feat(mqtt): per-topic control, CW keyer, radio state, AX.25, debug logging #3307

Workflow file for this run

	name: CI

	on:
	pull_request:
	branches: [main]
	# Post-merge guard. We dropped the "Require branches up to date before
	# merging" branch-protection bit because it forced a CI rerun on every
	# PR whenever main moved (~15–25 min × N stale PRs per batch day).
	# The trade-off is: CI now runs on the merged commit itself, so any
	# semantic conflict that slipped past three-way merge surfaces on main
	# within ~10 min and triggers normal failure notifications.
	push:
	branches: [main]

	# Least-privilege GITHUB_TOKEN. CI only needs to read repo contents to
	# build and test — no writes, no token-bearing API calls. Any future job
	# that needs more (e.g. publishing a comment) must opt in at the job level.
	permissions:
	contents: read

	jobs:
	build:
	runs-on: ubuntu-latest
	container: ghcr.io/ten9876/aethersdr-ci:latest

	steps:
	- uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6

	- name: Setup DeepFilterNet3 (DFNR)
	run: bash scripts/setup/setup-deepfilter.sh

	- name: Configure
	run: cmake -B build -G Ninja -DCMAKE_BUILD_TYPE=RelWithDebInfo

	- name: Build
	run: cmake --build build -j$(nproc)

	# Cross-platform build check — runs on every PR (no path gating).
	#
	# History: previously gated by a `check-paths` job on a hand-maintained
	# allow-list of files known to carry `#ifdef Q_OS_WIN` / `Q_OS_MAC`
	# branches (CMakeLists, third_party/, workflows, MainWindow,
	# AudioEngine). The allow-list was a chronic leaker — every time a
	# new platform-guarded path got added to a different file, the gap
	# surfaced as a release-time regression: #796 (origin), #2671
	# (MainWindow added after a miss), #3052 (AudioEngine added after a
	# miss), #3210 (re-added after relapse), the v26.5.3
	# ClientPhaseRotator.cpp M_PI MSVC break, and the CwSidetonePortAudioSink
	# WASAPI fix in #3241 (Windows-only logic never compiled by CI). An
	# always-on rule trades ~15-20 min of runner time per PR for a category
	# of regression we kept paying for.
	#
	# Caching strategy:
	# * Qt install — handled by jurplel/install-qt-action's cache: true
	# * FFTW3 third_party — actions/cache, keyed on the setup script + version
	# * MSVC object files — sccache via GitHub Actions cache backend
	# Net effect: cold cache ~5 min, warm cache ~2-3 min (was ~14 min).
	check-windows:
	runs-on: windows-latest
	env:
	SCCACHE_GHA_ENABLED: 'true'
	steps:
	- uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6

	- name: Install Qt
	uses: jurplel/install-qt-action@48d3ad6db93f3627c8ee7a0454bc6f3744f7e730 # v4.3.1
	with:
	# 6.8.3 matches what community Windows builders are running and
	# exposes Qt 6.8.x MOC volume so we catch issues like #1910 (MSVC
	# COFF section limit, fixed by /bigobj) before they reach users.
	version: '6.8.3'
	modules: 'qtmultimedia qtwebsockets qtserialport qtshadertools'
	cache: true

	- name: Cache FFTW3
	uses: actions/cache@27d5ce7f107fe9357f9df03efb73ab90386fccae # v5
	with:
	path: third_party/fftw3
	key: ${{ runner.os }}-fftw3-3.3.5-${{ hashFiles('scripts/setup/setup-fftw.ps1') }}

	- name: Install FFTW3
	shell: pwsh
	run: .\scripts\setup\setup-fftw.ps1

	- name: Setup DeepFilterNet3 (DFNR)
	shell: pwsh
	run: .\scripts\setup\setup-deepfilter.ps1

	- name: Setup sccache
	uses: mozilla-actions/sccache-action@9e7fa8a12102821edf02ca5dbea1acd0f89a2696 # v0.0.10

	# CMAKE_CXX_COMPILER_LAUNCHER is honoured by Ninja and Make
	# generators, NOT by the Visual Studio / MSBuild generator that
	# CMake selects by default on windows-latest. Switch to Ninja so
	# cl.exe invocations actually route through sccache. msvc-dev-cmd
	# puts the MSVC toolchain on PATH (cl.exe, link.exe, etc.) so
	# Ninja can find them.
	- name: Setup MSVC environment
	uses: ilammy/msvc-dev-cmd@0b201ec74fa43914dc39ae48a89fd1d8cb592756 # v1
	with:
	arch: x64

	- name: Setup hidapi (HID encoder support)
	# Needed for the Ulanzi Dial Windows backend (#3239) and the
	# existing USB HID encoders (#3171). Without this, HAVE_HIDAPI
	# is undefined and the dial code falls back to a stub — CI
	# would miss real compile errors in that path. Runs after
	# msvc-dev-cmd so the script has cl.exe + link.exe on PATH.
	shell: pwsh
	run: .\scripts\setup\setup-hidapi.ps1

