Something went wrong. Try again.
Play and record audio on Windows from Zig through WASAPI and COM, without linking a C library.
Something went wrong. Try again.
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869# SPDX-FileCopyrightText: © 2026 Jeffrey C. Ollie <jeff@ocjtech.us># SPDX-License-Identifier: MIT
# The GitHub mirror exists for this one runner.## Forgejo is where this repository lives and where the real CI is; what it cannot# do is run a Windows machine. `zig build check` on the Forgejo side proves the# whole COM half still *type-checks*, cross-compiled for two Windows# architectures, which is the cheap half. This is the other half: it links and# executes it, which is the only way to catch a symbol that does not exist in# `ole32` or `avrt`, a calling convention written down wrongly, a COM vtable laid# out differently than declared, or apartment handling that is subtly wrong.## There is no audio endpoint on a hosted runner, so every test that wants to# actually make a sound reports SkipZigTest. What runs is the whole# operating-system-independent suite -- the ring buffer, the channel# permutations, the format encoder, the sample conversions -- plus the COM tests# that do not need hardware: creating the device enumerator, enumerating nothing,# registering and unregistering a notification client, and the apartment# decision table. Enumerating an empty machine is worth having: it is a path that# has to work and is easy to get wrong.## The audible half cannot be automated here at all. See "What the tests cover,# and what they cannot" in the README.## Actions are pinned to a commit with the tag they came from in a trailing# comment, so that a moving tag cannot change what CI executes.
name: test
on: push: workflow_dispatch:
jobs: test: strategy: # Not fail-fast: when something breaks in one optimize mode it is useful to # know at once whether it broke in the others too. A release build in # particular exercises different arithmetic in the sample converters, where # a safe build would have trapped instead. fail-fast: false matrix: optimize: [Debug, ReleaseSafe, ReleaseFast] runs-on: windows-latest steps: - name: Check out uses: actions/checkout@fbc6f3992d24b796d5a048ff273f7fcc4a7b6c09 # v5
# No Nix here, unlike the Forgejo workflow, so the compiler comes from an # action instead of from the flake's dev shell. The version is pinned to # the same one `build.zig.zon` requires. - name: Set up Zig uses: mlugg/setup-zig@d1434d08867e3ee9daa34448df10607b98908d29 # v2.2.1 with: version: 0.17.0
- name: Test run: zig build test -Doptimize=${{ matrix.optimize }} --summary all
- name: Build for every supported target run: zig build check -Doptimize=${{ matrix.optimize }} --summary all
# The examples are the worked code a reader copies, so they are built # rather than only type-checked. Not run: they would either make a noise or # fail for want of an endpoint, and neither tells us anything here. - name: Build the examples run: zig build -Doptimize=${{ matrix.optimize }} --summary all