5 releases (3 breaking)

0.5.1 Sep 1, 2024
0.5.0 Feb 25, 2024
0.4.0 Jan 7, 2024
0.3.0 Dec 1, 2023
0.2.0 May 27, 2023

#93 in Science

MIT/Apache

79KB
1K SLoC

Diman is a library for zero-cost compile time unit checking.

use diman::si::dimensions::{Length, Time, Velocity};
use diman::si::units::{seconds, meters, kilometers, hours, hour};

fn get_velocity(x: Length<f64>, t: Time<f64>) -> Velocity<f64> {
    x / t
}

let v1 = get_velocity(36.0 * kilometers, 1.0 * hours);
let v2 = get_velocity(10.0 * meters, 1.0 * seconds);

assert_eq!(v1, v2);
assert_eq!(format!("{} km/h", v1.value_in(kilometers / hour)), "36 km/h");

Diman prevents unit errors at compile time:

let time = 1.0 * seconds;
let length = 10.0 * meters;
let sum = length + time;

This results in a compiler error:

let sum = length + time;
                   ^^^^
= note: expected struct `Quantity<_, Dimension { length: 1, time: 0, mass: 0, temperature: 0, current: 0, amount_of_substance: 0, luminous_intensity: 0 }>`
        found struct `Quantity<_, Dimension { length: 0, time: 1, mass: 0, temperature: 0, current: 0, amount_of_substance: 0, luminous_intensity: 0 }>`

Disclaimer

Diman is implemented using Rust's const generics feature. While min_const_generics has been stabilized since Rust 1.51, Diman uses more complex generic expressions and therefore requires the two currently unstable features generic_const_exprs and adt_const_params.

Moreover, Diman is in its early stages of development and APIs will change.

If you cannot use unstable Rust for your project or require a stable library, consider using uom or dimensioned, both of which do not require any experimental features and are much more mature libraries in general.

Features

  • Invalid operations between physical quantities (adding length and time, for example) turn into compile errors.
  • Newly created quantities are automatically converted to an underlying base representation. This means that the used types are dimensions (such as Length) instead of concrete units (such as meters) which makes for more meaningful code.
  • Systems of dimensions and units can be user defined via the unit_system! macro. This gives the user complete freedom over the choice of dimensions and makes them part of the user's library, so that arbitrary new methods can be implemented on them.
  • The rational-dimensions features allows the usage of quantities and units with rational exponents.
  • f32 and f64 float storage types (behind the f32 and f64 feature gate respectively).
  • The std feature is enabled by default. If disabled, Diman will be a no_std crate, thus suitable for use on embedded devices such as GPU device kernels.
  • The num-traits-libm feature uses libm to provide math functions in no_std environments. While one can use libm in std, the libm implementations are generally slower so this is unlikely to be desirable.
  • Vector storage types via glam (behind the glam-vec2, glam-vec3, glam-dvec2 and glam-dvec3 features).
  • Serialization and Deserialization via serde (behind the serde feature gate, see the official documentation for more info).
  • HDF5 support using hdf5-rs (behind the hdf5 feature gate).
  • Quantities implement the Equivalence trait so that they can be sent via MPI using mpi (behind the mpi feature gate).
  • Random quantities can be generated via rand (behind the rand feature gate, see the official documentation for more info).

The Quantity type

Physical quantities are represented by the Quantity<S, D> struct, where S is the underlying storage type (f32, f64, ...) and D is the dimension of the quantity. Quantity should behave like its underlying storage type whenever allowed by the dimensions.

Arithmetics and math

Addition and subtraction of two quantities is allowed if the dimensions match:

let l = 5.0 * meters + 10.0 * kilometers;

Multiplication and division of two quantities produces a new quantity:

let l = 5.0 * meters;
let t = 2.0 * seconds;
let v: Velocity<f64> = l / t;

Addition and subtraction of a Quantity and a storage type is possible if and only if D is dimensionless:

let l1 = 5.0 * meters;
let l2 = 10.0 * kilometers;
let x = l1 / l2 - 0.5;
let y = 0.5 - l1 / l2;

Quantity implements the dimensionless methods of S, such as sin, cos, etc. for dimensionless quantities:

let l1 = 5.0f64 * meters;
let l2 = 10.0f64 * kilometers;
let angle_radians = (l1 / l2).asin();

