026 -0700 [full ci] Merge branch 'v3.0-dev' into ms-rade-v2 commit 944900d949624f878c089eff56c2529fe9c58888 Author: Mooneer Salem Date: Fri Oct 16 16:59:14 2026 -0700 Test the TX/RX threads without sound cards, start transmissions cleanly, and instrument rade_loss (#1562) * Test a whole transmission and a steady tone through the TX/RX threads without sound cards or a clock Speech is sent and received again through the real threads, with a model of two sound cards on one clock in place of the real ones: in even and uneven blocks, at 48, 44.1 and 8 kHz, and from and to files as the rade_loss test does it. Sync has to be found promptly and kept, nothing lost on the way in, no holes in the signal once it's going, and what comes back has to be what was sent (by RADE's own measure of loss). A tone through each thread in analog mode, with the sound card taking its output stalling at random, may only be broken where there was nowhere to put it or nothing to give. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Start transmissions without silence mixed into the first words or holes in the signal When the microphone hadn't yet delivered a full frame, the TX thread sent a frame of silence in its place, which is how it built up the 640 ms it kept ready for the radio's sound card: the first half second or so of each transmission went out a frame at a time with silence in between. And the radio's sound card was given the transmission from the start, with nothing in hand, so it could ask for more than had been made and get silence (a hole in the signal). The TX thread now waits for the rest of a frame instead, and the radio's sound card isn't given a transmission until 240 ms of it (two RADE modem frames) is ready. Until then it gets silence, with the VOX tone if that's on. What's said now goes out whole, about 240 ms after it's said rather than 640 ms. The round trip and tone tests now require that: nothing but the speech sent, no holes in the signal and the radio's sound card never going short. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Report the sound card FIFO counters at the end of a test run, and decode rade_loss's recording again when it fails FreeDV's command-line test mode now logs how often each sound card's FIFO overflowed or ran short, so that a failing run says whether the sound cards were involved. When rade_loss fails, it decodes the same recording a second time (telling damage in the recording from something lost while receiving it) and keeps the recording, the feature files and both halves' logs. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * TEMPORARY: run only a macOS build and repeated rade_loss tests for this branch To be removed before merging. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Count the radio going short only while transmitting, and check rade_loss's recording against what FreeDV sent The radio's output sound card counted as short whenever FreeDV wasn't transmitting, so the count was much the same for every run. Now only the times while transmitting are reported. rade_loss now has FreeDV keep a copy of what it sent (-txoutfile) and compares the recording with it: stretches missing or extra, and holes of silence, are reported with where they are. FreeDV's plot taps dropping samples while sending or receiving are reported alongside: the TX/RX thread carrying on while ordinary threads (the recorder's among them) don't get to run, as when the machine stalls. The temporary validation workflow keeps the copy and the check's report, and flags runs where the machine looks to have stalled. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Report how promptly the radio's sound card was served while transmitting, and tell gaps in FreeDV's own copy from ones in the recording At the end of a test run FreeDV now reports, for the radio's output while transmitting: the length of its blocks, the longest time between its asking for more and how often that was more than twice a block, and how many blocks of silence it was given while a transmission was starting or while there was signal to send that couldn't be read. That tells FreeDV not keeping up from the sound card not asking. check_rade_recording.py no longer takes a stretch missing from FreeDV's own copy of what it sent, where its tap dropped samples, for the recording having something extra. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Tell a stalled machine from a stalled sound system in rade_loss's failures When the radio's sound card goes a while without asking FreeDV for audio, this records what's needed to say why: * Core Audio's own timestamps for each device (how far its sample position moved on across the longest gap between calls, how often it jumped ahead) and the overloads it reported, logged when the device stops. * A watchdog thread in test mode that does nothing but wake every 5 ms, at real-time priority on macOS, reporting how late it woke while transmitting and when, next to when the radio's longest gap was. * Core Audio's log (overloads, skipped cycles) over the transmission, saved by test_rade_loss.sh on macOS. check_rade_recording.py reports all of these and says which it was. It also now recognises a gap in FreeDV's own copy that only partly matches the recording (reporting the rest as lost from the recording), and says when the sound card was slow to ask while FreeDV had signal ready. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Say the machine stalled in rade_loss's check even when audio went missing afterwards After the whole machine stalls, the sound system may make up for lost time by skipping, so audio can be missing from the recording as well as added. The check now gives the machine's verdict whenever the watchdog was held up at the same moment, and only says "real-time" of the watchdog when it ran at real-time priority. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Run the regular CI for this branch again, and add release notes Removes the temporary macOS-only workflow and the regular workflows' exclusion of this branch. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Remove the rade_loss instrumentation that's done its job The validation runs found every rade_loss failure to be the machine or the sound system stalling, or the recorder falling behind, never FreeDV. So take out what was there to find that out: the Core Audio callback timing in MacAudioDevice, the radio output callback's gap timing, the watchdog thread, the capture of Core Audio's log, and check_rade_recording.py with FreeDV's copy of what it sent. Keep the sound card FIFO summary at the end of a test run and the second decode of a failed rade_loss recording. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp --------- Co-authored-by: Claude Opus 5.5 commit d35ee753b0156180414f5a4d72b845fa6aa5fe89 Merge: a1bfbb4a 1219e9b5 Author: Mooneer Salem Date: Fri Oct 16 16:58:56 2026 -0700 Merge branch 'v3.0-dev' into ms-pipeline-offline-tests commit a78a4252939f41881bb62976f3f451f304b1b77b Author: Mooneer Salem Date: Fri Oct 16 23:53:37 2026 +0000 latest user manual PDF commit 1219e9b54b9c2fa37ee61bc55486cc921e951b6f Merge: 44096994 e4507b4c Author: Mooneer Salem Date: Fri Oct 16 16:52:59 2026 -0700 Merge branch 'master' into v3.0-dev commit e4507b4c5dadffaf7756c81143a2c8b958d2be3f Author: Mooneer Salem Date: Fri Oct 16 16:51:03 2026 -0700 Start transmissions without silence mixed into the first words or holes in the signal (#1567) When the microphone hadn't delivered a whole frame, the TX thread sent a frame of silence in its place, so the first half second or so of each transmission went out with silence mixed in. The radio's sound card was also given the transmission with nothing in hand, so it could ask for more than had been made and send a hole in the signal. The TX thread now waits for the rest of a frame, and the radio's sound card isn't given a transmission until 240 ms of it is ready (silence, with the VOX tone if that's on, until then). Adds TxRxThreadTest cases that send RADE's speech recording through the TX and RX threads, and a steady tone through each, with a model of two sound cards on one clock. From #1562 on v3.0-dev. Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp Co-authored-by: Claude Opus 5.5 commit a1bfbb4adc1f34ccd73eb8f2af1c7e60ab2d173e Merge: 0f81e378 44096994 Author: Mooneer Salem Date: Fri Oct 16 13:56:52 2026 -0700 [full ci] Merge branch 'v3.0-dev' into ms-pipeline-offline-tests Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 0f81e378d838f0036dc08720e7a40c3bf6e7e4e9 Author: Mooneer Salem Date: Fri Oct 16 13:56:51 2026 -0700 Remove the rade_loss instrumentation that's done its job The validation runs found every rade_loss failure to be the machine or the sound system stalling, or the recorder falling behind, never FreeDV. So take out what was there to find that out: the Core Audio callback timing in MacAudioDevice, the radio output callback's gap timing, the watchdog thread, the capture of Core Audio's log, and check_rade_recording.py with FreeDV's copy of what it sent. Keep the sound card FIFO summary at the end of a test run and the second decode of a failed rade_loss recording. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit ba8b12784c37faa9d79aeadade0daf7f48c95131 Author: Mooneer Salem Date: Fri Oct 16 11:33:44 2026 -0700 Start transmissions without silence mixed into the first words or holes in the signal When the microphone hadn't delivered a whole frame, the TX thread sent a frame of silence in its place, so the first half second or so of each transmission went out with silence mixed in. The radio's sound card was also given the transmission with nothing in hand, so it could ask for more than had been made and send a hole in the signal. The TX thread now waits for the rest of a frame, and the radio's sound card isn't given a transmission until 240 ms of it is ready (silence, with the VOX tone if that's on, until then). Adds TxRxThreadTest cases that send RADE's speech recording through the TX and RX threads, and a steady tone through each, with a model of two sound cards on one clock. From #1562 on v3.0-dev. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 440969942d2f8153e7f6282afe12bfe0878bc808 Author: Mooneer Salem Date: Fri Oct 16 17:25:17 2026 +0000 latest user manual PDF commit 6af1d294559511181bf2a46fc4eec5e7b28277df Merge: 4f391e28 00744313 Author: Mooneer Salem Date: Fri Oct 16 10:25:14 2026 -0700 [full ci] Merge branch 'v3.0-dev' into ms-pipeline-offline-tests commit 00744313e4e578c758df66d3bbaca0ca60ca2371 Author: Mooneer Salem Date: Fri Oct 16 10:23:09 2026 -0700 Run the GUI tests on Windows, and fix the two Windows bugs they found (#1566) * Fix five bugs found by running the GUI tests on macOS (from #1563) (#1564) * FreeDV Reporter: don't give a direction of 0 to stations without a usable grid square Redoing the list (e.g. on changing the band filter) recalculated every direction from the stored heading, which is 0 for a station whose grid square (or one's own) can't be used, rather than leaving it blank. The display_options GUI test now changes the band filter and checks again. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * macOS: send silence when there's nothing to send, rather than the last block again When an output callback left its buffer alone (as FreeDV's does for the radio's sound card with nothing to send), the block before was sent over again, on and on: after a transmission with the VOX tone on, that kept a tone going into the radio. The buffer is now cleared before each callback, as on Windows. Found by the main window's test_tone GUI scenario. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Use the default main window size in place of a saved one too small to be of use Only negative and enormous saved sizes were replaced. On Linux the window manager won't make the window smaller than it can be, but macOS made a 10x10 window when asked to. Sizes below 300x200 now get the default too. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Don't let a frequency read from the radio before a change undo the change What the radio reports is read on the rig control thread and acted on later on the UI thread. A reading taken just before FreeDV asked for a new frequency could then arrive just after, putting FreeDV back on the old one until the next reading: anything done meanwhile went by the old frequency. Switching to analog just after changing band chose the old band's sideband (USB rather than LSB on 40 m), which the rig_control:analog_sideband GUI test caught on macOS CI every time. The rig controllers now note what frequency was last asked for, and readings that differ from it are ignored until the radio's seen to be there (or for 3 s at most, for radios that can't be tuned exactly where they're asked). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Don't pass built-up messages to the logger as its format The logger formats messages later, on a thread of its own, keeping only a pointer to the format (which it expects to be a string literal) while copying the arguments. 42 places in the macOS and Windows audio code and ParallelStep passed a string they'd just built as the format, so it could be gone by the time it was read: AddressSanitizer caught it on macOS (heap-use-after-free in ulog_async's drain(), from MacAudioEngine::start(), when the Settings window's audio device lists were refreshed). Those strings now go in as a "%s" argument, which also stops any '%' in them being taken as a conversion. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Release notes for the fixes from the macOS GUI tests Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Reword a comment codespell took for a typo Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * macOS CI: make RADE's venv with Python 3.13 The macos-26-arm64 runner image of 2026-10-13 has Python 3.15 as python3, which torch has no packages for yet, so the RADE venv couldn't be set up (test-app-folder failed on macOS 15 and 26). It's now made with Python 3.13, from actions/setup-python (which the images keep a copy of). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Forgot to include OpenSSL devel packages in README. * Add pulseaudio-utils to allow unit tests to work. --------- Co-authored-by: Claude Opus 5.5 * latest user manual PDF * Scale the main window's default size to the display The size used when none has been saved (800x780), and the smallest saved size that's kept, were in pixels, which on Windows are the screen's own: at 200%, 800 of them are only as wide as 400 elsewhere. The controls on either side then took all of the window, leaving the waterfall and spectrum no room at all (0 pixels wide on a laptop at 200%). Both are now scaled to suit the display. Nothing changes on Linux or macOS, where sizes are already scaled. The saved size's default is now -1 (none saved), so that a first start gets the scaled default rather than 800x780. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * FreeDV Reporter: remove what's left of a flag from the status message on Windows too A flag that's lost its end (as backspacing over it does), and stray parts of one, are taken out of the status message. On Windows, wxString holds UTF-16, in which each of the characters a flag is made of takes two places, so looking at it a place at a time never found them and nothing was taken out. It's now looked at a whole character at a time. Found by the reporter:status_message_flags GUI test on Windows. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Build the GUI test agent for Windows, and keep test builds out of the registry - The test agent didn't compile for Windows: LibreSSL's headers (from TcpConnectionHandler.h) warn about replacing wincrypt.h's names, which wxWidgets has already brought in, and warnings are errors. Nothing there uses wincrypt.h's names, so the warning is turned off for that include. - Even when given a configuration file, FreeDV on Windows reads its default configuration from the registry, and keeps the last one used there. Test builds keep both as files in the folder named by $FREEDV_TEST_CONFIG_DIR instead, if it's set, as they are on Linux and macOS, so that tests never touch the user's settings. - Closing a popup "as clicking elsewhere would" now works on Windows, where wxWidgets dismisses one when it stops being the active window rather than on a click. - get() reports a combo control's popupShown (its list ignores the mouse until it's finished sliding open, on Windows) and every window's clientSize. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Run the GUI tests on Windows freedvtest: - Audio: existing loopback devices stand in for the ones tests ask for, as on macOS: Virtual Audio Cable's lines, then VB-Cable (which now and then leaves a few milliseconds out). Played into and recorded from with the sounddevice module (wasapi.py), in the test's own process: in a venv, Windows' python.exe runs the real one as a program of its own, which stopping the first leaves running. With fewer than four devices, as on Windows CI, they're shared (one for the radio's input and output, one for the microphone and speaker), and shared() says so, for the few checks that can't be made then. - start(): a stand-in home folder for $FREEDV_TEST_CONFIG_DIR and %LOCALAPPDATA% (where the Setup Wizard imports other programs' settings from, each in a folder of its own: user_config_path()). - Configuration files are escaped as wxFileConfig reads them: a backslash starts an escape sequence there, which mangled every Windows path. - The FreeDV Reporter server's files are copied when they can't be linked; serial ports are Unavailable (they'd need com0com); combo_popup() waits for the list to have opened. run_tests.py runs scenarios as CTest does, for Windows, where FreeDV is cross-compiled elsewhere and there's no build folder. Tests: checks that need separate devices are left out when they're shared; parts that need a device to be plugged in, taken away or run at another rate are left out on Windows as on macOS; unwritable folders are made with icacls; the Use/Load Default Configuration steps wait for the change; the FreeDV Reporter window is given room as on macOS; errors Windows logs about missing serial ports are expected; bad_command_line can't see the usage text (written to the console window, if any) on Windows. CI: build-unit-tests also builds FreeDV for testing (x86_64), and a new gui-tests job runs the GUI tests on Windows with VAC Lite and VB-Cable. setup-windows-audio-drivers can leave out installing FreeDV and SoX. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Describe how the GUI tests work Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Release note for the Windows GUI tests Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Don't copy the status message just to look through it [full ci] clang-tidy (performance-unnecessary-copy-initialization) caught it. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Keep building with wxWidgets 3.0, which has no FromDIP() [full ci] Ubuntu 22.04's wxWidgets is 3.0. Only Linux builds use it, and sizes are already scaled there, so it keeps the unscaled sizes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Ask for a window that fits on CI's screen in window_geometry [full ci] Windows makes no window much taller than the screen, and the Windows runners' screens are 1024x768: a 900x800 window became 900x788. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Keep and print the output of GUI test tries that failed before a retry [full ci] run_tests.py overwrote a scenario's log when it ran it again, so a scenario that only passed on a retry left nothing to look into. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Don't close the FreeDV Reporter window twice when asked to exit twice [full ci] Closing the main window yields to handle the Reporter dialog's last messages. A second request to close handled meanwhile closed and destroyed the dialog and set m_reporterDialog to null, which the first then used. It's now refused while the first finishes. Seen as a crash on exit in reporter:status_message_flags on Windows, where the test agent asks the main window to close every 20 ms until it does. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Don't leave Test PTT greyed out when the port can't be opened The test's thread waited on a condition variable that the controller's error callback could signal before the wait had begun (a serial port that can't be opened fails at once), and so waited forever, leaving Test PTT disabled. Whether it's over is now kept track of under the mutex and checked by the wait. Seen as settings:ptt_test timing out on Windows CI. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Listen for the changed tone in test_tone until it's heard [full ci] On Windows CI, a single recording right after the change was once nearly silent. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Windows: fix the main window being too small on high DPI displays, and broken flags left in FreeDV Reporter's status message (#1565) * Scale the main window's default size to the display The size used when none has been saved (800x780), and the smallest saved size that's kept, were in pixels, which on Windows are the screen's own: at 200%, 800 of them are only as wide as 400 elsewhere. The controls on either side then took all of the window, leaving the waterfall and spectrum no room at all (0 pixels wide on a laptop at 200%). Both are now scaled to suit the display. Nothing changes on Linux or macOS, where sizes are already scaled. The saved size's default is now -1 (none saved), so that a first start gets the scaled default rather than 800x780. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * FreeDV Reporter: remove what's left of a flag from the status message on Windows too A flag that's lost its end (as backspacing over it does), and stray parts of one, are taken out of the status message. On Windows, wxString holds UTF-16, in which each of the characters a flag is made of takes two places, so looking at it a place at a time never found them and nothing was taken out. It's now looked at a whole character at a time. Found by the reporter:status_message_flags GUI test on Windows. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Release notes for the Windows fixes Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Don't copy the status message just to look through it clang-tidy (performance-unnecessary-copy-initialization) caught it. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Keep building with wxWidgets 3.0, which has no FromDIP() Ubuntu 22.04's wxWidgets is 3.0. Only Linux builds use it, and sizes are already scaled there, so it keeps the unscaled sizes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Don't close the FreeDV Reporter window twice when asked to exit twice Closing the main window yields to handle the Reporter dialog's last messages. A second request to close handled meanwhile closed and destroyed the dialog and set m_reporterDialog to null, which the first then used. It's now refused while the first finishes. Seen as a crash on exit in reporter:status_message_flags on Windows, where the test agent asks the main window to close every 20 ms until it does. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Don't leave Test PTT greyed out when the port can't be opened The test's thread waited on a condition variable that the controller's error callback could signal before the wait had begun (a serial port that can't be opened fails at once), and so waited forever, leaving Test PTT disabled. Whether it's over is now kept track of under the mutex and checked by the wait. Seen as settings:ptt_test timing out on Windows CI. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp --------- Co-authored-by: Claude Opus 5.5 * Use ubuntu-slim for 'scope' to allow other builds to proceed. --------- Co-authored-by: Claude Opus 5.5 commit 4957a4d913e50aaa157b38a65afea3604bdc3d08 Author: Mooneer Salem Date: Fri Oct 16 10:14:55 2026 -0700 Use ubuntu-slim for 'scope' to allow other builds to proceed. commit 6aa261065e8ff30d89489e3d8f682a546953abd2 Merge: 2ed694e1 392fe4c3 Author: Mooneer Salem Date: Fri Oct 16 09:51:13 2026 -0700 Merge branch 'master' into ms-windows-gui-tests commit 50965081b22b0eba9a89c286b2b6641dbd9a3deb Author: Mooneer Salem Date: Fri Oct 16 16:43:31 2026 +0000 latest user manual PDF commit 392fe4c31afa951e9ba90ef72662c6737f64ec92 Author: Mooneer Salem Date: Fri Oct 16 09:40:44 2026 -0700 Windows: fix the main window being too small on high DPI displays, and broken flags left in FreeDV Reporter's status message (#1565) * Scale the main window's default size to the display The size used when none has been saved (800x780), and the smallest saved size that's kept, were in pixels, which on Windows are the screen's own: at 200%, 800 of them are only as wide as 400 elsewhere. The controls on either side then took all of the window, leaving the waterfall and spectrum no room at all (0 pixels wide on a laptop at 200%). Both are now scaled to suit the display. Nothing changes on Linux or macOS, where sizes are already scaled. The saved size's default is now -1 (none saved), so that a first start gets the scaled default rather than 800x780. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * FreeDV Reporter: remove what's left of a flag from the status message on Windows too A flag that's lost its end (as backspacing over it does), and stray parts of one, are taken out of the status message. On Windows, wxString holds UTF-16, in which each of the characters a flag is made of takes two places, so looking at it a place at a time never found them and nothing was taken out. It's now looked at a whole character at a time. Found by the reporter:status_message_flags GUI test on Windows. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Release notes for the Windows fixes Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Don't copy the status message just to look through it clang-tidy (performance-unnecessary-copy-initialization) caught it. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Keep building with wxWidgets 3.0, which has no FromDIP() Ubuntu 22.04's wxWidgets is 3.0. Only Linux builds use it, and sizes are already scaled there, so it keeps the unscaled sizes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Don't close the FreeDV Reporter window twice when asked to exit twice Closing the main window yields to handle the Reporter dialog's last messages. A second request to close handled meanwhile closed and destroyed the dialog and set m_reporterDialog to null, which the first then used. It's now refused while the first finishes. Seen as a crash on exit in reporter:status_message_flags on Windows, where the test agent asks the main window to close every 20 ms until it does. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Don't leave Test PTT greyed out when the port can't be opened The test's thread waited on a condition variable that the controller's error callback could signal before the wait had begun (a serial port that can't be opened fails at once), and so waited forever, leaving Test PTT disabled. Whether it's over is now kept track of under the mutex and checked by the wait. Seen as settings:ptt_test timing out on Windows CI. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp --------- Co-authored-by: Claude Opus 5.5 commit f97dba81aa349e23dd2917be729f5c2cc7f781b9 Author: Mooneer Salem Date: Fri Oct 16 07:13:14 2026 -0700 Don't leave Test PTT greyed out when the port can't be opened The test's thread waited on a condition variable that the controller's error callback could signal before the wait had begun (a serial port that can't be opened fails at once), and so waited forever, leaving Test PTT disabled. Whether it's over is now kept track of under the mutex and checked by the wait. Seen as settings:ptt_test timing out on Windows CI. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 2ed694e1df0e6b51475216044e69ccf9dd4d09eb Author: Mooneer Salem Date: Fri Oct 16 07:13:14 2026 -0700 Listen for the changed tone in test_tone until it's heard [full ci] On Windows CI, a single recording right after the change was once nearly silent. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit c2dbac8b0dbefda4f92962f90fc94a3534c3d20e Author: Mooneer Salem Date: Fri Oct 16 07:13:14 2026 -0700 Don't leave Test PTT greyed out when the port can't be opened The test's thread waited on a condition variable that the controller's error callback could signal before the wait had begun (a serial port that can't be opened fails at once), and so waited forever, leaving Test PTT disabled. Whether it's over is now kept track of under the mutex and checked by the wait. Seen as settings:ptt_test timing out on Windows CI. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 296e6cfdace6928dc1d6155bc192c9921eeddca5 Author: Mooneer Salem Date: Fri Oct 16 06:51:58 2026 -0700 Don't close the FreeDV Reporter window twice when asked to exit twice Closing the main window yields to handle the Reporter dialog's last messages. A second request to close handled meanwhile closed and destroyed the dialog and set m_reporterDialog to null, which the first then used. It's now refused while the first finishes. Seen as a crash on exit in reporter:status_message_flags on Windows, where the test agent asks the main window to close every 20 ms until it does. