[LucasAntunesdeAlmeida]

What

The no-audio-device probe (contract item 9) without cgo, following oto 3.5's new order: PulseAudio first, then ALSA. With it the engine has no cgo left, and since Ebitengine v2.10 needs none on Linux, a Collider game builds for Linux with CGO_ENABLED=0, from any OS.

Before this, a cgo-free Linux build compiled the probe to a no-op and panicked at start on a machine without sound (the v0.6.1 bug back again). Checked in Valve's sniper SDK container under Xvfb with no /dev/snd, examples/horde built with CGO_ENABLED=0 GOOS=linux on Windows:

panic: audio: audio error: oto: initialization failed: PulseAudio: ... ALSA: oto: ALSA device initialization failed: "default": ...

Design

• device_unix.go (oto's unix build tags): probeDevice tries PulseAudio, then probeALSA when the platform has it (set by device_alsa.go, as oto does), and names both failures in the one log line.
• PulseAudio through github.com/jfreymuth/pulse (already in the module graph through oto, now a direct dependency): connect, create oto's stream (stereo float at 48 kHz) corked so nothing plays, close. The library's one-second request timeout keeps a server that does not answer from hanging the frame.
• ALSA through purego (also already there, through Ebitengine and oto): libasound opened like oto (libasound.so.2, then libasound.so), oto 3.5's candidates (default, plug:default, then every output device except null, each followed by its plug: variant), opened non-blocking (busy counts as present: oto waits for it), first to take FLOAT_LE/2ch/48k wins. oto 3.5 moved past a device that opens but cannot be configured; the probe does too. A machine without libasound is an answer, not a crash.
• ALSA and its plugins (JACK, OSS, pulse) print each failed attempt to stderr. The cgo probe replaced ALSA's error handler, which does not reach what a plugin's own library (libjack) prints. Now fd 2 points at /dev/null while the ALSA probe runs and is restored after, so the engine's line is the only output.
• Other platforms: device_other.go no-op, as before.

No API change. README contract item 9 names PulseAudio/PipeWire and ALSA; PUBLISHING.md's Linux section now says CGO_ENABLED=0 GOOS=linux GOARCH=amd64 go build, no dev packages, and lists the runtime libraries. CI also runs CGO_ENABLED=0 GOOS=linux go vet . && CGO_ENABLED=0 GOOS=linux go build ./....

Proof

• Tests (device_unix_test.go, device_alsa_test.go): oto's candidate order, stderr muted to the null device and restored, a missing server failing in well under a second, both errors named, and the probe answering on whatever machine runs it (CI has no server or card).
• Local gate on Windows: gofmt, vet, staticcheck (also GOOS=linux), tests, js/wasm, windows, and the new cgo-free Linux vet/build. Linux tests in the sniper SDK container under Xvfb pass with CGO_ENABLED=0 and =1.
• Real runs of the cgo-free examples/horde (ldd: libc, libdl, libpthread only):
  • sniper SDK, no device: one line, collider: no audio device, playing silently: PulseAudio: dial unix pulse/native: connect: no such file or directory; ALSA: no playback device among 16 ("default": No such file or directory), and the game keeps running. No ALSA or JACK noise.
  • sniper SDK, ALSA null as default: no line, plays.
  • Debian bookworm with a PulseAudio server on a null sink: no line; the server lists sink input application.name = "horde", Corked: no (oto is playing through PulseAudio).