[DNM] Fix wide-string override_ty

[fw-immunant]

This isn't suitable for merging, but shows the constraints we're working in--namely, we need to synthesize a new CTypeId (for the appropriately-sized array type of wide-chars, to replace the same array of c_ints) in a place where the TypedAstContext is not mutable.

[Rua]

I was thinking about this but had a different approach in mind. Instead of trying to synthesize a new type, just pass an override for the pointee/element type to convert_array_to_pointer_decay (and only there). For array decay it's always the same pattern between source and target types, where [foo; N] turns into *const/mut foo. It's not even technically a cast, anyway.

[fw-immunant]

I was thinking about this but had a different approach in mind. Instead of trying to synthesize a new type, just pass an override for the pointee/element type to convert_array_to_pointer_decay (and only there). For array decay it's always the same pattern between source and target types, where [foo; N] turns into *const/mut foo. It's not even technically a cast, anyway.

I believe the problem with that approach is that the val argument to convert_array_to_pointer_decay has already been translated (its type is WithStmts<Box<Expr>>) with the wrong override_ty.

Of course, we could address that via inversion of control, changing convert_cast to be responsible for the translation of the value to cast.

[fw-immunant]

This isn't suitable for merging, but shows the constraints we're working in--namely, we need to synthesize a new CTypeId (for the appropriately-sized array type of wide-chars, to replace the same array of c_ints) in a place where the TypedAstContext is not mutable.

One note - I used dashmap here in the hopes that it would let us insert new CTypeIds to the c_types map without disturbing the required Index impl which must return &CType, but since it doesn't actually do that (we have to leak a boxed CType to do so because DashMap returns its own Ref<V> type rather than &V), dashmap is not really helping compared to a RefCell<HashMap>, and what would actually solve the Index problem is an append-only, address-stable hashmap. That should fix the hackiness side of things here.

[Rua]

Rather than trying to synthesize a completely new array type here, could convert_array_to_pointer_decay not be modified to take an element type instead of source and target types, and then called directly?