I ask this because I've been eager to get my hands dirty with Clojure. I have experience with other functional languages (Haskell, OCaml, F# and some Emacs Lisp), so the paradigm isn't new.
I've looked at core.typed, but doesn't seem as neat as the syntax of Typed Racket. It's something, though. Is it checked at compile time?
What irks me is that static typing and type inference can make code feel really robust, and for the lack of a better word, safe. In Clojure I see a lot of nice things but the lack of typing, though common for Lisps, always bothers me a bit. I don't like runtime errors that happen because the compiler wasn't able to tell me that this object doesn't have that method. JavaScript's undefined is not a function or its kin in Python are examples of this behaviour, which I don't like.
Though, I see that Clojure's answer to this is rapid REPL development--which is great!--and unit testing, and what I've recently discovered, pre- and post-conditions. But it still feels somehow inferior.
I am torn between learning Scala or Clojure. Knowing OCaml and F#, Scala doesn't look that interesting, messy and multi-paradigm. Clojure appeals to me because its a Lisp and has one paradigm, but on the other hand I'm scared by runtime exceptions.
Should I just ignore my trepidations and proceed?
Yap give it a try.
From what I understand Core.typed while not as rigorous as static type system in Haskell or OCaml can let you gradually add typing to your application. They more you add the greater the benefit.
But also personally haven't used it.
I used Dialyzer in Erlang, which is a similar concept. You annotate your code with types and the more you do the more type errors it will find for you:
http://learnyousomeerlang.com/dialyzer
It is surprisingly good. Here is example of production code in Erlang with some type annotations (it is a websocket handler from Cowboy a webserver):
https://github.com/ninenines/cowboy/blob/master/src/cowboy_w...
Notice the type and -callback declarations at the top then the -spec lines before some functions.
While they may have a small number of engineers (20 or max 20?), their engineers are probably laser focus in delivering the product (re: no web-app for the longest of time, no RoR vs Django vs Clojure vs Java vs PHP, no React vs Backbone vs etc etc), just back-end: Erlang, front-end: various phones that needed to be supported.
I would attribute their success at their skill set and experience. If you watch Rick Reed presentation in MeetBSD, a few key points to note: Rick Reed is very smart and knows full-stack (minus the browser): from the OS (internal, driver, etc) up all the way to Erlang VM and the WhatsApp backend.
The team had to modify FreeBSD and patch Erlang VM whenever necessary. How many companies have _that_ kind of talent? (e.g.: modify Linux and JVM to make them run faster? or Ruby implementation and PostgreSQL/MySQL while supporting the actual product?). Rick himself has tons of experience writing distributed systems at Yahoo! (using C++ nonetheless...).
While Erlang helps them but in reality, their skill + experience matter more.
So when you watch cat videos on your smart phone, there is about 50% some non-statically compiled functional code has done its job in setting up the necessary signaling.
WhatsApp has also recently been the poster child for running a successful distributed platform, with a tiny number of engineers (about 10 or 20) by also leveraging a non-statically compiled and functional language.
In general static type checking is nice, I like it. But clearly not a deal breaker. O often hear "I won't touch anything unless it has static typing" or "large applications should have static types".
But maybe the question to ask, should applications be that large? Lately we have been all upvoting microservices. And if anything those encourage decoupling large monstrous application into smaller independent services. At that point there are just a soup of dynamically typed components.