Revision control

Copy as Markdown

Other Tools

// Copyright 2013 The Servo Project Developers. See the COPYRIGHT
// file at the top-level directory of this distribution.
//
// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
// option. This file may not be copied, modified, or distributed
// except according to those terms.
use super::UnknownUnit;
use crate::approxord::{max, min};
use crate::length::Length;
use crate::num::*;
use crate::scale::Scale;
use crate::vector::{vec2, BoolVector2D, Vector2D};
use crate::vector::{vec3, BoolVector3D, Vector3D};
use core::cmp::{Eq, PartialEq};
use core::fmt;
use core::hash::Hash;
use core::iter::Sum;
use core::marker::PhantomData;
use core::ops::{Add, AddAssign, Div, DivAssign, Mul, MulAssign, Neg, Sub, SubAssign};
#[cfg(feature = "bytemuck")]
use bytemuck::{Pod, Zeroable};
#[cfg(feature = "mint")]
use mint;
use num_traits::{Float, NumCast, Signed};
#[cfg(feature = "serde")]
use serde;
/// A 2d size tagged with a unit.
#[repr(C)]
pub struct Size2D<T, U> {
/// The extent of the element in the `U` units along the `x` axis (usually horizontal).
pub width: T,
/// The extent of the element in the `U` units along the `y` axis (usually vertical).
pub height: T,
#[doc(hidden)]
pub _unit: PhantomData<U>,
}
impl<T: Copy, U> Copy for Size2D<T, U> {}
impl<T: Clone, U> Clone for Size2D<T, U> {
fn clone(&self) -> Self {
Size2D {
width: self.width.clone(),
height: self.height.clone(),
_unit: PhantomData,
}
}
}
#[cfg(feature = "serde")]
impl<'de, T, U> serde::Deserialize<'de> for Size2D<T, U>
where
T: serde::Deserialize<'de>,
{
/// Deserializes 2d size from tuple of width and height.
fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
where
D: serde::Deserializer<'de>,
{
let (width, height) = serde::Deserialize::deserialize(deserializer)?;
Ok(Size2D {
width,
height,
_unit: PhantomData,
})
}
}
#[cfg(feature = "serde")]
impl<T, U> serde::Serialize for Size2D<T, U>
where
T: serde::Serialize,
{
/// Serializes 2d size to tuple of width and height.
fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
where
S: serde::Serializer,
{
(&self.width, &self.height).serialize(serializer)
}
}
#[cfg(feature = "arbitrary")]
impl<'a, T, U> arbitrary::Arbitrary<'a> for Size2D<T, U>
where
T: arbitrary::Arbitrary<'a>,
{
fn arbitrary(u: &mut arbitrary::Unstructured<'a>) -> arbitrary::Result<Self> {
let (width, height) = arbitrary::Arbitrary::arbitrary(u)?;
Ok(Size2D {
width,
height,
_unit: PhantomData,
})
}
}
#[cfg(feature = "bytemuck")]
unsafe impl<T: Zeroable, U> Zeroable for Size2D<T, U> {}
#[cfg(feature = "bytemuck")]
unsafe impl<T: Pod, U: 'static> Pod for Size2D<T, U> {}
impl<T, U> Eq for Size2D<T, U> where T: Eq {}
impl<T, U> PartialEq for Size2D<T, U>
where
T: PartialEq,
{
fn eq(&self, other: &Self) -> bool {
self.width == other.width && self.height == other.height
}
}
impl<T, U> Hash for Size2D<T, U>
where
T: Hash,
{
fn hash<H: core::hash::Hasher>(&self, h: &mut H) {
self.width.hash(h);
self.height.hash(h);
}
}
impl<T: fmt::Debug, U> fmt::Debug for Size2D<T, U> {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
fmt::Debug::fmt(&self.width, f)?;
write!(f, "x")?;
fmt::Debug::fmt(&self.height, f)
}
}
impl<T: Default, U> Default for Size2D<T, U> {
fn default() -> Self {
Size2D::new(Default::default(), Default::default())
}
}
impl<T, U> Size2D<T, U> {
/// The same as [`Zero::zero()`] but available without importing trait.
///
/// [`Zero::zero()`]: ./num/trait.Zero.html#tymethod.zero
#[inline]
pub fn zero() -> Self
where
T: Zero,
{
Size2D::new(Zero::zero(), Zero::zero())
}
/// Constructor taking scalar values.
#[inline]
pub const fn new(width: T, height: T) -> Self {
Size2D {
width,
height,
_unit: PhantomData,
}
}
/// Constructor taking scalar strongly typed lengths.
#[inline]
pub fn from_lengths(width: Length<T, U>, height: Length<T, U>) -> Self {
Size2D::new(width.0, height.0)
}
/// Constructor setting all components to the same value.
#[inline]
pub fn splat(v: T) -> Self
where
T: Clone,
{
Size2D {
width: v.clone(),
height: v,
_unit: PhantomData,
}
}
/// Tag a unitless value with units.
#[inline]
pub fn from_untyped(p: Size2D<T, UnknownUnit>) -> Self {
Size2D::new(p.width, p.height)
}
}
impl<T: Copy, U> Size2D<T, U> {
/// Return this size as an array of two elements (width, then height).
#[inline]
pub fn to_array(self) -> [T; 2] {
[self.width, self.height]
}
/// Return this size as a tuple of two elements (width, then height).
#[inline]
pub fn to_tuple(self) -> (T, T) {
(self.width, self.height)
}
/// Return this size as a vector with width and height.
#[inline]
pub fn to_vector(self) -> Vector2D<T, U> {
vec2(self.width, self.height)
}
/// Drop the units, preserving only the numeric value.
#[inline]
pub fn to_untyped(self) -> Size2D<T, UnknownUnit> {
self.cast_unit()
}
/// Cast the unit
#[inline]
pub fn cast_unit<V>(self) -> Size2D<T, V> {
Size2D::new(self.width, self.height)
}
/// Rounds each component to the nearest integer value.
///
/// This behavior is preserved for negative values (unlike the basic cast).
///
/// ```rust
/// # use euclid::size2;
/// enum Mm {}
///
/// assert_eq!(size2::<_, Mm>(-0.1, -0.8).round(), size2::<_, Mm>(0.0, -1.0))
/// ```
#[inline]
#[must_use]
pub fn round(self) -> Self
where
T: Round,
{
Size2D::new(self.width.round(), self.height.round())
}
/// Rounds each component to the smallest integer equal or greater than the original value.
///
/// This behavior is preserved for negative values (unlike the basic cast).
///
/// ```rust
/// # use euclid::size2;
/// enum Mm {}
///
/// assert_eq!(size2::<_, Mm>(-0.1, -0.8).ceil(), size2::<_, Mm>(0.0, 0.0))
/// ```
#[inline]
#[must_use]
pub fn ceil(self) -> Self
where
T: Ceil,
{
Size2D::new(self.width.ceil(), self.height.ceil())
}
/// Rounds each component to the biggest integer equal or lower than the original value.