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 1c44737f2726326f1f0fe40b1383242f1d4f3e3a Author: Mooneer Salem Date: Fri Oct 16 06:51:58 2026 -0700 Don't close the FreeDV Reporter window twice when asked to exit twice [full ci] Closing the main window yields to handle the Reporter dialog's last messages. A second request to close handled meanwhile closed and destroyed the dialog and set m_reporterDialog to null, which the first then used. It's now refused while the first finishes. Seen as a crash on exit in reporter:status_message_flags on Windows, where the test agent asks the main window to close every 20 ms until it does. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 2808b858f44eefdeaafd08e6fb6c6472d54aba9c Author: Mooneer Salem Date: Fri Oct 16 05:34:24 2026 -0700 Keep and print the output of GUI test tries that failed before a retry [full ci] run_tests.py overwrote a scenario's log when it ran it again, so a scenario that only passed on a retry left nothing to look into. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 2ec15493e53a6aabdcec4f81262315bf491d0f83 Author: Mooneer Salem Date: Fri Oct 16 03:10:52 2026 -0700 Ask for a window that fits on CI's screen in window_geometry [full ci] Windows makes no window much taller than the screen, and the Windows runners' screens are 1024x768: a 900x800 window became 900x788. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit af643561f96a374a99c78d76d9aaa6832ceb52a5 Author: Mooneer Salem Date: Fri Oct 16 00:52:12 2026 -0700 Keep building with wxWidgets 3.0, which has no FromDIP() Ubuntu 22.04's wxWidgets is 3.0. Only Linux builds use it, and sizes are already scaled there, so it keeps the unscaled sizes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit e6cf3f8922c5eff99afcce44587d3f665fbb6f7a Author: Mooneer Salem Date: Fri Oct 16 00:52:12 2026 -0700 Keep building with wxWidgets 3.0, which has no FromDIP() [full ci] Ubuntu 22.04's wxWidgets is 3.0. Only Linux builds use it, and sizes are already scaled there, so it keeps the unscaled sizes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 11440ae0c5395b6bc05c4591a89a859b275bd594 Author: Mooneer Salem Date: Thu Oct 15 23:29:36 2026 -0700 Don't copy the status message just to look through it clang-tidy (performance-unnecessary-copy-initialization) caught it. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 569a398e16c8a258b7348db33475b1a8c3c267f3 Author: Mooneer Salem Date: Thu Oct 15 23:29:36 2026 -0700 Don't copy the status message just to look through it [full ci] clang-tidy (performance-unnecessary-copy-initialization) caught it. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 27eb3cab793ecb56ebee97ef444e14becb13ddc2 Author: Mooneer Salem Date: Thu Oct 15 23:14:31 2026 -0700 Release notes for the Windows fixes Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 5df4964673cd72957805a7a561ed4647d3ea82cf Author: Mooneer Salem Date: Thu Oct 15 23:16:16 2026 -0700 Release note for the Windows GUI tests Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 50b0a21f5b2bc5e51fdc532e6c329fcc711d400b Author: Mooneer Salem Date: Thu Oct 15 23:14:31 2026 -0700 FreeDV Reporter: remove what's left of a flag from the status message on Windows too A flag that's lost its end (as backspacing over it does), and stray parts of one, are taken out of the status message. On Windows, wxString holds UTF-16, in which each of the characters a flag is made of takes two places, so looking at it a place at a time never found them and nothing was taken out. It's now looked at a whole character at a time. Found by the reporter:status_message_flags GUI test on Windows. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 5d3900871d7060707b7a3c36e08392fb38bee05c Author: Mooneer Salem Date: Thu Oct 15 23:14:21 2026 -0700 Scale the main window's default size to the display The size used when none has been saved (800x780), and the smallest saved size that's kept, were in pixels, which on Windows are the screen's own: at 200%, 800 of them are only as wide as 400 elsewhere. The controls on either side then took all of the window, leaving the waterfall and spectrum no room at all (0 pixels wide on a laptop at 200%). Both are now scaled to suit the display. Nothing changes on Linux or macOS, where sizes are already scaled. The saved size's default is now -1 (none saved), so that a first start gets the scaled default rather than 800x780. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 6f2e660a3efe9f9dfe71be1de64f396630f9a693 Author: Mooneer Salem Date: Thu Oct 15 23:14:53 2026 -0700 Describe how the GUI tests work Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 2331feda33f4e611a5519744b561a6e66788c3c1 Author: Mooneer Salem Date: Thu Oct 15 23:14:53 2026 -0700 Run the GUI tests on Windows freedvtest: - Audio: existing loopback devices stand in for the ones tests ask for, as on macOS: Virtual Audio Cable's lines, then VB-Cable (which now and then leaves a few milliseconds out). Played into and recorded from with the sounddevice module (wasapi.py), in the test's own process: in a venv, Windows' python.exe runs the real one as a program of its own, which stopping the first leaves running. With fewer than four devices, as on Windows CI, they're shared (one for the radio's input and output, one for the microphone and speaker), and shared() says so, for the few checks that can't be made then. - start(): a stand-in home folder for $FREEDV_TEST_CONFIG_DIR and %LOCALAPPDATA% (where the Setup Wizard imports other programs' settings from, each in a folder of its own: user_config_path()). - Configuration files are escaped as wxFileConfig reads them: a backslash starts an escape sequence there, which mangled every Windows path. - The FreeDV Reporter server's files are copied when they can't be linked; serial ports are Unavailable (they'd need com0com); combo_popup() waits for the list to have opened. run_tests.py runs scenarios as CTest does, for Windows, where FreeDV is cross-compiled elsewhere and there's no build folder. Tests: checks that need separate devices are left out when they're shared; parts that need a device to be plugged in, taken away or run at another rate are left out on Windows as on macOS; unwritable folders are made with icacls; the Use/Load Default Configuration steps wait for the change; the FreeDV Reporter window is given room as on macOS; errors Windows logs about missing serial ports are expected; bad_command_line can't see the usage text (written to the console window, if any) on Windows. CI: build-unit-tests also builds FreeDV for testing (x86_64), and a new gui-tests job runs the GUI tests on Windows with VAC Lite and VB-Cable. setup-windows-audio-drivers can leave out installing FreeDV and SoX. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit ddc407e8e182a539c6e7d2fef70dda26c23a9311 Author: Mooneer Salem Date: Thu Oct 15 23:14:53 2026 -0700 Build the GUI test agent for Windows, and keep test builds out of the registry - The test agent didn't compile for Windows: LibreSSL's headers (from TcpConnectionHandler.h) warn about replacing wincrypt.h's names, which wxWidgets has already brought in, and warnings are errors. Nothing there uses wincrypt.h's names, so the warning is turned off for that include. - Even when given a configuration file, FreeDV on Windows reads its default configuration from the registry, and keeps the last one used there. Test builds keep both as files in the folder named by $FREEDV_TEST_CONFIG_DIR instead, if it's set, as they are on Linux and macOS, so that tests never touch the user's settings. - Closing a popup "as clicking elsewhere would" now works on Windows, where wxWidgets dismisses one when it stops being the active window rather than on a click. - get() reports a combo control's popupShown (its list ignores the mouse until it's finished sliding open, on Windows) and every window's clientSize. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit f8af45cdaa4fffb304afe9bbe8b22b1c2f849d31 Author: Mooneer Salem Date: Thu Oct 15 23:14:31 2026 -0700 FreeDV Reporter: remove what's left of a flag from the status message on Windows too A flag that's lost its end (as backspacing over it does), and stray parts of one, are taken out of the status message. On Windows, wxString holds UTF-16, in which each of the characters a flag is made of takes two places, so looking at it a place at a time never found them and nothing was taken out. It's now looked at a whole character at a time. Found by the reporter:status_message_flags GUI test on Windows. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit cc70cac4a81ba8335b3c0657f76d961d04006efc Author: Mooneer Salem Date: Thu Oct 15 23:14:21 2026 -0700 Scale the main window's default size to the display The size used when none has been saved (800x780), and the smallest saved size that's kept, were in pixels, which on Windows are the screen's own: at 200%, 800 of them are only as wide as 400 elsewhere. The controls on either side then took all of the window, leaving the waterfall and spectrum no room at all (0 pixels wide on a laptop at 200%). Both are now scaled to suit the display. Nothing changes on Linux or macOS, where sizes are already scaled. The saved size's default is now -1 (none saved), so that a first start gets the scaled default rather than 800x780. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit a59468c1674ab25e747b06f6baa7bc2ed1f36323 Author: Mooneer Salem Date: Fri Oct 16 04:40:06 2026 +0000 latest user manual PDF commit 7b70d34f3ed346183c77609dcdca1497dd125b3f Author: Mooneer Salem Date: Thu Oct 15 21:39:49 2026 -0700 Run the GUI tests on macOS and fix the bugs they found; run PGO builds, tests and coverage only on merges (#1563) * Run the GUI tests that don't need audio devices on macOS 34 scenarios (Setup Wizard, Settings, main window, Filter and FreeDV Reporter) now run there, in CI as well. What it took: * The tests' stand-in home folder is used on macOS too, by way of CFFIXED_USER_HOME: macOS finds the home folder (and so where FreeDV keeps its settings) without looking at $HOME. Without it, scenarios that load FreeDV's default configuration used the real one. * The test agent brings FreeDV to the front before activating one of its windows, which can't otherwise become active while another application is in front. * FreeDV Reporter's test stations create their client in the thread it runs in: with Python 3.9 (macOS's own), what they sent never went. * Checks allow for numbers grouped by thousands as the user's region has them (LC_ALL doesn't apply on macOS), for macOS's own empty folders in the stand-in home folder, and for "Usage: FreeDV". Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * FreeDV Reporter: don't give a direction of 0 to stations without a usable grid square Redoing the list (e.g. on changing the band filter) recalculated every direction from the stored heading, which is 0 for a station whose grid square (or one's own) can't be used, rather than leaving it blank. The display_options GUI test now changes the band filter and checks again. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Add release notes Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * macOS: send silence when there's nothing to send, rather than the last block again When an output callback left its buffer alone (as FreeDV's does for the radio's sound card with nothing to send), the block before was sent over again, on and on: after a transmission with the VOX tone on, that kept a tone going into the radio. The buffer is now cleared before each callback, as on Windows. Found by the main window's test_tone GUI scenario. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Use the default main window size in place of a saved one too small to be of use Only negative and enormous saved sizes were replaced. On Linux the window manager won't make the window smaller than it can be, but macOS made a 10x10 window when asked to. Sizes below 300x200 now get the default too. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Run all of the GUI tests on macOS, audio included macOS has no devices to create for each test, so freedvtest.audio gives tests existing loopback devices (BlackHole, VB-Cable) in place of the ones they ask for, leaving out a driver's extra devices, which carry what's played into its first. Most need four; CI now builds four BlackHole drivers (with a deployment target this Xcode still accepts). Tests that need what can't be done there (devices that are only inputs, taking one away, the sound server itself) exit 77 and are skipped, which every GUI test now allows for. * sox plays and records, with buffer sizes that BlackHole doesn't lose audio with alongside FreeDV. * Silence is played into VB-Cable devices while in use, as they otherwise repeat what was last played into them. * Where FreeDV is to hear its own transmission, it listens to the device it sends to: feeding one BlackHole device to another loses most of it. * Sample rates are set (and put back) through Core Audio (freedvtest/coreaudio.py), and the Setup Wizard's import finds other programs' settings in the stand-in home folder's Library/Preferences. * The serial port shim is loaded with DYLD_INSERT_LIBRARIES and replaces ioctl() by interposing, for macOS's /dev/ttysN pseudo-terminals. * The test agent can choose items from macOS's application menu (File > Exit is Quit FreeDV there), and finds windows that are being shown before ones that have closed but not yet gone. * Checks allow for "Monitor of" being PulseAudio's, the FreeDV Reporter list not scrolling sideways to a cell (the window is given room for every column), the Spectrum's axis labels taking more room, and the Setup Wizard's import being given device names too short to cut down. * window_geometry checks that a tiny saved size now gives a usable window. macOS CI runs them along with everything else, apart from live_view, which uses the public FreeDV Reporter server, as on Linux. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Release notes for the macOS GUI tests and fixes Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Stop rig control tests failing when rigctld is slow to answer them rigctld's stand-in radio takes its time over each question, and with FreeDV asking it things as well, a test's own question sometimes went unanswered for longer than the second it was given: that failed the test there and then (analog_sideband, about one run in fifteen on macOS). Questions now get 3 s and are asked again if unanswered. 60 runs in a row passed after. Also, wait_for()'s messages were worked out before waiting, so a failure reported the radio as it was beforehand ("mode ... is PKTUSB" just after switching to analog) rather than as it was at the end. They're now worked out when the wait fails. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * macOS CI: make RADE's venv with Python 3.13 The macos-26-arm64 runner image of 2026-10-13 has Python 3.15 as python3, which torch has no packages for yet, so the RADE venv couldn't be set up (test-app-folder failed on macOS 15 and 26). It's now made with Python 3.13, from actions/setup-python (which the images keep a copy of). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Don't use ccache in the Linux coverage job either Like the lint and test-application jobs (#1561), it builds against the system's wxWidgets and OpenSSL, so there's little worth caching, and its cache would crowd those of the builds that compile them out of the repository's 10 GB. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * macOS CI: build four BlackHole drivers only where the GUI tests run Building four rather than two took the Intel PGO job past its 3 minute limit for setting up the virtual audio devices. build_macos_sound_drivers.sh (and the action that runs it) now take how many to build, two by default; the test job builds four and has 6 minutes to do it in. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Don't let a frequency read from the radio before a change undo the change What the radio reports is read on the rig control thread and acted on later on the UI thread. A reading taken just before FreeDV asked for a new frequency could then arrive just after, putting FreeDV back on the old one until the next reading: anything done meanwhile went by the old frequency. Switching to analog just after changing band chose the old band's sideband (USB rather than LSB on 40 m), which the rig_control:analog_sideband GUI test caught on macOS CI every time. The rig controllers now note what frequency was last asked for, and readings that differ from it are ignored until the radio's seen to be there (or for 3 s at most, for radios that can't be tuned exactly where they're asked). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Test playing back through a speaker at 44.1 kHz rather than 22.05 kHz BlackHole and VB-Cable can't run at 22.05 kHz, so on macOS the scenario was skipped at that point. 44.1 kHz is just as different from the 48 kHz recorded at, and every device used runs at it. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Release note for the rig control frequency fix Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * CI: PGO builds, tests and coverage only on merges; packages to try out otherwise With the GUI tests, a push took about an hour on macOS and 45 minutes on Linux. Now only pushes to master and the v*-dev branches (merges, in effect) get PGO builds, the tests and the coverage report. Other branches get lint and plain, unsigned packages to try out: AppImages, a universal macOS disk image and Windows installers, without PGO or LTO. A push whose commit message says "[full ci]", or a run started by hand, gets the full set anywhere. Each workflow decides which with a small "scope" job. On branches, the macOS packaging job builds without a profile rather than waiting for one, the Linux AppImage job builds plainly, and Windows has a new job for the installers (without ccache, as no merge would seed its cache). Signing only happens on full runs. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Reword a comment codespell took for a typo [full ci] Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Don't pass built-up messages to the logger as its format [full ci] The logger formats messages later, on a thread of its own, keeping only a pointer to the format (which it expects to be a string literal) while copying the arguments. 41 places in the macOS and Windows audio code passed a string they'd just built as the format, so it could be gone by the time it was read: AddressSanitizer caught it on macOS (heap-use-after-free in ulog_async's drain(), from MacAudioEngine::start(), when the Settings window's audio device lists were refreshed). Those strings now go in as a "%s" argument, which also stops any '%' in them being taken as a conversion. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Release note for the logging fix [full ci] Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * macOS CI: give the GUI tests a second try under the sanitizers too [full ci] Some of them measure levels and timing, and under TSan and ASan the macOS runners can be slowed down past what they allow: settings:audio_tests (48% of the tone, against 50%), filter:equalizer (-2.0 dB rather than 0) and main_window:voice_keyer_reply each failed once that way, having passed everywhere else. A sanitizer's own report fails every try, so nothing it finds is hidden by this. The other tests keep one try under the sanitizers, as before. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Remove bugfix section from 3.0 since bugfixes in this PR will be applied by master/2.4.1. * Fix five bugs found by running the GUI tests on macOS (from #1563) (#1564) * FreeDV Reporter: don't give a direction of 0 to stations without a usable grid square Redoing the list (e.g. on changing the band filter) recalculated every direction from the stored heading, which is 0 for a station whose grid square (or one's own) can't be used, rather than leaving it blank. The display_options GUI test now changes the band filter and checks again. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * macOS: send silence when there's nothing to send, rather than the last block again When an output callback left its buffer alone (as FreeDV's does for the radio's sound card with nothing to send), the block before was sent over again, on and on: after a transmission with the VOX tone on, that kept a tone going into the radio. The buffer is now cleared before each callback, as on Windows. Found by the main window's test_tone GUI scenario. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Use the default main window size in place of a saved one too small to be of use Only negative and enormous saved sizes were replaced. On Linux the window manager won't make the window smaller than it can be, but macOS made a 10x10 window when asked to. Sizes below 300x200 now get the default too. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Don't let a frequency read from the radio before a change undo the change What the radio reports is read on the rig control thread and acted on later on the UI thread. A reading taken just before FreeDV asked for a new frequency could then arrive just after, putting FreeDV back on the old one until the next reading: anything done meanwhile went by the old frequency. Switching to analog just after changing band chose the old band's sideband (USB rather than LSB on 40 m), which the rig_control:analog_sideband GUI test caught on macOS CI every time. The rig controllers now note what frequency was last asked for, and readings that differ from it are ignored until the radio's seen to be there (or for 3 s at most, for radios that can't be tuned exactly where they're asked). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Don't pass built-up messages to the logger as its format The logger formats messages later, on a thread of its own, keeping only a pointer to the format (which it expects to be a string literal) while copying the arguments. 42 places in the macOS and Windows audio code and ParallelStep passed a string they'd just built as the format, so it could be gone by the time it was read: AddressSanitizer caught it on macOS (heap-use-after-free in ulog_async's drain(), from MacAudioEngine::start(), when the Settings window's audio device lists were refreshed). Those strings now go in as a "%s" argument, which also stops any '%' in them being taken as a conversion. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Release notes for the fixes from the macOS GUI tests Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Reword a comment codespell took for a typo Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * macOS CI: make RADE's venv with Python 3.13 The macos-26-arm64 runner image of 2026-10-13 has Python 3.15 as python3, which torch has no packages for yet, so the RADE venv couldn't be set up (test-app-folder failed on macOS 15 and 26). It's now made with Python 3.13, from actions/setup-python (which the images keep a copy of). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Forgot to include OpenSSL devel packages in README. * Add pulseaudio-utils to allow unit tests to work. --------- Co-authored-by: Claude Opus 5.5 --------- Co-authored-by: Claude Opus 5.5 commit 46c12df93fa31a2c340d6f5f9ea34a7adb135a5e Merge: 6c27816a 7859c2c8 Author: Mooneer Salem Date: Thu Oct 15 21:39:03 2026 -0700 [skip ci] Merge branch 'master' into ms-macos-gui-tests commit 7859c2c84dfadc46f5ace5692acbb6411240d5ca Author: Mooneer Salem Date: Thu Oct 15 21:37:25 2026 -0700 Fix five bugs found by running the GUI tests on macOS (from #1563) (#1564) * FreeDV Reporter: don't give a direction of 0 to stations without a usable grid square Redoing the list (e.g. on changing the band filter) recalculated every direction from the stored heading, which is 0 for a station whose grid square (or one's own) can't be used, rather than leaving it blank. The display_options GUI test now changes the band filter and checks again. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * macOS: send silence when there's nothing to send, rather than the last block again When an output callback left its buffer alone (as FreeDV's does for the radio's sound card with nothing to send), the block before was sent over again, on and on: after a transmission with the VOX tone on, that kept a tone going into the radio. The buffer is now cleared before each callback, as on Windows. Found by the main window's test_tone GUI scenario. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Use the default main window size in place of a saved one too small to be of use Only negative and enormous saved sizes were replaced. On Linux the window manager won't make the window smaller than it can be, but macOS made a 10x10 window when asked to. Sizes below 300x200 now get the default too. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Don't let a frequency read from the radio before a change undo the change What the radio reports is read on the rig control thread and acted on later on the UI thread. A reading taken just before FreeDV asked for a new frequency could then arrive just after, putting FreeDV back on the old one until the next reading: anything done meanwhile went by the old frequency. Switching to analog just after changing band chose the old band's sideband (USB rather than LSB on 40 m), which the rig_control:analog_sideband GUI test caught on macOS CI every time. The rig controllers now note what frequency was last asked for, and readings that differ from it are ignored until the radio's seen to be there (or for 3 s at most, for radios that can't be tuned exactly where they're asked). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Don't pass built-up messages to the logger as its format The logger formats messages later, on a thread of its own, keeping only a pointer to the format (which it expects to be a string literal) while copying the arguments. 42 places in the macOS and Windows audio code and ParallelStep passed a string they'd just built as the format, so it could be gone by the time it was read: AddressSanitizer caught it on macOS (heap-use-after-free in ulog_async's drain(), from MacAudioEngine::start(), when the Settings window's audio device lists were refreshed). Those strings now go in as a "%s" argument, which also stops any '%' in them being taken as a conversion. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Release notes for the fixes from the macOS GUI tests Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Reword a comment codespell took for a typo Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * macOS CI: make RADE's venv with Python 3.13 The macos-26-arm64 runner image of 2026-10-13 has Python 3.15 as python3, which torch has no packages for yet, so the RADE venv couldn't be set up (test-app-folder failed on macOS 15 and 26). It's now made with Python 3.13, from actions/setup-python (which the images keep a copy of). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Forgot to include OpenSSL devel packages in README. * Add pulseaudio-utils to allow unit tests to work. --------- Co-authored-by: Claude Opus 5.5 commit 6c27816a28cef8cecfd9bdf15954f437c8f030c7 Author: Mooneer Salem Date: Thu Oct 15 21:37:04 2026 -0700 Remove bugfix section from 3.0 since bugfixes in this PR will be applied by master/2.4.1. commit 50b7aaeae31d600209fd9c91a5755156130b25ed Author: Mooneer Salem Date: Thu Oct 15 21:34:16 2026 -0700 Add pulseaudio-utils to allow unit tests to work. commit f780e80bfb7fd4c09c7a62769d21a7486856b7e3 Author: Mooneer Salem Date: Thu Oct 15 21:04:10 2026 -0700 Forgot to include OpenSSL devel packages in README. commit 5393f321f362018f97b3a9f66bb9417e305438c3 Author: Mooneer Salem Date: Thu Oct 15 10:44:50 2026 -0700 macOS CI: make RADE's venv with Python 3.13 The macos-26-arm64 runner image of 2026-10-13 has Python 3.15 as python3, which torch has no packages for yet, so the RADE venv couldn't be set up (test-app-folder failed on macOS 15 and 26). It's now made with Python 3.13, from actions/setup-python (which the images keep a copy of). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit c416c355c43a46bf54e10113b0e8bf92f2c59657 Author: Mooneer Salem Date: Thu Oct 15 20:26:54 2026 -0700 Reword a comment codespell took for a typo Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit ffc6fb74a26ee3e0b7f7f3c21ab88523b270ebcf Author: Mooneer Salem Date: Thu Oct 15 19:01:02 2026 -0700 Release notes for the fixes from the macOS GUI tests Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 5dd0359f5d41fb043b8dcd830331eb3fcc8dba65 Author: Mooneer Salem Date: Thu Oct 15 13:39:49 2026 -0700 Don't pass built-up messages to the logger as its format The logger formats messages later, on a thread of its own, keeping only a pointer to the format (which it expects to be a string literal) while copying the arguments. 