Fix GDExtension Variant type conversion

[Pylgos]

Fixes #75757
Fixes godotengine/godot-cpp#1132
Fixes godotengine/godot-cpp#907
Fixes godotengine/godot-cpp#1106

With this PR, VariantTypeConstructor<T>::type_from_variant now does the type conversion from Variant to T. Previously, it only reinterpret_cast the variant data by VariantInternalAccessor<T>::get.

[dsnopek]

To check if I'm understanding this correctly:

Presently, VariantTypeConstructor<T>::type_from_variant() will assume that the Variant internally already has a value that is type T. When it isn't actually type T, then weird things happen (ala #75757).

With this PR, rather than using VariantInternalAccessor<T>::get(), it'll rely on the operator T() method on Variant to get the value, possibly converting it to T if it isn't already internally T.

Do I have that right?

[dsnopek]

I just tested the reproduction project on #75757 and I'm able to reproduce the issue. Using the changes in this PR fixes it!

Assuming I'm understanding the changes (as described in my previous comment), this seems good to me. :-)

However, it should still be reviewed by someone on the @godotengine/core or @godotengine/gdextension teams in case this would cause some collateral damage that I'm not aware of.

[Pylgos]

@dsnopek
Your understanding is correct. Thanks for reviewing this PR.

[touilleMan]

Is there a reason for the C-style cast instead of reinterpret_cast everywhere ?

*reinterpret_cast<T *>p_value = *reinterpret_cast<Variant *>(p_variant);

[dsnopek]

I don't know the answer to the question :-) but because that cast isn't added by this PR, I don't think this should block this PR.

There's loads and loads of casts in GDExtension, and more than one bug has been caused by using the inappropriate cast for the situation, so perhaps we need an issue/PR that's dedicated to sorting out all the different casts?

[dsnopek]

Given my increased understanding of GDExtension since I wrote my last comment, and the additional testing of this change from #1132, I'm feeling pretty confident that this change is correct. Let's see if the CI agrees with me :-)

Probably should still have someone else from the GDExtension team take a look at it, though - I've added it to the agenda for the next GDExtension meeting.

[akien-mga]

Would need a rebase to pass CI, last time this PR was updated godot-cpp was mid update and it's now locked to that state.

[dsnopek]

A quick explanation for why I think this change is right for future reviewers:

The functions generated by VariantTypeConstructor<T>::type_from_variant() are used in godot-cpp exclusively for all the T Variant::operator T() methods. This change makes it so that we're also calling the T Variant::operator T() methods on the Godot side (hence converting to the requested type), as opposed to calling VariantInternalAccessor<T>::get() which is used currently (and hence not converting, but just assuming the data is internally already the requested type).

Despite coming from variant_internal.h and the generic name, VariantTypeConstructor<T> appears to only be used by GDExtension, so this change won't have any ramifications outside of GDExtension.

[dsnopek]

Discussed at the GDExtension meeting, and this change seems fine from the perspective of godot-cpp. It would be good if @Bromeon and @touilleMan could confirm that this won't cause any problems in the Rust or Python bindings. Approved if they say it looks fine!

[Bromeon]

I ran our full CI on commit d7eb710.

All green! ✔️

[akien-mga]

Thanks!