///
/// This behavior is preserved for negative values (unlike the basic cast).
///
/// ```rust
/// # use euclid::size2;
/// enum Mm {}
///
/// assert_eq!(size2::<_, Mm>(-0.1, -0.8).floor(), size2::<_, Mm>(-1.0, -1.0))
/// ```
#[inline]
#[must_use]
pub fn floor(self) -> Self
where
T: Floor,
{
Size2D::new(self.width.floor(), self.height.floor())
}
/// Returns result of multiplication of both components
pub fn area(self) -> T::Output
where
T: Mul,
{
self.width * self.height
}
/// Linearly interpolate each component between this size and another size.
///
/// # Example
///
/// ```rust
/// use euclid::size2;
/// use euclid::default::Size2D;
///
/// let from: Size2D<_> = size2(0.0, 10.0);
/// let to: Size2D<_> = size2(8.0, -4.0);
///
/// assert_eq!(from.lerp(to, -1.0), size2(-8.0, 24.0));
/// assert_eq!(from.lerp(to, 0.0), size2( 0.0, 10.0));
/// assert_eq!(from.lerp(to, 0.5), size2( 4.0, 3.0));
/// assert_eq!(from.lerp(to, 1.0), size2( 8.0, -4.0));
/// assert_eq!(from.lerp(to, 2.0), size2(16.0, -18.0));
/// ```
#[inline]
pub fn lerp(self, other: Self, t: T) -> Self
where
T: One + Sub<Output = T> + Mul<Output = T> + Add<Output = T>,
{
let one_t = T::one() - t;
self * one_t + other * t
}
}
impl<T: NumCast + Copy, U> Size2D<T, U> {
/// Cast from one numeric representation to another, preserving the units.
///
/// When casting from floating point to integer coordinates, the decimals are truncated
/// as one would expect from a simple cast, but this behavior does not always make sense
/// geometrically. Consider using `round()`, `ceil()` or `floor()` before casting.
#[inline]
pub fn cast<NewT: NumCast>(self) -> Size2D<NewT, U> {
self.try_cast().unwrap()
}
/// Fallible cast from one numeric representation to another, preserving the units.
///
/// When casting from floating point to integer coordinates, the decimals are truncated
/// as one would expect from a simple cast, but this behavior does not always make sense
/// geometrically. Consider using `round()`, `ceil()` or `floor()` before casting.
pub fn try_cast<NewT: NumCast>(self) -> Option<Size2D<NewT, U>> {
match (NumCast::from(self.width), NumCast::from(self.height)) {
(Some(w), Some(h)) => Some(Size2D::new(w, h)),
_ => None,
}
}
// Convenience functions for common casts
/// Cast into an `f32` size.
#[inline]
pub fn to_f32(self) -> Size2D<f32, U> {
self.cast()
}
/// Cast into an `f64` size.
#[inline]
pub fn to_f64(self) -> Size2D<f64, U> {
self.cast()
}
/// Cast into an `uint` size, truncating decimals if any.
///
/// When casting from floating point sizes, it is worth considering whether
/// to `round()`, `ceil()` or `floor()` before the cast in order to obtain
/// the desired conversion behavior.
#[inline]
pub fn to_usize(self) -> Size2D<usize, U> {
self.cast()
}
/// Cast into an `u32` size, truncating decimals if any.
///
/// When casting from floating point sizes, it is worth considering whether
/// to `round()`, `ceil()` or `floor()` before the cast in order to obtain
/// the desired conversion behavior.
#[inline]
pub fn to_u32(self) -> Size2D<u32, U> {
self.cast()
}
/// Cast into an `u64` size, truncating decimals if any.
///
/// When casting from floating point sizes, it is worth considering whether
/// to `round()`, `ceil()` or `floor()` before the cast in order to obtain
/// the desired conversion behavior.
#[inline]
pub fn to_u64(self) -> Size2D<u64, U> {
self.cast()
}
/// Cast into an `i32` size, truncating decimals if any.
///
/// When casting from floating point sizes, it is worth considering whether
/// to `round()`, `ceil()` or `floor()` before the cast in order to obtain
/// the desired conversion behavior.
#[inline]
pub fn to_i32(self) -> Size2D<i32, U> {
self.cast()
}
/// Cast into an `i64` size, truncating decimals if any.
///
/// When casting from floating point sizes, it is worth considering whether
/// to `round()`, `ceil()` or `floor()` before the cast in order to obtain
/// the desired conversion behavior.
#[inline]
pub fn to_i64(self) -> Size2D<i64, U> {
self.cast()
}
}
impl<T: Float, U> Size2D<T, U> {
/// Returns true if all members are finite.
#[inline]
pub fn is_finite(self) -> bool {
self.width.is_finite() && self.height.is_finite()
}
}
impl<T: Signed, U> Size2D<T, U> {
/// Computes the absolute value of each component.
///
/// For `f32` and `f64`, `NaN` will be returned for component if the component is `NaN`.
///
/// For signed integers, `::MIN` will be returned for component if the component is `::MIN`.
pub fn abs(self) -> Self {
size2(self.width.abs(), self.height.abs())
}
/// Returns `true` if both components is positive and `false` any component is zero or negative.
pub fn is_positive(self) -> bool {
self.width.is_positive() && self.height.is_positive()
}
}
impl<T: PartialOrd, U> Size2D<T, U> {
/// Returns the size each component of which are minimum of this size and another.
#[inline]
pub fn min(self, other: Self) -> Self {
size2(min(self.width, other.width), min(self.height, other.height))
}
/// Returns the size each component of which are maximum of this size and another.
#[inline]
pub fn max(self, other: Self) -> Self {
size2(max(self.width, other.width), max(self.height, other.height))
}
/// Returns the size each component of which clamped by corresponding
/// components of `start` and `end`.
///
/// Shortcut for `self.max(start).min(end)`.
#[inline]
pub fn clamp(self, start: Self, end: Self) -> Self
where
T: Copy,
{
self.max(start).min(end)
}
// Returns true if this size is larger or equal to the other size in all dimensions.
#[inline]
pub fn contains(self, other: Self) -> bool {
self.width >= other.width && self.height >= other.height
}
/// Returns vector with results of "greater then" operation on each component.
pub fn greater_than(self, other: Self) -> BoolVector2D {
BoolVector2D {
x: self.width > other.width,
y: self.height > other.height,
}
}
/// Returns vector with results of "lower then" operation on each component.
pub fn lower_than(self, other: Self) -> BoolVector2D {
BoolVector2D {
x: self.width < other.width,
y: self.height < other.height,
}
}
/// Returns `true` if any component of size is zero, negative, or NaN.
pub fn is_empty(self) -> bool
where
T: Zero,
{
let zero = T::zero();
// The condition is expressed this way so that we return true in
// the presence of NaN.
