I wasn't able to do this without using Box - happy to take any suggestions if you can think of a better implementation as I only have very nascent Rust knowledge.

make the builder generic over the error callback (you will need to use OutputStreamBuilder<ErrorCallback> instead of Self<ErrorCallback in with_error_callback). There should be no need to make OutputStream generic.

I wasn't able to do this without using Box - happy to take any suggestions if you can think of a better implementation as I only have very nascent Rust knowledge.

make the builder generic over the error callback (you will need to use OutputStreamBuilder<ErrorCallback> instead of Self<ErrorCallback in with_error_callback). There should be no need to make OutputStream generic.

I tried that for a while but was unable to get it working with a fallback to the default error callback but will try again next week.

I think a dyn Fn makes sense here, the callback is not performance-critical.

The problem with opening default stream is that it may try to open stream multiple times. At the same time the call back gets owned at each attempt. See also my comment in open_stream_or_fallback.

I think a dyn Fn makes sense here, the callback is not performance-critical.

I worry about the underflow error (when the source can not deliver samples fast enough to the OS). The a slow error handler could make the problem worse, a single underflow error could cause many more. I do not know if the delay caused by dyn is severe enough to cause that.

I think a dyn Fn makes sense here, the callback is not performance-critical.

I worry about the underflow error (when the source can not deliver samples fast enough to the OS). The a slow error handler could make the problem worse, a single underflow error could cause many more. I do not know if the delay caused by dyn is severe enough to cause that.

@dvdsk I have now implemented this using generics instead without dyn dispatch. Let me know your feedback.

Looks great, if you want to you could make the example a little more in-depth (see comment) but this is good to merge.

All done - feel free to merge once you're happy!

O.o. CI is broken it seems....

well I'll see if I can fix that in a bit. PR looks great, ill try to get CI sorted and then merge this.

O.o. CI is broken it seems....

well I'll see if I can fix that in a bit. PR looks great, ill try to get CI sorted and then merge this.

CI is all fixed - I broke it by forgetting a cargo fmt. Thanks for your help on this PR!

I wonder if anyone can simulate/test underflow error. I could not. On my system no matter how slow the source is pipewire/ALSA does not complain (at least not through CPAL).

On my system (Linux+ALSA+Pipewire). When I stop audio server I get BackendSpecificError instead.
Generally, I do not believe there are any recoverable errors that we get in the callback.

On my system (Linux+ALSA+Pipewire). When I stop audio server I get BackendSpecificError instead.

Generally, I do not believe there are any recoverable errors that we get in the callback.

On CoreAudio/Mac we get the correct error and can destroy & recreate the output stream with a different device.

@will3942 Are there any recoverable errors you get from this callback?

@will3942 Are there any recoverable errors you get from this callback?

How do you define recoverable? Recover without having to tear down and destroy the resources?

I wonder if anyone can simulate/test underflow error. I could not. On my system no matter how slow the source is pipewire/ALSA does not complain (at least not through CPAL).

One way to do it is to use an RPi with SD card and do a lot of I/O on it, writes in particular.

@will3942 Yes, by "recoverable" here I mean being able to proceed without re-initializing the stream.

@roderickvd The output stream can not know where delay comes from. So I believe that just artificially stalling the sample callback should have the same effect.

@will3942 Yes, by "recoverable" here I mean being able to proceed without re-initializing the stream.

Not that I'm aware of, no.

In my application code I will be using this to remove and recreate the OutputStream, checking if the device is still available. It would be nice if Rodio could optionally provide this functionality out of the box but it feels like there's quite a few toggles/options/timeouts you'd want to provide the user in determining how that logic is provided.