Exponentiation and related operations are supported via squared, cubed, powi, sqrt, cbrt:

let length = 2.0f64 * meters;
let area = length.squared();
assert_eq!(area, 4.0 * square_meters);
assert_eq!(area.sqrt(), length);
let vol = length.cubed();
assert_eq!(vol, 8.0 * cubic_meters);
assert_eq!(vol.cbrt(), length);
let foo = length.powi::<4>();

Note that unlike its float equivalent, powi receives its exponent as a generic instead of as a normal function argument. Exponentiation of dimensionful quantities with an non-constant integer is not supported, since the compiler cannot infer the dimension of the return type. However, dimensionless quantities can be raised to arbitrary powers using powf:

let l1 = 2.0f64 * meters;
let l2 = 5.0f64 * kilometers;
let x = (l1 / l2).powf(2.71);

Creation and conversion

New quantities can be created either by multiplying with a unit, or by calling the .new function on the unit:

let l1 = 2.0 * meters;
let l2 = meters.new(2.0);
assert_eq!(l1, l2);

For a full list of the units supported by dimans SI module, see the definitions. Composite units can be defined on the spot via multiplication/division of units:

let v1 = (kilometers / hour).new(3.6);
let v2 = 3.6 * kilometers / hour;
assert_eq!(v1, 1.0 * meters_per_second);
assert_eq!(v2, 1.0 * meters_per_second);

Note that at the moment, the creation of quantities via units defined in this composite way incurs a small performance overhead compared to creation from just a single unit (which is just a single multiplication). This will be fixed once const_fn_floating_point_arithmetic or a similar feature is stabilized.

Conversion into the underlying storage type can be done using the value_in function:

let length = 2.0f64 * kilometers;
assert_eq!(format!("{} m", length.value_in(meters)), "2000 m");

This also works for composite units:

let vel = 10.0f64 * meters_per_second;
assert_eq!(format!("{} km/h", vel.value_in(kilometers / hour)), "36 km/h");

For dimensionless quantities, .value() provides access to the underlying storage types. Alternatively, dimensionless quantities also implement Deref for the same operation.

let l1: Length<f64> = 5.0 * meters;
let l2: Length<f64> = 10.0 * kilometers;
let ratio_value: f64 = (l1 / l2).value();
let ratio_deref: f64 = *(l1 / l2);
assert_eq!(ratio_value, ratio_deref);

Unchecked creation and conversion

If absolutely required, .value_unchecked() provides access to the underlying storage type for all quantities. This is not unit-safe since the return value will depend on the unit system!

let length: Length<f64> = 5.0 * kilometers;
let value: f64 = length.value_unchecked();
assert_eq!(value, 5000.0); // This only holds in SI units!

Similarly, if absolutely required, new quantities can be constructed from storage types using Quantity::new_unchecked. This operation is also not unit-safe!

let length: Length<f64> = Length::new_unchecked(5000.0);
assert_eq!(length, 5.0 * kilometers); // This only holds in SI units!

The combination of value_unchecked and new_unchecked comes in handy when using third party libraries that only takes the raw storage type as argument. As an example, suppose we have a function foo that takes a Vec<f64> and returns a Vec<f64>, and suppose it sorts the numbers or does some other unit safe operation. Then we could reasonably write:

   let lengths: Vec<Length<f64>> = vec![
       1.0 * meters,
       2.0 * kilometers,
       3.0 * meters,
       4.0 * kilometers,
   ];
   let unchecked = lengths.into_iter().map(|x| x.value_unchecked()).collect();
   let fooed = foo(unchecked);
   let result: Vec<_> = fooed
       .into_iter()
       .map(|x| Length::new_unchecked(x))
       .collect();

Debug

Debug is implemented and will print the quantity in its base representation.

let length: Length<f64> = 5.0 * kilometers;
let time: Time<f64> = 1.0 * seconds;
assert_eq!(format!("{:?}", length / time), "5000 m s^-1")

Custom unit systems

The unit_system macro

Diman also provides the unit_system macro for defining custom unit systems for everything that is not covered by SI alone. The macro will add a new quantity type and implement all the required methods and traits to make it usable. As an example, consider the following macro call:

diman::unit_system!(
    quantity_type Quantity;
    dimension_type Dimension;

    dimension Length;
    dimension Time;
    dimension Mass;

    dimension Velocity = Length / Time;
    dimension Frequency = 1 / Time;
    dimension Energy = Mass * Velocity^2;