!(self.width > zero && self.height > zero)
}
}
impl<T: PartialEq, U> Size2D<T, U> {
/// Returns vector with results of "equal" operation on each component.
pub fn equal(self, other: Self) -> BoolVector2D {
BoolVector2D {
x: self.width == other.width,
y: self.height == other.height,
}
}
/// Returns vector with results of "not equal" operation on each component.
pub fn not_equal(self, other: Self) -> BoolVector2D {
BoolVector2D {
x: self.width != other.width,
y: self.height != other.height,
}
}
}
impl<T: Round, U> Round for Size2D<T, U> {
/// See [`Size2D::round()`](#method.round).
#[inline]
fn round(self) -> Self {
self.round()
}
}
impl<T: Ceil, U> Ceil for Size2D<T, U> {
/// See [`Size2D::ceil()`](#method.ceil).
#[inline]
fn ceil(self) -> Self {
self.ceil()
}
}
impl<T: Floor, U> Floor for Size2D<T, U> {
/// See [`Size2D::floor()`](#method.floor).
#[inline]
fn floor(self) -> Self {
self.floor()
}
}
impl<T: Zero, U> Zero for Size2D<T, U> {
#[inline]
fn zero() -> Self {
Size2D::new(Zero::zero(), Zero::zero())
}
}
impl<T: Neg, U> Neg for Size2D<T, U> {
type Output = Size2D<T::Output, U>;
#[inline]
fn neg(self) -> Self::Output {
Size2D::new(-self.width, -self.height)
}
}
impl<T: Add, U> Add for Size2D<T, U> {
type Output = Size2D<T::Output, U>;
#[inline]
fn add(self, other: Self) -> Self::Output {
Size2D::new(self.width + other.width, self.height + other.height)
}
}
impl<T: Copy + Add<T, Output = T>, U> Add<&Self> for Size2D<T, U> {
type Output = Self;
fn add(self, other: &Self) -> Self {
Size2D::new(self.width + other.width, self.height + other.height)
}
}
impl<T: Add<Output = T> + Zero, U> Sum for Size2D<T, U> {
fn sum<I: Iterator<Item = Self>>(iter: I) -> Self {
iter.fold(Self::zero(), Add::add)
}
}
impl<'a, T: 'a + Add<Output = T> + Copy + Zero, U: 'a> Sum<&'a Self> for Size2D<T, U> {
fn sum<I: Iterator<Item = &'a Self>>(iter: I) -> Self {
iter.fold(Self::zero(), Add::add)
}
}
impl<T: AddAssign, U> AddAssign for Size2D<T, U> {
#[inline]
fn add_assign(&mut self, other: Self) {
self.width += other.width;
self.height += other.height;
}
}
impl<T: Sub, U> Sub for Size2D<T, U> {
type Output = Size2D<T::Output, U>;
#[inline]
fn sub(self, other: Self) -> Self::Output {
Size2D::new(self.width - other.width, self.height - other.height)
}
}
impl<T: SubAssign, U> SubAssign for Size2D<T, U> {
#[inline]
fn sub_assign(&mut self, other: Self) {
self.width -= other.width;
self.height -= other.height;
}
}
impl<T: Copy + Mul, U> Mul<T> for Size2D<T, U> {
type Output = Size2D<T::Output, U>;
#[inline]
fn mul(self, scale: T) -> Self::Output {
Size2D::new(self.width * scale, self.height * scale)
}
}
impl<T: Copy + MulAssign, U> MulAssign<T> for Size2D<T, U> {
#[inline]
fn mul_assign(&mut self, other: T) {
self.width *= other;
self.height *= other;
}
}
impl<T: Copy + Mul, U1, U2> Mul<Scale<T, U1, U2>> for Size2D<T, U1> {
type Output = Size2D<T::Output, U2>;
#[inline]
fn mul(self, scale: Scale<T, U1, U2>) -> Self::Output {
Size2D::new(self.width * scale.0, self.height * scale.0)
}
}
impl<T: Copy + MulAssign, U> MulAssign<Scale<T, U, U>> for Size2D<T, U> {
#[inline]
fn mul_assign(&mut self, other: Scale<T, U, U>) {
*self *= other.0;
}
}
impl<T: Copy + Div, U> Div<T> for Size2D<T, U> {
type Output = Size2D<T::Output, U>;
#[inline]
fn div(self, scale: T) -> Self::Output {
Size2D::new(self.width / scale, self.height / scale)
}
}
impl<T: Copy + DivAssign, U> DivAssign<T> for Size2D<T, U> {
#[inline]
fn div_assign(&mut self, other: T) {
self.width /= other;
self.height /= other;
}
}
impl<T: Copy + Div, U1, U2> Div<Scale<T, U1, U2>> for Size2D<T, U2> {
type Output = Size2D<T::Output, U1>;
#[inline]
fn div(self, scale: Scale<T, U1, U2>) -> Self::Output {
Size2D::new(self.width / scale.0, self.height / scale.0)
}
}
impl<T: Copy + DivAssign, U> DivAssign<Scale<T, U, U>> for Size2D<T, U> {
#[inline]
fn div_assign(&mut self, other: Scale<T, U, U>) {
*self /= other.0;
}
}
/// Shorthand for `Size2D::new(w, h)`.
