[didibus]

There's just one type at compile time, but many more at runtime. This is still a strongly typed language. The only problem is our static analysers are too dumb to prove things without providing ample explicit hints, or changing the way we code to restrict certain ambiguities that it can not resolve at compile time. Haskell has chosen to try and push the boundaries of such static analyzer, but there's still limits, and it can't infer everything, and still restricts certain designs. I admire it for its efforts.

Clojure has a different strategy, it creates a new time, REPL time. So you can test your types at REPL time. Not when it compiles, but a little before it runs. It won't prove what you don't try though. So in practice, its using a statistical model where the programmer is the heuristic. You best guess the edge cases, and try them at REPL time. This will not catch all static errors, but will also catch some runtime errors. So it creates a disjoint set of errors that it catches. This is a trade off. Static types and REPL time will catch some of the same things, but also different errors.

Now both adding static type info, and doing REPL time testing comes to a cost to the programmer. Its one more thing we have to do. Some like me, fond more value most often at the REPL, it helps me explore and innovate my code, and is just more fun to me. I also prefer the kind of bugs it catches. Others think the opposite.

What most people seem to agree on though, is that doing both is way too much effort. That's why you don't have REPL time be a popular activity in Haskell, or core.typed be popular in Clojure.

[wtetzner]

> That's why you don't have REPL time be a popular activity in Haskell

Is it not a popular activity to use the Haskell REPL (ghci)? I though it was pretty common to use it when developing code, though I admit I don't have any hard data.

[brandonbloom]

The Haskell REPL is pretty good for a static language, but the experience is dramatically different to how a Clojure programmer would use want to use it. To be fair, Node and Python also have totally not usable REPLs for this style.

[bad_user]

GHC's repl is completely fine.

The thing with dynamic languages is that the development style is basically println-driven.

It goes like this: because you can't keep anything longer than a 1-page script in your head and because you can't remember the APIs of other people and hence you can't trust anything you write, in order to keep some sanity, you have to execute every freaking line of code that you write in order to verify that what you wrote actually works — and the sooner you execute, the better, because if your program crashes, the triggered error can happen far away from where the mistake is actually made.

This happens for every dynamic language, not just Clojure. This is why the read–eval–print loop is so important.

However the development experience changes dramatically in a good static language (no, not talking of Java or Go), because you can write more than one line of code before feeling the need to verify it — when compiler type checks a piece of code, at the very least you can be sure that the APIs you used, or the shape of the data you're interacting with are correct.

Refactoring is also painless. Ever done refactoring of projects built on dynamic languages? It's a freaking nightmare and no, the tests don't help that much, the tests actually become part of the problem.

This is also why dynamic languages folks complaining about long compile times are missing the point — those long compile times are necessary to give you guarantees that in a dynamic language you don't get at all, changing the experience, because in turn you don't have to run your code that often.

[didibus]

Since your specific about your statically typed language, which I'm assuming you mean Haskell. Are you also specific about your dynamically typed language? Are you talking specifically of Clojure?

Its unfair to club Clojure and imperative object oriented dynamic languages together. The same way its unfair to club Java and Haskell together.

You're right about the print-ln style. You do run your code everytime you touch a single line. That's what I like about it. But its a personal preference, like some people prefer to compose music on a sheet, others rather have their instrument in hand.

And you're forgetting the trade offs. With haskell, you wrestle the compiler, and every line you write has a compile error at first, until you get it right. This takes as much time if not more, at least for me, then it does running each of my lines of code in my REPL.

I guess I fall in that category where I kind of enjoy the beauty of both, though at the end of the day, I find myself having more fun coding when writing Clojure.

I've never suffered from a Clojure refactoring. You have to be a little more careful, but its never been that painful to me. Again, could be how I perceive "coding pain" is different from others.

[brandonbloom]

These are old tired arguments.

I prefer being forced to keep my program simple by making complexity intolerable over encapsulating it. Your preference may differ.

I find I have to refactor my dynamically typed programs less frequently than my statically typed ones. Your mileage may vary.

