Applicatives still make sense with only monoidal structure: they're monoidal functors `C -> Set`, or more generally `C -> V` for `C` enriched over `V`.
Applicatives still make sense with only monoidal structure: they're monoidal functors `C -> Set`, or more generally `C -> V` for `C` enriched over `V`.
Regardless I hope you can see what I mean when I say the mathematics you need for Haskell is wonky.
Applicatives are monoid objects for Day convolution of functors (in the same way that monads are monoid objects for composition of functors) given by the coend `F \otimes G (x) = \int^{i, j} Hom_C(i \otimes j, x) \otimes_V F(i) \otimes_V F(j)`. These coincide with closed functors by currying if we have closed monoidal structure, but remain well-defined without it.
Using `<*>` instead of `liftA2` in the typeclass definition is basically a historical accident, and imo a mistake.
> Regardless I hope you can see what I mean when I say the mathematics you need for Haskell is wonky.
Wonky is too strong, but I agree that it's usually not treated with anything near the appropriate level of detail.