#[inline]
pub const fn size2<T, U>(w: T, h: T) -> Size2D<T, U> {
Size2D::new(w, h)
}
#[cfg(feature = "mint")]
impl<T, U> From<mint::Vector2<T>> for Size2D<T, U> {
#[inline]
fn from(v: mint::Vector2<T>) -> Self {
Size2D {
width: v.x,
height: v.y,
_unit: PhantomData,
}
}
}
#[cfg(feature = "mint")]
impl<T, U> From<Size2D<T, U>> for mint::Vector2<T> {
#[inline]
fn from(s: Size2D<T, U>) -> Self {
mint::Vector2 {
x: s.width,
y: s.height,
}
}
}
impl<T, U> From<Vector2D<T, U>> for Size2D<T, U> {
#[inline]
fn from(v: Vector2D<T, U>) -> Self {
size2(v.x, v.y)
}
}
impl<T, U> From<Size2D<T, U>> for [T; 2] {
#[inline]
fn from(s: Size2D<T, U>) -> Self {
[s.width, s.height]
}
}
impl<T, U> From<[T; 2]> for Size2D<T, U> {
#[inline]
fn from([w, h]: [T; 2]) -> Self {
size2(w, h)
}
}
impl<T, U> From<Size2D<T, U>> for (T, T) {
#[inline]
fn from(s: Size2D<T, U>) -> Self {
(s.width, s.height)
}
}
impl<T, U> From<(T, T)> for Size2D<T, U> {
#[inline]
fn from(tuple: (T, T)) -> Self {
size2(tuple.0, tuple.1)
}
}
#[cfg(test)]
mod size2d {
use crate::default::Size2D;
#[cfg(feature = "mint")]
use mint;
#[test]
pub fn test_area() {
let p = Size2D::new(1.5, 2.0);
assert_eq!(p.area(), 3.0);
}
#[cfg(feature = "mint")]
#[test]
pub fn test_mint() {
let s1 = Size2D::new(1.0, 2.0);
let sm: mint::Vector2<_> = s1.into();
let s2 = Size2D::from(sm);
assert_eq!(s1, s2);
}
mod ops {
use crate::default::Size2D;
use crate::scale::Scale;
pub enum Mm {}
pub enum Cm {}
pub type Size2DMm<T> = crate::Size2D<T, Mm>;
pub type Size2DCm<T> = crate::Size2D<T, Cm>;
#[test]
pub fn test_neg() {
assert_eq!(-Size2D::new(1.0, 2.0), Size2D::new(-1.0, -2.0));
assert_eq!(-Size2D::new(0.0, 0.0), Size2D::new(-0.0, -0.0));
assert_eq!(-Size2D::new(-1.0, -2.0), Size2D::new(1.0, 2.0));
}
#[test]
pub fn test_add() {
let s1 = Size2D::new(1.0, 2.0);
let s2 = Size2D::new(3.0, 4.0);
assert_eq!(s1 + s2, Size2D::new(4.0, 6.0));
assert_eq!(s1 + &s2, Size2D::new(4.0, 6.0));
let s1 = Size2D::new(1.0, 2.0);
let s2 = Size2D::new(0.0, 0.0);
assert_eq!(s1 + s2, Size2D::new(1.0, 2.0));
assert_eq!(s1 + &s2, Size2D::new(1.0, 2.0));
let s1 = Size2D::new(1.0, 2.0);
let s2 = Size2D::new(-3.0, -4.0);
assert_eq!(s1 + s2, Size2D::new(-2.0, -2.0));
assert_eq!(s1 + &s2, Size2D::new(-2.0, -2.0));
let s1 = Size2D::new(0.0, 0.0);
let s2 = Size2D::new(0.0, 0.0);
assert_eq!(s1 + s2, Size2D::new(0.0, 0.0));
assert_eq!(s1 + &s2, Size2D::new(0.0, 0.0));
}
#[test]
pub fn test_add_assign() {
let mut s = Size2D::new(1.0, 2.0);
s += Size2D::new(3.0, 4.0);
assert_eq!(s, Size2D::new(4.0, 6.0));
let mut s = Size2D::new(1.0, 2.0);
s += Size2D::new(0.0, 0.0);
assert_eq!(s, Size2D::new(1.0, 2.0));
let mut s = Size2D::new(1.0, 2.0);
s += Size2D::new(-3.0, -4.0);
assert_eq!(s, Size2D::new(-2.0, -2.0));
let mut s = Size2D::new(0.0, 0.0);
s += Size2D::new(0.0, 0.0);
assert_eq!(s, Size2D::new(0.0, 0.0));
}
#[test]
pub fn test_sum() {
let sizes = [
Size2D::new(0.0, 1.0),
Size2D::new(1.0, 2.0),
Size2D::new(2.0, 3.0),
];
let sum = Size2D::new(3.0, 6.0);
assert_eq!(sizes.iter().sum::<Size2D<_>>(), sum);
}
#[test]
pub fn test_sub() {
let s1 = Size2D::new(1.0, 2.0);
let s2 = Size2D::new(3.0, 4.0);
assert_eq!(s1 - s2, Size2D::new(-2.0, -2.0));
let s1 = Size2D::new(1.0, 2.0);
let s2 = Size2D::new(0.0, 0.0);
assert_eq!(s1 - s2, Size2D::new(1.0, 2.0));
let s1 = Size2D::new(1.0, 2.0);
let s2 = Size2D::new(-3.0, -4.0);
assert_eq!(s1 - s2, Size2D::new(4.0, 6.0));
let s1 = Size2D::new(0.0, 0.0);
let s2 = Size2D::new(0.0, 0.0);
assert_eq!(s1 - s2, Size2D::new(0.0, 0.0));
}
#[test]
pub fn test_sub_assign() {
let mut s = Size2D::new(1.0, 2.0);
s -= Size2D::new(3.0, 4.0);
assert_eq!(s, Size2D::new(-2.0, -2.0));
let mut s = Size2D::new(1.0, 2.0);
s -= Size2D::new(0.0, 0.0);
assert_eq!(s, Size2D::new(1.0, 2.0));
let mut s = Size2D::new(1.0, 2.0);
s -= Size2D::new(-3.0, -4.0);
assert_eq!(s, Size2D::new(4.0, 6.0));
let mut s = Size2D::new(0.0, 0.0);
s -= Size2D::new(0.0, 0.0);
assert_eq!(s, Size2D::new(0.0, 0.0));
}
#[test]
pub fn test_mul_scalar() {
let s1: Size2D<f32> = Size2D::new(3.0, 5.0);
let result = s1 * 5.0;
assert_eq!(result, Size2D::new(15.0, 25.0));
}
#[test]
pub fn test_mul_assign_scalar() {
let mut s1 = Size2D::new(3.0, 5.0);
s1 *= 5.0;
assert_eq!(s1, Size2D::new(15.0, 25.0));
}
#[test]
pub fn test_mul_scale() {
let s1 = Size2DMm::new(1.0, 2.0);
let cm_per_mm: Scale<f32, Mm, Cm> = Scale::new(0.1);
let result = s1 * cm_per_mm;
assert_eq!(result, Size2DCm::new(0.1, 0.2));
}
#[test]
pub fn test_mul_assign_scale() {
let mut s1 = Size2DMm::new(1.0, 2.0);
let scale: Scale<f32, Mm, Mm> = Scale::new(0.1);
s1 *= scale;
assert_eq!(s1, Size2DMm::new(0.1, 0.2));
}
#[test]
pub fn test_div_scalar() {
let s1: Size2D<f32> = Size2D::new(15.0, 25.0);
let result = s1 / 5.0;
assert_eq!(result, Size2D::new(3.0, 5.0));
}
#[test]
pub fn test_div_assign_scalar() {
let mut s1: Size2D<f32> = Size2D::new(15.0, 25.0);
s1 /= 5.0;
assert_eq!(s1, Size2D::new(3.0, 5.0));
}
#[test]
pub fn test_div_scale() {
let s1 = Size2DCm::new(0.1, 0.2);
let cm_per_mm: Scale<f32, Mm, Cm> = Scale::new(0.1);
let result = s1 / cm_per_mm;
assert_eq!(result, Size2DMm::new(1.0, 2.0));
}
#[test]
pub fn test_div_assign_scale() {
let mut s1 = Size2DMm::new(0.1, 0.2);
let scale: Scale<f32, Mm, Mm> = Scale::new(0.1);
s1 /= scale;
assert_eq!(s1, Size2DMm::new(1.0, 2.0));
}
#[test]
pub fn test_nan_empty() {
use std::f32::NAN;
assert!(Size2D::new(NAN, 2.0).is_empty());
assert!(Size2D::new(0.0, NAN).is_empty());
assert!(Size2D::new(NAN, -2.0).is_empty());
}
}
}
/// A 3d size tagged with a unit.
