[aidenn0]

Static typing does not prevent this; the fact that static typed languages can parse freeform formats like JSON works as a simple example.

If you statically prove that you check that they keys are there before using them, then the type system can be satisfied.

[yogthos]

You don't know what the keys are up front because each library is an independent module. Of course, you could just use Any type, but then you're not doing static typing.

[aidenn0]

Example:

Module X expects the "foo" key, then write a function that checks if it has the foo key and returns a type Maybe(HasFoo). It's less good than just treating the type as the union of HasFoo and HasBar or whatever, but if you fully decouple, then you have to be able to handle the case in which the mapping lacks a foo key anyways, and this will statically froce you to.

[yogthos]

But the point is that you don't know the totality of the keys up front. Module A might deal with keys X and Y, while module B deals with keys W and Z. These two modules know absolutely nothing about each other and they're developed by different people.

[aidenn0]

Right, so Module A defines a "TypeA" (which describes a map containing keys X and Y) and Module B defines a "TypeB" (which describes a map containing keys W and Z) and they each define their respective functions that return a (Maybe Type A) and (Maybe Type B). If the map has keys X and Y then you will get a (Just Type A) and if it does not you will get (Nothing). The type system will then enforce that you handle the possibility of (Nothing) which is something you would need to do in a dynamically typed language as well.

Remember, in the sense we are currently talking about[1], objects are untyped, it's variables that are typed. So there is no need to describe the type of this map that is passing around in concrete terms; it's perfectly cromulent to treat the map as a TypeA in one place and a TypeB in another place, so long as you ensure that the prerequisites for those types are satisfied by the object.

It is true that it is usually preferable in statically typed languages to demonstrate that the map satisfies the requirements for TypeA before type erasure (i.e. at compile time), but if you want total decoupling, then that is not possible (since the requirement that the map have keys X and Y would need to be enforced outside of Module A). This does not mean that a type system cannot help you though.

I think we are getting to the limits of what can be easily communicated via HN comments, so if you still don't understand, then perhaps I'll write a blog article explaining it more fully.

[yogthos]

I'm saying that there is no way to know what all the keys in the map are going to be up front, or what are all the types these keys can have.

For example, there is middleware for parsing our request params into a :params key. There's another piece of middleware that parses the values of these keys. So, you may or may not have a :params key, and the types inside the params can be absolutely anything. Then you could just have completely separate middleware that might add something like a CSRF token to the request map. And so forth. Your only real option here is to treat the entire structure as Any type.

If you still don't understand the problem, I really don't know how else to explain this.