Fix regression 70154 caused by my prior CCD fix.

[Geekotron]

My prior commit (foolishly) trusted the comment that said from - mnormal * mlen * 0.1 would be "a little inside the bounding box". It's not necessarily; at high velocities this calculation puts the point fully outside the body.

So my attempt to use that same point later to determine the length of motion we needed did not work.

I've partially reverted, it's back to only using the -10% motion for calculating the cast start point and NOT using it to determine length of velocity needed. So to fix collision fails, I realized at the end where it was previously subtracting 1% of body length it actually just needed to add it to the distance between from (support point) and to (collision point) instead.

I've confirmed this fix gets both my original issue repro working and the official physics tests: (https://github.com/fabriceci/Godot-Physics-Tests)

(with the exception of 2d cast shape which seems to have already been broken before my change)

fix #70154

[rburing]

Makes sense and passes the tests. Thanks! Do you plan to look into the 2D shape cast CCD? (Separately from this PR.)

[akien-mga]

Thanks!

[LakshayaG73]

@Geekotron can you elaborate a bit on this 2D shape cast CCD issue?

[Geekotron]

@Geekotron can you elaborate a bit on this 2D shape cast CCD issue?

@LakshayaG73 - fabriceci has a test project which is how he located regression 70154. The test project is (https://github.com/fabriceci/Godot-Physics-Tests

The 2d test shows this as of Godot4beta7 (which was before my regression was introduced):

It appears that cast shape CCD does not currently work, at least at very high speeds. @fabriceci could maybe say how long this has been the case since he set up the test.

Makes sense and passes the tests. Thanks! Do you plan to look into the 2D shape cast CCD? (Separately from this PR.)

@rburing - Sorry, I've taken a quick look at the cast_shape code and it would take a while to dig through. Given I'm mainly a 3D guy (didn't even know we had a cast_shape option in 2D), I'd probably not be the best to fix it right now. At the moment I'm actually looking into trying to eliminate the shaking reported in issue #61827 and a few others.

[fabriceci]

@Geekotron The shape cast is bugged since a long time.

I added in the tests the rotation and collision cases, I came across another bug, @Geekotron could you take a look at it too?

It only happens at very high speed, but it happens in 2D as well, and in 3D, only when there are rotations:

Here is a video frame by frame:

failed.mp4

Here is in 3D:

Tips 1: in the test project if you press P you enter in Frame by Frame/Pause mode, and then press O to go to the next frame
Tips 2: If you edit base/pause.gd and change the first variable "paused" to true, you can start a scene in Pause mode