#[repr(C)]
pub struct Size3D<T, U> {
/// The extent of the element in the `U` units along the `x` axis.
pub width: T,
/// The extent of the element in the `U` units along the `y` axis.
pub height: T,
/// The extent of the element in the `U` units along the `z` axis.
pub depth: T,
#[doc(hidden)]
pub _unit: PhantomData<U>,
}
impl<T: Copy, U> Copy for Size3D<T, U> {}
impl<T: Clone, U> Clone for Size3D<T, U> {
fn clone(&self) -> Self {
Size3D {
width: self.width.clone(),
height: self.height.clone(),
depth: self.depth.clone(),
_unit: PhantomData,
}
}
}
#[cfg(feature = "serde")]
impl<'de, T, U> serde::Deserialize<'de> for Size3D<T, U>
where
T: serde::Deserialize<'de>,
{
fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
where
D: serde::Deserializer<'de>,
{
let (width, height, depth) = serde::Deserialize::deserialize(deserializer)?;
Ok(Size3D {
width,
height,
depth,
_unit: PhantomData,
})
}
}
#[cfg(feature = "serde")]
impl<T, U> serde::Serialize for Size3D<T, U>
where
T: serde::Serialize,
{
fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
where
S: serde::Serializer,
{
(&self.width, &self.height, &self.depth).serialize(serializer)
}
}
#[cfg(feature = "bytemuck")]
unsafe impl<T: Zeroable, U> Zeroable for Size3D<T, U> {}
#[cfg(feature = "bytemuck")]
unsafe impl<T: Pod, U: 'static> Pod for Size3D<T, U> {}
impl<T, U> Eq for Size3D<T, U> where T: Eq {}
impl<T, U> PartialEq for Size3D<T, U>
where
T: PartialEq,
{
fn eq(&self, other: &Self) -> bool {
self.width == other.width && self.height == other.height && self.depth == other.depth
}
}
impl<T, U> Hash for Size3D<T, U>
where
T: Hash,
{
fn hash<H: core::hash::Hasher>(&self, h: &mut H) {
self.width.hash(h);
self.height.hash(h);
self.depth.hash(h);
}
}
impl<T: fmt::Debug, U> fmt::Debug for Size3D<T, U> {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
fmt::Debug::fmt(&self.width, f)?;
write!(f, "x")?;
fmt::Debug::fmt(&self.height, f)?;
write!(f, "x")?;
fmt::Debug::fmt(&self.depth, f)
}
}
impl<T: Default, U> Default for Size3D<T, U> {
fn default() -> Self {
Size3D::new(Default::default(), Default::default(), Default::default())
}
}
impl<T, U> Size3D<T, U> {
/// The same as [`Zero::zero()`] but available without importing trait.
///
/// [`Zero::zero()`]: ./num/trait.Zero.html#tymethod.zero
pub fn zero() -> Self
where
T: Zero,
{
Size3D::new(Zero::zero(), Zero::zero(), Zero::zero())
}
/// Constructor taking scalar values.
#[inline]
pub const fn new(width: T, height: T, depth: T) -> Self {
Size3D {
width,
height,
depth,
_unit: PhantomData,
}
}
/// Constructor taking scalar strongly typed lengths.
#[inline]
pub fn from_lengths(width: Length<T, U>, height: Length<T, U>, depth: Length<T, U>) -> Self {
Size3D::new(width.0, height.0, depth.0)
}
/// Constructor setting all components to the same value.
#[inline]
pub fn splat(v: T) -> Self
where
T: Clone,
{
Size3D {
width: v.clone(),
height: v.clone(),
depth: v,
_unit: PhantomData,
}
}
/// Tag a unitless value with units.
#[inline]
pub fn from_untyped(p: Size3D<T, UnknownUnit>) -> Self {
Size3D::new(p.width, p.height, p.depth)
}
}
impl<T: Copy, U> Size3D<T, U> {
/// Return this size as an array of three elements (width, then height, then depth).
#[inline]
pub fn to_array(self) -> [T; 3] {
[self.width, self.height, self.depth]
}
/// Return this size as an array of three elements (width, then height, then depth).
#[inline]
pub fn to_tuple(self) -> (T, T, T) {
(self.width, self.height, self.depth)
}
/// Return this size as a vector with width, height and depth.
#[inline]
pub fn to_vector(self) -> Vector3D<T, U> {
vec3(self.width, self.height, self.depth)
}
/// Drop the units, preserving only the numeric value.
#[inline]
pub fn to_untyped(self) -> Size3D<T, UnknownUnit> {
self.cast_unit()
}
/// Cast the unit
#[inline]
pub fn cast_unit<V>(self) -> Size3D<T, V> {
Size3D::new(self.width, self.height, self.depth)
}
/// Rounds each component to the nearest integer value.
///
/// This behavior is preserved for negative values (unlike the basic cast).
///
/// ```rust
/// # use euclid::size3;
/// enum Mm {}
///
/// assert_eq!(size3::<_, Mm>(-0.1, -0.8, 0.4).round(), size3::<_, Mm>(0.0, -1.0, 0.0))
/// ```
#[inline]
#[must_use]
pub fn round(self) -> Self
where
T: Round,
{
Size3D::new(self.width.round(), self.height.round(), self.depth.round())
}
/// Rounds each component to the smallest integer equal or greater than the original value.
///
/// This behavior is preserved for negative values (unlike the basic cast).
///
/// ```rust
/// # use euclid::size3;
/// enum Mm {}
///
/// assert_eq!(size3::<_, Mm>(-0.1, -0.8, 0.4).ceil(), size3::<_, Mm>(0.0, 0.0, 1.0))
/// ```
#[inline]
#[must_use]
pub fn ceil(self) -> Self
where
T: Ceil,
{
Size3D::new(self.width.ceil(), self.height.ceil(), self.depth.ceil())
}
/// Rounds each component to the biggest integer equal or lower than the original value.
///
/// This behavior is preserved for negative values (unlike the basic cast).
///
/// ```rust
/// # use euclid::size3;
/// enum Mm {}
///
/// assert_eq!(size3::<_, Mm>(-0.1, -0.8, 0.4).floor(), size3::<_, Mm>(-1.0, -1.0, 0.0))
/// ```
#[inline]
#[must_use]
pub fn floor(self) -> Self
where
T: Floor,
{
Size3D::new(self.width.floor(), self.height.floor(), self.depth.floor())
}
/// Returns result of multiplication of all components
pub fn volume(self) -> T
where
T: Mul<Output = T>,
{
self.width * self.height * self.depth
}
/// Linearly interpolate between this size and another size.
