A long-standing TODO item for SDL was audio capture: the ability for your app to record sound from a microphone. We designed it into the API for 2.0.0, but it didn't exist at all beyond that. Not only was it not implemented in any of the backends, the higher-level infrastructure was missing in the library core too, so trying to open an audio device for recording would always fail.
No longer! Now you can record audio from your SDL2-based programs.
A fundamental difficulty with SDL--for me at least--is that adding a new feature to the audio subsystem means implementing it twenty-three times. There are three different audio APIs on Windows, seven on Linux, not to mention Android, iOS, macOS, Emscripten, and various other Unix variants and miscellaneous fringe targets. It took some effort to support as many of these as possible! Overall, we did okay; all of the major platforms have support: Windows (winmm, DirectSound), Linux (ALSA, OSS, PulseAudio), macOS, iOS, Android, and Emscripten. Some of the others are beyond my reach (XAudio2 on Windows doesn't support capture at all, and I don't have an AIX install, for crying out loud!), but I'll definitely welcome external patches to fill any gaps.
Using the new capture support should be very familiar to those that have used SDL's usual audio APIs. You enumerate and open capture devices the same way. The familiar audio callback fires regularly, as you'd expect, although you read data in the callback instead of writing it. If you want a simpler interface, there's also an equivalent of 2.0.4's SDL_QueueAudio() API to grab samples from the device at your leisure.
This is in revision control now, and will officially ship with SDL 2.0.5.
Want to see it in action? Fire up a web browser that supports recording from a microphone, and point it here. That's a C program, compiled with Emscripten, that uses SDL's new audio capture support.