    #[prefix(kilo, milli)]
    #[base(Length)]
    #[symbol(m)]
    unit meters;

    #[base(Time)]
    #[symbol(s)]
    unit seconds;

    unit hours: Time = 3600 * seconds;
    unit meters_per_second: Velocity = meters / seconds;
    unit kilometers_per_hour: Velocity = kilometers / hours;
    constant SPEED_OF_LIGHT = 299792458 * meters_per_second;
);


fn too_fast(x: Length<f64>, t: Time<f64>) -> bool {
    x / t > 0.1f64 * SPEED_OF_LIGHT
}

too_fast(100.0 * kilometers, 0.3 * hours);

The macro accepts five different keywords:

  1. quantity_type specifies the name of the quantity type. Required for compiler error messages to have something to point to.
  2. dimension_type specifies the name of the dimension type. Required for compiler error messages to have something to point to.
  3. dimension defines a new dimension which is a type. Dimensions without a right hand side are base dimensions (such as Length and Time in this example), whereas dimensions with a right hand side are derived dimensions (such as Velocity in this example).
  4. unit defines a new units, which are methods on the corresponding quantities and constant defines constants. Units without a right-hand side are the base units to one specific base dimension, meaning that they are the unit that will internally be represented with a conversion factor of 1. Base units require the #[base(...)] attribute in order to specify which dimension they are the base unit of. Units with a right hand side are derived from other units.
  5. constant defines a new constant.

SI Prefixes

Unit prefixes can automatically be generated with the #[prefix(...)] attribute for unit statements. For example

#[base(Length)]
#[prefix(kilo, milli)]
#[symbol(m)]
unit meters;

will automatically generate the unit meters with symbol m, as well as kilometers and millimeters with symbols km and mm corresponding to 1e3 m and 1e-3 m. For simplicity, the attribute #[metric_prefixes] is provided, which will generate all metric prefixes from atto- to exa- automatically.

Aliases

Unit aliases can automatically be generated with the #[alias(...)] macro. For example

#[alias(metres)]
unit meters;

will automatically generate a unit metres that has exactly the same definition as meters. This works with prefixes as expected (i.e. an alias is generated for every prefixed unit).

Quantity products and quotients

Sometimes, intermediate types in computations are quantities that don't really have a nice name and are also not needed too many times. Having to add a definition to the unit system for this case can be cumbersome. This is why the Product and Quotient types are provided:

use diman::si::dimensions::{Length, Time};
use diman::{Product, Quotient};

fn foo(l: Length<f64>, t: Time<f64>) -> Product<Length<f64>, Time<f64>> {
    l * t
}

fn bar(l: Length<f64>, t: Time<f64>) -> Quotient<Length<f64>, Time<f64>> {
    l / t
}

Rational dimensions

The rational-dimensions feature allows using quantities with rational exponents in their base dimensions, as opposed to just integer values. This allows expressing defining dimensions and units such as:

unit_system!(
    // ...
    dimension Sorptivity = Length Time^(-1/2);
    unit meters_per_sqrt_second: Sorptivity = meters / seconds^(1/2);
    // ...
);
let l = 2.0 * micrometers;
let t = 5.0 * milliseconds;
let sorptivity: Sorptivity = l / t.sqrt();

The unit system generated with rational-dimensions supports a superset of features of a unit system generated without them. Still, this feature should be enabled only when necessary, since the compiler errors in case of dimension mismatches will be harder to read.

serde

Serialization and deserialization of the units is provided via serde if the serde feature gate is enabled:

#[derive(Serialize, Deserialize, Debug, PartialEq)]
struct Parameters {
    my_length: Length<f64>,
    my_vel: Velocity<f64>,
}

let params: Parameters =
     serde_yaml::from_str("
        my_length: 100 m
        my_vel: 10 m s^-1
    ").unwrap();
assert_eq!(
    params,
    Parameters {
        my_length: 100.0 * meters,
        my_vel: 10.0 * meters_per_second,
    }
)

rand

Diman allows generating random quantities via rand if the rand feature gate is enabled:


let mut rng = rand::thread_rng();
for _ in 0..100 {
    let start = 0.0 * meters;
    let end = 1.0 * kilometers;
    let x = rng.gen_range(start..end);
    assert!(Length::meters(0.0) <= x);
    assert!(x < Length::meters(1000.0));
}

Dependencies

~0.3–7.5MB
~77K SLoC