///
/// # Example
///
/// ```rust
/// use euclid::size3;
/// use euclid::default::Size3D;
///
/// let from: Size3D<_> = size3(0.0, 10.0, -1.0);
/// let to: Size3D<_> = size3(8.0, -4.0, 0.0);
///
/// assert_eq!(from.lerp(to, -1.0), size3(-8.0, 24.0, -2.0));
/// assert_eq!(from.lerp(to, 0.0), size3( 0.0, 10.0, -1.0));
/// assert_eq!(from.lerp(to, 0.5), size3( 4.0, 3.0, -0.5));
/// assert_eq!(from.lerp(to, 1.0), size3( 8.0, -4.0, 0.0));
/// assert_eq!(from.lerp(to, 2.0), size3(16.0, -18.0, 1.0));
/// ```
#[inline]
pub fn lerp(self, other: Self, t: T) -> Self
where
T: One + Sub<Output = T> + Mul<Output = T> + Add<Output = T>,
{
let one_t = T::one() - t;
self * one_t + other * t
}
}
impl<T: NumCast + Copy, U> Size3D<T, U> {
/// Cast from one numeric representation to another, preserving the units.
///
/// When casting from floating point to integer coordinates, the decimals are truncated
/// as one would expect from a simple cast, but this behavior does not always make sense
/// geometrically. Consider using `round()`, `ceil()` or `floor()` before casting.
#[inline]
pub fn cast<NewT: NumCast>(self) -> Size3D<NewT, U> {
self.try_cast().unwrap()
}
/// Fallible cast from one numeric representation to another, preserving the units.
///
/// When casting from floating point to integer coordinates, the decimals are truncated
/// as one would expect from a simple cast, but this behavior does not always make sense
/// geometrically. Consider using `round()`, `ceil()` or `floor()` before casting.
pub fn try_cast<NewT: NumCast>(self) -> Option<Size3D<NewT, U>> {
match (
NumCast::from(self.width),
NumCast::from(self.height),
NumCast::from(self.depth),
) {
(Some(w), Some(h), Some(d)) => Some(Size3D::new(w, h, d)),
_ => None,
}
}
// Convenience functions for common casts
/// Cast into an `f32` size.
#[inline]
pub fn to_f32(self) -> Size3D<f32, U> {
self.cast()
}
/// Cast into an `f64` size.
#[inline]
pub fn to_f64(self) -> Size3D<f64, U> {
self.cast()
}
/// Cast into an `uint` size, truncating decimals if any.
///
/// When casting from floating point sizes, it is worth considering whether
/// to `round()`, `ceil()` or `floor()` before the cast in order to obtain
/// the desired conversion behavior.
#[inline]
pub fn to_usize(self) -> Size3D<usize, U> {
self.cast()
}
/// Cast into an `u32` size, truncating decimals if any.
///
/// When casting from floating point sizes, it is worth considering whether
/// to `round()`, `ceil()` or `floor()` before the cast in order to obtain
/// the desired conversion behavior.
#[inline]
pub fn to_u32(self) -> Size3D<u32, U> {
self.cast()
}
/// Cast into an `i32` size, truncating decimals if any.
///
/// When casting from floating point sizes, it is worth considering whether
/// to `round()`, `ceil()` or `floor()` before the cast in order to obtain
/// the desired conversion behavior.
#[inline]
pub fn to_i32(self) -> Size3D<i32, U> {
self.cast()
}
/// Cast into an `i64` size, truncating decimals if any.
///
/// When casting from floating point sizes, it is worth considering whether
/// to `round()`, `ceil()` or `floor()` before the cast in order to obtain
/// the desired conversion behavior.
#[inline]
pub fn to_i64(self) -> Size3D<i64, U> {
self.cast()
}
}
impl<T: Float, U> Size3D<T, U> {
/// Returns true if all members are finite.
#[inline]
pub fn is_finite(self) -> bool {
self.width.is_finite() && self.height.is_finite() && self.depth.is_finite()
}
}
impl<T: Signed, U> Size3D<T, U> {
/// Computes the absolute value of each component.
///
/// For `f32` and `f64`, `NaN` will be returned for component if the component is `NaN`.
///
/// For signed integers, `::MIN` will be returned for component if the component is `::MIN`.
pub fn abs(self) -> Self {
size3(self.width.abs(), self.height.abs(), self.depth.abs())
}
/// Returns `true` if all components is positive and `false` any component is zero or negative.
pub fn is_positive(self) -> bool {
self.width.is_positive() && self.height.is_positive() && self.depth.is_positive()
}
}
impl<T: PartialOrd, U> Size3D<T, U> {
/// Returns the size each component of which are minimum of this size and another.
#[inline]
pub fn min(self, other: Self) -> Self {
size3(
min(self.width, other.width),
min(self.height, other.height),
min(self.depth, other.depth),
)
}
/// Returns the size each component of which are maximum of this size and another.
#[inline]
pub fn max(self, other: Self) -> Self {
size3(
max(self.width, other.width),
max(self.height, other.height),
max(self.depth, other.depth),
)
}
/// Returns the size each component of which clamped by corresponding
/// components of `start` and `end`.
///
/// Shortcut for `self.max(start).min(end)`.
#[inline]
pub fn clamp(self, start: Self, end: Self) -> Self
where
T: Copy,
{
self.max(start).min(end)
}
// Returns true if this size is larger or equal to the other size in all dimensions.
#[inline]
pub fn contains(self, other: Self) -> bool {
self.width >= other.width && self.height >= other.height && self.depth >= other.depth
}
/// Returns vector with results of "greater than" operation on each component.
pub fn greater_than(self, other: Self) -> BoolVector3D {
BoolVector3D {
x: self.width > other.width,
y: self.height > other.height,
z: self.depth > other.depth,
}
}
/// Returns vector with results of "lower than" operation on each component.
pub fn lower_than(self, other: Self) -> BoolVector3D {
BoolVector3D {
x: self.width < other.width,
y: self.height < other.height,
z: self.depth < other.depth,
}
}
/// Returns `true` if any component of size is zero, negative or NaN.
pub fn is_empty(self) -> bool
where
T: Zero,
{
let zero = T::zero();
!(self.width > zero && self.height > zero && self.depth > zero)
}
}
impl<T: PartialEq, U> Size3D<T, U> {
/// Returns vector with results of "equal" operation on each component.
pub fn equal(self, other: Self) -> BoolVector3D {
BoolVector3D {
x: self.width == other.width,
y: self.height == other.height,
z: self.depth == other.depth,
}
}
/// Returns vector with results of "not equal" operation on each component.
pub fn not_equal(self, other: Self) -> BoolVector3D {
BoolVector3D {
x: self.width != other.width,
y: self.height != other.height,
z: self.depth != other.depth,
}
}
}
impl<T: Round, U> Round for Size3D<T, U> {
/// See [`Size3D::round()`](#method.round).
#[inline]
fn round(self) -> Self {
self.round()
}
}
impl<T: Ceil, U> Ceil for Size3D<T, U> {
/// See [`Size3D::ceil()`](#method.ceil).
#[inline]
fn ceil(self) -> Self {
self.ceil()
}
}
impl<T: Floor, U> Floor for Size3D<T, U> {
/// See [`Size3D::floor()`](#method.floor).
