Specialize types pass

[rtfeldman]

This isn't trying to do anything with expressions yet; all it does is that you give it a Variable and Subs and it monomorphizes that variable.

This is based on the cor implementation at https://github.com/ayazhafiz/cor/blob/b67174d98b6dde16809b3606f9f88d4863d8743f/experiments/lss/monotype/lower_type.ml#L62-L63

[ayazhafiz]

Looks good generally. Is the purpose of monomorphize that you will eventually do all the cloning of expressions/specialization to exact types there?

In monomorphize_type, I would be sure to clone the type variable. You don't want to reuse the same one, especially if the variable is generic.

I would consider removing most of the tests for now, or using the uitest framework instead. I'm not sure specializing \x, y -> x y to ([] -> []), [] -> [] says very much. really, you'll want to see that the specialization produces specific instantiations.

[rtfeldman]

Yeah I plan to add specialize_expr which does the expr cloning part next, at which point I'll throw out these tests. I mainly just wanted to give myself a sense of this part working in isolation before getting exprs involved.

Great point about cloning the variable! I think the way we want to do that is to make a new var here: https://github.com/roc-lang/roc/pull/7149/files#diff-17f90baf2650b85b580fc6e1b09ed4277812b36c92697d17f2473dba5a5c615cR77

...and then change the next line to cache.set_monomorphic(new_var); return new_var; yeah?