[Bullet] Godot 3.x with Bullet, collision layer & mask choice affects collision accuracy

[m4nu3lf]

Godot version

3.x build 65ab913

System information

OS: Arch Linux, CPU: Ryzen 5, GUP: RX 480

Issue description

Collisions between two physics bodies behave better when the collision layer & mask are chosen so that
(collision_layer_a & collision_mask_b) != 0 AND (collision_layer_b & collision_mask_a) !=0 with respect to when
(collision_layer_a & collision_mask_b) != 0 AND (collision_layer_b & collision_mask_a) ==0
or
(collision_layer_a & collision_mask_b) == 0 AND (collision_layer_b & collision_mask_a) !=0
but two bodies should either not collide or properly collide.

This could also be affecting CCD.

Steps to reproduce

Attached is a project with two scenarios in which the only difference is the collision layer of the two rigid bodies.
The two rigid bodies have the same properties except for position & initial velocity (so same layer & collision mask).
In the well-behaved one, the rigid bodies are both stopped by the static one.
In the not well behaved one, the slow rigid body is stopped, but the fast one is not.

Minimal reproduction project

godot_bullet_collision_bug.zip

[pouleyKetchoupp]

Looks like a bug in Bullet with CCD not using custom collision filtering, but I'm not sure what the proper fix is.

This line is using default collision filter logic in Bullet CCD:

godot/thirdparty/bullet/BulletDynamics/Dynamics/btDiscreteDynamicsWorld.cpp

Lines 801 to 802 in f91afeb

	if (!ClosestConvexResultCallback::needsCollision(proxy0))
	return false;

Which forces (collision_layer_a & collision_mask_b) != 0 AND (collision_layer_b & collision_mask_a) !=0 in order to work correctly.

I've just opened a discussion upstream:
bulletphysics/bullet3#3983