42 places in the macOS and Windows audio code and ParallelStep passed a string they'd just built as the format, so it could be gone by the time it was read: AddressSanitizer caught it on macOS (heap-use-after-free in ulog_async's drain(), from MacAudioEngine::start(), when the Settings window's audio device lists were refreshed). Those strings now go in as a "%s" argument, which also stops any '%' in them being taken as a conversion. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 0c9f87b95e89f164107115085aec1b96bac02263 Author: Mooneer Salem Date: Thu Oct 15 12:32:27 2026 -0700 Don't let a frequency read from the radio before a change undo the change What the radio reports is read on the rig control thread and acted on later on the UI thread. A reading taken just before FreeDV asked for a new frequency could then arrive just after, putting FreeDV back on the old one until the next reading: anything done meanwhile went by the old frequency. Switching to analog just after changing band chose the old band's sideband (USB rather than LSB on 40 m), which the rig_control:analog_sideband GUI test caught on macOS CI every time. The rig controllers now note what frequency was last asked for, and readings that differ from it are ignored until the radio's seen to be there (or for 3 s at most, for radios that can't be tuned exactly where they're asked). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 204ec8760cab582e12b3f42292d762992db2b64f Author: Mooneer Salem Date: Thu Oct 15 09:01:51 2026 -0700 Use the default main window size in place of a saved one too small to be of use Only negative and enormous saved sizes were replaced. On Linux the window manager won't make the window smaller than it can be, but macOS made a 10x10 window when asked to. Sizes below 300x200 now get the default too. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit cbdfa5bea83eca4d335bbf09c04b10ec425cb5e0 Author: Mooneer Salem Date: Thu Oct 15 09:01:51 2026 -0700 macOS: send silence when there's nothing to send, rather than the last block again When an output callback left its buffer alone (as FreeDV's does for the radio's sound card with nothing to send), the block before was sent over again, on and on: after a transmission with the VOX tone on, that kept a tone going into the radio. The buffer is now cleared before each callback, as on Windows. Found by the main window's test_tone GUI scenario. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit b236eae686e950c032e9aae97e667e463c1a2610 Author: Mooneer Salem Date: Thu Oct 15 02:05:27 2026 -0700 FreeDV Reporter: don't give a direction of 0 to stations without a usable grid square Redoing the list (e.g. on changing the band filter) recalculated every direction from the stored heading, which is 0 for a station whose grid square (or one's own) can't be used, rather than leaving it blank. The display_options GUI test now changes the band filter and checks again. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 43dd04a9bf417630cd340e60aa507ace940cdd73 Author: Mooneer Salem Date: Thu Oct 15 15:22:53 2026 -0700 macOS CI: give the GUI tests a second try under the sanitizers too [full ci] Some of them measure levels and timing, and under TSan and ASan the macOS runners can be slowed down past what they allow: settings:audio_tests (48% of the tone, against 50%), filter:equalizer (-2.0 dB rather than 0) and main_window:voice_keyer_reply each failed once that way, having passed everywhere else. A sanitizer's own report fails every try, so nothing it finds is hidden by this. The other tests keep one try under the sanitizers, as before. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit dc3379ff0338590bc319276154acb1c94eeb1d1f Author: Mooneer Salem Date: Thu Oct 15 13:39:55 2026 -0700 Release note for the logging fix [full ci] Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit f7819f831a92c1840a45bd6e67fec72fea457254 Author: Mooneer Salem Date: Thu Oct 15 13:39:49 2026 -0700 Don't pass built-up messages to the logger as its format [full ci] The logger formats messages later, on a thread of its own, keeping only a pointer to the format (which it expects to be a string literal) while copying the arguments. 41 places in the macOS and Windows audio code passed a string they'd just built as the format, so it could be gone by the time it was read: AddressSanitizer caught it on macOS (heap-use-after-free in ulog_async's drain(), from MacAudioEngine::start(), when the Settings window's audio device lists were refreshed). Those strings now go in as a "%s" argument, which also stops any '%' in them being taken as a conversion. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit bb936774ed15f964e936d085a77aa4cc01ed7f51 Author: Mooneer Salem Date: Thu Oct 15 12:57:41 2026 -0700 Reword a comment codespell took for a typo [full ci] Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 40228e9edfeaa753c7b905d6d4490fd232990a36 Author: Mooneer Salem Date: Thu Oct 15 12:46:55 2026 -0700 CI: PGO builds, tests and coverage only on merges; packages to try out otherwise With the GUI tests, a push took about an hour on macOS and 45 minutes on Linux. Now only pushes to master and the v*-dev branches (merges, in effect) get PGO builds, the tests and the coverage report. Other branches get lint and plain, unsigned packages to try out: AppImages, a universal macOS disk image and Windows installers, without PGO or LTO. A push whose commit message says "[full ci]", or a run started by hand, gets the full set anywhere. Each workflow decides which with a small "scope" job. On branches, the macOS packaging job builds without a profile rather than waiting for one, the Linux AppImage job builds plainly, and Windows has a new job for the installers (without ccache, as no merge would seed its cache). Signing only happens on full runs. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 7fc912be1b4249f5d83022602f19433c0e613413 Author: Mooneer Salem Date: Thu Oct 15 12:33:06 2026 -0700 Release note for the rig control frequency fix Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 8f32f98e35022c50252cf17a28f0eb7a0265cfd1 Author: Mooneer Salem Date: Thu Oct 15 12:32:27 2026 -0700 Test playing back through a speaker at 44.1 kHz rather than 22.05 kHz BlackHole and VB-Cable can't run at 22.05 kHz, so on macOS the scenario was skipped at that point. 44.1 kHz is just as different from the 48 kHz recorded at, and every device used runs at it. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 70b0cdfb6b900ed394bf38562c8478b028e3f9e0 Author: Mooneer Salem Date: Thu Oct 15 12:32:27 2026 -0700 Don't let a frequency read from the radio before a change undo the change What the radio reports is read on the rig control thread and acted on later on the UI thread. A reading taken just before FreeDV asked for a new frequency could then arrive just after, putting FreeDV back on the old one until the next reading: anything done meanwhile went by the old frequency. Switching to analog just after changing band chose the old band's sideband (USB rather than LSB on 40 m), which the rig_control:analog_sideband GUI test caught on macOS CI every time. The rig controllers now note what frequency was last asked for, and readings that differ from it are ignored until the radio's seen to be there (or for 3 s at most, for radios that can't be tuned exactly where they're asked). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 70a25251f88806317597f270f2a4af29126268a7 Author: Mooneer Salem Date: Thu Oct 15 11:16:31 2026 -0700 macOS CI: build four BlackHole drivers only where the GUI tests run Building four rather than two took the Intel PGO job past its 3 minute limit for setting up the virtual audio devices. build_macos_sound_drivers.sh (and the action that runs it) now take how many to build, two by default; the test job builds four and has 6 minutes to do it in. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit f283ca8e3870092010c66304088524954450c4c0 Author: Mooneer Salem Date: Thu Oct 15 11:05:32 2026 -0700 Don't use ccache in the Linux coverage job either Like the lint and test-application jobs (#1561), it builds against the system's wxWidgets and OpenSSL, so there's little worth caching, and its cache would crowd those of the builds that compile them out of the repository's 10 GB. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 860fde0127d609e381441a0b1e759e27002faea0 Author: Mooneer Salem Date: Thu Oct 15 10:44:50 2026 -0700 macOS CI: make RADE's venv with Python 3.13 The macos-26-arm64 runner image of 2026-10-13 has Python 3.15 as python3, which torch has no packages for yet, so the RADE venv couldn't be set up (test-app-folder failed on macOS 15 and 26). It's now made with Python 3.13, from actions/setup-python (which the images keep a copy of). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 940fadf740572ae845093ce03f09f76159b5592c Author: Mooneer Salem Date: Thu Oct 15 10:37:31 2026 -0700 Stop rig control tests failing when rigctld is slow to answer them rigctld's stand-in radio takes its time over each question, and with FreeDV asking it things as well, a test's own question sometimes went unanswered for longer than the second it was given: that failed the test there and then (analog_sideband, about one run in fifteen on macOS). Questions now get 3 s and are asked again if unanswered. 60 runs in a row passed after. Also, wait_for()'s messages were worked out before waiting, so a failure reported the radio as it was beforehand ("mode ... is PKTUSB" just after switching to analog) rather than as it was at the end. They're now worked out when the wait fails. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 7f25e8e1bff70958ee09b27626375f87d8fc3fac Author: Mooneer Salem Date: Thu Oct 15 09:02:05 2026 -0700 Release notes for the macOS GUI tests and fixes Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 459e167abf8dfc640b413cfb80602db9baff3855 Author: Mooneer Salem Date: Thu Oct 15 09:01:51 2026 -0700 Run all of the GUI tests on macOS, audio included macOS has no devices to create for each test, so freedvtest.audio gives tests existing loopback devices (BlackHole, VB-Cable) in place of the ones they ask for, leaving out a driver's extra devices, which carry what's played into its first. Most need four; CI now builds four BlackHole drivers (with a deployment target this Xcode still accepts). Tests that need what can't be done there (devices that are only inputs, taking one away, the sound server itself) exit 77 and are skipped, which every GUI test now allows for. * sox plays and records, with buffer sizes that BlackHole doesn't lose audio with alongside FreeDV. * Silence is played into VB-Cable devices while in use, as they otherwise repeat what was last played into them. * Where FreeDV is to hear its own transmission, it listens to the device it sends to: feeding one BlackHole device to another loses most of it. * Sample rates are set (and put back) through Core Audio (freedvtest/coreaudio.py), and the Setup Wizard's import finds other programs' settings in the stand-in home folder's Library/Preferences. * The serial port shim is loaded with DYLD_INSERT_LIBRARIES and replaces ioctl() by interposing, for macOS's /dev/ttysN pseudo-terminals. * The test agent can choose items from macOS's application menu (File > Exit is Quit FreeDV there), and finds windows that are being shown before ones that have closed but not yet gone. * Checks allow for "Monitor of" being PulseAudio's, the FreeDV Reporter list not scrolling sideways to a cell (the window is given room for every column), the Spectrum's axis labels taking more room, and the Setup Wizard's import being given device names too short to cut down. * window_geometry checks that a tiny saved size now gives a usable window. macOS CI runs them along with everything else, apart from live_view, which uses the public FreeDV Reporter server, as on Linux. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit d4878ec0d0bb51df422f2d998d558d09f7209743 Author: Mooneer Salem Date: Thu Oct 15 09:01:51 2026 -0700 Use the default main window size in place of a saved one too small to be of use Only negative and enormous saved sizes were replaced. On Linux the window manager won't make the window smaller than it can be, but macOS made a 10x10 window when asked to. Sizes below 300x200 now get the default too. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 5b3f304946d4370c15646f0beeae50250ad97c60 Author: Mooneer Salem Date: Thu Oct 15 09:01:51 2026 -0700 macOS: send silence when there's nothing to send, rather than the last block again When an output callback left its buffer alone (as FreeDV's does for the radio's sound card with nothing to send), the block before was sent over again, on and on: after a transmission with the VOX tone on, that kept a tone going into the radio. The buffer is now cleared before each callback, as on Windows. Found by the main window's test_tone GUI scenario. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 4830dbeea228043bd8c3d4ca0241289dd4bbdef6 Author: Mooneer Salem Date: Thu Oct 15 02:06:02 2026 -0700 Add release notes Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 03f0d388c963a3af228585308f8b7af66c9e8ee9 Author: Mooneer Salem Date: Thu Oct 15 02:05:27 2026 -0700 FreeDV Reporter: don't give a direction of 0 to stations without a usable grid square Redoing the list (e.g. on changing the band filter) recalculated every direction from the stored heading, which is 0 for a station whose grid square (or one's own) can't be used, rather than leaving it blank. The display_options GUI test now changes the band filter and checks again. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 9fa33e3ba59b9003bec343d3d6af1fb8192d09e5 Author: Mooneer Salem Date: Thu Oct 15 02:05:27 2026 -0700 Run the GUI tests that don't need audio devices on macOS 34 scenarios (Setup Wizard, Settings, main window, Filter and FreeDV Reporter) now run there, in CI as well. What it took: * The tests' stand-in home folder is used on macOS too, by way of CFFIXED_USER_HOME: macOS finds the home folder (and so where FreeDV keeps its settings) without looking at $HOME. Without it, scenarios that load FreeDV's default configuration used the real one. * The test agent brings FreeDV to the front before activating one of its windows, which can't otherwise become active while another application is in front. * FreeDV Reporter's test stations create their client in the thread it runs in: with Python 3.9 (macOS's own), what they sent never went. * Checks allow for numbers grouped by thousands as the user's region has them (LC_ALL doesn't apply on macOS), for macOS's own empty folders in the stand-in home folder, and for "Usage: FreeDV". Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit afd36a969ce001f8d110194cd4e34e180721a321 Author: Mooneer Salem Date: Thu Oct 15 00:48:30 2026 -0700 Fix compiler error. commit 3a90df3a2e8f17bfd3ceede9aa18a81440214188 Merge: 89323ad0 fc9be200 Author: Mooneer Salem Date: Thu Oct 15 00:44:56 2026 -0700 Merge branch 'v3.0-dev' into ms-rade-v2 commit fc9be2008699fc4a2e04b73df83f928e6dd3ac7c Merge: aef873ef c2ca064c Author: Mooneer Salem Date: Thu Oct 15 00:44:44 2026 -0700 Merge branch 'master' into v3.0-dev commit 4f391e280261e68aaf309e0d28088ec151883b5a Author: Mooneer Salem Date: Thu Oct 15 00:13:18 2026 -0700 Run the regular CI for this branch again, and add release notes Removes the temporary macOS-only workflow and the regular workflows' exclusion of this branch. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 9a754cd2b86ba19dec2c058215327b4938c718d1 Author: Mooneer Salem Date: Thu Oct 15 00:13:18 2026 -0700 Say the machine stalled in rade_loss's check even when audio went missing afterwards After the whole machine stalls, the sound system may make up for lost time by skipping, so audio can be missing from the recording as well as added. The check now gives the machine's verdict whenever the watchdog was held up at the same moment, and only says "real-time" of the watchdog when it ran at real-time priority. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 8565a5145a53ce9e0530f7602b4fb4b84669a5ef Author: Mooneer Salem Date: Wed Oct 14 00:36:50 2026 -0700 Tell a stalled machine from a stalled sound system in rade_loss's failures When the radio's sound card goes a while without asking FreeDV for audio, this records what's needed to say why: * Core Audio's own timestamps for each device (how far its sample position moved on across the longest gap between calls, how often it jumped ahead) and the overloads it reported, logged when the device stops. * A watchdog thread in test mode that does nothing but wake every 5 ms, at real-time priority on macOS, reporting how late it woke while transmitting and when, next to when the radio's longest gap was. * Core Audio's log (overloads, skipped cycles) over the transmission, saved by test_rade_loss.sh on macOS. check_rade_recording.py reports all of these and says which it was. It also now recognises a gap in FreeDV's own copy that only partly matches the recording (reporting the rest as lost from the recording), and says when the sound card was slow to ask while FreeDV had signal ready. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 182174a2779788ad8d21a9d05104f47b35e979b8 Author: Mooneer Salem Date: Tue Oct 13 23:21:04 2026 -0700 Report how promptly the radio's sound card was served while transmitting, and tell gaps in FreeDV's own copy from ones in the recording At the end of a test run FreeDV now reports, for the radio's output while transmitting: the length of its blocks, the longest time between its asking for more and how often that was more than twice a block, and how many blocks of silence it was given while a transmission was starting or while there was signal to send that couldn't be read. That tells FreeDV not keeping up from the sound card not asking. check_rade_recording.py no longer takes a stretch missing from FreeDV's own copy of what it sent, where its tap dropped samples, for the recording having something extra. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 1720ac6dfd2606b2b8e30cba997a3ade44ebb83f Author: Mooneer Salem Date: Tue Oct 13 22:47:12 2026 -0700 Count the radio going short only while transmitting, and check rade_loss's recording against what FreeDV sent The radio's output sound card counted as short whenever FreeDV wasn't transmitting, so the count was much the same for every run. Now only the times while transmitting are reported. rade_loss now has FreeDV keep a copy of what it sent (-txoutfile) and compares the recording with it: stretches missing or extra, and holes of silence, are reported with where they are. FreeDV's plot taps dropping samples while sending or receiving are reported alongside: the TX/RX thread carrying on while ordinary threads (the recorder's among them) don't get to run, as when the machine stalls. The temporary validation workflow keeps the copy and the check's report, and flags runs where the machine looks to have stalled. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit f5a5a8a31081b4285591c448a73a5f0ee679949a Author: Mooneer Salem Date: Tue Oct 13 21:17:17 2026 -0700 TEMPORARY: run only a macOS build and repeated rade_loss tests for this branch To be removed before merging. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 6a17ccc5369f5e984cd23e5c30cb33476b54cf6c Author: Mooneer Salem Date: Tue Oct 13 21:17:17 2026 -0700 Report the sound card FIFO counters at the end of a test run, and decode rade_loss's recording again when it fails FreeDV's command-line test mode now logs how often each sound card's FIFO overflowed or ran short, so that a failing run says whether the sound cards were involved. When rade_loss fails, it decodes the same recording a second time (telling damage in the recording from something lost while receiving it) and keeps the recording, the feature files and both halves' logs. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 13d4eeff6ce7ef73027f264634e4f28bb3884fd8 Merge: fcb94c16 c2ca064c Author: Mooneer Salem Date: Tue Oct 13 21:15:00 2026 -0700 Merge remote-tracking branch 'origin/master' into ms-pipeline-offline-tests # Conflicts: # CMakeLists.txt # USER_MANUAL.pdf commit c2ca064c7e8110e66b332646f380e0cec35b57c4 Author: Mooneer Salem Date: Wed Oct 14 04:11:27 2026 +0000 latest user manual PDF commit 7596752e7315adf855e61ff8329d2288f457eb2a Author: Mooneer Salem Date: Tue Oct 13 21:08:57 2026 -0700 Cache only the CI builds that compile wxWidgets and LibreSSL, and build those two without debugging information (#1561) * Cache only the CI builds that compile wxWidgets and LibreSSL, and build those two without debugging information One branch's compiler caches didn't fit in the repository's 10 GB of Actions caches, so they evicted each other and builds started cold. * ccache is now only used by the builds that compile wxWidgets and LibreSSL (the AppImage, Windows and macOS PGO-instrumented builds). The Linux and macOS lint and test jobs use the system's or Homebrew's, compile about 250 files and gain a minute or two from a cache. * Each cache is limited to 1 GB rather than 5. * wxWidgets and LibreSSL are built optimised and without debugging information whatever FreeDV itself is built as. Their debugging information was most of what was cached. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Add PR #1561 to the release notes Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Save each shared Windows compiler cache from one job only The sanitizer and unit test builds of the same sanitizer and architecture share a cache key, and both saved it: two copies of nearly the same thing from every run, only the newer of which is ever restored. The sanitizer build (which compiles the whole application) now saves it alone. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp --------- Co-authored-by: Claude Opus 5.5 commit fcb94c16619801a186db9c19a8c7c02236f18c1d Author: Mooneer Salem Date: Tue Oct 13 20:41:34 2026 -0700 Start transmissions without silence mixed into the first words or holes in the signal When the microphone hadn't yet delivered a full frame, the TX thread sent a frame of silence in its place, which is how it built up the 640 ms it kept ready for the radio's sound card: the first half second or so of each transmission went out a frame at a time with silence in between. And the radio's sound card was given the transmission from the start, with nothing in hand, so it could ask for more than had been made and get silence (a hole in the signal). The TX thread now waits for the rest of a frame instead, and the radio's sound card isn't given a transmission until 240 ms of it (two RADE modem frames) is ready. Until then it gets silence, with the VOX tone if that's on. What's said now goes out whole, about 240 ms after it's said rather than 640 ms. The round trip and tone tests now require that: nothing but the speech sent, no holes in the signal and the radio's sound card never going short. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 1f790148d1eaf415875428dd9ce8ec6403387895 Author: Mooneer Salem Date: Tue Oct 13 19:02:07 2026 -0700 Save each shared Windows compiler cache from one job only The sanitizer and unit test builds of the same sanitizer and architecture share a cache key, and both saved it: two copies of nearly the same thing from every run, only the newer of which is ever restored. The sanitizer build (which compiles the whole application) now saves it alone. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 00de825f1db3cb50fec358e024241dc3aca766d7 Author: Mooneer Salem Date: Tue Oct 13 18:58:05 2026 -0700 Test a whole transmission and a steady tone through the TX/RX threads without sound cards or a clock Speech is sent and received again through the real threads, with a model of two sound cards on one clock in place of the real ones: in even and uneven blocks, at 48, 44.1 and 8 kHz, and from and to files as the rade_loss test does it. Sync has to be found promptly and kept, nothing lost on the way in, no holes in the signal once it's going, and what comes back has to be what was sent (by RADE's own measure of loss). A tone through each thread in analog mode, with the sound card taking its output stalling at random, may only be broken where there was nowhere to put it or nothing to give. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit aef873ef2f99c9e5ea8e69f53fbc74539d26ac98 Author: Mooneer Salem Date: Wed Oct 14 01:25:45 2026 +0000 latest user manual PDF commit c66a38e11fd7dc5e7eb8c7adff857d070374695d Author: Mooneer Salem Date: Tue Oct 13 18:23:04 2026 -0700 Run the GUI tests in CI, report coverage, add more GUI tests and let Settings be reopened with wxWidgets 3.3.4 (#1556) * Run the GUI tests in CI, report coverage, and add more GUI tests CI (Linux): * The ubuntu-26.04 test-application jobs (plain and each sanitizer) run the GUI tests after the unit tests. * A new coverage job builds with --coverage (WITH_COVERAGE=1 in build_linux.sh), runs every test and reports what was executed in the job summary, with the HTML report as an artifact. Tests: * gui_main_window_default_config, exit_menu, tune_level_per_band, frequency_units and appearance_change. * gui_reporter_sort_every_column, numeric_filters and status_message_flags. Fixes: * Load Default Configuration loaded an empty configuration on Linux builds with wxWidgets 3.3: the default configuration's location was worked out again, and wxWidgets gives a different answer by then. The location used at startup is now remembered. * TxRxThreadTest aborted under RealtimeSanitizer: a test that turns on FIFO state logging made the logger's first use on its thread happen on a real-time path. The test now logs once before it starts. Also removes plot_scatter.cpp/.h, which nothing used. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Add PR #1556 to release notes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Fix build with wxWidgets older than 3.3 wxCONFIG_USE_XDG doesn't exist before wxWidgets 3.3, and the code that uses it in Load Default Configuration had lost the check for that. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Fix what the first CI run of the GUI tests found FreeDV: * Load Default Configuration still loaded an empty configuration with wxWidgets older than 3.3 on Linux: wxWidgets looks for the default configuration under the application's name, which is "freedv" when it's opened at startup and "FreeDV" afterwards. The configuration opened at startup is now kept and reused. * FreeDV Reporter sorted stations wrongly when their frequencies were more than 2^31 Hz apart (e.g. someone on a 2.4 GHz satellite uplink and someone on HF), as it subtracted them into an int. Distance, heading and SNR are compared rather than subtracted now too. Test agent and tests: * Build with wxWidgets 3.0 (Ubuntu 22.04). * Sorting by another column left the previous one marked as sorted by with wxWidgets 3.2. * Start rigctld with -P RIG, without which Hamlib 4.6 doesn't answer whether the radio is transmitting. * Don't assume what's selected once the selected station is filtered out, or that everyone on the public server has a frequency. * Give connection_burst longer for slow (sanitizer) builds. CI: * Leave the test that depends on who's on the public FreeDV Reporter server out of CI. * Print the coverage summary in the same layout as freedv-backend's, and don't count the branches that compilers add for exceptions. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Leave more compiler-generated branches out of coverage, and add tests for real ones Coverage report: branches for cleaning up after a constructor that throws ("new Something(...)") and for asserts failing aren't counted any more, alongside those for exceptions. None of them are decisions in the code, and together they were most of what was reported as untested. Tests: * TxRxThreadTest: what's transmitted is dropped, not waited for or written over what's queued, when the radio's FIFO is full. * gui_audio_errors_sound_server_restarted: the sound server being restarted under FreeDV, and FreeDV carrying on after Stop and Start. * gui_rig_control_mode_display and editing_frequency: the radio's mode being displayed, and its dial being ignored while the frequency is being typed in. * gui_main_window_reporting_without_callsign and band_names. * gui_settings_ptt_key_names. Test agent: a "focus" command. Test library: the private sound server can be stopped and started again. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Test what FreeDV does when the sound server isn't there Tests: * audio_PulseAudioDeviceTest: an engine or a device whose start() fails because there is no sound server says so, and names the device it was asked for; a device whose start() failed has nothing left to stop; an underflow or an overflow is reported to the device it belongs to, with that device's state. * gui_audio_errors_sound_server_gone_while_starting: the sound server going away between FreeDV checking its sound cards and opening its streams, and the two errors that come out of that. Test library: AudioServerGateway, which stands between FreeDV and the sound server and can be made to stop letting connections through. Fix: IAudioDevice's audio callbacks had no initialiser, and IAudioDevice has no constructor, so PulseAudioDevice::start()'s "if (onAudioErrorFunction)" guard read garbage and called through a stale function pointer (seen as std::bad_alloc). All five function/state pairs are nullptr by default now; this affected the PulseAudio, PortAudio, WASAPI and Mac devices alike. * Test what FreeDV does when a sound card goes away while it's starting FreeDV checks its sound cards over one connection and closes it before opening any streams, so AudioServerGateway can now take a device away in between: after_next_connection_close() runs a test's own action when the next connection closes, and holds the connections after it back until that's done. VirtualAudioDevices.remove() unloads one device and waits for the sound server to stop listing it. test_sound_card_gone_while_starting covers the resulting paths in MainFrame::startRxStream(): the "Could not find ..." boxes for the TX output, RX input and RX output of a two sound card setup, for the RX input and RX output of a receive-only one, and the cleanup that follows them in both cases. * Test what FreeDV does when the sound server goes away as it opens its streams Pressing Start with a sound card that couldn't be reached left FreeDV in a state it couldn't leave. startRxStream() started the TX processing thread before it opened the RX streams; when opening those failed, the cleanup reset the sound devices and returned without ever signalling that thread. Nothing else stopped it either: the cleanup sets m_RxRunning to false, and stopRxStream() only does anything when it's true. On exit, ~TxRxThread then joined a thread still waiting at startSem_.wait() for a start signal that never comes, so FreeDV hung. Where it got that far it died instead in wxLog recursion, a RecordStep worker logging while the pipeline was being destroyed during exit. startRxStream() already waited for the RX devices before starting the RX thread; it now does the same before starting the TX one. Two scenarios in test/gui/test_audio_errors.py: - sound_server_gone_while_opening_streams refuses FreeDV's connections to the sound server from its (2 + n)th onwards, for n = 0..3, which is what the server going away looks like to each of the four devices in turn. FreeDV names the ones it couldn't reach, closes the ones it had opened, and starts none of them. AudioServerGateway gains refuse_connections_after() for this; the 1 + 1 + 4 connections FreeDV makes on Start were measured first. - receive_only_device_moved moves the radio's input and the speaker's output of a one-pair-of-devices setup to other devices, and checks that FreeDV carries on using and remembering the new ones. Measured against the fixed code, the two scenarios cover 36 lines of src/main.cpp (1751/1941, 90.2% -> 1787/1941, 92.1%): the four device-start failures and the two one-sound-card device-changed handlers, 3193-3195, 3209-3211, 3536-3541, 3546-3553, 3562-3570 and 3576-3585, with nothing newly uncovered. What's left inside startRxStream() is the zero-sound-card branch (3163-3172, unreachable), the xrun counters (3459-3482, 3517-3523) and the FIFO race guard at 3503. * Test what the PTT key does when the time-out timer cuts TX short Nothing tested the momentary PTT key and the time-out timer together: which key presses the time-out is allowed to leave in force, and what happens to a press that is let go while TX is still starting. Both are now covered by a new main window scenario, ptt_key_time_out: - With the key still held when the time-out stops TX, pressing it again must not restart TX, and letting go must bring it back. FreeDV blocks the key until it is genuinely released (ongui.cpp:1225-1227) and a poll timer watches for the release (1232-1240); neither had ever run. - With a delay configured before TX starts, the button goes down as soon as the key is pressed while TX is still waiting. Letting go in that window is deferred (ongui.cpp:1071, 1078-1079), TX starts anyway, and is then stopped by itself (1684-1690) rather than being left keyed. The delay also exercises the waits before TX starts and after it stops (1618-1629, 1479-1490), which no test had reached. Running the test suite now covers 33 more lines of ongui.cpp (90 lines uncovered down to 57, 82.7% of the project covered up to 83.1%), with nothing that was covered becoming uncovered. The existing ptt_key and time_out_timer scenarios still pass. ongui.cpp:1287-1292, which would clear a time-out warning that is no longer wanted, is left untested: m_totWarningDialog_ is only ever set while the time-out has 15 seconds or less to go, and the Extend callback clears it (1265-1266) before raising the remaining time (1271), so the branch cannot be reached from the interface. * Add the remaining FreeDV Reporter tests, and don't reuse threads from a failed start Tests: * gui_reporter_text_filters: filtering by each column that holds text. * gui_reporter_saved_filters: the window opening with band tracking, a custom idle time and the saved width of the Add Message window. * gui_reporter_selection: what unselects a station (clicking elsewhere, sorting, the station leaving). startRxStream() forgets any processing threads left by an earlier start that failed before starting them. Starting again as receive-only would otherwise have found the old transmit thread and started it, with the devices it was made for long gone. gui_reporter_row_actions waits longer for FreeDV's own frequency to reach the server, which took more than 15 seconds once under UBSan. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Keep GUI tests off the public FreeDV Reporter server, and wait for FreeDV to carry on after a dialog closes FreeDV connects to FreeDV Reporter as soon as it's launched, so every test that didn't say where to find a server (the Setup Wizard ones and several Settings ones) connected to the public one. Configurations written by the test library now point at somewhere nothing is listening unless told otherwise, including for first-run launches; only the live view test names the public server. wait_for_window_closed() also now waits for FreeDV to leave the dialog's event loop. gui_settings_frequency_list failed once locally because its next command was answered before FreeDV had acted on Settings being closed. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Wait for the radio before double-clicking a station in the FreeDV Reporter test gui_reporter_row_actions double-clicked a station as soon as FreeDV had started. If FreeDV hadn't connected to the radio by then, the radio went to that station's frequency and then straight back: on connecting, FreeDV tunes the radio to the frequency chosen in the main window. The test now waits for the connection first. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Remove unused plot, main window and FreeDV Reporter code, and test more of the main window Removed, as nothing used them: PlotScalar's bar graph and log scale modes and two of its ways of adding samples; MainFrame's OnExitClick, OnCloseFrame and OnPaint, m_zoom and m_schedule_restore; and FreeDV Reporter's filtersEnabled() and hasAnyColumnFilter(). New scenarios: gui_main_window_tab_layout, heard_list, receive_reports (what's told to other programs over UDP and logged when a callsign is received), test_tone (the interference tone and the left channel VOX tone), exit_while_tuning, old_settings and bad_command_line. The test library can now return what it records, per channel, and say how much of it is a tone. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Keep four decimal places when Settings converts a typed-in frequency from kHz to MHz, and add more FreeDV Reporter, Filter and Settings tests Fix: with a frequency typed into the Rig Control tab's entry box, turning off "frequencies in kHz" and pressing Apply left the box with one decimal place of MHz (21313.5 kHz became 21.3). New scenarios: gui_reporter_off_at_start, qsy_with_radio and new_callsign, and gui_filter_equalizer_receive_only. gui_settings_ptt_key_names covers the rest of the keys FreeDV has names for, and gui_settings_ptt_test a radio that Hamlib can't reach and a port that can't be opened. The test agent takes Insert, Home, End, PageUp, PageDown and numeric key codes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Don't make the audio thread wait for seconds after it's been held up for seconds, and test more Setup Wizard imports Fix (Linux): PulseAudioDevice kept how far a round of work overran, in nanoseconds, in an int. Work held up for more than about 2.1 seconds (a machine too busy to run it) overflowed that, and the next round then waited up to 2 seconds more instead of not waiting at all. It's kept in 64 bits now. audio_PulseAudioDeviceTest covers how rounds of work are paced: to the stream's latency, with overruns coming out of the waits that follow, and with nothing registered for underflows and overflows. gui_setup_wizard_import_more covers device names that were cut short, radios keyed by DTR or their own VOX, USB radios, and settings with nothing usable. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Test when the speaker is muted around transmissions pipeline_TxRxThreadTest: what's received is played; not while transmitting in half duplex, nor until the radio's had time to go back to receiving; in full duplex it carries on, with no such wait, except while the voice keyer is sending. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Allow for a late wait being made up for in the PulseAudio pacing test A wait that ends late is deliberately taken off the next one, so a round after a late one is shorter than the stream's latency. Check the rounds so far rather than each by itself. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Test files that can't be written, ports that aren't serial ports and more of recording, playback and the audio tests Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Don't count more of the branches that compilers add, and remove unused plot, notebook and dialog code Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Test the other side of more conditions in the main window, FreeDV Reporter, rig control and audio set-up Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Choose the station to log by its callsign in the Log QSO test By its position, the list could re-sort itself (FreeDV's own entry arriving) between working the position out and selecting it, and someone else was logged. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Let the Settings window be opened more than once with wxWidgets 3.3.4 on Linux The same dialog is shown each time. wxWidgets 3.3.4's GTK wxDialog::Show() takes a return code that's already set to mean the dialog ended itself while being set up and doesn't show it, so after the first time Settings was closed, ShowModal() returned straight away. Clear the return code before showing it. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Make two GUI tests independent of key handling and of timing * Typed into the recording suffix, a key doesn't get as far as being typed with some versions of wxWidgets (CI's 3.2), so check each key for either being replaced or having done nothing. * Cut the sound server off after a count of connections rather than on seeing one close, which the gateway could find out about after FreeDV had made the next. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Add PR #1556's wxWidgets 3.3.4 fix and code removal to the release notes Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Don't type the PTT key into the recording suffix in the GUI test In CI, where the main window still counts as the active one while the Start Recording window is up, a space keyed the transmitter rather than being typed, and the test then couldn't stop FreeDV. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp --------- Co-authored-by: Claude Opus 5.5 commit a622c647987fd36f7ff454dd8776f2e0e487f50e Author: Mooneer Salem Date: Tue Oct 13 17:39:27 2026 -0700 Add PR #1561 to the release notes Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit b301b87761a8c432b7de1cbd63548fd66a05114f Author: Mooneer Salem Date: Tue Oct 13 17:38:57 2026 -0700 Cache only the CI builds that compile wxWidgets and LibreSSL, and build those two without debugging information One branch's compiler caches didn't fit in the repository's 10 GB of Actions caches, so they evicted each other and builds started cold. * ccache is now only used by the builds that compile wxWidgets and LibreSSL (the AppImage, Windows and macOS PGO-instrumented builds). The Linux and macOS lint and test jobs use the system's or Homebrew's, compile about 250 files and gain a minute or two from a cache. * Each cache is limited to 1 GB rather than 5. * wxWidgets and LibreSSL are built optimised and without debugging information whatever FreeDV itself is built as. Their debugging information was most of what was cached. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit ae053a7db8a917d9c7edb20f9deb005977e4a25f Author: Mooneer Salem Date: Tue Oct 13 16:44:57 2026 -0700 Don't type the PTT key into the recording suffix in the GUI test In CI, where the main window still counts as the active one while the Start Recording window is up, a space keyed the transmitter rather than being typed, and the test then couldn't stop FreeDV. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 64eb804ece77de89a5cd5a7fe694995036141dc2 Author: Mooneer Salem Date: Tue Oct 13 15:57:29 2026 -0700 Add PR #1556's wxWidgets 3.3.4 fix and code removal to the release notes Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 0270b1668110591a27cdf0437bdd45772d97403c Author: Mooneer Salem Date: Tue Oct 13 15:46:00 2026 -0700 Make two GUI tests independent of key handling and of timing * Typed into the recording suffix, a key doesn't get as far as being typed with some versions of wxWidgets (CI's 3.2), so check each key for either being replaced or having done nothing. * Cut the sound server off after a count of connections rather than on seeing one close, which the gateway could find out about after FreeDV had made the next. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit ba7ebabf13cbd4d1c45d60d916b7e4cb2f9caf74 Author: Mooneer Salem Date: Tue Oct 13 15:06:48 2026 -0700 Let the Settings window be opened more than once with wxWidgets 3.3.4 on Linux The same dialog is shown each time. wxWidgets 3.3.4's GTK wxDialog::Show() takes a return code that's already set to mean the dialog ended itself while being set up and doesn't show it, so after the first time Settings was closed, ShowModal() returned straight away. Clear the return code before showing it. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 2380b232d5c073350090a26e06fc50197cadf9bd Author: Mooneer Salem Date: Tue Oct 13 15:06:48 2026 -0700 Choose the station to log by its callsign in the Log QSO test By its position, the list could re-sort itself (FreeDV's own entry arriving) between working the position out and selecting it, and someone else was logged. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit fce0bdc26df4e51ee6342fc038a640454cbe3e55 Merge: 5740940e d1117c9d Author: Mooneer Salem Date: Tue Oct 13 14:17:41 2026 -0700 Merge remote-tracking branch 'origin/v3.0-dev' into ms-ci-gui-tests commit 5740940ef90b4ac1e73b50ea6defe9a54775919a Author: Mooneer Salem Date: Tue Oct 13 14:17:40 2026 -0700 Test the other side of more conditions in the main window, FreeDV Reporter, rig control and audio set-up Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit d1117c9d5207792618b1456dfbeb505d2d961c2a Author: Mooneer Salem Date: Tue Oct 13 14:17:14 2026 -0700 [skip ci] Switch to official wxWidgets 3.3.4. (#1560) * Switch to official wxWidgets 3.3.4. * [skip ci] Add PR #1560 to changelog. commit 9e4e2b97178e141144653a3aea8672358fab51bc Author: Mooneer Salem Date: Tue Oct 13 14:16:12 2026 -0700 [skip ci] Add PR #1560 to changelog. commit 1dd7571051d16dbdc97f3abf769919739d3165e6 Author: Mooneer Salem Date: Tue Oct 13 19:38:26 2026 +0000 latest user manual PDF commit 679913bf7d8a471e4b29d13580aaf17125dccd8f Merge: 651f1282 fd75a158 Author: Mooneer Salem Date: Tue Oct 13 12:36:35 2026 -0700 Merge branch 'v3.0-dev' into ms-wxwidgets-334 commit fd75a158181e2b0b212a49d778f7bbf32a1cba37 Merge: 54bd3619 e90765d1 Author: Mooneer Salem Date: Tue Oct 13 12:36:16 2026 -0700 Merge branch 'master' into v3.0-dev commit 651f12823bd3a868bfbf883fa78c848bfd340917 Author: Mooneer Salem Date: Tue Oct 13 12:35:44 2026 -0700 Switch to official wxWidgets 3.3.4. commit e90765d15ffbe4840e81ade3e7477a15a5db8558 Author: Mooneer Salem Date: Tue Oct 13 12:35:30 2026 -0700 Fix hang on exit after a failed start, uninitialised audio callbacks, an audio stall on Linux and a Settings frequency conversion; make CI caches smaller (#1559) * Fix FreeDV hanging on exit after a failed start, and uninitialised audio callbacks * startRxStream() started the transmit processing thread before it opened the receive streams. When opening those failed, nothing ever signalled or stopped that thread, and FreeDV hung on exit waiting for it. Both processing threads are now only started once every device is open, and any left by a start that failed aren't reused by the next. * IAudioDevice's callbacks had no initialisers (and it has no constructor), so a device that failed to start before the application had registered an error handler called through a garbage pointer. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Add PR #1559 to release notes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Fix a two second audio stall after processing is held up, and a typed-in frequency losing digits in Settings Linux: PulseAudioDevice kept how far a round of work overran, in nanoseconds, in an int. Work held up for more than about 2.1 seconds overflowed that, and the next round then waited up to 2 seconds more instead of not waiting at all. It's kept in 64 bits now. Settings > Rig Control: with a frequency typed into the entry box, turning off "frequencies in kHz" and pressing Apply left it with one decimal place of MHz (21313.5 kHz became 21.3) rather than four. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Make the CI's compiler caches smaller The repository's 10 GB of Actions caches doesn't hold one branch's worth of them, so branches evict each other's and builds start cold. * The Windows sanitizer and unit test builds of the same sanitizer and architecture compile the same things with the same flags, so they now share a cache rather than each keeping its own copy. * ccache compresses what it stores harder (zstd level 10 rather than 1). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Commits containing only docs shouldn't cause a CI build. * Test: make sure ignore paths actually work. * Revert "Test: make sure ignore paths actually work." This reverts commit 088688e9d62af41e4d4e1aa50b5e6fdbebaa53fd. --------- Co-authored-by: Claude Opus 5.5 commit 5857f6a83ea4edb77cf0baf2c85f46691f5ef3bf Author: Mooneer Salem Date: Tue Oct 13 11:33:55 2026 -0700 Revert "Test: make sure ignore paths actually work." This reverts commit 088688e9d62af41e4d4e1aa50b5e6fdbebaa53fd. commit 088688e9d62af41e4d4e1aa50b5e6fdbebaa53fd Author: Mooneer Salem Date: Tue Oct 13 11:32:57 2026 -0700 Test: make sure ignore paths actually work. commit 4036650b59268c724c6587eb39e2a389fe73bf8c Author: Mooneer Salem Date: Tue Oct 13 11:32:07 2026 -0700 Commits containing only docs shouldn't cause a CI build. commit 2e5e3262d8a79318c936696d70b7fcf07cba1205 Author: Mooneer Salem Date: Tue Oct 13 10:44:16 2026 -0700 Make the CI's compiler caches smaller The repository's 10 GB of Actions caches doesn't hold one branch's worth of them, so branches evict each other's and builds start cold. * The Windows sanitizer and unit test builds of the same sanitizer and architecture compile the same things with the same flags, so they now share a cache rather than each keeping its own copy. * ccache compresses what it stores harder (zstd level 10 rather than 1). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit a10bf0bca57c6b406799f49e08e403e51b783fe0 Author: Mooneer Salem Date: Tue Oct 13 10:43:30 2026 -0700 Don't count more of the branches that compilers add, and remove unused plot, notebook and dialog code Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 8e7426bb978c7965da6d171fb5f3642a4c308307 Author: Mooneer Salem Date: Tue Oct 13 10:43:30 2026 -0700 Test files that can't be written, ports that aren't serial ports and more of recording, playback and the audio tests Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 93d7a84bdc469774d64c6c077f08bb59ac73a96e Author: Mooneer Salem Date: Tue Oct 13 08:36:48 2026 -0700 Allow for a late wait being made up for in the PulseAudio pacing test A wait that ends late is deliberately taken off the next one, so a round after a late one is shorter than the stream's latency. Check the rounds so far rather than each by itself. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 71770e53b5497c449c608498915a6fd7478d3026 Author: Mooneer Salem Date: Tue Oct 13 00:45:53 2026 -0700 Fix a two second audio stall after processing is held up, and a typed-in frequency losing digits in Settings Linux: PulseAudioDevice kept how far a round of work overran, in nanoseconds, in an int. Work held up for more than about 2.1 seconds overflowed that, and the next round then waited up to 2 seconds more instead of not waiting at all. It's kept in 64 bits now. Settings > Rig Control: with a frequency typed into the entry box, turning off "frequencies in kHz" and pressing Apply left it with one decimal place of MHz (21313.5 kHz became 21.3) rather than four. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit c211ad8d03228790e97b8a52c330815fb0b124f8 Author: Mooneer Salem Date: Mon Oct 12 23:53:10 2026 -0700 Test when the speaker is muted around transmissions pipeline_TxRxThreadTest: what's received is played; not while transmitting in half duplex, nor until the radio's had time to go back to receiving; in full duplex it carries on, with no such wait, except while the voice keyer is sending. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 1f7504f8488ebff2ef4f357591628df12404a8f4 Author: Mooneer Salem Date: Mon Oct 12 23:52:03 2026 -0700 Don't make the audio thread wait for seconds after it's been held up for seconds, and test more Setup Wizard imports Fix (Linux): PulseAudioDevice kept how far a round of work overran, in nanoseconds, in an int. Work held up for more than about 2.1 seconds (a machine too busy to run it) overflowed that, and the next round then waited up to 2 seconds more instead of not waiting at all. It's kept in 64 bits now. audio_PulseAudioDeviceTest covers how rounds of work are paced: to the stream's latency, with overruns coming out of the waits that follow, and with nothing registered for underflows and overflows. gui_setup_wizard_import_more covers device names that were cut short, radios keyed by DTR or their own VOX, USB radios, and settings with nothing usable. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 534214606b1b193e253b8698d842c714f8b30dee Author: Mooneer Salem Date: Mon Oct 12 23:49:07 2026 -0700 Keep four decimal places when Settings converts a typed-in frequency from kHz to MHz, and add more FreeDV Reporter, Filter and Settings tests Fix: with a frequency typed into the Rig Control tab's entry box, turning off "frequencies in kHz" and pressing Apply left the box with one decimal place of MHz (21313.5 kHz became 21.3). New scenarios: gui_reporter_off_at_start, qsy_with_radio and new_callsign, and gui_filter_equalizer_receive_only. gui_settings_ptt_key_names covers the rest of the keys FreeDV has names for, and gui_settings_ptt_test a radio that Hamlib can't reach and a port that can't be opened. The test agent takes Insert, Home, End, PageUp, PageDown and numeric key codes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit c4b3fdf81041b414d0bd8511d4eb10467f8e2eaa Author: Mooneer Salem Date: Mon Oct 12 23:43:27 2026 -0700 Remove unused plot, main window and FreeDV Reporter code, and test more of the main window Removed, as nothing used them: PlotScalar's bar graph and log scale modes and two of its ways of adding samples; MainFrame's OnExitClick, OnCloseFrame and OnPaint, m_zoom and m_schedule_restore; and FreeDV Reporter's filtersEnabled() and hasAnyColumnFilter(). New scenarios: gui_main_window_tab_layout, heard_list, receive_reports (what's told to other programs over UDP and logged when a callsign is received), test_tone (the interference tone and the left channel VOX tone), exit_while_tuning, old_settings and bad_command_line. The test library can now return what it records, per channel, and say how much of it is a tone. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 8a7fff9fa8a21f606db4f3586f8821cc5b260f49 Author: Mooneer Salem Date: Mon Oct 12 23:12:02 2026 -0700 Wait for the radio before double-clicking a station in the FreeDV Reporter test gui_reporter_row_actions double-clicked a station as soon as FreeDV had started. If FreeDV hadn't connected to the radio by then, the radio went to that station's frequency and then straight back: on connecting, FreeDV tunes the radio to the frequency chosen in the main window. The test now waits for the connection first. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 12fd132a72ee4a182e54895e5ed01600c28bec1c Author: Mooneer Salem Date: Mon Oct 12 22:40:30 2026 -0700 Keep GUI tests off the public FreeDV Reporter server, and wait for FreeDV to carry on after a dialog closes FreeDV connects to FreeDV Reporter as soon as it's launched, so every test that didn't say where to find a server (the Setup Wizard ones and several Settings ones) connected to the public one. Configurations written by the test library now point at somewhere nothing is listening unless told otherwise, including for first-run launches; only the live view test names the public server. wait_for_window_closed() also now waits for FreeDV to leave the dialog's event loop. gui_settings_frequency_list failed once locally because its next command was answered before FreeDV had acted on Settings being closed. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 60f71ebb9ff76b3f9bdd955d9904ae609891157b Author: Mooneer Salem Date: Mon Oct 12 22:10:57 2026 -0700 Add PR #1559 to release notes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit ac51e6a753753daf91257779a65c144c29797f93 Author: Mooneer Salem Date: Mon Oct 12 22:10:52 2026 -0700 Fix FreeDV hanging on exit after a failed start, and uninitialised audio callbacks * startRxStream() started the transmit processing thread before it opened the receive streams. When opening those failed, nothing ever signalled or stopped that thread, and FreeDV hung on exit waiting for it. Both processing threads are now only started once every device is open, and any left by a start that failed aren't reused by the next. * IAudioDevice's callbacks had no initialisers (and it has no constructor), so a device that failed to start before the application had registered an error handler called through a garbage pointer. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 1a73d862c71c86dc65db43dd65f0f73cf23b5c8e Author: Mooneer Salem Date: Mon Oct 12 22:09:50 2026 -0700 Add the remaining FreeDV Reporter tests, and don't reuse threads from a failed start Tests: * gui_reporter_text_filters: filtering by each column that holds text. * gui_reporter_saved_filters: the window opening with band tracking, a custom idle time and the saved width of the Add Message window. * gui_reporter_selection: what unselects a station (clicking elsewhere, sorting, the station leaving). startRxStream() forgets any processing threads left by an earlier start that failed before starting them. Starting again as receive-only would otherwise have found the old transmit thread and started it, with the devices it was made for long gone. gui_reporter_row_actions waits longer for FreeDV's own frequency to reach the server, which took more than 15 seconds once under UBSan. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 6ba2b9cc120e493ff67e9b377f2f709d21648eb8 Author: Mooneer Salem Date: Mon Oct 12 21:59:27 2026 -0700 Test what the PTT key does when the time-out timer cuts TX short Nothing tested the momentary PTT key and the time-out timer together: which key presses the time-out is allowed to leave in force, and what happens to a press that is let go while TX is still starting. Both are now covered by a new main window scenario, ptt_key_time_out: - With the key still held when the time-out stops TX, pressing it again must not restart TX, and letting go must bring it back. FreeDV blocks the key until it is genuinely released (ongui.cpp:1225-1227) and a poll timer watches for the release (1232-1240); neither had ever run. - With a delay configured before TX starts, the button goes down as soon as the key is pressed while TX is still waiting. Letting go in that window is deferred (ongui.cpp:1071, 1078-1079), TX starts anyway, and is then stopped by itself (1684-1690) rather than being left keyed. The delay also exercises the waits before TX starts and after it stops (1618-1629, 1479-1490), which no test had reached. Running the test suite now covers 33 more lines of ongui.cpp (90 lines uncovered down to 57, 82.7% of the project covered up to 83.1%), with nothing that was covered becoming uncovered. The existing ptt_key and time_out_timer scenarios still pass. ongui.cpp:1287-1292, which would clear a time-out warning that is no longer wanted, is left untested: m_totWarningDialog_ is only ever set while the time-out has 15 seconds or less to go, and the Extend callback clears it (1265-1266) before raising the remaining time (1271), so the branch cannot be reached from the interface. commit 7f7c58a18d9bb9e204db20ffb9ce07543116587b Author: Mooneer Salem Date: Mon Oct 12 20:59:26 2026 -0700 Test what FreeDV does when the sound server goes away as it opens its streams Pressing Start with a sound card that couldn't be reached left FreeDV in a state it couldn't leave. startRxStream() started the TX processing thread before it opened the RX streams; when opening those failed, the cleanup reset the sound devices and returned without ever signalling that thread. Nothing else stopped it either: the cleanup sets m_RxRunning to false, and stopRxStream() only does anything when it's true. On exit, ~TxRxThread then joined a thread still waiting at startSem_.wait() for a start signal that never comes, so FreeDV hung. Where it got that far it died instead in wxLog recursion, a RecordStep worker logging while the pipeline was being destroyed during exit. startRxStream() already waited for the RX devices before starting the RX thread; it now does the same before starting the TX one. Two scenarios in test/gui/test_audio_errors.py: - sound_server_gone_while_opening_streams refuses FreeDV's connections to the sound server from its (2 + n)th onwards, for n = 0..3, which is what the server going away looks like to each of the four devices in turn. FreeDV names the ones it couldn't reach, closes the ones it had opened, and starts none of them. AudioServerGateway gains refuse_connections_after() for this; the 1 + 1 + 4 connections FreeDV makes on Start were measured first. - receive_only_device_moved moves the radio's input and the speaker's output of a one-pair-of-devices setup to other devices, and checks that FreeDV carries on using and remembering the new ones. Measured against the fixed code, the two scenarios cover 36 lines of src/main.cpp (1751/1941, 90.2% -> 1787/1941, 92.1%): the four device-start failures and the two one-sound-card device-changed handlers, 3193-3195, 3209-3211, 3536-3541, 3546-3553, 3562-3570 and 3576-3585, with nothing newly uncovered. What's left inside startRxStream() is the zero-sound-card branch (3163-3172, unreachable), the xrun counters (3459-3482, 3517-3523) and the FIFO race guard at 3503. commit 35300b23ae69c7679e0004760bd2243d1a2628f9 Author: Mooneer Salem Date: Mon Oct 12 18:43:00 2026 -0700 Test what FreeDV does when a sound card goes away while it's starting FreeDV checks its sound cards over one connection and closes it before opening any streams, so AudioServerGateway can now take a device away in between: after_next_connection_close() runs a test's own action when the next connection closes, and holds the connections after it back until that's done. VirtualAudioDevices.remove() unloads one device and waits for the sound server to stop listing it. test_sound_card_gone_while_starting covers the resulting paths in MainFrame::startRxStream(): the "Could not find ..." boxes for the TX output, RX input and RX output of a two sound card setup, for the RX input and RX output of a receive-only one, and the cleanup that follows them in both cases. commit 84b2923bd1a39f66014e94e84892c5bf2ac82f42 Author: Mooneer Salem Date: Mon Oct 12 18:05:41 2026 -0700 Test what FreeDV does when the sound server isn't there Tests: * audio_PulseAudioDeviceTest: an engine or a device whose start() fails because there is no sound server says so, and names the device it was asked for; a device whose start() failed has nothing left to stop; an underflow or an overflow is reported to the device it belongs to, with that device's state. * gui_audio_errors_sound_server_gone_while_starting: the sound server going away between FreeDV checking its sound cards and opening its streams, and the two errors that come out of that. Test library: AudioServerGateway, which stands between FreeDV and the sound server and can be made to stop letting connections through. Fix: IAudioDevice's audio callbacks had no initialiser, and IAudioDevice has no constructor, so PulseAudioDevice::start()'s "if (onAudioErrorFunction)" guard read garbage and called through a stale function pointer (seen as std::bad_alloc). All five function/state pairs are nullptr by default now; this affected the PulseAudio, PortAudio, WASAPI and Mac devices alike. commit 8c88a4d663a24172190b3de0cf95e44d4eb9ab35 Merge: f207fbb3 54bd3619 Author: Mooneer Salem Date: Mon Oct 12 13:07:23 2026 -0700 Merge remote-tracking branch 'origin/v3.0-dev' into ms-ci-gui-tests commit f207fbb3e0e754b76caa90fbf15da44be6d892f6 Author: Mooneer Salem Date: Mon Oct 12 13:07:23 2026 -0700 Leave more compiler-generated branches out of coverage, and add tests for real ones Coverage report: branches for cleaning up after a constructor that throws ("new Something(...)") and for asserts failing aren't counted any more, alongside those for exceptions. None of them are decisions in the code, and together they were most of what was reported as untested. Tests: * TxRxThreadTest: what's transmitted is dropped, not waited for or written over what's queued, when the radio's FIFO is full. * gui_audio_errors_sound_server_restarted: the sound server being restarted under FreeDV, and FreeDV carrying on after Stop and Start. * gui_rig_control_mode_display and editing_frequency: the radio's mode being displayed, and its dial being ignored while the frequency is being typed in. * gui_main_window_reporting_without_callsign and band_names. * gui_settings_ptt_key_names. Test agent: a "focus" command. Test library: the private sound server can be stopped and started again. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit f81dfab42e0a426955677e51cd02990922c21b87 Author: Mooneer Salem Date: Mon Oct 12 20:05:34 2026 +0000 latest user manual PDF commit 54bd36192dfda83b3866b229fb0410510fe77320 Merge: 5751cbe4 7d80acae Author: Mooneer Salem Date: Mon Oct 12 13:03:49 2026 -0700 Merge branch 'master' into v3.0-dev commit 7d80acae41e91116cf41b68850ad2f58636ad7f5 Author: Mooneer Salem Date: Mon Oct 12 13:02:06 2026 -0700 FreeDV Reporter: fix sorting of stations whose frequencies are far apart (#1558) * FreeDV Reporter: fix sorting of stations whose frequencies are far apart Sorting by frequency subtracted one frequency from the other and kept the result in an int, which it doesn't fit in once they're more than 2^31 Hz (about 2.1 GHz) apart: someone on a 2.4 GHz satellite uplink was listed below everyone on HF. Frequencies are now compared instead, as are distances, headings and signal reports (whose differences were rounded to whole numbers). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Add PR #1558 to release notes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp --------- Co-authored-by: Claude Opus 5.5 commit e702ff41d47a8c6b3572f436a28ba1df38317eb4 Author: Mooneer Salem Date: Mon Oct 12 13:01:23 2026 -0700 [skip ci] Linux: fix Load Default Configuration loading an empty configuration (#1557) * Linux: fix Load Default Configuration loading an empty configuration with wxWidgets 3.3 The fix in #1549 worked out where the default configuration is in the same way that startup does. By then wxWidgets has been set up to use that location, though, and gives a different answer, so an empty configuration was loaded from somewhere else. The location used at startup is now remembered and used instead. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Add PR #1557 to release notes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Fix build with wxWidgets older than 3.3 wxCONFIG_USE_XDG doesn't exist before wxWidgets 3.3, and the code that uses it in Load Default Configuration had lost the check for that. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Keep the default configuration that was opened at startup With wxWidgets older than 3.3 on Linux, Load Default Configuration still loaded an empty configuration: wxWidgets looks for the default one under the application's name, which is "freedv" when it's opened at startup and "FreeDV" afterwards. The configuration opened at startup is now kept and reused. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp --------- Co-authored-by: Claude Opus 5.5 commit e9a0208ba157c6f059408bc448e1ec5fecbd8676 Author: Mooneer Salem Date: Mon Oct 12 11:21:50 2026 -0700 Add PR #1558 to release notes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit c96f26560ebcd559fa4682e45cbc753f1959d429 Author: Mooneer Salem Date: Mon Oct 12 11:21:46 2026 -0700 FreeDV Reporter: fix sorting of stations whose frequencies are far apart Sorting by frequency subtracted one frequency from the other and kept the result in an int, which it doesn't fit in once they're more than 2^31 Hz (about 2.1 GHz) apart: someone on a 2.4 GHz satellite uplink was listed below everyone on HF. Frequencies are now compared instead, as are distances, headings and signal reports (whose differences were rounded to whole numbers). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit b878b323500d7597059a1819e4d2271916c7aa87 Author: Mooneer Salem Date: Mon Oct 12 11:08:59 2026 -0700 Keep the default configuration that was opened at startup With wxWidgets older than 3.3 on Linux, Load Default Configuration still loaded an empty configuration: wxWidgets looks for the default one under the application's name, which is "freedv" when it's opened at startup and "FreeDV" afterwards. The configuration opened at startup is now kept and reused. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 9eafbb566056e055aca6b63faa31887415a9786e Author: Mooneer Salem Date: Mon Oct 12 11:08:17 2026 -0700 Fix what the first CI run of the GUI tests found FreeDV: * Load Default Configuration still loaded an empty configuration with wxWidgets older than 3.3 on Linux: wxWidgets looks for the default configuration under the application's name, which is "freedv" when it's opened at startup and "FreeDV" afterwards. The configuration opened at startup is now kept and reused. * FreeDV Reporter sorted stations wrongly when their frequencies were more than 2^31 Hz apart (e.g. someone on a 2.4 GHz satellite uplink and someone on HF), as it subtracted them into an int. Distance, heading and SNR are compared rather than subtracted now too. Test agent and tests: * Build with wxWidgets 3.0 (Ubuntu 22.04). * Sorting by another column left the previous one marked as sorted by with wxWidgets 3.2. * Start rigctld with -P RIG, without which Hamlib 4.6 doesn't answer whether the radio is transmitting. * Don't assume what's selected once the selected station is filtered out, or that everyone on the public server has a frequency. * Give connection_burst longer for slow (sanitizer) builds. CI: * Leave the test that depends on who's on the public FreeDV Reporter server out of CI. * Print the coverage summary in the same layout as freedv-backend's, and don't count the branches that compilers add for exceptions. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit b73d6379a98215b93adcd97967679d219b8e9aab Author: Mooneer Salem Date: Mon Oct 12 09:14:31 2026 -0700 Fix build with wxWidgets older than 3.3 wxCONFIG_USE_XDG doesn't exist before wxWidgets 3.3, and the code that uses it in Load Default Configuration had lost the check for that. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 1195785ae020b0df3ef5669761e450624aa0ed81 Author: Mooneer Salem Date: Mon Oct 12 09:14:26 2026 -0700 Fix build with wxWidgets older than 3.3 wxCONFIG_USE_XDG doesn't exist before wxWidgets 3.3, and the code that uses it in Load Default Configuration had lost the check for that. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit a4c8d4f54d418a9d3cef2ee110d2901977ad12c3 Author: Mooneer Salem Date: Mon Oct 12 09:04:09 2026 -0700 Add PR #1557 to release notes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 2b9ed861a0329cdea8d5fe5376e5e379ced45fc1 Author: Mooneer Salem Date: Mon Oct 12 09:04:05 2026 -0700 Linux: fix Load Default Configuration loading an empty configuration with wxWidgets 3.3 The fix in #1549 worked out where the default configuration is in the same way that startup does. By then wxWidgets has been set up to use that location, though, and gives a different answer, so an empty configuration was loaded from somewhere else. The location used at startup is now remembered and used instead. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 13f18323f81cb161390e89806833fd89cf4eebb0 Author: Mooneer Salem Date: Mon Oct 12 09:03:26 2026 -0700 Add PR #1556 to release notes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit b7cd11f66785c6b16d31330cea30a65fcd1a6961 Author: Mooneer Salem Date: Mon Oct 12 09:03:11 2026 -0700 Run the GUI tests in CI, report coverage, and add more GUI tests CI (Linux): * The ubuntu-26.04 test-application jobs (plain and each sanitizer) run the GUI tests after the unit tests. * A new coverage job builds with --coverage (WITH_COVERAGE=1 in build_linux.sh), runs every test and reports what was executed in the job summary, with the HTML report as an artifact. Tests: * gui_main_window_default_config, exit_menu, tune_level_per_band, frequency_units and appearance_change. * gui_reporter_sort_every_column, numeric_filters and status_message_flags. Fixes: * Load Default Configuration loaded an empty configuration on Linux builds with wxWidgets 3.3: the default configuration's location was worked out again, and wxWidgets gives a different answer by then. The location used at startup is now remembered. * TxRxThreadTest aborted under RealtimeSanitizer: a test that turns on FIFO state logging made the logger's first use on its thread happen on a real-time path. The test now logs once before it starts. Also removes plot_scatter.cpp/.h, which nothing used. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit add48d70592e2aa94540f69c8202370970d59230 Author: Mooneer Salem Date: Mon Oct 12 15:30:01 2026 +0000 latest user manual PDF commit f305b6970dc6aa69bbbdfb44760ba5ba81e0b004 Author: Mooneer Salem Date: Mon Oct 12 08:26:34 2026 -0700 Linux: fix crash on Start with no sound server and handle losing it; fix leftover monitor volume popups (#1555) * Linux: fix crash on Start with no sound server and handle losing it; fix leftover monitor volume popups * Pressing Start with no sound server (PipeWire/PulseAudio) running aborted FreeDV, as the audio engine went on to use a connection that was never made. It now reports no devices instead. * Losing the sound server while running went unreported. Streams that fail are now reported like other audio errors. * Stopping after that waited for a stream to confirm a disconnect that it could no longer make. * Each use of Adjust Monitor Volume left a hidden popup behind; it now destroys itself once dismissed. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Add PR #1555 to release notes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp --------- Co-authored-by: Claude Opus 5.5 commit 5751cbe480acf92f5f84688497d2930fc34a8660 Author: Mooneer Salem Date: Mon Oct 12 15:04:19 2026 +0000 latest user manual PDF commit d94fc54368741f5b70dc6fcf7d9d535da37db4dc Author: Mooneer Salem Date: Mon Oct 12 07:57:14 2026 -0700 Add GUI tests for plots, rig control, sound server loss and more, and fix the defects they found (#1554) * Add GUI tests for plots, rig control, sound server loss and more, and fix the defects they found Tests: * test/gui/test_plots.py (gui_plots_*): tuning with the mouse and keyboard on the waterfall and spectrum, and every plot being drawn. * test/gui/test_rig_control.py (gui_rig_control_*): frequency, mode and PTT against Hamlib's rigctld, and the radio being unreachable or lost. * gui_audio_errors_device_removed: a device, then the whole sound server, going away while FreeDV is running. Uses a sound server of its own. * gui_reporter_row_actions and gui_reporter_turn_off: callsign lookups, copying a message, the website button, double-click to tune and the main window's Turn Off button. * gui_main_window_monitor_volume, mouse_controls, play_file_loop and help_and_restore. Defects fixed: * Linux: pressing Start with no sound server running aborted FreeDV. * Linux: losing the sound server while running went unreported, and stopping afterwards waited for a stream shutdown that would never be confirmed. * Restore defaults reported its result through wxLogMessage/wxLogError rather than a message box. * Each use of Adjust Monitor Volume left a hidden popup behind. To support the tests: web pages are opened through LaunchBrowser(), the plots and the monitor volume popup get names, and the test agent gains mouse events, arrow keys, the clipboard, options for file choosers and dismissing popups. Its quit command no longer gets lost when it arrives while a dialog that's just been dismissed is still finishing up, which was behind tests occasionally failing with FreeDV not exiting. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Add PR #1554 to release notes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Test agent: fix build on Ubuntu 22.04 Sorting FreeDV Reporter's rows with std::stable_sort uses std::get_temporary_buffer in GCC 12's standard library, which is deprecated in C++20. clang warns about that, and warnings are errors here. Sort positions with std::sort instead, keeping equal rows in the model's order. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp --------- Co-authored-by: Claude Opus 5.5 commit ff019b2004ddfc3976e29894cfa12e3ade624924 Author: Mooneer Salem Date: Mon Oct 12 06:34:56 2026 -0700 Test agent: fix build on Ubuntu 22.04 Sorting FreeDV Reporter's rows with std::stable_sort uses std::get_temporary_buffer in GCC 12's standard library, which is deprecated in C++20. clang warns about that, and warnings are errors here. Sort positions with std::sort instead, keeping equal rows in the model's order. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 7dee8e84bc11b6366ca8f630d9df721fd4eb5a52 Author: Mooneer Salem Date: Mon Oct 12 06:13:30 2026 -0700 Add PR #1554 to release notes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 00dbc8db941d7c226c224d5376917e9bd570ff5c Author: Mooneer Salem Date: Mon Oct 12 06:13:27 2026 -0700 Add PR #1555 to release notes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 2bdddfd5c0a543630d485626ef6ccbc9fbef6470 Author: Mooneer Salem Date: Mon Oct 12 06:13:11 2026 -0700 Linux: fix crash on Start with no sound server and handle losing it; fix leftover monitor volume popups * Pressing Start with no sound server (PipeWire/PulseAudio) running aborted FreeDV, as the audio engine went on to use a connection that was never made. It now reports no devices instead. * Losing the sound server while running went unreported. Streams that fail are now reported like other audio errors. * Stopping after that waited for a stream to confirm a disconnect that it could no longer make. * Each use of Adjust Monitor Volume left a hidden popup behind; it now destroys itself once dismissed. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 4457b7b6c797631ac586c751b099c05f58d26cba Author: Mooneer Salem Date: Mon Oct 12 06:11:53 2026 -0700 Add GUI tests for plots, rig control, sound server loss and more, and fix the defects they found Tests: * test/gui/test_plots.py (gui_plots_*): tuning with the mouse and keyboard on the waterfall and spectrum, and every plot being drawn. * test/gui/test_rig_control.py (gui_rig_control_*): frequency, mode and PTT against Hamlib's rigctld, and the radio being unreachable or lost. * gui_audio_errors_device_removed: a device, then the whole sound server, going away while FreeDV is running. Uses a sound server of its own. * gui_reporter_row_actions and gui_reporter_turn_off: callsign lookups, copying a message, the website button, double-click to tune and the main window's Turn Off button. * gui_main_window_monitor_volume, mouse_controls, play_file_loop and help_and_restore. Defects fixed: * Linux: pressing Start with no sound server running aborted FreeDV. * Linux: losing the sound server while running went unreported, and stopping afterwards waited for a stream shutdown that would never be confirmed. * Restore defaults reported its result through wxLogMessage/wxLogError rather than a message box. * Each use of Adjust Monitor Volume left a hidden popup behind. To support the tests: web pages are opened through LaunchBrowser(), the plots and the monitor volume popup get names, and the test agent gains mouse events, arrow keys, the clipboard, options for file choosers and dismissing popups. Its quit command no longer gets lost when it arrives while a dialog that's just been dismissed is still finishing up, which was behind tests occasionally failing with FreeDV not exiting. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 9891cc83bdb5db42aeb0c4802e32d8da5dc30d31 Author: Mooneer Salem Date: Mon Oct 12 08:48:14 2026 +0000 latest user manual PDF commit 96cc4594925ac668de6f560bb668c4a914f836fb Author: Mooneer Salem Date: Mon Oct 12 08:44:54 2026 +0000 latest user manual PDF commit cce409fbd197b8fe5cb22be7937fd44f3d37655d Author: Mooneer Salem Date: Mon Oct 12 01:08:01 2026 -0700 Add GUI tests for audio device errors and serial port PTT (#1553) * Add GUI tests for audio device errors and serial port PTT Tests: * test/gui/test_audio_errors.py (gui_audio_errors_*): starting with a device that's missing, set to too low a sample rate or not chosen at all, receive only setups, and streams being moved to other devices while FreeDV is running. * test/gui/test_serial_ptt.py (gui_serial_ptt_*): keying a radio with RTS and/or DTR in either polarity, transmitting when CTS says to, ports that can't be opened and the Settings window's Test PTT button. Pseudo-terminals stand in for serial ports. They have no control lines, so FreeDV is run with a small library (built by the test, so a C compiler is needed) that keeps track of the ones it sets and supplies the one it reads. Also: * gui_settings_auto_start no longer lets FreeDV connect to the public FreeDV Reporter server. * If FreeDV doesn't exit when told to, the test library reports what each of its threads was doing. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Add PR #1553 to release notes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp --------- Co-authored-by: Claude Opus 5.5 commit ba47e6a5a3b882bc9a9430880f18f97c2876ebc8 Merge: a259bc84 fced2711 Author: Mooneer Salem Date: Mon Oct 12 01:02:44 2026 -0700 Merge branch 'master' into v3.0-dev commit 5209f291fb10edac03464d1a3e92828203f9e822 Author: Mooneer Salem Date: Mon Oct 12 01:01:10 2026 -0700 Add PR #1553 to release notes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 2daf29cdc1b5953c2fb2ea40866880f0bbf87c22 Author: Mooneer Salem Date: Mon Oct 12 01:00:59 2026 -0700 Add GUI tests for audio device errors and serial port PTT Tests: * test/gui/test_audio_errors.py (gui_audio_errors_*): starting with a device that's missing, set to too low a sample rate or not chosen at all, receive only setups, and streams being moved to other devices while FreeDV is running. * test/gui/test_serial_ptt.py (gui_serial_ptt_*): keying a radio with RTS and/or DTR in either polarity, transmitting when CTS says to, ports that can't be opened and the Settings window's Test PTT button. Pseudo-terminals stand in for serial ports. They have no control lines, so FreeDV is run with a small library (built by the test, so a C compiler is needed) that keeps track of the ones it sets and supplies the one it reads. Also: * gui_settings_auto_start no longer lets FreeDV connect to the public FreeDV Reporter server. * If FreeDV doesn't exit when told to, the test library reports what each of its threads was doing. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit fced271183dba46d5cb864920d24eef2975eb484 Author: Mooneer Salem Date: Mon Oct 12 00:59:05 2026 -0700 Fix crash when closing FreeDV right after stopping with a radio connected (#1543) * Fix crash when closing FreeDV right after stopping with a radio connected Stopping the modem hands the rig controllers to detached threads to finish disconnecting. When they're done, they call back into the main window (onRadioDisconnected_()), which may have been destroyed by then if FreeDV was closed straight after stopping. Wait a bounded amount of time for those disconnects in MainFrame's destructor, as is already done when closing while the modem is running. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Add PR #1543 to release notes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp --------- Co-authored-by: Claude Opus 5.5 commit b392daadb5dc6f189efa0096023d6d40101cba97 Merge: 958812c7 75bfa636 Author: Mooneer Salem Date: Mon Oct 12 00:58:07 2026 -0700 Merge branch 'master' into ms-rig-disconnect-close-crash commit 75bfa636ff7805dbe5739494a76e9904a051a6aa Author: Mooneer Salem Date: Mon Oct 12 00:57:46 2026 -0700 [skip ci] FreeDV Reporter: fix idle filter check marks, stale Request QSY button and Show menu handler (#1547) * FreeDV Reporter: fix idle filter check marks, stale Request QSY button and Show menu handler * Cancelling the idle filter's "Custom..." dialog left "Custom..." checked alongside the setting actually in effect, or "120" checked if the filter was disabled. Put the check marks back to what's in effect. * Request QSY stayed enabled after a request was sent, as the row is deselected without a selection change being reported. Disable it whenever the selection is cleared from code. * The Show menu's handler cast the event's object to wxMenuItem*, but it's the wxMenu. Look the item up by ID instead. Found by GUI tests of the FreeDV Reporter window on v3.0-dev (#1546). Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Add PR #1547 to release notes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp --------- Co-authored-by: Claude Opus 5.5 commit 05a7b2d1d1be37522b166583a3a3fd696b0fa987 Merge: 937e9156 cb0881e2 Author: Mooneer Salem Date: Mon Oct 12 00:57:08 2026 -0700 Merge branch 'master' into ms-reporter-fixes commit cb0881e2506730252a66ff2ff7d43f5f5201b9bd Author: Mooneer Salem Date: Mon Oct 12 00:56:18 2026 -0700 [skip ci] Fix voice keyer button label, recorded voice keyer file name and Load Default Configuration (#1549) * Fix voice keyer button label, recorded voice keyer file name and Load Default Configuration * The voice keyer button never changed to "Stop Voice Keyer" as the text being replaced lacked the accelerator that the label has. * With that fixed, make sure the label goes back to "Start Voice Keyer" if the chosen file turns out to be unusable. * Recording a new voice keyer file to a name ending in .wav saved the name without its extension, so the file couldn't be found the next time FreeDV started. * Load Default Configuration reloaded the configuration already in use (wxConfigBase::Create() returns the existing one if there is one) rather than the default one. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Add PR #1549 to release notes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp --------- Co-authored-by: Claude Opus 5.5 commit d549489fb1a8f9602706ed318d6e69c60d9b34bc Merge: 22bff8a5 3844954c Author: Mooneer Salem Date: Mon Oct 12 00:55:28 2026 -0700 Merge branch 'master' into ms-voice-keyer-config-fixes commit 3844954c7d4711fd2c2c31118ed922c5fe3faca2 Author: Mooneer Salem Date: Mon Oct 12 00:55:01 2026 -0700 [skip ci] Fix logged contact times, Filter window equaliser changes, leftover Log QSO windows and pasted recording suffixes (#1552) * Fix logged contact times, Filter window equaliser changes, leftover Log QSO windows and pasted recording suffixes * Logged contacts could be timed a millisecond (so, as displayed, a second) early, as the time of day was truncated rather than rounded. The unit test that allowed for this now doesn't. * Changes to the equalisers were ignored if the Filter window had been opened before FreeDV was started. * Each use of Log QSO left a hidden window behind. * A recording suffix that wasn't typed (e.g. pasted) wasn't restricted to what's safe in a file name. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Add PR #1552 to release notes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp --------- Co-authored-by: Claude Opus 5.5 commit a259bc84a2523ac1a21d1f4d434adf8b173c16e6 Author: Mooneer Salem Date: Sun Oct 11 23:59:40 2026 -0700 Add GUI tests for the Log QSO, Start Recording and Filter windows and fix the defects they found (#1551) * Add GUI tests for the Log QSO, Start Recording and Filter windows and fix the defects they found Tests: * test/gui/test_log_qso.py (gui_log_qso_*): logging a contact typed in from scratch, one from the list of callsigns heard and one chosen in FreeDV Reporter, checking what's sent to the logging program. * test/gui/test_filter.py (gui_filter_*): the Filter window's controls and the effect of its volumes, equalisers and noise suppression on what's heard. * gui_main_window_recording_options: the rest of the Start Recording window. Defects fixed: * Logged contacts could be timed a millisecond (so, as displayed, a second) early, as the time of day was truncated rather than rounded. The unit test that allowed for this now doesn't. * Changes to the equalisers were ignored if the Filter window had been opened before FreeDV was started. * The main window's speaker level didn't follow the Filter window's while FreeDV wasn't running. * Each use of Log QSO left a hidden window behind. * A recording suffix that wasn't typed (e.g. pasted) wasn't restricted to what's safe in a file name. To support the tests: names for the widgets involved, clicking rows of lists with columns in the test agent, filter settings in its state and a stand-in for a WSJT-X compatible logging program in the test library. