Jaro ReindersIn particular, if we know a function will always be applied at a monomorphic type then we can always specialize it to that type. That is what we usually want for type variables standing for monads (which are terribly slow if unspecialised).
Jesper Agdakx 🔸@jaror Oh this would be a really nice feature indeed! If there was a way to annotate a type class with "actually make sure that these dictionaries are never passed at run-time" I'd use that a lot.
I would be fine with such a feature requiring mandatory SPECIALIZE pragmas for any specializations that are not used in the same file.
I would be fine with such a feature requiring mandatory SPECIALIZE pragmas for any specializations that are not used in the same file.
MangoIVI think the correct place to insert such a check would be when doing staged programming while going down a level - say we have some argument that represents code, it should be possible to declaratively insert assertions about what the result of evaluating that splice is; that is
- this expression evaluates to a concrete dictionary
- this expression evaluates to a concrete type
- this expression evaluates to a concrete runtime rep (to be able to do register allocation)
And it would not only make sense in the last stage - cross compilation could contain assertions like "the host and target platform's width of the following primitive types should agree"
An issue with this is that "monomorphised" is sadly not enough - you also need to make sure that you e.g. don't have any stuck type families.