• Stars
    star
    119
  • Rank 297,930 (Top 6 %)
  • Language
    Haskell
  • Created about 8 years ago
  • Updated 4 months ago

Reviews

There are no reviews yet. Be the first to send feedback to the community and the maintainers!

Repository Details

Haskell Functional Reactive Programming framework with type-level clocks

README


Build Status Version on Hackage

Rhine is a library for synchronous and asynchronous Functional Reactive Programming (FRP). It separates the aspects of clocking, scheduling and resampling from each other, and ensures clock-safety on the type level.

Versions 1.* vs. 0.*

Confused because some examples from the article don't work anymore? As a big simplification and breaking change, explicit schedules were removed in version 1.0. For an overview of the required changes, see this page.

Concept

Complex reactive programs often process data at different rates. For example, games, GUIs and media applications may output audio and video signals, or receive user input at unpredictable times. Coordinating these different rates is a hard problem in general. If not enough care is taken, buffer underruns and overflows, space and time leaks, accidental synchronisation of independent sub-systems, and concurrency issues, such as deadlocks, may all occur.

Rhine tackles these problems by annotating the signal processing components with clocks, which hold the information when data will be input, processed and output. Different components of the signal network will become active at different times, or work at different rates. If components running under different clocks need to communicate, it has to be decided when each component becomes active ("scheduling"), and how data is transferred between the different rates ("resampling"). Rhine separates all these aspects from each other, and from the individual signal processing of each subsystem. It offers a flexible API to all of them and implements several reusable standard solutions. In the places where these aspects need to intertwine, typing constraints on clocks come into effect, enforcing clock safety.

Example

A typical example, which can be run as cd rhine-examples/ && cabal run Demonstration, (or using nix flakes with nix develop followed cabal run Demonstration), would be:

  -- | Create a simple message containing the time stamp since initialisation,
  --   for each tick of the clock.
  --   Since 'createMessage' works for arbitrary clocks (and doesn't need further input data),
  --   it is a 'Behaviour'.
  --   @time@ is the 'TimeDomain' of any clock used to sample,
  --   and it needs to be constrained in order for time differences
  --   to have a 'Show' instance.
  createMessage
    :: (Monad m, Show (Diff time))
    => String
    -> Behaviour m time String
  createMessage str
    =   timeInfoOf sinceInit >-> arr show
    >-> arr (("Clock " ++ str ++ " has ticked at: ") ++)

  -- | Output a message /every second/ (= every 1000 milliseconds).
  --   Let us assume we want to assure that 'printEverySecond'
  --   is only called every second,
  --   then we constrain its type signature with the clock @Millisecond 1000@.
  printEverySecond :: Show a => ClSF IO (Millisecond 1000) a ()
  printEverySecond = arrMCl print

  -- | Specialise 'createMessage' to a specific clock.
  ms500 :: ClSF IO (Millisecond 500) () String
  ms500 = createMessage "500 MS"


  ms1200 :: ClSF IO (Millisecond 1200) () String
  ms1200 = createMessage "1200 MS"

  -- | Create messages every 500 ms and every 1200 ms,
  --   collecting all of them in a list,
  --   which is output every second.
  main :: IO ()
  main = flow $
    ms500 @@ waitClock            --  a Rhine = a ClSF in the context of a Clock
    |@|                           --  compose 2 Rhines in parallel
    ms1200 @@ waitClock           --  a Rhine at a different clock
    >-- collect -->               --  buffer results from both Rhines into a list
    printEverySecond @@ waitClock --  the final Rhine

  -- | Uncomment the following for a type error (the clocks don't match):

  -- typeError = ms500 >>> printEverySecond

This repository

  • rhine/: The main library, which is also mirrored on hackage.
  • rhine-gloss/: A wrapper library to gloss, a functional OpenGL library.
  • rhine-bayes/: A library for stochastic processes and online machine learning, using monad-bayes.
  • rhine-examples/: Different examples as a starting point to learn Rhine.

Documentation

The best way to learn about Rhine is currently the article Rhine: FRP with Type-Level Clocks.

For a quick reference of the most important concepts, see the cheatsheet.

Additional documentation

FAQ

  • Why does my blocking code, e.g. arrMCl readLn, behave erratically?

Clocks must be the only things that block a thread, not ClSFs. So for example, you can fix:

arrMCl readLn

by using:

tagS >>> arr read :: ClSF IO StdinClock () Int

tagS contains the string that the StdinClock grabbed from stdin, and only the clock has been allowed to block the thread!

  • Can a sampling schedule dynamically change, e.g. depend on a signal?

Yes, for instance you could implement a distance-dependent collision detector.

  • How to handle slow computations, i.e. computations that take longer than the sample rate?

Several strategies exist and it depends on your use case. For FixedStep clocks, it won't matter since the execution of the program isn't tied to a realtime clock. For ClSFs running on UTCTime clocks, you can execute the slow code in a separate thread and coordinate merging the results back into the signal network.

Development

See Contributing.md for details.

  • Rhine usually follows up-to-date GHC versions.
  • Contributions are welcome! There are always a few issues labelled help needed, in case you're looking for an easy way to get started.
  • Rhine is a beginner-friendly Haskell project! Even if you're new to Haskell and FRP, you can contribute. This is a good place to start contributing to open-source projects. Have a look at issues labelled good first issue. If you have questions, don't hesitate to ask on Github.

Related projects