For context, there’s currently a big thread about the type of ($), which is going to be more complicated in GHC 8.0 due to levity/runtime rep polymorphism. As I understand it, the intention is that this more complex type will be hidden unless a levity polymorphism flag is active. From my perspective, makes the question of whether this type is too complicated for beginners moot.

What puzzles me is that I still don’t understand what this feature is *for*. According to [1], levity polymorphism is a more principled replacement for OpenKind, which was a bit of hack. Fine.

[1] https://ghc.haskell.org/trac/ghc/wiki/NoSubKinds

OpenKind exists because developers wanted to use ($), undefined, and error with unboxed types. That’s understandable, but it doesn’t seem worth all this effort. Surely, having to use error# instead of error when working with unboxed or unlifed types is a small thing next to all the other differences.

OpenKind is also used when doing type inference, because arguments to functions might be * or #. That’s more compelling, but doesn’t seem like something that needs to be exposed to the programmer.

So, is that it? Are there less-trivial levity-polymorphic functions? Is this a step on a road to new features, like [2]? Is there an article or something that makes the case for this?

[2] https://ghc.haskell.org/trac/ghc/wiki/UnliftedDataTypes

Dave Menendez <dave@zednenem.com>
<http://www.eyrie.org/~zednenem/>