riottracker / rtmidi

Haskell wrapper for RtMidi, the lightweight, cross-platform MIDI I/O library.

RtMidi

Haskell wrapper for RtMidi, the lightweight, cross-platform MIDI I/O library.

Version

The version of this library does not correspond to the RtMidi version (sorry). You can look at the justfile to see that we vendor sources from version 6.0.0.

How to use

See Hackage for the latest released version and add RtMidi and vector to your library build-depends.

Follow the callback example to receive messages.

Follow the playback example to send messages.

Development

This project is tested with Cabal (latest versions of last two compiler lines) and Stack (latest LTS).

You can get started with development like so (if you have just installed):

# Build and test on OSX CoreMIDI support or Linux with ALSA support.
# (You can also manually invoke stack and pass `--flag RtMidi:jack` for Jack support.)
just test

# Print information about all accessible MIDI devices and all compiled APIs.
just example-report

TODO

• See if there is a way to autodetect Jack in the Cabal file.
• See if there is a way to make OSX not take 3 seconds when initializing the MIDI subsystem.

rtmidi's Issues

Proposed maintenance

Here are some maintenance tasks I'm planning to do:

• Bump base bounds for 8.10 (and 8.8?)
• Add executable targets in the cabal file for example programs
• Update rtmidi sources
• Add CI for multiple GHC versions (and Stack LTS) (example)
• Maybe: Set default backends based on OS, not flags
• Maybe: Add additional "unlifted" layer (as in MonadUnliftIO, but maybe just IO) with "safe" withPort methods.
• Release to Hackage
• Release to Stackage

wrap all results in Either

first step: marshal error messages

make this thread-safe

Strictness and Vectors

Now we represent messages as [Word8], but we can probably make much better use of the heap with Vector Word8.
Also we yield messages to the callback as the lazy tuple (Double, [Word8]), but a strict tuple or datatype would be better (ensuring that it implements NFData).

cc-options are not passed in 8.10.2

I've been doing all my testing with 8.8.4 on Linux and Mac, and things work great (including with jack support). We test that we can BUILD with 8.8.4 and 8.10.2 in CI, but we can't run the test suite (since CI doesn't have a sound device).

When I tried to add to nightly stackage, I realized that the test suite fails in 8.10.2 because it tries to use the "dummy" API. This only happens when no other APIs are configured! (See https://github.com/riottracker/RtMidi/blob/master/rtmidi/RtMidi.cpp#L61 ) I have since repo'ed this by setting my ghc version with ghcup and cabal testing.

This problem indicates that cabal or ghc are not correctly passing the cc-options we set to our C++ compiler, and that this has changed between ghc versions. I have tried updating the cabal-version and using cxx-{options,sources} but that has no effect.