[argothiel]

There is an issue with pulse_delay() which calls pa_stream_get_latency(). PulseAudio documents that this function returns -PA_ERR_NODATA when timing information is not available yet. The ALSA Pulse plugin compares the result against positive PA_ERR_NODATA; it therefore treats that temporary condition as an unrelated failure and returns -EIO.

Change PA_ERR_NODATA to -PA_ERR_NODATA in the comparison to not miss that branch.

Fixes #65.

[ZmoleCristian]

Independent confirmation of the impact and the proposed fix. We encountered repeated voice-session termination in a Linux application using CPAL's ALSA backend through the PulseAudio ALSA plugin. The native error was snd_pcm_avail_delay: I/O error (5), observed both shortly after opening playback and after reopening the output stream. For the user this repeatedly interrupted voice interaction.

We independently arrived at the identical one-line correction in this PR and tested the actual pulse_delay() function with controlled dependency stubs: the original code returns false -EIO for -PA_ERR_NODATA; with the correction, all 101 cases (0 through 100 consecutive timing-unavailable responses before success) pass. Other errors and a disconnected stream still propagate as errors. These are deterministic function-level tests, not 101 hardware runs.

We also installed a privately rebuilt plugin with this change. The user reports no further interruptions in the current observation period; longer real-world validation is ongoing. Our historical ALSA logs do not expose the underlying PulseAudio return code, so we cannot attribute every past EIO to this branch.

The positive comparison appears to have been introduced in c2456f2 (2008-09-03), when the wait-for-timing loop was added. Before that, the function already returned EIO on any latency-query error, so this dates the faulty comparison rather than establishing the first occurrence of the symptom.

This has a substantial user-visible impact in applications that treat the error as fatal. Please consider prioritizing review of this fix. Thank you for preparing it.

[gmelikov]

Tested this exact patch locally. The reproducer failed with EIO in 3/3 unpatched runs and succeeded in 3/3 patched runs.

My case is - pactl load-module module-native-protocol-tcp listen=127.0.0.1 port=4713 auth-anonymous=1 on laptop and port-forward it via SSH.

[argothiel]

@perexg @tiwai

Could you please advise who can approve and merge it?

[argothiel]

@i-garrison @poettering
Maybe you could help? It's about a one-line fix in the PulseAudio plugin. Two users have independently confirmed it resolves the issue.

[i-garrison]

I can confirm that checking for negative value of error constant is valid fix indeed!

@perexg could you please merge this? Thank you!

[fayerman-source]

I tried the change on my WSL2 setup. The installed plugin had the positive-16 comparison; silent snd_pcm_avail_delay probe failed three times with the installed plugin, then passed three out of three using a copy with the comparison changed to negative 16. With that copy in asoundrc, playback and capture opened. The first Codex slash voice still failed at OpenDevices/25; later it worked. I can't explain that first failure, but wanted to add another WSL2 repro.

[argothiel]

Is @perexg the only one with merging rights?

[perexg]

Apologize, busy with other things. The signed-off-by / commit author e-mail is anonymous.

[perexg]

Sorry, wrong push. The bd8eaa1 was not applied.

[argothiel]

@perexg No worries, I just wanted to make sure if we have the right person! 😊

The e-mail sign-off address has been corrected.