GDExtension call functions without hash

[ParadoxV5]

GDExtension classdb_get_method_bind and variant_get_ptr_utility_function both require hash values from the JSON. This is meant to guard any forward incompatibility when upgrading Godot version.

Howëver, object_method_bind_call has a GDExtensionCallError *r_error return arg that (persumably) knows how to report wrong arg counts or types, so the hash check is redundant here. (It’s good for object_method_bind_ptrcall tho.)

OTOH, GDExtensionPtrUtilityFunction takes an argc-argv pair for args, similar to “regular call”; but doesn’t return a GDExtensionCallError *r_error, similar to “ptrcall”. I much prefer having a GDExtensionCallError return for a complete and consistent error handling mechanism.

Language binders like my WIP currently had to bundle hashes for godot knows how many functions (or at least those of @GlobalScope, which rely on variant_get_ptr_utility_function) in our GDExtensions.

(Better yet, instead of returning function pointers, let us call utility functions directly – function name, argc, argv, return, etc. all in one go, similar to how MethodBind has variant_call.)

[ParadoxV5]

similar to how MethodBind has variant_call

Edit: not exactly. variant_call prefers the script, whereas classdb_get_method_bind persumably only accesses the class. A(n extension) script persumably can use classdb_get_method_bind to call super, but using variant_call will surely (I found out the hard way) cause infinite recursion.

[dsnopek]

Howëver, object_method_bind_call has a GDExtensionCallError *r_error return arg that (persumably) knows how to report wrong arg counts or types, so the hash check is redundant here.

Unfortunately, when making a ptrcall we don't (and in most cases can't) verify the argument types.

Firstly, we don't check the argument types because ptrcalls aim to be the most performant way to call a native function, so we don't do any checking, leaving it up to the language binding to check before calling. Otherwise, you can end up checking the types twice: once in the binding (for type safe languages) and once in the call.

But, also, arguments in ptrcalls are passed as pointers to the raw value, so int *, float *, Object **, etc, not as Variants which can store type information. So, on the receiving end, we can't tell the difference between a int * and Object ** - there's no way to check the type, it's just a random address in memory.

[ParadoxV5]

(It’s good for object_method_bind_ptrcall tho.)

[ParadoxV5]

(emphasis mine)

Otherwise, you can end up checking the types twice: once in the binding (for type safe languages) and once in the call.

Indeed. Godot-side type checks therefore aligns more with dynamic type languages like GDScript, Python, JavaScript, Ruby…

[ParadoxV5]

Oh, now I see it. GDExtensionPtrUtilityFunction also takes raw type pointers as args, making it more align with “ptrcall”. I suppose the argc is there because utility functions may be vararg?

Utility functions doesn’t have a “regular call” counterpart (object_method_bind_call to object_method_bind_ptrcall) for consistency. Well, MethodBind’s got a way to call static methods right?