Move Over Free Monads: Make Way for Free Applicatives! — John de Goes

Optimizing interpreters for FreeR Monads by using a No-Op Monad & 2 passes? **Background** I've been thinking a lot recently about applications needing to be as state-machine-like as possible, and the necessity/benefits of algebras for expressing (and in effect helping limit) what applications can do. These concepts most directly translate to `Free`/`FreeR` `Monad`/`Applicative` in Haskell. The concept can be achieved with `mtl` + typeclasses as a polymorphic `m` give near identical functionality -- you can constrict what the `m` can do with typeclass-based constraints -- this has very similar benefits (and a bit of an escape hatch) but isn't as cleanly expressed IMO. Recently while readi…

Move Over Free Monads: Make Way for Free Applicatives! 00:45:29 by **John de Goes** (2016)

IMHO, OOP design patterns lack a punch compared to functional design patterns. Compare and contrast: - The Actor Model for concurrency; - Higher-order combinators for collections (`.map`, `.fold`, `.choose`, `.collect`, ...) which are the ancestor of LINQ; - ~~Monads~~ Burritos; - Applicatives, used for example in `optparse-applicative`; - Free structures as DSLs, particularly free monads and free applicatives; ...and probably many more which I forget right now. These are effective solutions to some very hard problems in software engineering, mostly relating to separation of concerns and testability.

We just finished converting an ad hoc collection of elasticsearch/geotrellis/spark rdd code to Free and while the long term results aren't in, the code is orders of magnitude more composable and modular. From what I've read/watched there are some limitations compared to other methods, but I think we'll be able to work around them creatively.

Sure, we'll look a few popular ways of doing effects in Scala right now. With the Free Monad you'll see `Coproduct` is used to present a list of possible capabilities(called algebras in this context). The `Inject` typeclass is then used to assert that an effect has all the capabilities in the list. The Freestyle project replaces `Coproduct` with it's own fast tagging system but contains the same idea. There's a coproduct which represents a lists of capabilities. `Inject` proves that certain capabilities are in the effect. With the Eff Monad we'll see the same thing. Similarly to Freestyle, Eff provides it's own coproduct. But instead of listing capabilities, Eff lists other monads. In the…

I really liked John de Goes' presentation on Free Applicatives: [video . It might seem that taking things up to the Free level would complicate things further, but I found the conceptual insights presented there very helpful in clearly distinguishing the underlying concepts.