Jon Pretty

tl;dw you can generate values of some unknown type along with evidence that the type is a monoid or whatever, but without knowing or caring what the type is

I want randomly generated, possibly recursive, data types as inputs for type class derivation.

AIUI @d6 technique is "suppose I have some unknown type T with along instances of F[T], G[T] etc. ...". OTOH, I want to generate types T, U, V ... then derive instances of F for them. That means I need the structure of the types, they can't be completely opaque.

This sounds super interesting. Maybe if the type class being witnessed is itself `Arbitrary`, this could work?

Well, ideally Arbitrary would be one of the F's we can derive instances of! The guts of my problem is a finite tree/graph generation problem. The tricky bit is mapping those into types which can feed into derivation infrastructure.

Presumably the actual combinatorics of this are well-understood, and it's the technicalities of performing the mapping in Scala that are the issue?

Exactly. We need to generate types which can feed into downstream compilation. It's doable with current mechanisms, but clunky and very time consuming.

So, I have an idea but it’s really horrible and you’re all going to question my sanity. And it’s probably wrong too. This is for tests, so generate source prior to compilation that will be compiled with the tests. Your generator can pick from the randomly generated types.