Publish one Linux AppImage instead of separate CPU and Vulkan builds #279

Workflow file for this run

	name: CI

	on:
	push:
	branches: [master]
	# Every pull request, whatever it targets — deliberately unfiltered.
	#
	# This used to be `branches: [master]`, which meant a pull request based on
	# another branch got no checks at all: not a compile, not a test, nothing. A
	# stacked change could then be merged down into its base and reach master
	# having never been built on its own. That is the same shape of hole as
	# excluding crates from the Windows check — work that looks reviewed because a
	# PR exists, and was never compiled.
	pull_request:

	env:
	CARGO_HTTP_MULTIPLEXING: false
	CARGO_HTTP_HTTP2: false

	jobs:
	# ── Rust compile check ────────────────────────────────────────────────────────
	cargo-check:
	name: cargo check (${{ matrix.os }})
	runs-on: ${{ matrix.os }}
	strategy:
	fail-fast: false
	matrix:
	os: [ubuntu-22.04, windows-latest]

	steps:
	- uses: actions/checkout@v4

	- name: Install Linux system dependencies
	if: runner.os == 'Linux'
	run: \|
	sudo apt-get update -q
	sudo apt-get install -y \
	libwebkit2gtk-4.1-dev \
	libayatana-appindicator3-dev \
	librsvg2-dev \
	libssl-dev \
	libasound2-dev \
	libspeechd-dev \
	libxdo-dev \
	patchelf \
	squashfs-tools \
	pkg-config \
	build-essential

	- uses: dtolnay/rust-toolchain@stable

	# whisper-rs builds whisper.cpp through the `cmake` crate, so the check
	# now needs a C++ toolchain that the excluded-crates version did not. The
	# Windows runners ship both MSVC and CMake; assert it rather than let a
	# missing one surface as a confusing build-script panic.
	- name: Verify Windows build tools
	if: runner.os == 'Windows'
	shell: pwsh
	run: \|
	cmake --version
	rustc -vV

	# glslc, for the ggml Vulkan backend's compute shaders, which the
	# `vulkan` check below compiles. Same source as the release workflow uses.
	- name: Install Vulkan SDK (Linux)
	if: runner.os == 'Linux'
	run: \|
	wget -qO- https://packages.lunarg.com/lunarg-signing-key-pub.asc \
	\| sudo tee /etc/apt/trusted.gpg.d/lunarg.asc >/dev/null
	sudo wget -qO /etc/apt/sources.list.d/lunarg-vulkan-jammy.list \
	https://packages.lunarg.com/vulkan/lunarg-vulkan-jammy.list
	sudo apt-get update -q
	sudo apt-get install -y vulkan-sdk
	glslc --version

	- uses: Swatinem/rust-cache@v2
	with:
	key: check-${{ matrix.os }}

	# The same full-workspace check on both platforms, deliberately with
	# nothing excluded.
	#
	# This job used to skip voxctrl-inference and voxctrl-app on Windows,
	# on the grounds that the Linux check covered their type-safety. It does
	# not: everything under `src-tauri/src` that is `cfg(target_os =
	# "windows")` is invisible to a Linux compiler, and that is most of the
	# platform code there is. With those two excluded the excluded set was the
	# speech engine and the entire application, so no Windows build was ever
	# verified — which is how the Windows hotkey backend shipped naming keys
	# in a vocabulary nothing else used.
	#
	# `--all-targets` so the test code is checked too, not just the libraries.
	# Default features include the two ONNX-backed engines, so this fetches
	# ONNX Runtime at build time and needs network access.
	- name: cargo check (workspace, default features)
	run: cargo check --workspace --all-targets

	# The GPU features are off by default, so the workspace check above never
	# compiles a line of their execution-provider code — and the first thing
	# that would is a release build, where a mistake costs a whole run. That
	# is the same shape of gap that let the Windows backend ship unbuildable,
	# so each shipped GPU variant gets checked here.
	#
	# `moonshine-webgpu` is what the Windows GPU release artifact is built
	# with; `vulkan` is what the Linux AppImage is built with. Both pull
	# their own prebuilt ONNX Runtime / compile their own shaders, so this is
	# not free, but it is far cheaper than finding out at release time.
	- name: cargo check (moonshine-webgpu)
	run: cargo check -p voxctrl-app --features moonshine-webgpu

