[3.x] Fix KinematicBody axis lock

[aaronfranke]

This fixes #25798 in 3.x.

[AndreaCatania]

I'm not strongly against this, and for the presented reason it seems a good change. However, the physics server API are generally not so fast especially if compared to locked_axis & (1 << i). So have 3 PhysicsServer calls on a function that is called recursively each frame may be bad, especially when you have more than 1 character.

This is not a blocker if the physics team still think worth it.

[AndreaCatania]

By the way, just noticed the type of locked_axis should not be uint16_t but uint32_t. All back-ends supports up to 32 flags.

[pouleyKetchoupp]

@AndreaCatania The rest of the KinematicBody api is an expensive process anyway, I don't think these calls really make a difference. So I would rather keep things clean and simple.

For info, #25798 is only present on 3.2, since on master set_axis_lock is setting both members. It still makes sense to remove this duplicate on both branches anyway.

[pouleyKetchoupp]

Just one thing I'd like to change. In order to clean things completely, let's add a helper method on KinematicBody3D instead of calling the physics server directly.

Something like:

void KinematicBody::_apply_axis_lock(Vector3 &p_motion) const {
	if (get_axis_lock(PhysicsServer::BODY_AXIS_LINEAR_X)) {
		p_motion.x = 0.0f;
	}
	if (get_axis_lock(PhysicsServer::BODY_AXIS_LINEAR_Y)) {
		p_motion.y = 0.0f;
	}
	if (get_axis_lock(PhysicsServer::BODY_AXIS_LINEAR_Z)) {
		p_motion.z = 0.0f;
	}
}

And then:

_apply_axis_lock(result.motion);

Instead of:

	for (int i = 0; i < 3; i++) { {
		if (PhysicsServer3D::get_singleton()->body_is_axis_locked(get_rid(), PhysicsServer3D::BodyAxis(1 << i))) {
			result.motion[i] = 0;
		}
	}

So the code in KinematicBody won't rely on actual values from the physics server to work properly.

[aaronfranke]

@AndreaCatania I considered if we could replace PhysicsServer::get_singleton()->body_is_axis_locked with PhysicsServer::get_singleton()->body_get_axis_lock or similar and then keep the locked_axis & (1 << i), but it looks like this API used to exist and you removed it so I figured it was for a good reason.

[aaronfranke]

@pouleyKetchoupp What if the helper method was located on PhysicsServer3D? This would avoid the repeated calls to body_is_axis_locked , assuming we don't implement body_get_axis_lock or similar, since we should be able to get locked_axis directly. It somewhat makes sense because locked_axis is located in PhysicsServer3D's bodies and PhysicsServer3D::BodyAxis is in PhysicsServer3D, though it's not as local of a solution as what you propose.

[pouleyKetchoupp]

@aaronfranke It's not a bad idea, but currently I don't think it's worth the extra changes in the Physics Server API, and having to implement it in both Bullet and Godot Physics. I would rather do that later, only if for some reason we need it for optimization purpose.

[madmiraal]

See #45175 (comment). #25798 was fixed with #38852, but never backported to 3.2. I think the better approach is a backport of #38852.

[aaronfranke]

I updated this PR so that now it's a 3.x backport of #38852. This also includes backporting the renames, but the old names are kept as aliases for compatibility, and this PR also updates the documentation.

[pouleyKetchoupp]

Looks good!