The original plan was to have the inference pass mutate the AST as it eagerly unifies sorts. However, this seems to be slightly brittle and conflates the thing we want to do now (typechecking) with what we want to do later (know the concrete instantiation of some program term).
Might be easier to just take a final pass after inference; for every symbol, and the Resolver bindings that we've accumulated in the first pass, patch everything up.
dijkstracula
commented
1 year ago
The original plan was to have the inference pass mutate the AST as it eagerly unifies sorts. However, this seems to be slightly brittle and conflates the thing we want to do now (typechecking) with what we want to do later (know the concrete instantiation of some program term).
Might be easier to just take a final pass after inference; for every symbol, and the Resolver bindings that we've accumulated in the first pass, patch everything up.