#[inline]
fn floor(self) -> Self {
self.floor()
}
}
impl<T: Zero, U> Zero for Size3D<T, U> {
#[inline]
fn zero() -> Self {
Size3D::new(Zero::zero(), Zero::zero(), Zero::zero())
}
}
impl<T: Neg, U> Neg for Size3D<T, U> {
type Output = Size3D<T::Output, U>;
#[inline]
fn neg(self) -> Self::Output {
Size3D::new(-self.width, -self.height, -self.depth)
}
}
impl<T: Add, U> Add for Size3D<T, U> {
type Output = Size3D<T::Output, U>;
#[inline]
fn add(self, other: Self) -> Self::Output {
Size3D::new(
self.width + other.width,
self.height + other.height,
self.depth + other.depth,
)
}
}
impl<T: Copy + Add<T, Output = T>, U> Add<&Self> for Size3D<T, U> {
type Output = Self;
fn add(self, other: &Self) -> Self {
Size3D::new(
self.width + other.width,
self.height + other.height,
self.depth + other.depth,
)
}
}
impl<T: Add<Output = T> + Zero, U> Sum for Size3D<T, U> {
fn sum<I: Iterator<Item = Self>>(iter: I) -> Self {
iter.fold(Self::zero(), Add::add)
}
}
impl<'a, T: 'a + Add<Output = T> + Copy + Zero, U: 'a> Sum<&'a Self> for Size3D<T, U> {
fn sum<I: Iterator<Item = &'a Self>>(iter: I) -> Self {
iter.fold(Self::zero(), Add::add)
}
}
impl<T: AddAssign, U> AddAssign for Size3D<T, U> {
#[inline]
fn add_assign(&mut self, other: Self) {
self.width += other.width;
self.height += other.height;
self.depth += other.depth;
}
}
impl<T: Sub, U> Sub for Size3D<T, U> {
type Output = Size3D<T::Output, U>;
#[inline]
fn sub(self, other: Self) -> Self::Output {
Size3D::new(
self.width - other.width,
self.height - other.height,
self.depth - other.depth,
)
}
}
impl<T: SubAssign, U> SubAssign for Size3D<T, U> {
#[inline]
fn sub_assign(&mut self, other: Self) {
self.width -= other.width;
self.height -= other.height;
self.depth -= other.depth;
}
}
impl<T: Copy + Mul, U> Mul<T> for Size3D<T, U> {
type Output = Size3D<T::Output, U>;
#[inline]
#[rustfmt::skip]
fn mul(self, scale: T) -> Self::Output {
Size3D::new(
self.width * scale,
self.height * scale,
self.depth * scale,
)
}
}
impl<T: Copy + MulAssign, U> MulAssign<T> for Size3D<T, U> {
#[inline]
fn mul_assign(&mut self, other: T) {
self.width *= other;
self.height *= other;
self.depth *= other;
}
}
impl<T: Copy + Mul, U1, U2> Mul<Scale<T, U1, U2>> for Size3D<T, U1> {
type Output = Size3D<T::Output, U2>;
#[inline]
fn mul(self, scale: Scale<T, U1, U2>) -> Self::Output {
Size3D::new(
self.width * scale.0,
self.height * scale.0,
self.depth * scale.0,
)
}
}
impl<T: Copy + MulAssign, U> MulAssign<Scale<T, U, U>> for Size3D<T, U> {
#[inline]
fn mul_assign(&mut self, other: Scale<T, U, U>) {
*self *= other.0;
}
}
impl<T: Copy + Div, U> Div<T> for Size3D<T, U> {
type Output = Size3D<T::Output, U>;
#[inline]
#[rustfmt::skip]
fn div(self, scale: T) -> Self::Output {
Size3D::new(
self.width / scale,
self.height / scale,
self.depth / scale,
)
}
}
impl<T: Copy + DivAssign, U> DivAssign<T> for Size3D<T, U> {
#[inline]
fn div_assign(&mut self, other: T) {
self.width /= other;
self.height /= other;
self.depth /= other;
}
}
impl<T: Copy + Div, U1, U2> Div<Scale<T, U1, U2>> for Size3D<T, U2> {
type Output = Size3D<T::Output, U1>;
#[inline]
fn div(self, scale: Scale<T, U1, U2>) -> Self::Output {
Size3D::new(
self.width / scale.0,
self.height / scale.0,
self.depth / scale.0,
)
}
}
impl<T: Copy + DivAssign, U> DivAssign<Scale<T, U, U>> for Size3D<T, U> {
#[inline]
fn div_assign(&mut self, other: Scale<T, U, U>) {
*self /= other.0;
}
}
#[cfg(feature = "mint")]
impl<T, U> From<mint::Vector3<T>> for Size3D<T, U> {
#[inline]
fn from(v: mint::Vector3<T>) -> Self {
size3(v.x, v.y, v.z)
}
}
#[cfg(feature = "mint")]
impl<T, U> From<Size3D<T, U>> for mint::Vector3<T> {
#[inline]
fn from(s: Size3D<T, U>) -> Self {
mint::Vector3 {
x: s.width,
y: s.height,
z: s.depth,
}
}
}
impl<T, U> From<Vector3D<T, U>> for Size3D<T, U> {
#[inline]
fn from(v: Vector3D<T, U>) -> Self {
size3(v.x, v.y, v.z)
}
}
impl<T, U> From<Size3D<T, U>> for [T; 3] {
#[inline]
fn from(s: Size3D<T, U>) -> Self {
[s.width, s.height, s.depth]
}
}
impl<T, U> From<[T; 3]> for Size3D<T, U> {
#[inline]
fn from([w, h, d]: [T; 3]) -> Self {
size3(w, h, d)
}
}
impl<T, U> From<Size3D<T, U>> for (T, T, T) {
#[inline]
fn from(s: Size3D<T, U>) -> Self {
(s.width, s.height, s.depth)
}
}
impl<T, U> From<(T, T, T)> for Size3D<T, U> {
#[inline]
fn from(tuple: (T, T, T)) -> Self {
size3(tuple.0, tuple.1, tuple.2)
}
}
/// Shorthand for `Size3D::new(w, h, d)`.
