Physics Interpolation affects Camera2D position smoothing on higher framerates

[TotoShampoin]

Tested versions

• Reproducible in 4.2.stable, 4.3.stable

System information

ArchLinux 6.10.2-arch1-1.1-g14 - KDE Plasma 6.1.4 - Godot 4.3.stable.arch_linux

Issue description

When physics interpolation is activated, the camera smoothing actually goes faster than when it's deactivated for the same speed.

Steps to reproduce

1. Have a monitor with a refresh rate higher than 60Hz
2. Set up a 2D platformer, with a Camera2D that has Position Smoothing turned on
3. Test the camera's physics
4. Go to settings, turn on physics interpolation
5. Test again

Minimal reproduction project (MRP)

test.zip

[rburing]

I don't have a monitor with a high refresh rate, but I confirm the issue is also visible when setting e.g. position_smoothing_speed = 0.1 px/s.

This must be related to Camera2D::get_camera_transform getting called internally, which has the (cursed) side-effect of also updating the smoothing. Ideally we would untangle that Camera2D code and fix this bug in the process.

[Calinou]

• Related to Camera2D smoothing calculation appears incorrect #49747.

[rburing]

In the MRP here (which sets the Camera2D node position in code) the speedup is caused by the fact that both on NOTIFICATION_INTERNAL_PYSICS_PROCESS and on NOTIFICATION_TRANSFORM_CHANGED there is a call to Camera2D::get_camera_transform() which has the side-effect of updating the smoothing (among other things).

cc @lawnjelly since this also affects 3.x

[TotoShampoin]

........... Why should a getter even have a side effect?

[lawnjelly]

Yes, I have a feeling this was noticed before (maybe my previous recommendation at the time was don't use smoothing with physics interpolation).

A while ago, I put a comment in the source that the get_camera_transform() was a historical bug that needs refactor. This needs some care, as fixing this could be compat breaking (especially for 3.x), which was probably why I left to fix in a later PR.

In the case of a smoothed physics interpolated camera, presumably either interpolation needs to be switched off (if using it on idle), or it needs to be used on physics tick and the interpolation take over in between. Trying to use it on idle with physics interpolation won't make sense anyway and there will be jitter (ideally we should prevent this by hard coding, we force to physics mode in a couple of other places).

This probably needs a better MRP (or better still a better issue) for the problems of smoothing. This MRP has a number of problems, detailed below:

Improvements

• Read the docs - https://docs.godotengine.org/en/3.6/tutorials/physics/interpolation/index.html
• Set ticks per second to low value (10) to see what is going on
• Don't move physics interpolated cameras in _process(), you must use _physics_process()
• Either turn off position smoothing, or set process callback to physics

We probably need more physics interpolation warnings to prevent people shooting themselves in the foot like this. There are warnings in 3D but not many in 2D so far.

Also when the docs are merged for 4.x that will be helpful.

Here is fixed MRP, with smoothing set to physics mode, and the camera moved in physics tick, and TPS set to 10:
Camera2DFTI.zip

UPDATE:
Turns out Camera2D is already forced to physics process mode when using physics interpolation:

void Camera2D::_update_process_mode() {
	if (is_physics_interpolated_and_enabled()) {
		set_process_internal(is_current());
		set_physics_process_internal(is_current());

#ifdef TOOLS_ENABLED
		if (process_mode == CAMERA2D_PROCESS_IDLE) {
			WARN_PRINT_ONCE("Camera2D overridden to physics process mode due to use of physics interpolation.");
		}
#endif
	} else {