Add native code support & InputNetowkEncoder.

[AndreaCatania]

Port of: #15

This commit add support to implement the NetwokedController class from C++;
also it add the InputNetworkEncoder that can be used to simplify the
character controller implementation.

// Usually called into the constructor, or during the controller
// initialization. Registers the inputs encoded by `input_encoder`.
uint32_t input_id = input_encoder->register_input(
	StringName("move"),
	Vector2(), // Default
	DataBuffer::DATA_TYPE_NORMALIZED_VECTOR2,
	DataBuffer::COMPRESSION_LEVEL_2);

// Usually this is called from `collect_inputs()`.
// It encodes the input_array into the buffer, according to the
// registered inputs.
input_encoder->encode(input_array, r_buffer);

// Usually this is called from `controller_process`.
// It decodes the buffer and put the inputs inside the `input_array`.
input_encoder->decode(p_buffer, input_array);

// Usually this is called from `are_inputs_different`.
input_encoder->are_different(p_buffer_a, p_buffer_b);

// Usually this is called from `count_input_size`.
input_encoder->count_size(p_buffer);

[kubecz3k]

Looks good, but leaved some api suggestion which you might consider + some questions mostly to learn how it works

[kubecz3k]

when I have such functions which are returning uncommon value to act as exception I always wonder how much do I expect for the person who is using this function to always check if that exception happened, couple thoughts:

• For sure such function documentation is always welcome
• other thing is the usage of INDEX_NONE in situation when it's not exposed to GdScript...
• For such ID thing I usually expect it will return -1 in case of error actually (since that value very clearly dont make any sense and it's very clearly intentional when debugging (some very big int feels like a pointer or smth)
• If I expect that its not super rare for user to check it I usually actually add new function called 'has_input_id' which is documented, and do an assertion inside find_input_id (like soft crash) with communicate that there is no such id and 'has_input_id' should be used - only by the fact such function is on a list when one is reading codumention/help makes him aware that it can happen

[AndreaCatania]

You are perfectly correct. Also I just realized INDEX_NONE is not exposed to GDS.

[kubecz3k]

I dont think NetworkedInputInfo is exposed to GdScript, so what's expected as input here?

[AndreaCatania]

The input is just an array of variant. Each variant holds the input value

[kubecz3k]

why NetworkedController instance is called "node"? It was not obvious for me what this member is representing (and if some type casting is not coming into play here)

[AndreaCatania]

Essentially ServerController is a class that operates on a target node. That's the reason why you don't need any cast, because node is a member and already casted.
A better name should be used, anyway, I agree.

[kubecz3k]

just to ensure: will it still collect inputs of gdscript implementation here or it's not needed?

[kubecz3k]

(apply also to other changes around)

[kubecz3k]

I assume it will do, since there is GDVIRTUAL_CALL(_collect_inputs, p_delta, &r_buffer); inside native_collect_inputs() so I think it's the new way of calling gdscript from c++?

[AndreaCatania]

native_collect_inputs is just a wrap function that can be overridden from C++. If not, the GDS function is normally called using GDVIRTUAL_CALL(_collect_inputs, p_delta, &r_buffer); inside native_collect_inputs()