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Add PR #1551 to release notes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp --------- Co-authored-by: Claude Opus 5.5 commit e4ee003859469aa67d9a8c04b1e5fe64f88722bb Author: Mooneer Salem Date: Mon Oct 12 06:10:01 2026 +0000 latest user manual PDF commit 05542f253ae4f0a90a0b1e024e81a3dfba4a9f06 Author: Mooneer Salem Date: Sun Oct 11 23:04:38 2026 -0700 Add PR #1552 to release notes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 991ccab0498c1069a3fb4159f3effd2602ce7286 Author: Mooneer Salem Date: Sun Oct 11 23:04:24 2026 -0700 Fix logged contact times, Filter window equaliser changes, leftover Log QSO windows and pasted recording suffixes * Logged contacts could be timed a millisecond (so, as displayed, a second) early, as the time of day was truncated rather than rounded. The unit test that allowed for this now doesn't. * Changes to the equalisers were ignored if the Filter window had been opened before FreeDV was started. * Each use of Log QSO left a hidden window behind. * A recording suffix that wasn't typed (e.g. pasted) wasn't restricted to what's safe in a file name. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 79d3e382561e6f0d916a0bd51912171ccb7b9377 Author: Mooneer Salem Date: Sun Oct 11 23:03:30 2026 -0700 Add PR #1551 to release notes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 62f96177ed91d46744eae216dfdc733b5b966594 Author: Mooneer Salem Date: Sun Oct 11 23:03:12 2026 -0700 Add GUI tests for the Log QSO, Start Recording and Filter windows and fix the defects they found Tests: * test/gui/test_log_qso.py (gui_log_qso_*): logging a contact typed in from scratch, one from the list of callsigns heard and one chosen in FreeDV Reporter, checking what's sent to the logging program. * test/gui/test_filter.py (gui_filter_*): the Filter window's controls and the effect of its volumes, equalisers and noise suppression on what's heard. * gui_main_window_recording_options: the rest of the Start Recording window. Defects fixed: * Logged contacts could be timed a millisecond (so, as displayed, a second) early, as the time of day was truncated rather than rounded. The unit test that allowed for this now doesn't. * Changes to the equalisers were ignored if the Filter window had been opened before FreeDV was started. * The main window's speaker level didn't follow the Filter window's while FreeDV wasn't running. * Each use of Log QSO left a hidden window behind. * A recording suffix that wasn't typed (e.g. pasted) wasn't restricted to what's safe in a file name. To support the tests: names for the widgets involved, clicking rows of lists with columns in the test agent, filter settings in its state and a stand-in for a WSJT-X compatible logging program in the test library. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 1ccad589348ce6dc8b8e4e5c09da676776418450 Author: Mooneer Salem Date: Sun Oct 11 22:27:57 2026 -0700 Add GUI tests for the main window and fix the defects they found (#1550) * Add GUI tests for the main window and fix the defects they found Tests (test/gui/test_main_window.py, run as gui_main_window_*): transmitting by button and by space bar, TX and tune attenuation, receiving, analog, playing and recording files, the voice keyer, the time-out timer and the Settings/Help menu items that don't open other windows. Defects fixed: * The voice keyer button never changed to "Stop Voice Keyer", and (once it did) stayed that way after a file that couldn't be used. * Recording a new voice keyer file to a name ending in .wav saved the name without its extension. * Load Default Configuration reloaded the configuration already in use rather than the default one. To support the tests: the remaining file choosers go through ShowPathDialog, the Begin Recording and time-out warning dialogs get widget names, the test agent handles sliders, key releases and the status bar, and the test library gives FreeDV a HOME of its own. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Add PR #1550 to release notes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp --------- Co-authored-by: Claude Opus 5.5 commit 50006950ba2891e72599e10b583339da94cd6146 Author: Mooneer Salem Date: Sun Oct 11 21:27:04 2026 -0700 Add PR #1550 to release notes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 560e3f5ebd9e1cf042d9539bb0506c54b0c31790 Author: Mooneer Salem Date: Sun Oct 11 21:26:52 2026 -0700 Add GUI tests for the main window and fix the defects they found Tests (test/gui/test_main_window.py, run as gui_main_window_*): transmitting by button and by space bar, TX and tune attenuation, receiving, analog, playing and recording files, the voice keyer, the time-out timer and the Settings/Help menu items that don't open other windows. Defects fixed: * The voice keyer button never changed to "Stop Voice Keyer", and (once it did) stayed that way after a file that couldn't be used. * Recording a new voice keyer file to a name ending in .wav saved the name without its extension. * Load Default Configuration reloaded the configuration already in use rather than the default one. To support the tests: the remaining file choosers go through ShowPathDialog, the Begin Recording and time-out warning dialogs get widget names, the test agent handles sliders, key releases and the status bar, and the test library gives FreeDV a HOME of its own. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 22bff8a5ceb3455b884b931be2f85e860439cdcc Author: Mooneer Salem Date: Sun Oct 11 21:26:29 2026 -0700 Add PR #1549 to release notes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 53775a27d13bc9a2001c879868f4774148276472 Author: Mooneer Salem Date: Sun Oct 11 21:26:17 2026 -0700 Fix voice keyer button label, recorded voice keyer file name and Load Default Configuration * The voice keyer button never changed to "Stop Voice Keyer" as the text being replaced lacked the accelerator that the label has. * With that fixed, make sure the label goes back to "Start Voice Keyer" if the chosen file turns out to be unusable. * Recording a new voice keyer file to a name ending in .wav saved the name without its extension, so the file couldn't be found the next time FreeDV started. * Load Default Configuration reloaded the configuration already in use (wxConfigBase::Create() returns the existing one if there is one) rather than the default one. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 475d8ddc336ae995cf9f9515ac1c25407a4ba5c8 Author: Mooneer Salem Date: Sun Oct 11 21:24:46 2026 -0700 Add GUI tests for the Settings window and fix the defect they found (#1548) * Add GUI tests for the Settings window and fix the defect they found test/gui/test_settings.py has thirteen scenarios covering each tab of the Settings window, its OK/Apply/Cancel buttons, the audio and PTT test buttons, starting automatically on launch and what's locked while FreeDV is running. They're registered with CTest as gui_settings_*. To support them: * The Settings window and its controls are named. * The window's file and folder choosers go through ShowPathDialog(), so that a test script can say what to choose (or cancel) in place of the chooser being displayed. * The test agent can also drive list controls, notebook tabs and color pickers, and press keys. * freedvtest can read back a saved configuration, play a tone into a virtual audio device and measure the level on one. Defect found by the tests and fixed here: the Test buttons for the two audio inputs played what they'd recorded to the radio's output device (when one was selected) rather than to the speaker/headphones, where it can be heard. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Add PR #1548 to release notes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp --------- Co-authored-by: Claude Opus 5.5 commit 5ab3257f2f3fec2bff045df59c75cc159e3f9eeb Author: Mooneer Salem Date: Sun Oct 11 20:41:18 2026 -0700 Add PR #1548 to release notes. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 8ab58ae4c24e0cb15e89082f4d94f0cb1e0d03d4 Author: Mooneer Salem Date: Sun Oct 11 20:41:08 2026 -0700 Add GUI tests for the Settings window and fix the defect they found test/gui/test_settings.py has thirteen scenarios covering each tab of the Settings window, its OK/Apply/Cancel buttons, the audio and PTT test buttons, starting automatically on launch and what's locked while FreeDV is running. They're registered with CTest as gui_settings_*. To support them: * The Settings window and its controls are named. * The window's file and folder choosers go through ShowPathDialog(), so that a test script can say what to choose (or cancel) in place of the chooser being displayed. * The test agent can also drive list controls, notebook tabs and color pickers, and press keys. * freedvtest can read back a saved configuration, play a tone into a virtual audio device and measure the level on one. Defect found by the tests and fixed here: the Test buttons for the two audio inputs played what they'd recorded to the radio's output device (when one was selected) rather than to the speaker/headphones, where it can be heard. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 7313addd61837649877b6a0d2a16ec11a1111302 Author: Mooneer Salem Date: Sun Oct 11 20:40:05 2026 -0700 Add GUI tests for the FreeDV Reporter window and fix the defects they found (#1546) * Add GUI tests for the FreeDV Reporter window and fix the defects they found Extends the GUI test support added for the Setup Wizard so that scripts can also drive: * Lists with columns (wxDataViewCtrl): reading rows, columns and sort order; hovering over, selecting, double-clicking and right-clicking rows; clicking and right-clicking column headings. * Popup menus. These now go through ShowPopupMenu(), which hands the menu to the test agent in place of displaying it so that a script can see what's on it and choose an item. * wxMessageDialog, by way of the hook already used for wxMessageBox(). Scripts can now also say how the next one should be answered. * Menus of windows other than the main one, including submenus. * Dialogs provided by wxWidgets (e.g. text entry), whose windows and controls have no names of their own, as "@/<label or class>". * wxComboCtrl, wxListBox and wxSpinCtrl, and window activation. test/gui/freedvtest/reporter.py runs the real FreeDV Reporter server locally (from its own source) and connects stations to it, so that tests control what's listed and nothing is reported publicly. test/gui/test_reporter.py has eleven scenarios, registered with CTest as gui_reporter_*. One only views the public server; another (connection_burst) is an attempt to reproduce the crash in #1495, which it doesn't. FreeDV is now also forced to use X11 when run by the tests on Linux, as GTK otherwise prefers a Wayland session if one exists, even under Xvfb. Defects found by the tests and fixed here: * Cancelling the custom idle time left "Custom..." checked alongside the setting in effect (or "120" checked if the filter was disabled). * Request QSY stayed enabled after a request was sent, with nothing selected. * The Show menu's handler treated the event's object as the menu item that was chosen; it's the menu. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Add PR #1546 to release notes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> commit bd1561f8eee478484af934d71a01390ff2c9d0f4 Author: Mooneer Salem <mooneer@gmail.com> Date: Sun Oct 11 19:51:15 2026 -0700 Add PR #1546 to release notes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 937e91562194d9063db6360be79461beaeb96fc2 Author: Mooneer Salem <mooneer@gmail.com> Date: Sun Oct 11 19:51:13 2026 -0700 Add PR #1547 to release notes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 1a57634f927a6ee7ae9097198c9290136ea1023e Author: Mooneer Salem <mooneer@gmail.com> Date: Sun Oct 11 19:50:58 2026 -0700 FreeDV Reporter: fix idle filter check marks, stale Request QSY button and Show menu handler * Cancelling the idle filter's "Custom..." dialog left "Custom..." checked alongside the setting actually in effect, or "120" checked if the filter was disabled. Put the check marks back to what's in effect. * Request QSY stayed enabled after a request was sent, as the row is deselected without a selection change being reported. Disable it whenever the selection is cleared from code. * The Show menu's handler cast the event's object to wxMenuItem*, but it's the wxMenu. Look the item up by ID instead. Found by GUI tests of the FreeDV Reporter window on v3.0-dev (#1546). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit ded98e6dc5a6bacb81be61cd81d65f95bfacdca1 Author: Mooneer Salem <mooneer@gmail.com> Date: Sun Oct 11 19:48:48 2026 -0700 Add GUI tests for the FreeDV Reporter window and fix the defects they found Extends the GUI test support added for the Setup Wizard so that scripts can also drive: * Lists with columns (wxDataViewCtrl): reading rows, columns and sort order; hovering over, selecting, double-clicking and right-clicking rows; clicking and right-clicking column headings. * Popup menus. These now go through ShowPopupMenu(), which hands the menu to the test agent in place of displaying it so that a script can see what's on it and choose an item. * wxMessageDialog, by way of the hook already used for wxMessageBox(). Scripts can now also say how the next one should be answered. * Menus of windows other than the main one, including submenus. * Dialogs provided by wxWidgets (e.g. text entry), whose windows and controls have no names of their own, as "@<title>/<label or class>". * wxComboCtrl, wxListBox and wxSpinCtrl, and window activation. test/gui/freedvtest/reporter.py runs the real FreeDV Reporter server locally (from its own source) and connects stations to it, so that tests control what's listed and nothing is reported publicly. test/gui/test_reporter.py has eleven scenarios, registered with CTest as gui_reporter_*. One only views the public server; another (connection_burst) is an attempt to reproduce the crash in #1495, which it doesn't. FreeDV is now also forced to use X11 when run by the tests on Linux, as GTK otherwise prefers a Wayland session if one exists, even under Xvfb. Defects found by the tests and fixed here: * Cancelling the custom idle time left "Custom..." checked alongside the setting in effect (or "120" checked if the filter was disabled). * Request QSY stayed enabled after a request was sent, with nothing selected. * The Show menu's handler treated the event's object as the menu item that was chosen; it's the menu. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 5e346b88a11e25b297911825131084af5af59c98 Author: Mooneer Salem <mooneer@gmail.com> Date: Mon Oct 12 01:47:21 2026 +0000 latest user manual PDF commit 57add43189e0c9c421db91e048bf20f55ed83a51 Author: Mooneer Salem <mooneer@gmail.com> Date: Sun Oct 11 18:13:38 2026 -0700 Add GUI tests for the Setup Wizard and fix the defects they found (#1544) * Add GUI tests for the Setup Wizard and fix the defects they found Adds a way for Python scripts to drive FreeDV's user interface and check the results: * When built with UNITTEST, FreeDV takes a -testport option and connects to a test script listening on that local port (TestAgent, built on freedv-backend's TcpConnectionHandler). The script sends JSON commands to click, set and read controls, select menu items and read FreeDV's state and configuration. Message boxes are recorded for the script rather than displayed. * Controls in the main window and Setup Wizard are named with SetName() so that scripts can refer to them. * test/gui/freedvtest is the Python side: start() launches FreeDV with its own configuration and returns an object for sending commands. * test/gui/test_setup_wizard.py covers the wizard in nine scenarios, each registered with CTest under the "gui" label (Linux only). They're excluded from the per-push CI run. Defects found by the tests and fixed here: * Closing FreeDV right after stopping with a Hamlib rig connected crashed, as the rig finished disconnecting after the main window was destroyed. * The Setup Wizard's PTT serial device wasn't enabled or disabled when the PTT method changed. The PTT serial device (Setup Wizard and settings) is also now only selectable when PTT uses RTS or DTR, the only methods that use it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp * Add PR #1544 to release notes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> commit a66e302a8a3a3e7860da97da668dda553089d64e Author: Mooneer Salem <mooneer@gmail.com> Date: Sun Oct 11 17:57:57 2026 -0700 Fix TxRxThreadTest recording test failing on busy machines (#1545) recordingFromRadioContinuesUntilStopped handed the receive pipeline 0.6s of audio as fast as it could. What's received reaches the file through TapStep, whose buffer holds 0.25s and discards the rest, so the test only passed if TapStep's thread happened to run while the audio was being handed over. When it didn't, exactly one buffer's worth was recorded (11520 samples, where at least 21600 were expected). Hand the audio over a tenth of a second at a time, waiting for it to be recorded in between, as the file's other recording tests already do. Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> commit 53fd7d98bc8573b2e8d5a08be2e17ab235da0be1 Author: Mooneer Salem <mooneer@gmail.com> Date: Sun Oct 11 17:40:22 2026 -0700 Fix TxRxThreadTest recording test failing on busy machines recordingFromRadioContinuesUntilStopped handed the receive pipeline 0.6s of audio as fast as it could. What's received reaches the file through TapStep, whose buffer holds 0.25s and discards the rest, so the test only passed if TapStep's thread happened to run while the audio was being handed over. When it didn't, exactly one buffer's worth was recorded (11520 samples, where at least 21600 were expected). Hand the audio over a tenth of a second at a time, waiting for it to be recorded in between, as the file's other recording tests already do. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit f6db4a20fed914d6d4de9ce19fc6a08681c3bb76 Author: Mooneer Salem <mooneer@gmail.com> Date: Sun Oct 11 17:14:46 2026 -0700 Add PR #1544 to release notes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 63ca20367c737b59761d317b47ca4f0087d65581 Author: Mooneer Salem <mooneer@gmail.com> Date: Sun Oct 11 17:14:18 2026 -0700 Add GUI tests for the Setup Wizard and fix the defects they found Adds a way for Python scripts to drive FreeDV's user interface and check the results: * When built with UNITTEST, FreeDV takes a -testport option and connects to a test script listening on that local port (TestAgent, built on freedv-backend's TcpConnectionHandler). The script sends JSON commands to click, set and read controls, select menu items and read FreeDV's state and configuration. Message boxes are recorded for the script rather than displayed. * Controls in the main window and Setup Wizard are named with SetName() so that scripts can refer to them. * test/gui/freedvtest is the Python side: start() launches FreeDV with its own configuration and returns an object for sending commands. * test/gui/test_setup_wizard.py covers the wizard in nine scenarios, each registered with CTest under the "gui" label (Linux only). They're excluded from the per-push CI run. Defects found by the tests and fixed here: * Closing FreeDV right after stopping with a Hamlib rig connected crashed, as the rig finished disconnecting after the main window was destroyed. * The Setup Wizard's PTT serial device wasn't enabled or disabled when the PTT method changed. The PTT serial device (Setup Wizard and settings) is also now only selectable when PTT uses RTS or DTR, the only methods that use it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit 958812c7c65f177ceb9418205035e39673909449 Author: Mooneer Salem <mooneer@gmail.com> Date: Sun Oct 11 17:08:04 2026 -0700 Add PR #1543 to release notes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit f216e310f6410dbf3b3ca8e47c8a97134d7a8793 Author: Mooneer Salem <mooneer@gmail.com> Date: Sun Oct 11 17:07:49 2026 -0700 Fix crash when closing FreeDV right after stopping with a radio connected Stopping the modem hands the rig controllers to detached threads to finish disconnecting. When they're done, they call back into the main window (onRadioDisconnected_()), which may have been destroyed by then if FreeDV was closed straight after stopping. Wait a bounded amount of time for those disconnects in MainFrame's destructor, as is already done when closing while the modem is running. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BpmjFvzRZPSFGZLhgvhqDp commit af35e424c75f712afd986dccd9e06712df90270a Merge: bd648231 1a3b24d1 Author: Mooneer Salem <mooneer@gmail.com> Date: Sun Oct 11 16:05:30 2026 -0700 Merge branch 'v3.0-dev' into ms-agc-tweaks-test commit 89323ad06a62305b14eb9766cfa4931a0ca9a294 Merge: 9323ba36 1a3b24d1 Author: Mooneer Salem <mooneer@gmail.com> Date: Sun Oct 11 16:04:31 2026 -0700 Merge branch 'v3.0-dev' into ms-rade-v2 commit 1a3b24d127a3065e25b7d536a1acc235cb4e6295 Author: Mooneer Salem <mooneer@gmail.com> Date: Sun Oct 11 16:02:15 2026 -0700 [skip ci] Re-sync language files. commit f5eb013dd3f5c3826eab93b456eade246adef80b Merge: e9f1e4ef b9aabc29 Author: Mooneer Salem <mooneer@gmail.com> Date: Sun Oct 11 15:56:43 2026 -0700 Merge branch 'master' into v3.0-dev commit 420114e61810123c1420e40093dc61650dc1b86a Author: Mooneer Salem <mooneer@gmail.com> Date: Sun Oct 11 21:42:23 2026 +0000 latest user manual PDF commit b9aabc29c390ef388cf7dd87d71bc31169183279 Author: Mooneer Salem <mooneer@gmail.com> Date: Sun Oct 11 14:39:11 2026 -0700 Add unit tests for non-GUI code and fix the defects they found (#1542) * Add unit tests for non-GUI code and fix the defects they found Adds a DefineFreeDVUnitTest CMake macro and 18 CTest unit test programs covering the pipeline steps, configuration storage and migration, WSJT-X logging, Hamlib rig control, the voice keyer file cache, tab layout serialization and equalizer filter design. Fixes found while writing them: * Equalizer bass/treble filters were always designed for 8 kHz rather than the device's sample rate. * Frequency list entries could be saved 1 Hz low due to truncation. * WSJT-X doubles were sent in host rather than network byte order. * Text settings resembling environment variables (e.g. $HOME) were expanded when loaded. * Hamlib: a frequency set before connecting was dropped, and stale frequency/mode was reapplied when reconnecting the same controller. * OmniRig: frequency/mode set before connecting was dropped. * "Play File - From Radio" accepted stereo files, which played at half speed; these are now rejected. * Beeps shorter than 10 ms did not fade out. The equalizer filter design moves out of MainFrame into eq_design.cpp so that it can be tested, and WSJTXNetworkLogger takes an optional heartbeat interval for the same reason. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WqtgkshQCiTgcMW4Tiu4jG * Add release notes for PR #1542 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WqtgkshQCiTgcMW4Tiu4jG * Fix wording flagged by codespell Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WqtgkshQCiTgcMW4Tiu4jG * Fix problems found by CI and by building on Linux Production code: * ParallelStep: wait for each worker thread to exit before destroying its semaphore. Previously the thread could go on to wait on a semaphore that had already been released, which either crashed or hung on shutdown. * HamlibRigController: make destroying_ and rigResponseTime_ atomic as both are accessed from more than one thread. * sox_biquad: avoid left shifting negative samples (undefined behavior). Tests and build: * Don't rely on wxString::utf8_string() or a stream operator for wxString, neither of which exist in wxWidgets 3.0. * Link the bundled Hamlib and wxWidgets into the tests when those are built as part of FreeDV, and set the build RPATH for all tests so that codec2/RADE can be found. * Accept the VFO names reported by older versions of Hamlib. * Construct objects that own threads in memory that isn't reused, and skip the voice keyer cache tests under thread sanitizer, to avoid false positives caused by threads that thread sanitizer can't tell have finished. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WqtgkshQCiTgcMW4Tiu4jG * Free memory retained by unit tests on exit LeakSanitizer otherwise reports it as leaked. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WqtgkshQCiTgcMW4Tiu4jG * Keep RecordStepTest clean under RealtimeSanitizer on macOS RecordStep::execute() logs when its queue fills up. On macOS the first log call on a thread allocates that thread's local storage, which RealtimeSanitizer rejects inside a real-time function. Log once from the test's main thread before any test case runs. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WqtgkshQCiTgcMW4Tiu4jG * Add additional missed coverage. * Extend unit test coverage of rig control and fix the defects found Tests: * HamlibRigController: error paths, mode changes while transmitting and the connection settings passed to Hamlib. * New SerialPortRigControllerTest covering the serial port PTT controllers using a pseudo-terminal. * New UtilTest for the non-GUI helpers in util.cpp. * Additional voice keyer cache and PlaybackStep cases. * XDG log path tests only run on Linux, with an equivalent check for the Documents folder elsewhere. * Remove the 800XA transmit test, as that mode isn't supported. Fixes: * Serial port PTT input: disconnecting when the port was never opened joined a thread that was never started, and the destructor cleared event handlers and joined the polling thread while the controller's thread could still be using them. * Serial port controllers: make fields shared between threads atomic. * Hamlib: reapply the 625 ms timeout after rig_open(), which replaces it with the radio's default in some versions of Hamlib. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WqtgkshQCiTgcMW4Tiu4jG * Add TxRxThreadTest. * Cover FreeDVInterface and the rest of TxRxThread with unit tests Add FreeDVInterfaceTest, which exercises the legacy modes (mode names, frame sizes, test frame statistics, typed and reliable text, multiple mode receive, squelch) and RADE (loopback that syncs and delivers the callsign in the end of over). Extend TxRxThreadTest with transmit and voice keyer monitoring, looping playback, playing and recording on the radio side, queued resync, the time-out beep and end of over handling in both legacy modes and RADE. To make RADE startable without the application object, FreeDVInterface::start() now takes the verbose setting as an argument rather than reading it from wxGetApp(). Fixes found along the way: * voicekeyer_cache: don't call memcpy() with a null source when reading from an empty in-memory file (reported by UBSAN in CI). * TxRxThreadTest: make the fake real-time helper's counters atomic, as they are read while the thread under test updates them (reported by ThreadSanitizer). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WqtgkshQCiTgcMW4Tiu4jG * Move grid square math and serial port listing out of dialogs and test them The grid square, distance, bearing and compass point calculations used by the FreeDV Reporter window now live in gui/util/GridSquare, and the band lookup sits beside the FilterFrequency enum. The serial port enumeration, device filtering and numeric-aware sorting that the PTT and Easy Setup windows each had an identical copy of now live in gui/util/SerialPortList. Add GridSquareTest, FilterFrequencyTest and SerialPortListTest. Fix found along the way: the user's own grid square was only checked for being non-empty, so an incomplete one (e.g. "FN") was read past its end and produced meaningless distances and headings. Both grid squares are now validated the same way. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WqtgkshQCiTgcMW4Tiu4jG * Test the remaining switchable stages of the transmit and receive pipelines TxRxThreadTest now covers the tone interferer, recording of microphone audio and of what's sent to the radio and to the speaker, automatic gain control, noise reduction, the transmit frequency offset and stopping early when the real-time helper says time is up. FreeDVInterfaceTest now covers transmit and receive frequency offsets passed through the interface, and RADE bandwidth expansion. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WqtgkshQCiTgcMW4Tiu4jG * Build the unit tests for Windows and run them in CI The Windows workflow gains two jobs: build-unit-tests cross-compiles the unit tests (plain, UBSAN and, on x86_64, ASAN) and gathers them with the DLLs they need, and unit-tests runs them on Windows itself. * A "unit_tests" target builds every unit test and only what they need. * A "collect_unit_tests" target (Windows builds) copies the test executables and the DLLs they import that aren't part of Windows into one folder (cmake/CollectUnitTests.cmake.in). * test/RunUnitTests.ps1 runs every *Test.exe in a folder and reports which failed or took too long. Changes needed for the tests to build and pass on Windows: * Temporary files go through a new unitTestTempPath() helper instead of assuming /tmp and getpid(). * VoiceKeyerFileCacheTest uses std::filesystem to resize files and to change their permissions and modification times; the unreadable file case is skipped on Windows. * HamlibRigControllerTest allows for the device prefix that Hamlib puts in front of port names on Windows. * M_PI is defined by the shared test header if the compiler doesn't, and TxRxThreadTest includes wxWidgets ahead of anything that may include windows.h. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WqtgkshQCiTgcMW4Tiu4jG * Fix collecting the Windows unit tests with CMake 3.31 The collection script runs on its own (cmake -P), so no policies are set for it unless it asks for a minimum version, and without them CMake 3.31 doesn't recognise IN_LIST in if(). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WqtgkshQCiTgcMW4Tiu4jG * Give unit tests that use wxWidgets a manifest on Windows wxWidgets' GUI library imports functions that only version 6 of the common controls library exports by name, and Windows only gives that version to programs whose manifest asks for it. Without one, the six tests that link it failed to start on Windows ("entry point not found"). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WqtgkshQCiTgcMW4Tiu4jG --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> commit 82c2e51ca0760f02cac5e9ac5e570bee375d0761 Author: Mooneer Salem <mooneer@gmail.com> Date: Sun Oct 11 13:38:27 2026 -0700 Give unit tests that use wxWidgets a manifest on Windows wxWidgets' GUI library imports functions that only version 6 of the common controls library exports by name, and Windows only gives that version to programs whose manifest asks for it. Without one, the six tests that link it failed to start on Windows ("entry point not found"). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WqtgkshQCiTgcMW4Tiu4jG commit f730fb17c35492d3fad15a993402af51c5f6f33c Author: Mooneer Salem <mooneer@gmail.com> Date: Sun Oct 11 13:17:39 2026 -0700 Fix collecting the Windows unit tests with CMake 3.31 The collection script runs on its own (cmake -P), so no policies are set for it unless it asks for a minimum version, and without them CMake 3.31 doesn't recognise IN_LIST in if(). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WqtgkshQCiTgcMW4Tiu4jG commit 61ae6ecc91ec2a2ae1a7ddf05ee463751136c9bc Author: Mooneer Salem <mooneer@gmail.com> Date: Sun Oct 11 12:58:47 2026 -0700 Build the unit tests for Windows and run them in CI The Windows workflow gains two jobs: build-unit-tests cross-compiles the unit tests (plain, UBSAN and, on x86_64, ASAN) and gathers them with the DLLs they need, and unit-tests runs them on Windows itself. * A "unit_tests" target builds every unit test and only what they need. * A "collect_unit_tests" target (Windows builds) copies the test executables and the DLLs they import that aren't part of Windows into one folder (cmake/CollectUnitTests.cmake.in). * test/RunUnitTests.ps1 runs every *Test.exe in a folder and reports which failed or took too long. Changes needed for the tests to build and pass on Windows: * Temporary files go through a new unitTestTempPath() helper instead of assuming /tmp and getpid(). * VoiceKeyerFileCacheTest uses std::filesystem to resize files and to change their permissions and modification times; the unreadable file case is skipped on Windows. * HamlibRigControllerTest allows for the device prefix that Hamlib puts in front of port names on Windows. * M_PI is defined by the shared test header if the compiler doesn't, and TxRxThreadTest includes wxWidgets ahead of anything that may include windows.h. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WqtgkshQCiTgcMW4Tiu4jG commit 1b851eb59def6e3e02e8896fd239256b92456ebf Author: Mooneer Salem <mooneer@gmail.com> Date: Sun Oct 11 11:38:29 2026 -0700 Test the remaining switchable stages of the transmit and receive pipelines TxRxThreadTest now covers the tone interferer, recording of microphone audio and of what's sent to the radio and to the speaker, automatic gain control, noise reduction, the transmit frequency offset and stopping early when the real-time helper says time is up. FreeDVInterfaceTest now covers transmit and receive frequency offsets passed through the interface, and RADE bandwidth expansion. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WqtgkshQCiTgcMW4Tiu4jG commit 4e85e6e5df63d86f498b6e8717471b14ef9ca596 Author: Mooneer Salem <mooneer@gmail.com> Date: Sun Oct 11 10:15:18 2026 -0700 Move grid square math and serial port listing out of dialogs and test them The grid square, distance, bearing and compass point calculations used by the FreeDV Reporter window now live in gui/util/GridSquare, and the band lookup sits beside the FilterFrequency enum. The serial port enumeration, device filtering and numeric-aware sorting that the PTT and Easy Setup windows each had an identical copy of now live in gui/util/SerialPortList. Add GridSquareTest, FilterFrequencyTest and SerialPortListTest. Fix found along the way: the user's own grid square was only checked for being non-empty, so an incomplete one (e.g. "FN") was read past its end and produced meaningless distances and headings. Both grid squares are now validated the same way. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WqtgkshQCiTgcMW4Tiu4jG commit 2f234beebbbd9d75d2147892dbfd9860b7f03562 Author: Mooneer Salem <mooneer@gmail.com> Date: Sun Oct 11 09:19:16 2026 -0700 Cover FreeDVInterface and the rest of TxRxThread with unit tests Add FreeDVInterfaceTest, which exercises the legacy modes (mode names, frame sizes, test frame statistics, typed and reliable text, multiple mode receive, squelch) and RADE (loopback that syncs and delivers the callsign in the end of over). Extend TxRxThreadTest with transmit and voice keyer monitoring, looping playback, playing and recording on the radio side, queued resync, the time-out beep and end of over handling in both legacy modes and RADE. To make RADE startable without the application object, FreeDVInterface::start() now takes the verbose setting as an argument rather than reading it from wxGetApp(). Fixes found along the way: * voicekeyer_cache: don't call memcpy() with a null source when reading from an empty in-memory file (reported by UBSAN in CI). * TxRxThreadTest: make the fake real-time helper's counters atomic, as they are read while the thread under test updates them (reported by ThreadSanitizer). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WqtgkshQCiTgcMW4Tiu4jG commit 14759de03cc6fda6091aadfadeadcfe8f60330ed Merge: 00be01f0 94d06040 Author: Mooneer Salem <mooneer@gmail.com> Date: Sun Oct 11 08:02:30 2026 -0700 Merge branch 'ms-unit-tests' of github.com:drowe67/freedv-gui into ms-unit-tests commit 00be01f0209f702b2052edd0953490f1e7f5659e Author: Mooneer Salem <mooneer@gmail.com> Date: Sun Oct 11 08:02:20 2026 -0700 Add TxRxThreadTest. commit 94d060409802a43bf96fdb045814e77b75b9d03d Author: Mooneer Salem <mooneer@gmail.com> Date: Sun Oct 11 00:32:57 2026 -0700 Extend unit test coverage of rig control and fix the defects found Tests: * HamlibRigController: error paths, mode changes while transmitting and the connection settings passed to Hamlib. * New SerialPortRigControllerTest covering the serial port PTT controllers using a pseudo-terminal. * New UtilTest for the non-GUI helpers in util.cpp. * Additional voice keyer cache and PlaybackStep cases. * XDG log path tests only run on Linux, with an equivalent check for the Documents folder elsewhere. * Remove the 800XA transmit test, as that mode isn't supported. Fixes: * Serial port PTT input: disconnecting when the port was never opened joined a thread that was never started, and the destructor cleared event handlers and joined the polling thread while the controller's thread could still be using them. * Serial port controllers: make fields shared between threads atomic. * Hamlib: reapply the 625 ms timeout after rig_open(), which replaces it with the radio's default in some versions of Hamlib. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WqtgkshQCiTgcMW4Tiu4jG commit 51cd11305985f5cd209cf9b94a8f5003ec9b8eb5 Merge: 00bfb74f 7eee29ab Author: Mooneer Salem <mooneer@gmail.com> Date: Sat Oct 10 22:59:16 2026 -0700 Merge branch 'ms-unit-tests' of github.com:drowe67/freedv-gui into ms-unit-tests commit 00bfb74fb079d6880ca2ecdc0303d51c5986dcf2 Author: Mooneer Salem <mooneer@gmail.com> Date: Sat Oct 10 22:58:59 2026 -0700 Add additional missed coverage. commit 7eee29ab6f6df2ad01d0df1704f9456f13928b6c Author: Mooneer Salem <mooneer@gmail.com> Date: Sat Oct 10 17:06:33 2026 -0700 Keep RecordStepTest clean under RealtimeSanitizer on macOS RecordStep::execute() logs when its queue fills up. On macOS the first log call on a thread allocates that thread's local storage, which RealtimeSanitizer rejects inside a real-time function. Log once from the test's main thread before any test case runs. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WqtgkshQCiTgcMW4Tiu4jG commit 5ac8eb7457001356bc468bbfe1d4d2bbb16d355c Author: Mooneer Salem <mooneer@gmail.com> Date: Sat Oct 10 13:24:35 2026 -0700 Free memory retained by unit tests on exit LeakSanitizer otherwise reports it as leaked. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WqtgkshQCiTgcMW4Tiu4jG commit ae12f51376a9d9c610db6e7b04939cff217d9fc3 Author: Mooneer Salem <mooneer@gmail.com> Date: Sat Oct 10 12:47:51 2026 -0700 Fix problems found by CI and by building on Linux Production code: * ParallelStep: wait for each worker thread to exit before destroying its semaphore. Previously the thread could go on to wait on a semaphore that had already been released, which either crashed or hung on shutdown. * HamlibRigController: make destroying_ and rigResponseTime_ atomic as both are accessed from more than one thread. * sox_biquad: avoid left shifting negative samples (undefined behavior). Tests and build: * Don't rely on wxString::utf8_string() or a stream operator for wxString, neither of which exist in wxWidgets 3.0. * Link the bundled Hamlib and wxWidgets into the tests when those are built as part of FreeDV, and set the build RPATH for all tests so that codec2/RADE can be found. * Accept the VFO names reported by older versions of Hamlib. * Construct objects that own threads in memory that isn't reused, and skip the voice keyer cache tests under thread sanitizer, to avoid false positives caused by threads that thread sanitizer can't tell have finished. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WqtgkshQCiTgcMW4Tiu4jG commit f54e6c71faba86b7e1c12703a704914226e38e1b Author: Mooneer Salem <mooneer@gmail.com> Date: Sat Oct 10 11:01:03 2026 -0700 Fix wording flagged by codespell Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WqtgkshQCiTgcMW4Tiu4jG commit d4186fd606ea1207af3ca36d5662a52a173e15b0 Author: Mooneer Salem <mooneer@gmail.com> Date: Sat Oct 10 10:38:01 2026 -0700 Add release notes for PR #1542 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WqtgkshQCiTgcMW4Tiu4jG commit 23f09beec9a89292b5aae7a431cbd9000a6d5951 Author: Mooneer Salem <mooneer@gmail.com> Date: Sat Oct 10 10:37:30 2026 -0700 Add unit tests for non-GUI code and fix the defects they found Adds a DefineFreeDVUnitTest CMake macro and 18 CTest unit test programs covering the pipeline steps, configuration storage and migration, WSJT-X logging, Hamlib rig control, the voice keyer file cache, tab layout serialization and equalizer filter design. Fixes found while writing them: * Equalizer bass/treble filters were always designed for 8 kHz rather than the device's sample rate. * Frequency list entries could be saved 1 Hz low due to truncation. * WSJT-X doubles were sent in host rather than network byte order. * Text settings resembling environment variables (e.g. $HOME) were expanded when loaded. * Hamlib: a frequency set before connecting was dropped, and stale frequency/mode was reapplied when reconnecting the same controller. * OmniRig: frequency/mode set before connecting was dropped. * "Play File - From Radio" accepted stereo files, which played at half speed; these are now rejected. * Beeps shorter than 10 ms did not fade out. The equalizer filter design moves out of MainFrame into eq_design.cpp so that it can be tested, and WSJTXNetworkLogger takes an optional heartbeat interval for the same reason. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WqtgkshQCiTgcMW4Tiu4jG commit e9f1e4efa58a2ac33e7b26f585a2cc84a2fd0ff9 Author: Mooneer Salem <mooneer@gmail.com> Date: Sat Oct 10 07:01:38 2026 +0000 latest user manual PDF commit 9e9b6b6f896015bd859f07a0d8c7fbb87faa8ac7 Author: Mooneer Salem <mooneer@gmail.com> Date: Fri Oct 9 23:53:19 2026 -0700 Setup Wizard: support importing settings from WS (formerly WSJT-X Improved) (#1539) * Setup Wizard: support importing settings from WS (formerly WSJT-X Improved). WS stores its settings in WS.ini rather than WSJT-X.ini (e.g. %LOCALAPPDATA%\WS\WS.ini on Windows) but otherwise uses the same keys. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011sR6sG6MEefCKp5bKJjWE2 * Add PR number to release notes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011sR6sG6MEefCKp5bKJjWE2 --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> commit a8d84248837561b618141d0e82d072533d6df127 Author: Mooneer Salem <mooneer@gmail.com> Date: Sat Oct 10 06:38:04 2026 +0000 latest user manual PDF commit 6210a1d343f1a730695e3894c6c96b606bb60c96 Author: Mooneer Salem <mooneer@gmail.com> Date: Fri Oct 9 23:35:22 2026 -0700 Ignore SIGPIPE; test against freedv-backend #70 (#1540) * Ignore SIGPIPE so a closed connection can't kill FreeDV. Writing to a socket whose peer has closed raises SIGPIPE, whose default action terminates the process. Neither freedv-gui nor freedv-backend ignored it or used MSG_NOSIGNAL/SO_NOSIGPIPE, so if a server (e.g. FreeDV Reporter) closed the connection just as we sent to it, FreeDV could exit silently. TLS library writes can't be flagged per call on Linux, so ignore SIGPIPE process-wide at startup; writes then fail with EPIPE and the connection code treats that as a disconnect. Not applicable on Windows. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011sR6sG6MEefCKp5bKJjWE2 * TEMPORARY: build against freedv-backend PR #70 (reporting-unit-tests). Verifies freedv-gui still builds and passes CI with the freedv-backend changes in https://github.com/tmiw/freedv-backend/pull/70. Revert to GIT_TAG main once that PR is merged, before merging this one. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011sR6sG6MEefCKp5bKJjWE2 * Add PR #1540 to changelog. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011sR6sG6MEefCKp5bKJjWE2 * Revert freedv-backend to main. --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> commit 4d9331ed53b33fed7d56e3c062bc1abc03724e00 Author: Mooneer Salem <mooneer@gmail.com> Date: Fri Oct 9 22:34:30 2026 -0700 Revert freedv-backend to main. commit e1a4da856ab4a27c194b4040e6be827a672bb781 Author: Mooneer Salem <mooneer@gmail.com> Date: Fri Oct 9 00:35:23 2026 -0700 Add PR #1540 to changelog. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011sR6sG6MEefCKp5bKJjWE2 commit 4f485bb669c2d97ba3b58462bdb707aca921db6e Author: Mooneer Salem <mooneer@gmail.com> Date: Fri Oct 9 00:34:49 2026 -0700 TEMPORARY: build against freedv-backend PR #70 (reporting-unit-tests). Verifies freedv-gui still builds and passes CI with the freedv-backend changes in https://github.com/tmiw/freedv-backend/pull/70. Revert to GIT_TAG main once that PR is merged, before merging this one. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011sR6sG6MEefCKp5bKJjWE2 commit 668874fa21194cb027ace861c8a1fdd4e7c023ea Author: Mooneer Salem <mooneer@gmail.com> Date: Fri Oct 9 00:34:49 2026 -0700 Ignore SIGPIPE so a closed connection can't kill FreeDV. Writing to a socket whose peer has closed raises SIGPIPE, whose default action terminates the process. Neither freedv-gui nor freedv-backend ignored it or used MSG_NOSIGNAL/SO_NOSIGPIPE, so if a server (e.g. FreeDV Reporter) closed the connection just as we sent to it, FreeDV could exit silently. TLS library writes can't be flagged per call on Linux, so ignore SIGPIPE process-wide at startup; writes then fail with EPIPE and the connection code treats that as a disconnect. Not applicable on Windows. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011sR6sG6MEefCKp5bKJjWE2 commit 5c92470bde059f6d247b0d62c538f1a1e819199b Author: Mooneer Salem <mooneer@gmail.com> Date: Thu Oct 8 16:48:02 2026 -0700 Add PR number to release notes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011sR6sG6MEefCKp5bKJjWE2 commit 50c047a9348adcf4eb0198630f9a6f1b11edfc4f Author: Mooneer Salem <mooneer@gmail.com> Date: Thu Oct 8 16:47:52 2026 -0700 Setup Wizard: support importing settings from WS (formerly WSJT-X Improved). WS stores its settings in WS.ini rather than WSJT-X.ini (e.g. %LOCALAPPDATA%\WS\WS.ini on Windows) but otherwise uses the same keys. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011sR6sG6MEefCKp5bKJjWE2 commit 75a1d8315a2dac99b68b617405d59c08579ad7b1 Author: Mooneer Salem <mooneer@gmail.com> Date: Thu Oct 8 23:30:12 2026 +0000 latest user manual PDF commit 0e877318a53fb525ae493f2d3802a0ec05a1d086 Author: Mooneer Salem <mooneer@gmail.com> Date: Thu Oct 8 16:27:52 2026 -0700 Add infrastructure for translating FreeDV into other languages (#1526) * Add infrastructure for translating FreeDV into other languages - Add po/ directory with gettext build support: update-pot and update-po targets, compilation of languages listed in po/LINGUAS, and installation into the macOS app bundle or share/locale (Linux/Windows/AppImage). wxWidgets' own catalogs are included when bootstrapping wxWidgets. - Load the freedv catalog at startup based on the system language. - Mark user-visible strings for translation, convert concatenated messages to format strings, and unmark config keys/defaults and other strings that must not be translated. - Compare PTT method by selection index rather than displayed text. - Fix PTT serial port error handlers capturing a local string by reference in CallAfter(). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GxwHpb9KR3ymB15RFPGtR6 * Suppress clang-analyzer false positive for wxTranslations allocation wxTranslations::Set() takes ownership of the object, but clang-analyzer reports it as a potential leak. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GxwHpb9KR3ymB15RFPGtR6 * Add missed Error translations. * Add PR #1526 to changelog. * Add initial pt_BR translation. * Add contact info, fix macOS compile failure. * Revise translation setup instructions in README (#1528) * Revise translation setup instructions in README A change to the deep command will generate a file that translates virtually the entire interface. * Update README with additional info. * Fix formatting. --------- Co-authored-by: Mooneer Salem <mooneer@gmail.com> * Windows: install the catalogs in the same folder as the .exe. * Update README to add gettext dependency. * Forgot to include gettext in CI. * Only force LTR layout if we don't have a translation for a RTL language available. * Make sure gettext is installed for macOS/Linux CI as well. * Fix LTR rendering override logic. * Remove unneeded main.h includes. --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> Co-authored-by: manotroll <manotroll@gmail.com> commit e495e9b73e5eaf9417eaec60ea3e7366f959a06f Author: Mooneer Salem <mooneer@gmail.com> Date: Thu Oct 8 14:52:31 2026 -0700 Remove unneeded main.h includes. commit 35ecd437d7194968d34d6798971329cca8cddf5b Author: Mooneer Salem <mooneer@gmail.com> Date: Thu Oct 8 14:49:27 2026 -0700 Fix LTR rendering override logic. commit a7ace481bcbeee8fdd5eab4ef3904be51016f336 Author: Mooneer Salem <mooneer@gmail.com> Date: Thu Oct 8 13:54:12 2026 -0700 Make sure gettext is installed for macOS/Linux CI as well. commit 44a11318c2bcbe33a942ad5a29affe6f681c7975 Author: Mooneer Salem <mooneer@gmail.com> Date: Thu Oct 8 13:52:27 2026 -0700 Only force LTR layout if we don't have a translation for a RTL language available. commit ae7a686dca88876b8845655f2fe0ccb68f2ba2f6 Author: Mooneer Salem <mooneer@gmail.com> Date: Thu Oct 8 12:48:56 2026 -0700 Forgot to include gettext in CI. commit ccf58a6693f0720cbdf7a01588f44e3ee42a9e5a Author: Mooneer Salem <mooneer@gmail.com> Date: Thu Oct 8 11:37:36 2026 -0700 Update README to add gettext dependency. commit 737a23c1be61d52b0b7283efd2116d2ba7eec1b7 Author: Mooneer Salem <mooneer@gmail.com> Date: Thu Oct 8 11:36:17 2026 -0700 Windows: install the catalogs in the same folder as the .exe. commit 7a1f821bc952ce92f147459d587291f7da26636d Author: Mooneer Salem <mooneer@gmail.com> Date: Thu Oct 8 01:20:05 2026 -0700 Add contact info, fix macOS compile failure. commit e73ff097adcfc0c882293857a6c0d0c661071ccc Author: Mooneer Salem <mooneer@gmail.com> Date: Wed Oct 7 18:22:34 2026 -0700 Add initial pt_BR translation. commit 91d4be8323e23982a8ad2a767cfeb151e5c9f3e2 Author: Mooneer Salem <mooneer@gmail.com> Date: Wed Oct 7 18:21:14 2026 -0700 Add PR #1526 to changelog. commit 08f0377fb6dbaee0ac50d439e36ba1aec064194a Merge: 166a04ad 15ebcfcf Author: Mooneer Salem <mooneer@gmail.com> Date: Wed Oct 7 18:20:13 2026 -0700 Merge branch 'v3.0-dev' into feature/i18n commit 15ebcfcf8fe651755bef66476c1d3d2471a8426c Merge: 52c0371d e04a28ee Author: Mooneer Salem <mooneer@gmail.com> Date: Wed Oct 7 17:14:25 2026 -0700 Merge branch 'master' into v3.0-dev commit feda8f276374101a2acb3ebf6a1dfeac1f5bad81 Author: Mooneer Salem <mooneer@gmail.com> Date: Thu Oct 8 00:09:03 2026 +0000 latest user manual PDF commit e04a28eef4a256b1a78f457502b693ba8bb0f166 Author: Mooneer Salem <mooneer@gmail.com> Date: Wed Oct 7 17:06:36 2026 -0700 Windows: add logic to detect audio device stalls. (#1537) * Windows: add logic to detect audio device stalls. * Add PR #1537 to changelog. commit 89fd7c4005a42fda4d513b24ec8962d3dcb5d574 Author: Mooneer Salem <mooneer@gmail.com> Date: Wed Oct 7 17:05:24 2026 -0700 Reduce full and incremental build times (#1538) * Reduce full and incremental build times. - Move the build date out of a global compile definition and into a generated build_date.cpp. The definition changed every compile command each day, forcing a full rebuild (wxWidgets included) after any reconfigure and defeating ccache across days. - Enable wxWidgets precompiled headers. - Download the wxWidgets release tarball instead of git cloning it, and apply our patch with patch(1). git apply silently skips every hunk when the sources have no .git of their own and sit inside another repo. - Stop building wx::xrc, which nothing links, and drop unused xrc includes. - Precompile <wx/wx.h> for FreeDV's GUI-facing targets. - Make ccache work with precompiled headers: set CCACHE_SLOPPINESS in CI and the build scripts, and pass -fno-pch-timestamp to Clang. - Fix CCACHE_ARGS in build_osx.sh, which was missing quotes, so ccache was never actually enabled for USE_CCACHE=1 builds. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011sR6sG6MEefCKp5bKJjWE2 * Disable precompiled headers in the Linux lint job. The lint build uses GCC, so compile_commands.json pointed clang-tidy at GCC .gch files, which it rejects (clang-diagnostic-ignored-gch, an error under WarningsAsErrors). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011sR6sG6MEefCKp5bKJjWE2 * [skip ci] Add PR #1538 to changelog. --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> commit a0efcb12f70b3e810a1fc2ddff5725780d92ecd2 Author: Mooneer Salem <mooneer@gmail.com> Date: Wed Oct 7 17:04:52 2026 -0700 [skip ci] Add PR #1538 to changelog. commit 1bfc91c9be00e0a772fa1bd17f3818b5d39c3516 Author: Mooneer Salem <mooneer@gmail.com> Date: Wed Oct 7 16:23:24 2026 -0700 Disable precompiled headers in the Linux lint job. The lint build uses GCC, so compile_commands.json pointed clang-tidy at GCC .gch files, which it rejects (clang-diagnostic-ignored-gch, an error under WarningsAsErrors). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011sR6sG6MEefCKp5bKJjWE2 commit 4945928a20e7ba35c8adb26b8a0ead5599ab8ed0 Author: Mooneer Salem <mooneer@gmail.com> Date: Wed Oct 7 15:16:04 2026 -0700 Reduce full and incremental build times. - Move the build date out of a global compile definition and into a generated build_date.cpp. The definition changed every compile command each day, forcing a full rebuild (wxWidgets included) after any reconfigure and defeating ccache across days. - Enable wxWidgets precompiled headers. - Download the wxWidgets release tarball instead of git cloning it, and apply our patch with patch(1). git apply silently skips every hunk when the sources have no .git of their own and sit inside another repo. - Stop building wx::xrc, which nothing links, and drop unused xrc includes. - Precompile <wx/wx.h> for FreeDV's GUI-facing targets. - Make ccache work with precompiled headers: set CCACHE_SLOPPINESS in CI and the build scripts, and pass -fno-pch-timestamp to Clang. - Fix CCACHE_ARGS in build_osx.sh, which was missing quotes, so ccache was never actually enabled for USE_CCACHE=1 builds. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011sR6sG6MEefCKp5bKJjWE2 commit b299e2bb339b19c9c8be9af58868f7b449c1de4b Author: Mooneer Salem <mooneer@gmail.com> Date: Wed Oct 7 21:45:42 2026 +0000 latest user manual PDF commit 51834251eecaacb4f5b495ac14d4a21248bdec29 Merge: c9b53c06 f88f58ca Author: Mooneer Salem <mooneer@gmail.com> Date: Wed Oct 7 14:42:28 2026 -0700 Merge branch 'master' into ms-wasapi-stall-detect commit f88f58ca424f3ad13e0c2c658a20102d7b70edd0 Author: Mooneer Salem <mooneer@gmail.com> Date: Wed Oct 7 14:41:53 2026 -0700 Keep RX muted after TX until the radio confirms PTT off (#1536) * Keep RX muted after TX until the radio confirms PTT off. When going from TX to RX, FreeDV switched back to RX (waterfall, spectrum and decoding) immediately before requesting PTT off from the radio. Since the radio keeps transmitting until it has processed that request (e.g. ~350ms for some network-connected radios), whatever the radio sends on its RX audio during that time (such as TX monitor audio) showed up on the waterfall right after returning to RX. In half duplex, keep RX muted after TX until the PTT controller reports that PTT is off (via onPttChange), plus the RX input device's latency so that audio captured while the radio was still transmitting is skipped. If the radio never confirms, resume RX after 2 seconds. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011sR6sG6MEefCKp5bKJjWE2 * Rename variable flagged by codespell. codespell reads "nowNs" as a misspelling of "knowns"/"nouns". Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011sR6sG6MEefCKp5bKJjWE2 * [skip ci] Add PR #1536 to changelog. --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> commit ea517f2db46b16f7766aba8695d1653f73436afd Author: Mooneer Salem <mooneer@gmail.com> Date: Wed Oct 7 18:23:16 2026 +0000 latest user manual PDF commit c2c4c086c52717622a97088e1f86041ba18f6b30 Merge: 2b5b1f62 1fb0a47a Author: Mooneer Salem <mooneer@gmail.com> Date: Wed Oct 7 11:20:53 2026 -0700 Merge branch 'master' into ms-rx-after-ptt-off commit c9b53c06fcb2633676fb363ce8a41a11ad0f2963 Merge: c5a0b829 1fb0a47a Author: Mooneer Salem <mooneer@gmail.com> Date: Wed Oct 7 11:19:52 2026 -0700 Merge branch 'master' into ms-wasapi-stall-detect commit 1fb0a47ab9bda8afc81957df35addae29ab7dbcd Author: Mooneer Salem <mooneer@gmail.com> Date: Wed Oct 7 11:19:27 2026 -0700 Fix FreeDV hanging when starting the Voice Keyer with a cloud-backed file (#1534) * Don't hold the callback data mutex during PlaybackStep file reads. PlaybackStep's file I/O thread held g_mutexProtectingCallbackData while reading and resampling up to several seconds of audio. The UI thread takes the same mutex on its regular timer ticks, so a slow read (or decode) could stall the UI. Replace this with a dedicated g_mutexProtectingPlayFiles that only protects the lifetime of the play file SNDFILE handles. PlaybackStep now holds it only across each sf_read_short() call, and the code that closes the play files holds it only around sf_close(). The loop completion function (which calls sf_seek()) is now also called with the lock held, fixing an unlocked race with the file being closed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011sR6sG6MEefCKp5bKJjWE2 * Cache the voice keyer file in memory to avoid delaying TX. VoiceKeyerStartTx() opened the voice keyer file with sf_open() on the UI thread right before keying up. If the file lives on cloud-backed storage (e.g. an iCloud-synced ~/Documents with "Optimize Mac Storage" enabled) and has been evicted locally, that open blocks while the OS downloads the file again, freezing FreeDV and delaying TX by 1-1.5s or more. Add VoiceKeyerFileCache, which reads the whole file into memory on a background thread at startup, when a new file is chosen and after a new one is recorded. VoiceKeyerStartTx() now opens the cached copy via sf_open_virtual(). The cached copy is used only if the file's size and modification time (a metadata-only check that doesn't trigger a download) still match; otherwise, or for files over 64 MB, we fall back to opening the file directly and reload the cache in the background. Also close any play file left over from a prior playback before starting a new one rather than leaking it, since its memory reader must outlive it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011sR6sG6MEefCKp5bKJjWE2 * Fix clang-tidy performance-unnecessary-value-param in VoiceKeyerFileCache. load_() only reads its parameters, so take them by const reference. std::thread keeps its own copies of the arguments for the lifetime of the thread, so the references remain valid while load_() runs. