Ctl::pcm_info expected errors

[quaternic]

If the device exists but has no subdevices for the selected stream direction, Ctl::pcm_info returns an error, specifically libc::ENOENT. I couldn't find documentation for this, but the underlying syscall traces back to https://github.com/torvalds/linux/blob/72761a7e31225971d0b29d9195e0ffa986b77867/sound/core/pcm.c#L124-L133

In the example pcm_enumerate, function list_devices_for_card immediately returns that error to the caller, so any further devices get silently ignored. The example can be fixed by doing something like

let pcm_info = match ctl.pcm_info(device as u32, 0, direction) {
    Ok(info) => info,
    Err(e) if e.errno() == libc::ENOENT => continue,
    err => return err,
};

It's not very obvious that you have to do that, especially when the error prints as "No such file or directory".

[diwic]

Thanks for the catch.
Maybe it would make sense that every error should lead to a continue, not only your specific error case?

[quaternic]

In some cases that might be reasonable for libc::ENXIO as well (which you'd get for devices or subdevices that don't exist), but silently ignoring all errors isn't something to recommend. In the same example the caller is doing just that, which was hiding the problem.

The weirdness with snd_ctl_pcm_info is that to query how many subdevices there are for a device, you start by querying for the information for subdevice 0, and ENOENT is how it tells you that the device exists but has no subdevices for that direction. See also how it's used in aplay.

Minimally I'd just document this for Ctl::pcm_info and fix the example, but there could be an API to e.g. iterate all devices, which could also be more efficient (mainly by reusing the same Info).

[diwic]

Fair enough. Doing what aplay does is a strong argument :-)

Would you like to submit a PR to fix it?