	# Since the CPU AppImage was dropped, `vulkan` is not a variant of the
	# Linux release — it is the Linux release. The default-features check
	# above used to double as its coverage because the two were built the
	# same way; now nothing else compiles a line of the ggml Vulkan backend
	# before the release run does.
	- name: cargo check (vulkan)
	if: runner.os == 'Linux'
	run: cargo check -p voxctrl-app --features vulkan

	# ── Unit tests ────────────────────────────────────────────────────────────────
	cargo-test:
	name: cargo test
	runs-on: ubuntu-22.04

	steps:
	- uses: actions/checkout@v4

	- name: Install Linux system dependencies
	run: \|
	sudo apt-get update -q
	sudo apt-get install -y \
	libwebkit2gtk-4.1-dev \
	libayatana-appindicator3-dev \
	librsvg2-dev \
	libssl-dev \
	libasound2-dev \
	libspeechd-dev \
	libxdo-dev \
	patchelf \
	squashfs-tools \
	pkg-config \
	build-essential \
	espeak-ng \
	xvfb

	- uses: dtolnay/rust-toolchain@stable

	- uses: Swatinem/rust-cache@v2
	with:
	key: test-linux

	# Run tests for crates that have unit tests and no heavy system deps.
	# voxctrl-tts's phoneme-frontend tests exercise the real `espeak-ng`
	# binary when it is present, and skip cleanly when it isn't — it is
	# installed above so they actually run here.
	#
	# Under xvfb-run for the same reason: the X11 hotkey backend's tests talk
	# to a real X server when DISPLAY is set, and skip when it isn't. Only a
	# real server rejects a malformed XISelectEvents request, and a version
	# of that backend shipped selecting raw and hierarchy events in one mask
	# — which every server refuses, so no key ever arrived — with the whole
	# suite passing. Without a display here those tests silently skip and the
	# gap reopens.
	- name: cargo test
	run: xvfb-run -a cargo test -p voxctrl-config -p voxctrl-routing -p voxctrl-hotkeys -p voxctrl-winput -p voxctrl-tts -p voxctrl-update -p voxctrl-bugreport -p voxctrl-app

	# ── Unit tests (Windows) ──────────────────────────────────────────────────────
	#
	# The Linux lane above runs the whole suite; this one runs the crates whose
	# behaviour differs by platform, so a Windows-only regression cannot hide
	# behind a green Linux run. Crates needing a display, a desktop portal or
	# espeak-ng are left to the Linux lane, where those exist.
	cargo-test-windows:
	name: cargo test (windows)
	runs-on: windows-latest

	steps:
	- uses: actions/checkout@v4

	- uses: dtolnay/rust-toolchain@stable

	- uses: Swatinem/rust-cache@v2
	with:
	key: test-windows

	- name: cargo test
	run: >
	cargo test
	-p voxctrl-config
	-p voxctrl-routing
	-p voxctrl-hotkeys
	-p voxctrl-winput
	-p voxctrl-update
	-p voxctrl-text
	-p voxctrl-bugreport

	# ── Frontend type check ───────────────────────────────────────────────────────
	frontend-check:
	name: Frontend type check
	runs-on: ubuntu-latest

	steps:
	- uses: actions/checkout@v4

	- uses: actions/setup-node@v4
	with:
	node-version: "20"
	cache: npm

	- run: npm install

	# svelte-check (Svelte + TypeScript type checking)
	- run: npm run check

	# Frontend unit tests (vitest)
	- run: npm run test:unit

Navigation Menu

Publish one Linux AppImage instead of separate CPU and Vulkan builds #279

Workflow file

Publish one Linux AppImage instead of separate CPU and Vulkan builds #279

Uh oh!

Workflow file for this run