Remove cubeb_stream_get_current_device

[marcin-serwin]

While looking into cubeb_stream_get_current_device I've noticed some issues:

• It seems like the function is supposed to return strings but src/cubeb_audiounit.cpp returns non-native endian integers. This could crash cubeb-rs since it asserts that the returned value is a valid UTF-8 string.
• Also in src/cubeb_audiounit.cpp, the function does not free resources in case of errors, e.g., if the call to audiounit_get_default_device_name fails then *device is not freed causing leak.
• The device_destroy function in cubeb-pulse-rs does not free the allocated strings, it only frees the cubeb_device so it always causes memory leaks.

Instead of fixing this issues one by one I figured that it may be beneficial to simplify the API usage and avoid the leaks by not allocating memory in the first place.

audiounit_stream_get_current_device is removed, it's not implemented in the Rust version so I assume it's not super necessary. Strings and the cubeb_device are returned directly instead of being allocated on the heap. This causes the destroying function to become unnecessary and it is removed in the last commit.

This is an API and ABI break but code changes for clients should be minor, comparable to the ones present in test/test_devices.cpp.

I'm a bit unsure about the removal of the destroying function. It would be a noop in the current backends, however, in the future it may be actually necessary to allocate something for a backend. What do you think?

[padenot]

This is very edge-casey. The only caller that I know of is in Firefox, and this is used to pan the audio to the right speaker if using the buildint speakers, on macbook pre-2016, to avoid feedback into the microphone. We might be able to remove everything at this point and we'll deal with it using enumaration API.

[padenot]

(https://searchfox.org/mozilla-central/source/dom/media/GraphDriver.cpp#1143-1198)

[marcin-serwin]

The only other caller I know is RPCS3 where it is used to get the name of the default device that supports given stream params. If we want to delete it we should also accommodate for their usecase.

this is used to pan the audio to the right speaker if using the buildint speakers, on macbook pre-2016, to avoid feedback into the microphone.

Is this workaround even functional given that this function is not supported by the current macOS backend?

[padenot]

The only other caller I know is RPCS3 where it is used to get the name of the default device that supports given stream params. If we want to delete it we should also accommodate for their usecase.

We can send a patch there (same fix, enumerate, find the default device, get the name).

this is used to pan the audio to the right speaker if using the buildint speakers, on macbook pre-2016, to avoid feedback into the microphone.

Is this workaround even functional given that this function is not supported by the current macOS backend?

No it hasn't worked for a long time, and it doesn't matter anymore (Firefox generally uses macOS's system noise suppression and echo cancellation, and Apple have fixed their hardware). We simply didn't remove the code. We can probably just remove everything then.

[marcin-serwin]

The only other caller I know is RPCS3 where it is used to get the name of the default device that supports given stream params. If we want to delete it we should also accommodate for their usecase.

We can send a patch there (same fix, enumerate, find the default device, get the name).

They are doing that in the GetDevice function, not sure why they implemented the stream fallback.

[marcin-serwin]

RPCS3 devs are fine with just using enumeration so I guess we can proceed with removal.

[marcin-serwin]

@padenot Any updates regarding this?