Crosscut

There's one last aspect to normalizing variant types (in addition to sorting and deduplication), that we'll probably need in the language: flattening of nested variants. That would mean that variant { A, variant { B, C } } would become variant { A, B, C }.

Assume we have a function A -> variant { B, NetworkError }, and follow that up with another function B -> variant { C, FileNotFound }. Due to automatic lifting, that would result in a variant { variant { C, FileNotFound }, NetworkError }, which seems unnecessarily complex. And consider, chaining more functions could make that much worse.

I think the problem becomes more pronounced, if both functions return the same error, for example resulting in variant { variant { C, FileNotFound }, FileNotFound }. What even is that? If we flatten it to variant { C, FileNotFound, FileNotFound }, we can then deduplicate, resulting in variant { C, FileNotFound }. Much more reasonable!