Possibly related to the existing hacks around ?? in the type checker. cc @hborla, @xedin

@swift-ci create

The problem here as correctly mentioned @CraigSiemens is related to the fact that we don't unify types by their protocol conformances unless that protocol is explicitly specified, that's why e.g. `A() ?? B()` can't be type-checked into ??<P>(A?, B?) because of that. This is expected behavior.

We have an explicitly-specified protocol: the result type.

For the simplest comparison, let value: P = B() ?? C() works.

@belkadan It might work for a single operator but not for linked operators, binding collections can't really propagate contextual type which means that bindings for arguments would never have `P` as a possible type. We have talked about attempting protocols before but it leads to other failures. Unfortunately we can't solve this without rework of how solver infers possible types for type variables.

I don't know what you mean by "linked operators" or "collections". There are no collection literals here and all of the working examples Craig provides show that this is clearly a bug. It may be hard to fix, but it's hardly invalid.

I mean expressions like "1 + 2 + 3" where result of one operator is an argument to another one. This is a problem which might always require specifying protocol explicitly because fixing it means a revamp of the solver.