Dear Heinrich, and all others!
Heinrich Apfelmus
Second, it may helpful to look at how others do it. Your writing indicates that are already very knowledgeable, so this advice may be not very helpful, but for the sake of completeness, I would like to mention it anyway. For example, the `haskell-src-exts` package implements a syntax tree for Haskell, maybe it can offer an inspiration or two.
Yes, I have looked it up, and iss design is very clean. Unfrotunately my data is not purely a syntax tree, but it contains lots of information related to data layout, code generation, etc. For example haskell-src-exts does not need to track type information, and it can refer to declared entities simply by their name. If I did the same, I would also need to implement full-blown C++ name lookup with symbol tables, which I don't want to do. GHC's sources might be useful though, but I haven't been corageous enough to try my teeth on them. Anyways, I'm very grateful for your pointers! I learned about anonymous sum types and about vault. GADTs composed with phantom types and type families seem to offer countless possibilities. I'm missing some introductory documentation though - something that is more practical than theoretical. (Yes, GADTs are quite well covered, but I haven't found a tutorial for practical use of data families or -XTypeInType, probably because they are quite new features.) -- Aura