Invalid time_of_impact result for Ball and rotated Polyline about parry

[elsid]

Not sure is it related but time_of_impact is not commutative for Ball and Segment. If ball* and segment* arguments are swapped in time_of_impact_should_return_toi_for_ball_and_rotated_segment then the test does not pass because time_of_impact returns None:

#[test]
fn time_of_impact_should_return_toi_for_rotated_segment_and_ball() {
    let ball_isometry = Isometry2::identity();
    let ball_velocity = Vector2::new(1.0, 0.0);
    let ball = Ball::new(0.5);
    let segment_isometry = Isometry2::rotation(-std::f32::consts::FRAC_PI_2);
    let segment_velocity = Vector2::zeros();
    let segment = Segment::new(Point2::new(1.0, 1.0), Point2::new(-1.0, 1.0));

    assert_eq!(segment_isometry.transform_point(&Point2::new(1.0, 1.0)), Point2::new(0.99999994, -1.0));
    assert_eq!(segment_isometry.transform_point(&Point2::new(-1.0, 1.0)), Point2::new(1.0, 0.99999994));

    let toi = query::time_of_impact(
        &segment_isometry, &segment_velocity, &segment,
        &ball_isometry, &ball_velocity, &ball,
        1.0, 0.0,
    ).unwrap();

    assert_eq!(toi.unwrap().toi, 0.49999994);
}

[elsid]

The same tests are passing with ncollide. The main difference I see between these libraries is an attempt to use only one transformation and velocity in parry:

parry/src/query/time_of_impact/time_of_impact.rs

Lines 87 to 88 in f48629e

	let pos12 = pos1.inv_mul(pos2);
	let vel12 = vel2 - vel1;

.

[sebcrozet]

Thank you for the test cases! You are correct, there is a bug as it should be let vel12 = pos1.inverse_transform_vector(&(vel2 - vel1)) because the velocity must be expressed in the local-space of the first shape.

This will be fixed at part of today's release.