mod float32array_fix {
#[test]
fn should_correctly_convert_typed_arrays_regression_check() {
let source = [1.5_f64, 2.5, 3.5, 4.5];
let buggy = vec![0.0_f32; source.len()];
assert_eq!(buggy[0], 0.0);
assert_eq!(buggy[1], 0.0);
let fixed = source
.iter()
.copied()
.map(|value| value as f32)
.collect::<Vec<_>>();
assert!((fixed[0] - 1.5).abs() < f32::EPSILON);
assert!((fixed[1] - 2.5).abs() < f32::EPSILON);
assert!((fixed[2] - 3.5).abs() < f32::EPSILON);
assert!((fixed[3] - 4.5).abs() < f32::EPSILON);
}
}