[mrkgnao]

Demonstrate a pair of tutorials aimed at the same level of reader, one in JS, and one in Haskell, where you end up with a better understanding of the idea of a monad reading the former (as opposed to the implementation of the Monad instance for a list).

[dmitriid]

TIL there's some "idea of a monad".

Basically this is (in my mind) what's wrong with Haskell: it's overly concerned with the Platonic ideal.

Meanwhile that one page on jargon has shown me that I effortlessly implement any and all of those things daily (and understanding what I'm doing) without the need to understand "an idea". I just use the tool that solves the problem. If someone insists on calling this "monadic composition", or "lifting over typeclasses", or "zygohistomorphic prepromorphisms", so be it.

[mrkgnao]

It's code reuse for concepts. Wouldn't you agree that reusing intuition about things is good? It's the same as knowing what big-O is instead of just memorizing "bubble sort is slower than insertion sort, insertion sort is sometimes faster than quicksort but usually not", or knowing what concurrency is instead of memorizing the API of a library in your favorite language.

[dmitriid]

The entire "concept" of a monad fits in that description I linked to. It can be easily reused (which I've done numerous times with it, and with other concepts on that page).

Haskell for some reasons insists that I should only go for "The concept of a monad, which arises from category theory, has been applied by Moggi to structure the denotational semantics of programming languages" and

  A monad is a triple (M,unit,⋆) consisting of a type constructor M and two operations of the given 
  polymorphic types. These operations must satisfy three laws given in Section 3.
  
  We will often write expressions in the form
    m ⋆ λa. n

Should I? Really?

[mrkgnao]

Yes, you should care about the laws!

They allow you to edit code without the aforementioned released-last-week test runner having to check all your code after a big refactor. I mean, you can't call something with a "flatMap" method and a "return" method a monad! There are tons of nonsensical definitions that fit that which are going to become very unpleasant to use quickly.

[dmitriid]

> They allow you to edit code without the aforementioned released-last-week test runner having to check all your code after a big refactor.

Oh wow. How did I ever live with big refactorings before?

> There are tons of nonsensical definitions that fit that which are going to become very unpleasant to use quickly.

I recently learned that, apparently, I've been using "monads" and "monadic composition" for years now, never knowing what it is. Can't remember any unpleasantness that would "quickly arise".

There are other things than blind following after a Platonic ideal.

[mrkgnao]

The unpleasantness arises when writing Monad instances that don't follow the laws. Setting your disingenuity aside, try using

  xs >>= f = concat (reverse (map f xs))

or

  Array.prototype.chain = function (f) {
    return reverse(this.reduce((acc, it) => acc.concat(f(it)), []))
  }

(I don't know how JS works, but you get it) instead of

    xs >>= f = concat (map f xs)

and enjoy refactoring your code, blissfully and consciously ignorant of the laws that make a monad a monad. The laws aren't just supposed to enrich the "life of the mind".

[KirinDave]

> Basically this is (in my mind) what's wrong with Haskell: it's overly concerned with the Platonic ideal.

I read this and I think what's got you rustled here is that Monad is such a generic concept. It's quite higher level and so you can do novel things like write functions that don't know how they're executing, just that they are.

As an example:

    -- Config is a typeclass that enables getting a
    -- keyval from a config. The return type is MonadIO
    -- because config might need IO.
    loadTarget :: (MonadIO m, Config a) => a -> m a
    loadTarget config = do
        v <- grabKeyVal "host" config
        w <- grabKeyVal "port" config
        return (v,w)

What does that code do? The answer (if it's written carefully) is that it depends on what the underlying monad is! And that's a good thing, in many cases. If the monad is Maybe + IO, then you have a conditional loader.

But if the monad is an array and IO then you can specify many hosts and many ports and this code enumerates them all. If that's passed to a ping function like so:

    loadTarget config >>= pingHostPort

    -- alternatively

    doPings config = do
      hostPort <- loadTarget config
      pingHostPort hostPort

Well then your code will do the right thing, but a different thing, based entirely on the types alone! And you can generalize this out to even more powerful types. For example, you could write a web app that server side could do local network ports for you (why? you're a maliclious hacker of course!). In that case it might make sense to use the Continuation monad.

tl;dr and finally:

You say this is stupid abstract stuff, but the folks delivering features to you in the Javascript world disagree. You have generators now, which are a much more real and fair explanation of how to model monads in Javascript than that silly code snippet you posted that doesn't capture the spirit of them at all.

What's more, careful application of these concepts leads to libraries which are just better than anything you can have without appealing to generators. A great example of this Purescript-config. Here is an actual (redacted) same of some code I use at work in an AWS Lambda function to read the environment:

https://gist.github.com/KirinDave/9af0fc90d005164743198692f3...

So I have complete error reporting (full sets of missing keys) just by describing how to fetch the values from the environment. I can transparently switch to any other file type under the covers by changing from using "fromEnv". I only know that there is a key-value store in there.

Doing this in OO is really, really hard to get right, because imperative OO code cannot easily parameterize the execution strategy hierarchically without appealing to generics and ad-hoc polymorphism. That's hard.

The applicative style adopted here is very simple to re-interpret, because we can parameterize code on monads and applicatives (which explain the execution strategy of this code beyond its coarse structure) as we see fit.

You can do that with generators but it's frustratingly hard. Doing it in a truly generic way? Even harder. For this approach, the results are free (and Free, but hey).

I can give you other examples of how Purescript makes certain difficult aspects of javascript simply vanish, if you'd like.

[dmitriid]

Thank you for examples.

I have one complaint though:

> Doing this in OO is really, really hard to get right,

Why do you equate FP with monads? Moreover, why do you equate FP with Haskell/Purescript (statically typed FP with monads)?

[mrkgnao]

The use of monads is a side-effect (ha!) of committing to purity throughout a language, and that's what FP is being equated to: pure statically-typed FP.

(You can argue about how justified that is, of course. I'm not going into that, but you might want to see what John Carmack has to say[0]. No, he doesn't end by saying "we should all convert to the church of Haskell now", but he does talk about how large-scale game programming refactors are made easier when you're working with no (or very disciplined) side effects.)

Monads are not the only way to deal with effects while keeping purity, although they were the first discovered and so on: algebraic effect systems as in Koka[1] (and as simulated in Idris or PureScript) are another alternative. Koka infers effects, so it's probably easier for a C-family programmer to pick up (I know little about it, though).

[0]: https://www.youtube.com/watch?v=1PhArSujR_A

[1]: https://www.microsoft.com/en-us/research/project/koka/