Add variant_internal.hpp.

[Ivorforce]

This module contains VariantInternalType, VariantInternal, VariantGetInternalPtr, VariantInternalAccessor and VariantDefaultInitializer, facilitating access and manipulation of a Variant's internal value in various ways.

The reference godot variant_internal.hpp is here. I chose a different, less boilerplate-y approach to implementation. It should work the same though. Doing it this way should decrease maintenance (and implementation) effort, since most of these are just copy-pasted code without any type specific logic (with the exception of Object *).

There are a few small differences:

• VariantInternalType does not exist in godot's implementation. I added it because it makes implementation of the other functions easier.
• get_object returns Object * const * in my implementation, but const Object ** in godot's. Technically speaking, godot's implementation is wrong, because it allows manipulation of the contained Object *, even though a const Variant was supplied. Godot uses a C-style cast that acts as a const_cast to avoid cast issues.
• VariantInternalAccessor is not implemented for Object *, even though it is in Godot. I believe it is not possible to replicate the intended behavior without access to Variant::object_assign (or internal ObjData). I made it fail statically using SFINAE.

I think the rest should be the same.

PR contains a unit test for one of the internal accessors.

Depends on #1654.

[dsnopek]

Thanks!

At a high-level, this looks great!

• VariantInternalType does not exist in godot's implementation. I added it because it makes implementation of the other functions easier.

In order to discourage developers from using it (and create incompatibilities with Godot modules) you could put it in the godot::internal namespace? We do that in some other places, although, we should probably use it even more than we do.

• get_object returns Object * const * in my implementation, but const Object ** in godot's. Technically speaking, godot's implementation is wrong, because it allows manipulation of the contained Object *, even though a const Variant was supplied. Godot uses a C-style cast that acts as a const_cast to avoid cast issues.

I don't know about this. Since auto is discouraged in Godot modules, this would potentially make code incompatible between godot-cpp and and a Godot module.

I'd personally prefer to match Godot (even if it's wrong), and separately make an issue or PR for Godot to fix it there, and then we can match it in godot-cpp after.

• VariantInternalAccessor is not implemented for Object *, even though it is in Godot. I believe it is not possible to replicate the intended behavior without access to Variant::object_assign (or internal ObjData). I made it fail statically using SFINAE.

Hm, it sounds like the problem is only for setting with Object *? Is there a way we could still support getting?

[Ivorforce]

In order to discourage developers from using it (and create incompatibilities with Godot modules) you could put it in the godot::internal namespace? We do that in some other places, although, we should probably use it even more than we do.

That's fine by me. Done!

• get_object returns Object * const * in my implementation, but const Object ** in godot's. Technically speaking, godot's implementation is wrong, because it allows manipulation of the contained Object *, even though a const Variant was supplied. Godot uses a C-style cast that acts as a const_cast to avoid cast issues.

I don't know about this. Since auto is discouraged in Godot modules, this would potentially make code incompatible between godot-cpp and and a Godot module.
I'd personally prefer to match Godot (even if it's wrong), and separately make an issue or PR for Godot to fix it there, and then we can match it in godot-cpp after.

Ok. I changed it to a cast-return to match godot's interface.

• VariantInternalAccessor is not implemented for Object *, even though it is in Godot. I believe it is not possible to replicate the intended behavior without access to Variant::object_assign (or internal ObjData). I made it fail statically using SFINAE.

Hm, it sounds like the problem is only for setting with Object *? Is there a way we could still support getting?

Yep. I adjusted the PR accordingly.

[dsnopek]

Thanks! The latest version of this looks good to me :-)