Improve error message for impossible JoinKinds constraint

[adamgundry]

Fixes #423.

WARNING: typeclass Prolog ahead.

[arybczak]

Thanks. Do we want to drop support for GHC 8.2 and 8.4 (as this introduces UndecidableSuperClasses)? I personally don't mind.

EDIT: Never mind, for some reason I thought that this extension was introduced in GHC 8.6.

Also, doctests fail, presumably due to a missing import.

[phadej]

What will happen if a kind is forgotten, we get a "wrong" custom type error?

[adamgundry]

Yes, I was a bit dissatisfied with the need to list all the optic kinds explicitly, but I couldn't immediately see how to avoid it. Having thought a bit more I've got a version that doesn't need this.

[adamgundry]

I was playing around with this branch and noticed this:

> view @A_Getter (folded % _) [Just "abc", Nothing, Just "def"]

<interactive>:3:17: error:
    • A_Fold composed with l0 produces p0,
      but the context expects A_Getter

We should be able to do better here, by checking if more parameters are concrete.

On the other hand, we're a bit stuck if the context doesn't force a particular optic kind (cf. #308):

> view (folded % _) [Just "abc", Nothing, Just "def"]

<interactive>:4:1: error:
    • Overlapping instances for Is k0 A_Getter
        arising from a use of ‘view’
...
    • Overlapping instances for JoinKinds A_Fold l0 k0
        arising from a use of ‘%’

[arybczak]

What's the status of this PR?

[adamgundry]

I've been away on holiday so apologies for the radio silence. I'm still planning to polish this a bit, but it may be a little while before I can get back to it.

[arybczak]

@adamgundry I posted an alternative solution in #439, please have a look if you have a moment. It's not as fancy, but should be more predictable :)