Churros

Chan + Arrow = Churro

Simple alternative to Conduit, Pipes, Streams, Machines, etc.

Use-case is CSP like scenarios where you need a graph of actors.

Developed from a history of attempting to use co-routines libraries for setting up complicated asynchronous processes such as collections of MIDI instruments, etc, but being frustrated by not easily being able to conditionally consume and emit events. In these situations I'd fall back on creating a bunch of Chans and piping events manually. Churros just formalises that strategy to help you get it right!

Advantages over other alternatives:

• Focus on IO processes
• Dynamic choice of consumption/production
• Arrow instance
• Choice of transport via the Transport class

Disadvantages:

• No pure interface!
• No expressive return type encoded in the Churro datatype (could this be addressed?)
• Limited ability to perform lock-step computation (although this is by design)

See Hackage for more info!

Examples

See ./test/ directory for more extensive examples.

import Control.Churro

main = do
   runWaitChan        $ sourceList [1..10] >>> processDebug "after source" >>> delay 1 {- seconds -} >>> arr succ >>> sinkPrint
   wait =<< run @Chan $ sourceIO (\cb -> cb 1 >> print "Doing whatever!" >> cb 5) >>> filterC (> 3) >>> sinkIO print

Testing

Cabal test-suite including doctests:

cabal exec cabal test

Or for itterative development:

find {src,test} | entr -- cabal exec -- doctest -isrc -itest test/Churro/Test/Examples.hs

TODO

• Recovery/Retry capability
• Fix await deadlock
• Generic Chan functions, then specific newtype
• Stop using list functions
• Different transport options, buffered, etc.
• Write doctests for functionality
• Get Doctests working as part of the cabal test-suite
• Bundle in/out channels in the Transport class to allow Unagi to implement it
• Create profunctor instance
• Create contravariant functor instance
• Create ArrowChoice instance
• Create ArrowLoop instance
• Allow returning of results from run functions
• Get haddocks rendering correctly - Including contents
• Different transports for sections of the graph
• Allow configurable parallelism
• Early termination if downstream consumer completes
  • Ensure that infinite lists work when partially consumed