Re-implement rIso?

[conklech]

The lens removal (#23) left rIso commented out, so you can't actually do anything with the knowledge that r1 :~: r2. (Edit: This is of course wrong; you can cast both ways.)

To export a lens-compatible Iso', we need to incur a dependency on profunctors. (See the lens wiki.) Alternately, we could export cast. (Edit: We've always exported cast.)

-- | In 'lens' parlance,
--
-- > rIso :: (r1 :~: r2) => Iso' r1 r2
rIso :: (r1 :~: r2, Profunctor p, Functor f) => p r1 (f r1) -> p r2 (f r2)
rIso = dimap cast (fmap cast)

[acowley]

First, thanks for this. This is clearly the right way to provide a lens-compatible iso. In general, though...

Is this a useful definition? I'm honestly not sure, which is why it is distinguished by living in a zombie commented out state. We do export cast, so the potential user is not stranded, but losing the definition didn't impact me at all. So we can add a profunctors dependency to support this definition, or we could make rIso part of the documentation.

If we pick up a profunctors dependency for other reasons, then this is a nice thing for vinyl to define.

[conklech]

...you're right. We've always exported cast, and rIso is dramatically less useful than cast. Sorry, my brain wasn't turned on.

The use case for rIso does seem a bit improbable. Perhaps libraryA uses Rec '["foo" ::: t, "bar" ::: t] and libraryB uses Rec '["bar" ::: t, "foo" ::: t] and you want to interoperate between them?

[jonsterling]

Casting should suffice for records which have different order, I think... Is there any other use-case we'd like to address?

[acowley]

I'm going to close this for now, but it should be reopened if more evidence for rIso's usefulness comes to light.