Add a function to upcast UniquePtr

[bsilver8192]

Upcasting can change the value of a pointer in C++. This means anything
upcasting needs both the parent and child class defined, which is
awkward in user code when wrapping functions that accept ownership of a
superclass object (it requires C++ code that depends on
autocxx-generated C++ code that depends on the main C++ code, which then
needs to be depended on by another iteration of autocxx to call it from
Rust).

Ideally this would be a static function in C++ to avoid some potential
name conflicts that this approach produces (with C++ types sharing a
name in separate namespaces), but I don't think cxx supports those yet
(dtolnay/cxx#1036 and dtolnay/cxx#447).

[adetaylor]

Thanks, this is great!

One question before I review it in detail.

Ideally this would be a static function in C++ to avoid some potential
name conflicts that this approach produces (with C++ types sharing a
name in separate namespaces), but I don't think cxx supports those yet

This is true, but autocxx does support them by making C++ wrapper functions. So you could do that too, if you felt it were worth it. I suspect that we'd just run into similar name conflict risks though.

[bsilver8192]

Ideally this would be a static function in C++ to avoid some potential
name conflicts that this approach produces (with C++ types sharing a
name in separate namespaces), but I don't think cxx supports those yet

This is true, but autocxx does support them by making C++ wrapper functions. So you could do that too, if you felt it were worth it. I suspect that we'd just run into similar name conflict risks though.

That's referring to the upcast wrapper function. I think using the same naming convention as wrappers for static functions would avoid name conflicts on the subclass type. However, now that I think about it, the same problem happens on the superclass type, which is different than static functions because the C++ name needs to be unique (vs static functions already have a class name that's unique within the namespace to use as part of their name). So it's actually more complicated to fully solve than I was thinking.

Do you think it's worth worrying about right now? If so, could you point me to the logic for naming static function wrappers?

[adetaylor]

I think this is fine as-is for now. Thanks again.