#[inline]
pub const fn size3<T, U>(w: T, h: T, d: T) -> Size3D<T, U> {
Size3D::new(w, h, d)
}
#[cfg(test)]
mod size3d {
mod ops {
use crate::default::{Size2D, Size3D};
use crate::scale::Scale;
pub enum Mm {}
pub enum Cm {}
pub type Size3DMm<T> = crate::Size3D<T, Mm>;
pub type Size3DCm<T> = crate::Size3D<T, Cm>;
#[test]
pub fn test_neg() {
assert_eq!(-Size3D::new(1.0, 2.0, 3.0), Size3D::new(-1.0, -2.0, -3.0));
assert_eq!(-Size3D::new(0.0, 0.0, 0.0), Size3D::new(-0.0, -0.0, -0.0));
assert_eq!(-Size3D::new(-1.0, -2.0, -3.0), Size3D::new(1.0, 2.0, 3.0));
}
#[test]
pub fn test_add() {
let s1 = Size3D::new(1.0, 2.0, 3.0);
let s2 = Size3D::new(4.0, 5.0, 6.0);
assert_eq!(s1 + s2, Size3D::new(5.0, 7.0, 9.0));
assert_eq!(s1 + &s2, Size3D::new(5.0, 7.0, 9.0));
let s1 = Size3D::new(1.0, 2.0, 3.0);
let s2 = Size3D::new(0.0, 0.0, 0.0);
assert_eq!(s1 + s2, Size3D::new(1.0, 2.0, 3.0));
assert_eq!(s1 + &s2, Size3D::new(1.0, 2.0, 3.0));
let s1 = Size3D::new(1.0, 2.0, 3.0);
let s2 = Size3D::new(-4.0, -5.0, -6.0);
assert_eq!(s1 + s2, Size3D::new(-3.0, -3.0, -3.0));
assert_eq!(s1 + &s2, Size3D::new(-3.0, -3.0, -3.0));
let s1 = Size3D::new(0.0, 0.0, 0.0);
let s2 = Size3D::new(0.0, 0.0, 0.0);
assert_eq!(s1 + s2, Size3D::new(0.0, 0.0, 0.0));
assert_eq!(s1 + &s2, Size3D::new(0.0, 0.0, 0.0));
}
#[test]
pub fn test_sum() {
let sizes = [
Size3D::new(0.0, 1.0, 2.0),
Size3D::new(1.0, 2.0, 3.0),
Size3D::new(2.0, 3.0, 4.0),
];
let sum = Size3D::new(3.0, 6.0, 9.0);
assert_eq!(sizes.iter().sum::<Size3D<_>>(), sum);
}
#[test]
pub fn test_add_assign() {
let mut s = Size3D::new(1.0, 2.0, 3.0);
s += Size3D::new(4.0, 5.0, 6.0);
assert_eq!(s, Size3D::new(5.0, 7.0, 9.0));
let mut s = Size3D::new(1.0, 2.0, 3.0);
s += Size3D::new(0.0, 0.0, 0.0);
assert_eq!(s, Size3D::new(1.0, 2.0, 3.0));
let mut s = Size3D::new(1.0, 2.0, 3.0);
s += Size3D::new(-4.0, -5.0, -6.0);
assert_eq!(s, Size3D::new(-3.0, -3.0, -3.0));
let mut s = Size3D::new(0.0, 0.0, 0.0);
s += Size3D::new(0.0, 0.0, 0.0);
assert_eq!(s, Size3D::new(0.0, 0.0, 0.0));
}
#[test]
pub fn test_sub() {
let s1 = Size3D::new(1.0, 2.0, 3.0);
let s2 = Size3D::new(4.0, 5.0, 6.0);
assert_eq!(s1 - s2, Size3D::new(-3.0, -3.0, -3.0));
let s1 = Size3D::new(1.0, 2.0, 3.0);
let s2 = Size3D::new(0.0, 0.0, 0.0);
assert_eq!(s1 - s2, Size3D::new(1.0, 2.0, 3.0));
let s1 = Size3D::new(1.0, 2.0, 3.0);
let s2 = Size3D::new(-4.0, -5.0, -6.0);
assert_eq!(s1 - s2, Size3D::new(5.0, 7.0, 9.0));
let s1 = Size3D::new(0.0, 0.0, 0.0);
let s2 = Size3D::new(0.0, 0.0, 0.0);
assert_eq!(s1 - s2, Size3D::new(0.0, 0.0, 0.0));
}
#[test]
pub fn test_sub_assign() {
let mut s = Size3D::new(1.0, 2.0, 3.0);
s -= Size3D::new(4.0, 5.0, 6.0);
assert_eq!(s, Size3D::new(-3.0, -3.0, -3.0));
let mut s = Size3D::new(1.0, 2.0, 3.0);
s -= Size3D::new(0.0, 0.0, 0.0);
assert_eq!(s, Size3D::new(1.0, 2.0, 3.0));
let mut s = Size3D::new(1.0, 2.0, 3.0);
s -= Size3D::new(-4.0, -5.0, -6.0);
assert_eq!(s, Size3D::new(5.0, 7.0, 9.0));
let mut s = Size3D::new(0.0, 0.0, 0.0);
s -= Size3D::new(0.0, 0.0, 0.0);
assert_eq!(s, Size3D::new(0.0, 0.0, 0.0));
}
#[test]
pub fn test_mul_scalar() {
let s1: Size3D<f32> = Size3D::new(3.0, 5.0, 7.0);
let result = s1 * 5.0;
assert_eq!(result, Size3D::new(15.0, 25.0, 35.0));
}
#[test]
pub fn test_mul_assign_scalar() {
let mut s1: Size3D<f32> = Size3D::new(3.0, 5.0, 7.0);
s1 *= 5.0;
assert_eq!(s1, Size3D::new(15.0, 25.0, 35.0));
}
#[test]
pub fn test_mul_scale() {
let s1 = Size3DMm::new(1.0, 2.0, 3.0);
let cm_per_mm: Scale<f32, Mm, Cm> = Scale::new(0.1);
let result = s1 * cm_per_mm;
assert_eq!(result, Size3DCm::new(0.1, 0.2, 0.3));
}
#[test]
pub fn test_mul_assign_scale() {
let mut s1 = Size3DMm::new(1.0, 2.0, 3.0);
let scale: Scale<f32, Mm, Mm> = Scale::new(0.1);
s1 *= scale;
assert_eq!(s1, Size3DMm::new(0.1, 0.2, 0.3));
}
#[test]
pub fn test_div_scalar() {
let s1: Size3D<f32> = Size3D::new(15.0, 25.0, 35.0);
let result = s1 / 5.0;
assert_eq!(result, Size3D::new(3.0, 5.0, 7.0));
}
#[test]
pub fn test_div_assign_scalar() {
let mut s1: Size3D<f32> = Size3D::new(15.0, 25.0, 35.0);
s1 /= 5.0;
assert_eq!(s1, Size3D::new(3.0, 5.0, 7.0));
}
#[test]
pub fn test_div_scale() {
let s1 = Size3DCm::new(0.1, 0.2, 0.3);
let cm_per_mm: Scale<f32, Mm, Cm> = Scale::new(0.1);
let result = s1 / cm_per_mm;
assert_eq!(result, Size3DMm::new(1.0, 2.0, 3.0));
}
#[test]
pub fn test_div_assign_scale() {
let mut s1 = Size3DMm::new(0.1, 0.2, 0.3);
let scale: Scale<f32, Mm, Mm> = Scale::new(0.1);
s1 /= scale;
assert_eq!(s1, Size3DMm::new(1.0, 2.0, 3.0));
}
#[test]
fn test_nonempty() {
assert!(!Size2D::new(1.0, 1.0).is_empty());
assert!(!Size3D::new(1.0, 1.0, 1.0).is_empty());
}
#[test]
pub fn test_nan_empty() {
use std::f32::NAN;
assert!(Size3D::new(NAN, 2.0, 3.0).is_empty());
assert!(Size3D::new(0.0, NAN, 0.0).is_empty());
assert!(Size3D::new(1.0, 2.0, NAN).is_empty());
}
}
}