Skip to content
forked from wkolowski/AlgEff

My seminar talk about algebraic effects

Notifications You must be signed in to change notification settings

DrRuisseau/AlgEff

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

24 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Algebraic Effects vs Monads in theory and practice

This repo contains slides and code from my seminar talk. For the best experience:

  • Read the slides.
  • Play with the code in the following order:
    • Interpreter/Pure/ - three interpreters for CBV lambda calculus with integers and addition written in Haskell, Idris and Koka. Compare these to get acquainted with the languages.
    • Interpreter/OneEffect - interpreters for LC with added error messages, operation count, nondeterminism and logging, taken from Wadler's "The essence of functional programming". First see the Haskell version for the "traditional" monadic approach, then the Idris version for poor man's algebraic effects version (using Idris' Effect library) and then the Koka version for full-blown algebraic effects.
    • Interpreter/AllEffects - interpreters for LC with all the above effects combined. In order, read:
      • All_Monads.hs - to see why monads don't compose too well.
      • All_Transformers.hs - for a solution using monad transformers (and why it's not that good).
      • All_Classes.hs - for a solution using typeclasses like MonadError, MonadState etc.
      • All.idr - the Idris solution showing that Idris's effects are not that easy to use.
      • All.kk - Koka solution which shows that built-in algebraic effects are awesome.
    • Free - a free monad example in Haskell (the package free is required for this) and a showcase for how much easier the same thing is in Koka.
  • Read the summary below.

Summary

  • Monads work well when we have only a single effect, but they don't compose.
  • Transformers compose, but they aren't very flexible, so refactoring them is a pain in the ass.
  • Classes compose and are flexible, but implementing them requires N^2 instance for N monad transformers. Running the effects in a different order than the most obvious one was difficult for me.
  • Idris' effects compose and are flexible, but they are very mysterious both to define and run properly. What's more, the type error messages Idris gives us are horrible and quite unhelpful.
  • Koka's algebraic effects are wonderful. They have a clean syntax, are conceptually simple, compose well, are very flexible and easy to program with in practice. They are the clear winner of this comparison.

My personal feelings: Koka's algebraic effects > Haskell's monadic classes > Idris' effects > Haskell's monad transformers.

TODO

  • Figure out how to get final state in Count.idr

About

My seminar talk about algebraic effects

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Haskell 47.9%
  • Idris 28.9%
  • TeX 23.2%