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011sR6sG6MEefCKp5bKJjWE2 * [skip ci] Add PR #1534 to changelog. --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> commit 2b5b1f62907ae0d03afd7b5f34c77aebe2a20bb2 Merge: df7d7076 5544169d Author: Mooneer Salem <mooneer@gmail.com> Date: Wed Oct 7 09:18:21 2026 -0700 Merge branch 'master' into ms-rx-after-ptt-off commit df7d7076047bf70cf311ef036f859a33ca33ca74 Author: Mooneer Salem <mooneer@gmail.com> Date: Wed Oct 7 09:15:50 2026 -0700 [skip ci] Add PR #1536 to changelog. commit 36defda89d87536146164e885af07834ae8fe9d4 Merge: 55bfa763 5544169d Author: Mooneer Salem <mooneer@gmail.com> Date: Wed Oct 7 08:40:58 2026 -0700 Merge branch 'master' into ms-vk-file-cache commit 55bfa7637e5712ca38cce95daba0660d5120dd42 Author: Mooneer Salem <mooneer@gmail.com> Date: Wed Oct 7 08:40:27 2026 -0700 [skip ci] Add PR #1534 to changelog. commit c5a0b829b5a4119e528da32038261abf63a94268 Merge: 27e1bb22 5544169d Author: Mooneer Salem <mooneer@gmail.com> Date: Wed Oct 7 08:34:57 2026 -0700 Merge branch 'master' into ms-wasapi-stall-detect commit 27e1bb228ad086091d248d68291d894f4bbce6a3 Author: Mooneer Salem <mooneer@gmail.com> Date: Wed Oct 7 00:28:59 2026 -0700 [skip ci] Add PR #1537 to changelog. commit 5544169da27566a8e2ca6f118fb01cc11363df06 Author: Mooneer Salem <mooneer@gmail.com> Date: Wed Oct 7 00:23:45 2026 -0700 Fix data races on settings read by the audio threads (#1535) * Fix data race on the TX noise suppression setting. The RNNoise step's predicate in the TX pipeline read filterConfiguration.noiseReductionEnable directly on the TX thread while the Filter dialog wrote it on the UI thread (in OnNoiseReductionEnable() and in ExchangeData(), which rewrites every filter setting on any change). ConfigurationDataElement has no synchronization, so ThreadSanitizer reported a data race whenever filter settings were changed during TX. Mirror the setting into a std::atomic<bool> (g_noiseReductionEnabled) like the AGC setting already does with g_agcEnabled, and have the pipeline read that instead. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011sR6sG6MEefCKp5bKJjWE2 * Fix data races on settings read by the audio threads. ThreadSanitizer reported data races between the UI thread changing the following settings and the RX/TX threads reading them every audio block: * TX and voice keyer audio monitoring (enable + volume), changed from the PTT and Voice Keyer button context menus. * Channel noise SNR, written by Tools->Options OK/Apply (which also reloads the whole configuration) and read when using legacy modes. * Simulated interference tone enable, frequency and amplitude, written by Tools->Options OK/Apply. ConfigurationDataElement has no synchronization, and the tone settings were plain members of the app object. Mirror the configuration values into atomics (like g_agcEnabled and g_noiseReductionEnabled) that are refreshed in loadConfiguration_() and wherever the UI changes them, and make the tone settings atomic, so that the audio threads no longer read non-atomic values that the UI thread may be writing. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011sR6sG6MEefCKp5bKJjWE2 * [skip ci] Add PR #1535 to changelog. --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> commit 36614ede587e777dfcc08409b1d4672a4dca5254 Author: Mooneer Salem <mooneer@gmail.com> Date: Wed Oct 7 00:23:27 2026 -0700 [skip ci] Add PR #1535 to changelog. commit 57c1b76a8355b23afc3ef92856ba20c73131cd79 Author: Mooneer Salem <mooneer@gmail.com> Date: Tue Oct 6 22:37:14 2026 -0700 Windows: add logic to detect audio device stalls. commit f5ff14912c9b7de8d4c8e5c242fe5f9d037eff4d Author: Mooneer Salem <mooneer@gmail.com> Date: Tue Oct 6 21:57:38 2026 -0700 Rename variable flagged by codespell. codespell reads "nowNs" as a misspelling of "knowns"/"nouns". Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011sR6sG6MEefCKp5bKJjWE2 commit 0223cf45c65ed681bad3d62405f67d2b45fbd630 Author: Mooneer Salem <mooneer@gmail.com> Date: Tue Oct 6 21:55:01 2026 -0700 Fix clang-tidy performance-unnecessary-value-param in VoiceKeyerFileCache. load_() only reads its parameters, so take them by const reference. std::thread keeps its own copies of the arguments for the lifetime of the thread, so the references remain valid while load_() runs. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011sR6sG6MEefCKp5bKJjWE2 commit 03bef8bd6aa77c907624bd8b0e9bf46f10cb3cce Author: Mooneer Salem <mooneer@gmail.com> Date: Tue Oct 6 21:45:56 2026 -0700 Keep RX muted after TX until the radio confirms PTT off. When going from TX to RX, FreeDV switched back to RX (waterfall, spectrum and decoding) immediately before requesting PTT off from the radio. Since the radio keeps transmitting until it has processed that request (e.g. ~350ms for some network-connected radios), whatever the radio sends on its RX audio during that time (such as TX monitor audio) showed up on the waterfall right after returning to RX. In half duplex, keep RX muted after TX until the PTT controller reports that PTT is off (via onPttChange), plus the RX input device's latency so that audio captured while the radio was still transmitting is skipped. If the radio never confirms, resume RX after 2 seconds. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011sR6sG6MEefCKp5bKJjWE2 commit 33c5ff81abd396a9d4ac70f9d0773cf823ecec8a Author: Mooneer Salem <mooneer@gmail.com> Date: Tue Oct 6 21:13:28 2026 -0700 Fix data races on settings read by the audio threads. ThreadSanitizer reported data races between the UI thread changing the following settings and the RX/TX threads reading them every audio block: * TX and voice keyer audio monitoring (enable + volume), changed from the PTT and Voice Keyer button context menus. * Channel noise SNR, written by Tools->Options OK/Apply (which also reloads the whole configuration) and read when using legacy modes. * Simulated interference tone enable, frequency and amplitude, written by Tools->Options OK/Apply. ConfigurationDataElement has no synchronization, and the tone settings were plain members of the app object. Mirror the configuration values into atomics (like g_agcEnabled and g_noiseReductionEnabled) that are refreshed in loadConfiguration_() and wherever the UI changes them, and make the tone settings atomic, so that the audio threads no longer read non-atomic values that the UI thread may be writing. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011sR6sG6MEefCKp5bKJjWE2 commit d7d64a467e90433dd81bb612076c4c1348b90bc7 Author: Mooneer Salem <mooneer@gmail.com> Date: Tue Oct 6 20:18:56 2026 -0700 Fix data race on the TX noise suppression setting. The RNNoise step's predicate in the TX pipeline read filterConfiguration.noiseReductionEnable directly on the TX thread while the Filter dialog wrote it on the UI thread (in OnNoiseReductionEnable() and in ExchangeData(), which rewrites every filter setting on any change). ConfigurationDataElement has no synchronization, so ThreadSanitizer reported a data race whenever filter settings were changed during TX. Mirror the setting into a std::atomic<bool> (g_noiseReductionEnabled) like the AGC setting already does with g_agcEnabled, and have the pipeline read that instead. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011sR6sG6MEefCKp5bKJjWE2 commit 214f18ef18b21f9dc71bb09ed71fdef58967c711 Author: Mooneer Salem <mooneer@gmail.com> Date: Tue Oct 6 19:57:28 2026 -0700 Cache the voice keyer file in memory to avoid delaying TX. VoiceKeyerStartTx() opened the voice keyer file with sf_open() on the UI thread right before keying up. If the file lives on cloud-backed storage (e.g. an iCloud-synced ~/Documents with "Optimize Mac Storage" enabled) and has been evicted locally, that open blocks while the OS downloads the file again, freezing FreeDV and delaying TX by 1-1.5s or more. Add VoiceKeyerFileCache, which reads the whole file into memory on a background thread at startup, when a new file is chosen and after a new one is recorded. VoiceKeyerStartTx() now opens the cached copy via sf_open_virtual(). The cached copy is used only if the file's size and modification time (a metadata-only check that doesn't trigger a download) still match; otherwise, or for files over 64 MB, we fall back to opening the file directly and reload the cache in the background. Also close any play file left over from a prior playback before starting a new one rather than leaking it, since its memory reader must outlive it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011sR6sG6MEefCKp5bKJjWE2 commit e1b4787ec534a10449d3eac2105431a089ce9c49 Author: Mooneer Salem <mooneer@gmail.com> Date: Tue Oct 6 19:56:54 2026 -0700 Don't hold the callback data mutex during PlaybackStep file reads. PlaybackStep's file I/O thread held g_mutexProtectingCallbackData while reading and resampling up to several seconds of audio. The UI thread takes the same mutex on its regular timer ticks, so a slow read (or decode) could stall the UI. Replace this with a dedicated g_mutexProtectingPlayFiles that only protects the lifetime of the play file SNDFILE handles. PlaybackStep now holds it only across each sf_read_short() call, and the code that closes the play files holds it only around sf_close(). The loop completion function (which calls sf_seek()) is now also called with the lock held, fixing an unlocked race with the file being closed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011sR6sG6MEefCKp5bKJjWE2 commit e31e4aafdbc20460a8c45859b3c16b2628314c07 Author: Mooneer Salem <mooneer@gmail.com> Date: Wed Oct 7 01:23:19 2026 +0000 latest user manual PDF commit 9323ba3691e826b553763ce477afa571c6a8f547 Merge: 6c19e58b 52c0371d Author: Mooneer Salem <mooneer@gmail.com> Date: Tue Oct 6 18:19:53 2026 -0700 Merge branch 'v3.0-dev' into ms-rade-v2 commit 52c0371d48e69bd9708b7319160fdbe3168d8c88 Merge: 092db4bb 5024ef6e Author: Mooneer Salem <mooneer@gmail.com> Date: Tue Oct 6 18:19:25 2026 -0700 Merge branch 'master' into v3.0-dev commit 5024ef6ef0ef6a40863a9f60897595ce414eff3d Author: Mooneer Salem <mooneer@gmail.com> Date: Tue Oct 6 18:14:31 2026 -0700 Fix possible deadlock in PlaybackStep when audio playback completes. (#1533) * Fix possible deadlock in PlaybackStep when audio playback completes. * Add PR #1533 to changelog. commit 2acc527b00c52eb3220973929efe7b2df89854b5 Author: Mooneer Salem <mooneer@gmail.com> Date: Tue Oct 6 18:14:18 2026 -0700 Add PR #1533 to changelog. commit 00b95adb42d7a66c6e47e73abe769274fd2d70d3 Author: Mooneer Salem <mooneer@gmail.com> Date: Tue Oct 6 16:55:29 2026 -0700 Fix possible deadlock in PlaybackStep when audio playback completes. commit 6c19e58b3378fe7cb01e48c10f061e60748de5af Author: Mooneer Salem <mooneer@gmail.com> Date: Mon Oct 5 17:21:17 2026 -0700 Temporary change to test rade_c build fixes. commit 17a7c99393a63445f3a15ab29a3317df7ed51836 Author: Mooneer Salem <mooneer@gmail.com> Date: Mon Oct 5 09:32:26 2026 -0700 Remove uneeded text from README. commit 64dadba35fbff7f230bec5011055d7d732d7fae8 Merge: fde77500 092db4bb Author: Mooneer Salem <mooneer@gmail.com> Date: Sat Oct 3 11:28:43 2026 -0700 Merge branch 'v3.0-dev' into ms-rade-v2 commit 166a04ade47800496e54de20bebf926f20e11204 Author: Mooneer Salem <mooneer@gmail.com> Date: Sat Oct 3 11:09:05 2026 -0700 Add missed Error translations. commit 03c876fe167b9702d3cc451507bcb55215106cf3 Author: Mooneer Salem <mooneer@gmail.com> Date: Sat Oct 3 17:42:27 2026 +0000 latest user manual PDF commit 22dfd54db81fa03f97c76e4562a726e2c8023017 Merge: 481d526c 092db4bb Author: Mooneer Salem <mooneer@gmail.com> Date: Sat Oct 3 10:42:08 2026 -0700 Merge branch 'v3.0-dev' into feature/i18n commit 092db4bb80c8b7041a7d8e5a25eb8835e7c55405 Merge: 7856d4f0 58fc7442 Author: Mooneer Salem <mooneer@gmail.com> Date: Sat Oct 3 10:40:06 2026 -0700 Merge branch 'master' into v3.0-dev commit 58fc7442b5901649f15ab6bd193aedf532c5e7eb Author: Mooneer Salem <mooneer@gmail.com> Date: Sat Oct 3 10:39:43 2026 -0700 Windows: Fix spectrum plot axis labels appearing bold in light mode (#1531) * Windows: Fix spectrum plot axis labels appearing bold in light mode. The cached graticule bitmap added in #1524 is cleared to transparent black, and GDI+ antialiases text drawn into it against that black and makes the edge pixels opaque. Dark text therefore picked up a dark fringe and looked bold on a light background. Only cache the gridlines and ticks; draw the axis labels directly on the paint context, as PlotScalar does. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FrH6w6RDtAsEFLQwsgwGPG * Add PR #1531 to changelog. --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> commit e6ba3f9ef7748f811e9078f5e4b07b3a38fb6f1e Author: Mooneer Salem <mooneer@gmail.com> Date: Fri Oct 2 22:19:34 2026 -0700 Add PR #1531 to changelog. commit af9cb586527868cda01cd9bf91bfd81724642c21 Author: Mooneer Salem <mooneer@gmail.com> Date: Fri Oct 2 20:23:58 2026 -0700 Windows: Fix spectrum plot axis labels appearing bold in light mode. The cached graticule bitmap added in #1524 is cleared to transparent black, and GDI+ antialiases text drawn into it against that black and makes the edge pixels opaque. Dark text therefore picked up a dark fringe and looked bold on a light background. Only cache the gridlines and ticks; draw the axis labels directly on the paint context, as PlotScalar does. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FrH6w6RDtAsEFLQwsgwGPG commit 7856d4f0190285cd2779dd8f679a21ee45bd1ba4 Merge: aa1c43c3 8ec1670d Author: Mooneer Salem <mooneer@gmail.com> Date: Fri Oct 2 19:24:43 2026 -0700 Merge branch 'master' into v3.0-dev commit 8ec1670d935397f9cf890325b4993c03244cd75b Author: Mooneer Salem <mooneer@gmail.com> Date: Sat Oct 3 02:13:53 2026 +0000 latest user manual PDF commit 3ea8a5009e2190fa70dfc3a4929ce251f0229a54 Author: Mooneer Salem <mooneer@gmail.com> Date: Fri Oct 2 18:46:12 2026 -0700 macOS: Fix issue with waterfall background not being the correct color. (#1530) * macOS: Fix issue with waterfall background not being the correct color. * Add PR #1530 to changelog. commit aa1c43c3d08c1e07d1e6082353bfcecc1def8a40 Author: Mooneer Salem <mooneer@gmail.com> Date: Sat Oct 3 01:45:40 2026 +0000 latest user manual PDF commit 26a129d97b38f21f3ea3ca0100c9ca01787ad119 Author: Mooneer Salem <mooneer@gmail.com> Date: Fri Oct 2 18:42:57 2026 -0700 Fix bug with macOS not following dark/light mode changes. (#1529) * Fix bug with macOS not following dark/light mode changes. * Add PR #1529 to changelog. commit 9ed8c08cd82db09de3f8b8042ae3bfcbe9a15ab6 Author: Mooneer Salem <mooneer@gmail.com> Date: Fri Oct 2 17:09:27 2026 -0700 Add PR #1530 to changelog. commit bb0a4f489ce56347fceb5185fae897b5dc4a66c7 Author: Mooneer Salem <mooneer@gmail.com> Date: Fri Oct 2 17:05:15 2026 -0700 macOS: Fix issue with waterfall background not being the correct color. commit cc9257ac44e67a9b2a63cca4f91ec62915ba4ef6 Author: Mooneer Salem <mooneer@gmail.com> Date: Fri Oct 2 17:03:35 2026 -0700 Add PR #1529 to changelog. commit 4f04c506000fe92a6ed8cfa07734a88050e5a43c Author: Mooneer Salem <mooneer@gmail.com> Date: Fri Oct 2 17:02:12 2026 -0700 Fix bug with macOS not following dark/light mode changes. commit fde77500a1df1d3297f72dff9c905ed8aa38d7d2 Author: Mooneer Salem <mooneer@gmail.com> Date: Thu Oct 1 22:08:47 2026 -0700 TX: keep encoded-silence padding, but cap outfifo1 depth at 200ms. Removing the padding entirely (626a63d1) left only an ~80ms cushion in outfifo1. On CI runners the TX thread can wake >130ms late, causing outfifo1 underflows: gaps in the modem stream that don't go through the encoder, breaking RADE sync and feature alignment (rade loss 0.105 and 2.837 on macos-15/26 vs ~0.081 before). Restore padding with encoded silence when the mic FIFO runs dry so the modem stream stays continuous, but only while outfifo1 holds less than 200ms (previously it filled to ~640ms). Drop the one-time silence prime, which the capped padding replaces. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GxwHpb9KR3ymB15RFPGtR6 commit 626a63d139777d7082de9a15a5c26e6805ccb76c Author: Mooneer Salem <mooneer@gmail.com> Date: Thu Oct 1 20:06:40 2026 -0700 Reduce delay before EOO goes out at the end of TX. - Stop padding outfifo1 with encoded silence whenever the mic FIFO runs dry. That kept outfifo1 ~640ms full for the whole over, so all TX audio (including the EOO) was delayed by that much. Instead, prime outfifo1 once with a small amount of silence at the start of TX to absorb scheduling jitter. - Size the EOO-enqueue and outfifo1-drain timeouts from the audio actually queued instead of fixed 2s/1s values, which could expire mid-EOO. - Clear g_eoo_enqueued before setting endingTx so a stale value from the previous over can't end TX before this over's EOO is queued. rade_loss and rade_reporting_clean pass; EOO drain time dropped from ~620ms to ~310ms. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GxwHpb9KR3ymB15RFPGtR6 commit bd648231ae477ec30128929183150dab4bbbce4c Merge: 1f690e2d 1dbb637b Author: Mooneer Salem <mooneer@gmail.com> Date: Thu Oct 1 16:11:35 2026 -0700 Merge branch 'v3.0-dev' into ms-agc-tweaks-test commit 481d526c400cc8d43aee4ecaa867ec708e1980ce Merge: 82fc25d8 1dbb637b Author: Mooneer Salem <mooneer@gmail.com> Date: Thu Oct 1 16:09:52 2026 -0700 Merge branch 'v3.0-dev' into feature/i18n commit c353dae06b51595be229508cbe0defa501dbbba0 Merge: 812a232a 1dbb637b Author: Mooneer Salem <mooneer@gmail.com> Date: Thu Oct 1 16:09:36 2026 -0700 Merge branch 'v3.0-dev' into ms-rade-v2 commit 1dbb637b619231b96ac1306ffc64838e3a116f14 Merge: e0ac94a5 1277bbb8 Author: Mooneer Salem <mooneer@gmail.com> Date: Thu Oct 1 16:09:23 2026 -0700 Merge branch 'master' into v3.0-dev commit 1277bbb838f4799e015672289808b2d62bde4317 Author: Mooneer Salem <mooneer@gmail.com> Date: Thu Oct 1 16:07:48 2026 -0700 Reduce GUI CPU use: waterfall rendering and FreeDV Reporter updates (#1527) * macOS: draw the waterfall from a single ring-buffer image. The waterfall kept one image per 0.1 s block (~100 on screen) and drew each of them on every paint. CoreGraphics allocated, cleared and converted a buffer for every block on every frame, which was most of the GUI thread's time while receiving. On macOS the blocks now go into one image, kept in a ring buffer twice its height so the newest rows are always one contiguous run; a new block only writes its own rows and each paint draws a single image that points straight at the buffer. The canvas and graticule bitmaps use opaque host-order premultiplied ARGB, CoreGraphics' own format, and the canvas is drawn in copy mode, so compositing it is a plain memory copy. fullduplex_RADEV1, 45 s Time Profiler, Apple M1, main thread: - total: 5331 ms -> 2912-3495 ms (two runs) - Core Animation commit: 4277 -> 1644 ms - ripc_DrawImage: 2400 -> ~205 ms Other platforms keep drawing the per-block images. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GxwHpb9KR3ymB15RFPGtR6 * macOS: rasterize PlotScalar's plot area on the CPU. While transmitting, the Frm Mic plot repaints ten times a second, and Core Animation replaying its vector drawing (filling the background and the waveform polygon at 2x, and stroking every dash of the gridlines) was about a third of the GUI thread's time. On macOS the background, waveform and gridlines are now rasterized into a pixel buffer at the display's pixel density and drawn as one image in copy mode, in CoreGraphics' own pixel format (see PlotPixelBuffer in plot_osx.mm). The waveform is a vertical span per pixel column; the gridlines' pixels are worked out once per size. Colors are converted once from sRGB to the window's color space. Axis labels are still text. The half-height (SNR) plot, bar graphs and other platforms are unchanged. Compared with CoreGraphics filling and stroking the same paths (unantialiased, 2x), the gridlines are pixel-identical and at most 0.035% of waveform pixels (edge pixels) differ. rade_loss TX pass, Apple M1, main thread: - PlotScalar code + Core Animation replay: 1112 -> 699 ms - main thread excluding FreeDV Reporter: 3826 -> 3140 ms Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GxwHpb9KR3ymB15RFPGtR6 * FreeDV Reporter: reduce GUI thread work per server update. - Refit only the columns whose widest text changed, rather than every column (macOS). A column's fitted width only depends on its widest text and its header; across 49 partial refits in a test run, every width matched a full refit. Each refit went from ~19 ms to ~3 ms. - Re-measure only the cells whose text changed when a row updates (macOS and Windows), by caching each cell's text alongside its width. - Compile makeValidTime_()'s regexes once, format SNRs with a cached decimal separator, check the sort column by model column instead of looking columns up, and look up the system colors once per highlight pass instead of per row. fullduplex_RADEV1, 45 s Time Profiler, Apple M1: FreeDV Reporter on the main thread 1034 ms -> 289 ms (traffic from qso.freedv.org varies between runs; ~1800 updates per run in both cases). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GxwHpb9KR3ymB15RFPGtR6 * FreeDV Reporter: fix Linux build (macOS-only refit code). The per-column refit path was in the non-Windows branch, which also compiles on Linux, where refitAll and refitChangedColumns_() don't exist. Limit it to macOS and keep Linux's original setColumnAutosize_(true). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GxwHpb9KR3ymB15RFPGtR6 * Revert "macOS: rasterize PlotScalar's plot area on the CPU." This reverts commit 26131bd13c52d9b77cb8c02c7e35af69493f9cb0. It only helps when Core Animation composites on the CPU, as it does with the screen locked. With the screen unlocked, Core Animation renders on the GPU, where the original vectors are cheap and uploading a new 3.5 MB image every frame is not: the GUI's total CPU while transmitting (main thread plus Core Animation's render threads) was ~12% higher than master in a rade_loss TX pass on an Apple M1 (4,968 -> 5,587 ms, median of 2-3 runs). It also didn't change the macOS CI rade_loss failure rate. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GxwHpb9KR3ymB15RFPGtR6 * Linux: build PlotScalar's bitmaps at full density under Wayland. Under GTK on Wayland, cairo composites in-process with pixman. On a scaled display (here a 4K monitor at 1.25x, which GTK renders at 2x), the 1x plotArea_ and plotLines_ bitmaps were upscaled bilinearly on every paint: 53% of the GUI thread while transmitting. Build them at the window's content scale instead, so they're drawn 1:1. plotLines_ becomes a wxBitmap drawn through a wxMemoryDC, since a wxGraphicsContext on a wxImage draws in physical pixels. Keep 1x bitmaps under X11. There cairo-xlib hands the upscale to the X server (on the GPU with Xwayland/glamor), and full-density bitmaps would mean converting and uploading four times the pixels per frame: doing so raised the TX GUI thread from 1.8% to 3.0% of a core. This needs wxBitmap::UseAlpha() and wxGetDisplayInfo(), which are new in wxWidgets 3.2; with older versions the bitmaps stay at 1x as before. Fedora 44 / GNOME, GUI thread as % of one core (mean of 3 runs, all PASS): rade_loss, Wayland TX: 6.60 -> 3.68 generate_pgo_profiles (spectrum etc. visible), Wayland TX/RX: 3.77 / 6.00 -> 3.28 / 3.89 X11 (GUI thread and Xwayland), rade_loss and PGO: unchanged within noise Drawing the plot directly as vectors (DRAW_DIRECTLY, as on macOS and Windows) was also tried and is worse on Linux: no gain under Wayland and Xwayland's CPU tripled under X11. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SVXhR26Xxzywe32LV4SuDB * Linux: keep PlotScalar's plot area in a cairo surface. The plot area was redrawn into a wxBitmap that then had to be converted into a new cairo surface (via a GdkPixbuf) on every frame, since cairo can't draw from a wxBitmap directly. With GTK's cairo renderer, keep it in a cairo image surface instead: wrap it once as a wxGraphicsBitmap (whose pattern refers to the surface rather than copying it), scroll it in place with memmove(), draw the new points into it through a wxGraphicsContext on the same surface, and draw it straight into the window. This needs cairo's headers, which are found with pkg-config. Unlike including GTK's own headers, that's safe whichever GTK version wxWidgets was built for. Without them, or with wxWidgets before 3.2, the plot area is drawn as before. rade_loss, Fedora 44 / GNOME, GUI thread as % of one core (mean of 3 runs, all PASS): Wayland TX: 3.70 -> 2.21 (gnome-shell 1.53 -> 1.38) X11 TX: 1.89 -> 1.58 (Xwayland unchanged) RX (waterfall only): unchanged on both Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SVXhR26Xxzywe32LV4SuDB * PlotSpectrum: fix the trace disappearing on wide plots. Points closer together than HZ_GRANULARITY are skipped, but the spacing was measured from the previous bin rather than from the last point drawn. Bins can be closer together than HZ_GRANULARITY (RADE has 384 across 3 kHz), so on a wide plot few or no bins were ever far enough from the one before them. With RADE, the trace lost detail from ~900 px wide (131 of 383 segments drawn, rather than 251), was down to 33 at ~1570 px, and was empty at 1500-1537, 1800-1921, 2100-2305, 2400-2689 and 2700+ px. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SVXhR26Xxzywe32LV4SuDB * PlotScalar: only create the overlay's memory DC when building it. 34447c11 made the wxMemoryDC used to build the plotLines_ overlay a local of drawGraticuleFast(), so one was created and destroyed on every paint of every scalar plot, including with DRAW_DIRECTLY (macOS and Windows), where the overlay isn't used at all. On Windows that's a CreateCompatibleDC()/DeleteDC() pair per plot per frame, about 1% of the GUI thread in a generate_pgo_profiles run. Create it only when the overlay is (re)built. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SVXhR26Xxzywe32LV4SuDB * Windows: build PlotScalar's gridlines once instead of on every paint. With DRAW_DIRECTLY, the gridlines were collected into paths and stroked with dashed pens on every paint. GDI+ turns a dashed line into its dash pattern again each time it's stroked, and with building the paths, this was 18% of the GUI thread while receiving in a generate_pgo_profiles run on Windows (the largest single cost of drawing the plots). Build the gridlines once from solid segments in GDI+'s own dash patterns for these pens, keep them until the plot area changes size, and stroke them with solid pens. GDI+ draws both end points of a 1 px line and nothing for a zero-length one, so the single-pixel dots of the dot-dash lines are filled as 1x1 squares. Screenshots of the gridlines match the previous rendering pixel for pixel. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SVXhR26Xxzywe32LV4SuDB * Windows: copy the plot paint buffer to the window with BitBlt(). wxBufferedPaintDC copies its buffer with AlphaBlend() when it has an alpha channel (as this one is marked to, see the buffer's setup), and otherwise with StretchDIBits() since it's a DIB section. Either cost 8-12% of the GUI thread while receiving in a generate_pgo_profiles run. Draw into a wxMemoryDC set up the same way as wxBufferedPaintDC's (so text is still laid out for the window's DPI) and BitBlt() just the invalidated area to the window. This needs wxDCImpl::SetWindow(), new in wxWidgets 3.2; older versions still use wxBufferedPaintDC. Also call wxBitmap::UseAlpha() without an argument, which wxWidgets 3.0 requires on Windows. Windows 11, 200% scaling, GUI thread as % of one core, with this and the previous two commits (mean of 3 interleaved runs, all PASS): generate_pgo_profiles TX: 4.67 -> 4.03 (DWM 6.14 -> 5.47) generate_pgo_profiles RX: 6.19 -> 5.65 (DWM 6.49 -> 6.12) rade_loss TX: 1.13 -> 0.93 (DWM 4.91 -> 4.36) rade_loss RX: unchanged Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SVXhR26Xxzywe32LV4SuDB * [skip ci] Add PR #1527 to changelog. --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> commit ff89c56bfbc935d523d3b5ea610f9f73d6d899d9 Author: Mooneer Salem <mooneer@gmail.com> Date: Thu Oct 1 16:07:34 2026 -0700 [skip ci] Add PR #1527 to changelog.