[akurilin]

For me it's mostly about codebase scaling, refactoring and maintainability. Clojure is liberating and exciting when you're writing a tiny little project, but it's a whole other experience when you need to refactor dozens of files because you changed the format of the data being passed around, or you're changing an internal API that's called from a hundred different places.

You better have perfect code coverage, or you'll have no clue why and where something broke (the sink/source problem) or perhaps you won't even find out for a while because that scenario wasn't sufficiently tested and it slips into production. Having a compiler nag you about type inconsistencies is incredibly helpful in these scenarios.

The other big one is working with large blobs of data. Our product has a large analytics component to it, and massaging giant, deeply nested maps representing a certain compendium of statistics is really tough without the compiler spotting you. None of this is an issue when you have to satisfy a certain type, the compiler will basically give you a checklist of things to fix when you change something.

With Haskell you're getting all of the benefits of Clojure (expressiveness, leverage etc), plus the really useful addition of types and enforced purity on top.

Here's another clojurian's experience with switching to Haskell: http://bitemyapp.com/posts/2014-04-29-meditations-on-learnin...

[platz]

Would Prismatic/Schema address those data typing issues?

[akurilin]

I'm actually a Schema user, I love it, it's pretty convenient for addressing the boundary issue, since you don't have types and cannot really enforce that outside input conforms perfectly to the expected schema without run-time validation.

Schema helps (I used to use Mississippi back in the day with all sorts of complex checks, it was ugly), but it's more of an "add-on" in the end. I actually have a reasonably convenient flow for validating outside API input, business logic constraints and performing the necessary operations against the data store all in one big error monad nowadays. The problems is that I still have to go out of my way to add these input/output validation checks to everything when I could just get it for free through the type system.

Look, don't take my word for it: I strongly encourage you to try the two paradigms yourself and see what you think. Unfortunately it's very difficult to see the downsides without a large project, but you can probably extrapolate.