Module linker

[klntsky]

Let's discuss how to handle modules.

PureScript compiles code on per-module basis, because it has to maintain per-module FFI interfaces.

However, in our context we are always interested in just one function - a validator script.

We need to find a way to tie modules together.

Let's start with an obvious approach and then try to improve on it.

I propose embedding modules into CoreFn. Conceptually, a module can be seen as a function from a record of its imports to a record of its exports. Modules can't be co-recursive, so every valid module set can be transformed into a chain of let-bindings.

The downside is that modules are not zero-cost with this approach. Can we improve it?

[gnumonik]

In light of some conversations with the IOG folks, I'd like to propose an alternative:

1. Instead of modifying the PureScript compiler to emit PLC directly, we modify it to emit PIR "Modules", where a PIR Module is (essentially) a list of PIR Bindings. (PIR "Modules" don't exist, I'm speculating about a data structure we'd have to create)
2. We implement our own, external, final compilation machinery (i.e. we don't modify the PS compiler to do this) which operates on PIR modules.

I think that it might be possible to derive a PIR "Module" from a PS ExternsFile (which is a build artifact produced regardless of the codegen target). If that is in fact possible, then we don't need to modify the PS compiler's compilation pipeline at all, which would be a great boon for maintainability.

[gnumonik]

Oh and also, all of their optimization tools work on PIR and I believe it is generally the case that PIR is amenable to more/better optimizations than we could do with PLC