	# zlib is bundled under third_party/zlib (1.3.1) — no vcpkg install
	# needed; CMake builds zlibstatic from source. (#2651)

	# Verify sccache can actually compile through the backend. GitHub
	# Actions Cache occasionally has outages; when that happens, sccache's
	# GHA backend can't read/write and every compile through sccache
	# fails. Pre-flight a tiny test compile so we can fall through to a
	# plain build instead of failing the whole job.
	- name: Test sccache health
	id: sccache_health
	shell: pwsh
	continue-on-error: true
	run: \|
	'int main(){return 0;}' \| Out-File -Encoding ASCII test.cpp
	sccache cl /c test.cpp /Fo:test.obj 2>&1 \| Tee-Object -Variable scOut
	$exit = $LASTEXITCODE
	Remove-Item test.cpp,test.obj -ErrorAction SilentlyContinue
	if ($exit -eq 0) {
	"healthy=true" \| Out-File -FilePath $env:GITHUB_OUTPUT -Append
	Write-Host "sccache OK"
	} else {
	"healthy=false" \| Out-File -FilePath $env:GITHUB_OUTPUT -Append
	Write-Host "::warning::sccache unhealthy (exit $exit) — building without compiler cache"
	}

	- name: Configure
	shell: pwsh
	run: \|
	$launcher = "${{ steps.sccache_health.outputs.healthy }}" -eq "true"
	$cargs = @(
	"-B","build","-G","Ninja",
	"-DCMAKE_BUILD_TYPE=Release",
	"-DCMAKE_MSVC_DEBUG_INFORMATION_FORMAT=Embedded",
	"-DCMAKE_TOOLCHAIN_FILE=$env:CMAKE_TOOLCHAIN_FILE",
	"-DVCPKG_TARGET_TRIPLET=$env:VCPKG_TARGET_TRIPLET"
	)
	if ($launcher) {
	$cargs += "-DCMAKE_C_COMPILER_LAUNCHER=sccache"
	$cargs += "-DCMAKE_CXX_COMPILER_LAUNCHER=sccache"
	}
	& cmake @cargs

	- name: Build Opus (RADE dependency)
	# ExternalProject dependency ordering via BUILD_BYPRODUCTS is correct,
	# but a separate step gives clearer CI failure attribution and avoids
	# any Ninja parallel-build ordering edge cases with IMPORTED targets.
	run: cmake --build build --target build_opus -j $env:NUMBER_OF_PROCESSORS

	- name: Build
	run: cmake --build build -j $env:NUMBER_OF_PROCESSORS

	- name: sccache stats
	if: always()
	run: sccache --show-stats

	# macOS build check — runs on every PR (no path gating), same rationale
	# as check-windows above. Catches Q_OS_MAC / CoreAudio / Cocoa-bridge
	# breakage that Linux clang misses, which is otherwise only discovered
	# at release tag time by macos-dmg.yml. Build-only (no code signing,
	# no DMG, no notarization — those stay in macos-dmg.yml on tag push).
	# Apple Silicon runner only; Intel coverage remains release-time-only.
	check-macos:
	runs-on: macos-15
	steps:
	- uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6

	- name: Install dependencies via Homebrew
	# Mirrors macos-dmg.yml's apple-silicon path. Homebrew Qt targets
	# the current OS, which is fine for the arm64 CI check; deployment-
	# target pinning is a release-time concern handled in macos-dmg.yml.
	run: \|
	brew install ninja qt@6 qtkeychain fftw portaudio hidapi \
	autoconf automake libtool

	- name: Setup DeepFilterNet3 (DFNR)
	run: bash scripts/setup/setup-deepfilter.sh

	- name: Set up ccache
	uses: hendrikmuhs/ccache-action@d62db5f07c26379fc4b4e0916f098a92573c3b03 # v1.2.23
	with:
	key: ccache-macos-ci
	max-size: 500M

	- name: Configure
	run: \|
	cmake -B build -G Ninja \
	-DCMAKE_BUILD_TYPE=Release \
	-DCMAKE_C_COMPILER_LAUNCHER=ccache \
	-DCMAKE_CXX_COMPILER_LAUNCHER=ccache \
	-DCMAKE_PREFIX_PATH="$(brew --prefix qt@6);$(brew --prefix)" \
	-DMQTT_TLS=OFF

	- name: Build Opus (RADE dependency)
	run: cmake --build build --target build_opus -j$(sysctl -n hw.ncpu)

	- name: Build
	run: cmake --build build -j$(sysctl -n hw.ncpu)

Navigation Menu

Uh oh!

Uh oh!

feat(mqtt): per-topic control, CW keyer, radio state, AX.25, debug logging #3307

Workflow file

feat(mqtt): per-topic control, CW keyer, radio state, AX.25, debug logging #3307

Uh oh!

Workflow file for this run