No amount of type safety will prove my game is fun, or that my user can understand the UI. I want fast iteration times, since I can’t wait on the compiler to test a new enemy behavior or GUI layout.

[wz1000]

> No amount of type safety will prove my game is fun, or that my user can understand the UI

No, but what it can ensure to some extent is that your game runs, and doesn't crash randomly. If the game crashes constantly, no one is going to play it no matter how fun it is.

[didibus]

to some extent

That's the keyword here. Maybe I'm missing critical data, but I've never perceived the reduction in defect from Clojure to Haskell. I've looked for studies on it, and they all point to either no difference or incredibly close. Never I've been shown a case where the reduction in defects would have an impact on the business I work for. Enterprise software is a domain that isn't that sensitive to defect. Anything less then 5% difference would go unnoticed, and affect in no way sales.

My conclusion, it comes down to your own enjoyment. Which one do you have more fun using and are the most productive in, that's the one you should be using.

I allow muself to change my mind if Haskell really proves to be 10% to 30% or more lower defect, maybe in a later version, with some GHC extension, maybe liquid haskell, I'm not closing my mind to it if it happens I'll be there.

[tome]

> since I can’t wait on the compiler to test a new enemy behavior or GUI layout.

Clojure is compiled, isn't it?

[brandonbloom]

Yes, but it’s fast!

[tome]

If no other language has any functionality similar to how Clojure does things then I think we'll need references to explanations or videos before we can even begin to understand your claims!

[brandonbloom]

Clojure is following in the tradition of lisps that do this right. The important bit is the extra indirection on top-level names. I wrote more about it here a long time ago: https://www.brandonbloom.name/blog/2012/12/21/the-nodejs-rep...

Also interesting is the other end of the spectrum: Forth. Instead of mutation, offers snapshots and restores of the “dictionary”. See this video: https://youtu.be/mvrE2ZGe-rs

[tome]

I don't fully understand what you're getting at. It would have been nice to see no examples with code entered at a REPL and the results, in both JS and Clojure, say.

Are you saying that you want to be able to make bindings that are refreshed on REPL reload? For example if I have a file that contains

    x = 1

and in my REPL I write

    y = x + 1

then I change my file to say

    x = 10

and reload the REPL then y is 11?

[didibus]

Well, the three key aspects, in my preference order, are:

1. Server repl with editor integrated clients. So your text buffers in your favorite editor is the repl. Look at the gifs here https://atom.io/packages/proto-repl to give you an idea for it.

2. Reifed language constructs. You can read about it here http://www.lispcast.com/reification . An easy example is if you have fn A depend on B. If you change B and call A, A will use new B. That's because the information is still availaible at runtime for A to figure out the latest version of B when calling it.

3. Functional programming / emphasis on small independent code blocks that compose. This is where you hear things like immutability, functions that take functions, purity, side effect free, managed references, etc. Basicly state in Clojure is hard to corrupt. That means if you alter state in your repl, it rarely messes up the full state, allowing you to keep working long sessions with your app state still being valid and usable.

I don't have a link for #3. So I'll give an example. Say you have a map you want to add data too. Say this map is read by something else, but you want to try adding something deeply nested to it. In Clojure, you can try as much as you want, experiment until you succeed to mold the map the way you wanted. The other thing reading the map never saw any of your changes, because it sees an immutable view of it. So after your done, if you use that other thing, it'll still work, because you didn't mess up the state it was depending on.

[brandonbloom]

    (require '[foo :refer [f]])
    ; edit f in foo.clj
    (require '[foo :reload])
    (f 1) ; should call NEW f.

Node doesn’t have a reload construct. If you hack it in by mucking with the module cache, you still won’t get the new f in your module’s local copy of it.

[didibus]

Other languages do, like Common Lisp, Racket, and most other Lisps. I hear dylan and smalltalk do to, but I don't have first hand experience.

This page https://clojure.org/about/dynamic explains it well.

The difference basically is that the mindset is to work within a running environment, swaping things out as its running. Its closer to a Jupyter notebook, or an excel sheet in some ways, if that helps you visualize it.