Comment by lifthrasiir

That's an interesting approach. So the variant name itself doesn't bear the type information but should be uniquely resolved in the type check time, right? It will be pretty much optimal unless you are doing metaprogramming stuffs.