mirror of
https://github.com/vlang/v.git
synced 2023-08-10 21:13:21 +03:00
91 lines
1.6 KiB
V
91 lines
1.6 KiB
V
import math.vec
|
|
|
|
fn test_vec3_int() {
|
|
mut v1 := vec.vec3(0, 0, 0)
|
|
mut v2 := vec.vec3(0, 0, 0)
|
|
assert v1 == v2
|
|
v1.one()
|
|
v2.one()
|
|
assert v1.x == 1
|
|
assert v1.y == 1
|
|
assert v1.z == 1
|
|
assert v1 == v2
|
|
|
|
v3 := v1 + v2
|
|
assert typeof(v3).name == 'vec.Vec3[int]'
|
|
assert v3.x == 2
|
|
assert v3.y == 2
|
|
assert v3.z == 2
|
|
}
|
|
|
|
fn test_vec3_f32() {
|
|
mut v1 := vec.vec3(f32(0), 0, 0)
|
|
mut v2 := vec.vec3(f32(0), 0, 0)
|
|
assert v1 == v2
|
|
v1.one()
|
|
v2.one()
|
|
assert v1.x == 1
|
|
assert v1.y == 1
|
|
assert v1.z == 1
|
|
assert v1 == v2
|
|
|
|
v3 := v1 + v2
|
|
assert typeof(v3).name == 'vec.Vec3[f32]'
|
|
assert v3.x == 2
|
|
assert v3.y == 2
|
|
assert v3.z == 2
|
|
}
|
|
|
|
fn test_vec3_f64() {
|
|
mut v1 := vec.vec3(0.0, 0, 0)
|
|
mut v2 := vec.vec3(0.0, 0, 0)
|
|
assert v1 == v2
|
|
v1.one()
|
|
v2.one()
|
|
assert v1.x == 1
|
|
assert v1.y == 1
|
|
assert v1.z == 1
|
|
assert v1 == v2
|
|
|
|
v3 := v1 + v2
|
|
assert typeof(v3).name == 'vec.Vec3[f64]'
|
|
assert v3.x == 2
|
|
assert v3.y == 2
|
|
assert v3.z == 2
|
|
}
|
|
|
|
fn test_vec3_f64_utils_1() {
|
|
mut v1 := vec.vec3(2.0, 3.0, 1.5)
|
|
mut v2 := vec.vec3(1.0, 4.0, 1.5)
|
|
|
|
mut zv := vec.vec3(5.0, 5.0, 5.0)
|
|
zv.zero()
|
|
|
|
v3 := v1 + v2
|
|
assert v3.x == 3
|
|
assert v3.y == 7
|
|
assert v3.z == 3
|
|
|
|
v1l := vec.vec3(6.0, 2.0, -3.0)
|
|
assert v1l.magnitude() == 7
|
|
|
|
mut ctv1 := vec.vec3(0.000001, 0.000001, 0.000001)
|
|
ctv1.clean_tolerance(0.00001)
|
|
assert ctv1 == zv
|
|
}
|
|
|
|
fn test_vec3_f64_utils_2() {
|
|
mut v1 := vec.vec3(4.0, 4.0, 8.0)
|
|
assert v1.unit().magnitude() == 1
|
|
v2 := v1.mul_scalar(0.5)
|
|
assert v2.x == 2
|
|
assert v2.y == 2
|
|
assert v2.z == 4
|
|
assert v2.unit().magnitude() == 1
|
|
|
|
invv2 := v2.inv()
|
|
assert invv2.x == 0.5
|
|
assert invv2.y == 0.5
|
|
assert invv2.z == 0.25
|
|
}
|