3 releases

0.1.2 Nov 26, 2021
0.1.1 Nov 26, 2021
0.1.0 Oct 29, 2021

#50 in #fixed-point


Used in 2 crates

Custom license

13KB
278 lines

Note

The developpment of this package has been ceased, in favor of fpdec.rs.

Being based on const generics, this implementation of a fixed-point Decimal type provides some advantages:

  • Compact memory representation (16 bytes),
  • Very good performance.

Having the number of fractional digits as a constant type parameter provides the compiler with some extra opportunities to optimize the generated code. For example, in the implementation of the Add trait:

impl<const P: u8, const Q: u8> Add<Decimal<Q>> for Decimal<P>
where
    PrecLimitCheck<{ P <= MAX_PREC }>: True,
    PrecLimitCheck<{ Q <= MAX_PREC }>: True,
    PrecLimitCheck<{ const_max_u8(P, Q) <= MAX_PREC }>: True,
{
    type Output = Decimal<{ const_max_u8(P, Q) }>;

    fn add(self, other: Decimal<Q>) -> Self::Output {
        match P.cmp(&Q) {
            Ordering::Equal => Self::Output {
                coeff: Add::add(self.coeff, other.coeff),
            },
            Ordering::Greater => Self::Output {
                coeff: Add::add(
                    self.coeff,
                    mul_pow_ten(other.coeff, P - Q),
                ),
            },
            Ordering::Less => Self::Output {
                coeff: Add::add(
                    mul_pow_ten(self.coeff, Q - P),
                    other.coeff,
                ),
            },
        }
    }
}

For each combination of P and Q the compiler can reduce the code for the match statement to just one case.

And the multiplication of two Decimals is reduced to the multiplication of two integers (i128), because the resulting number of fractional digits is already determined at compile time:

impl<const P: u8, const Q: u8> Mul<Decimal<Q>> for Decimal<P>
where
    PrecLimitCheck<{ P <= MAX_PREC }>: True,
    PrecLimitCheck<{ Q <= MAX_PREC }>: True,
    PrecLimitCheck<{ (const_sum_u8(P, Q)) <= MAX_PREC }>: True,
{
    type Output = Decimal<{ const_sum_u8(P, Q) }>;

    #[inline(always)]
    fn mul(self, other: Decimal<Q>) -> Self::Output {
        Self::Output {
            coeff: self.coeff * other.coeff,
        }
    }
}

But there are also some serious drawbacks:

  • The large number of variants of the generic functions results in large binary code files.
  • Because each Decimal<P> is a different type, there some unusual asymmetries. For example, multipliying two Decimal<P> does not result in a Decimal<P>. I.e. Decimal<P> does not satisfy Mul<Self, Output = Self> like other numerical types.
  • Depends on nightly features.

Overall, the performance gains stemming from the use of const generics do not outweigh the disadvantages.

The package fpdec.rs follows the same objectives as this package. It does not provide the same performance, but avoids the drawbacks mentioned above.

----------

This crate strives to provide a fast implementation of Decimal fixed-point arithmetics.

It is targeted at typical business applications, dealing with numbers representing quantities, money and the like, not at scientific computations, for which the accuracy of floating point math is - in most cases - sufficient.

Objectives

  • "Exact" representation of decimal numbers (no deviation as with binary floating point numbers)
  • No hidden rounding errors (as inherent to floating point math)
  • Very fast operations (by mapping them to integer ops)
  • Range of representable decimal numbers sufficient for typical business applications

At the binary level a Decimal number is represented as a coefficient (stored as an i128 value) combined with a type parameter specifying the number of fractional decimal digits. I. e., the whole implementation is based on "const generics" and needs a rust version supporting this feature.

Status

Experimental (work in progess)

Getting started

Add rust-fixed-point-decimal to your Cargo.toml:

[dependencies]
rust-fixed-point-decimal = "0.1"

Note:

Because the implementation of "const generics" is still incomplete, you have to put the following at the start of your main.rs or lib.rs file:

#![allow(incomplete_features)]
#![feature(generic_const_exprs)]

Usage

A Decimal number can be created in different ways.

The easiest method is to use the procedural macro Dec:

# use rust_fixed_point_decimal::Dec;
let d = Dec!(-17.5);
assert_eq!(d.to_string(), "-17.5");

Alternatively you can convert an integer, a float or a string to a Decimal:

# use rust_fixed_point_decimal::Decimal;
let d = Decimal::<2>::from(297_i32);
assert_eq!(d.to_string(), "297.00");
# #![allow(incomplete_features)]
# #![feature(generic_const_exprs)]
# use rust_fixed_point_decimal::{Decimal, DecimalError};
# use std::convert::TryFrom;
let d = Decimal::<5>::try_from(83.0025)?;
assert_eq!(d.to_string(), "83.00250");
# Ok::<(), DecimalError>(())
# #![allow(incomplete_features)]
# #![feature(generic_const_exprs)]
# use rust_fixed_point_decimal::{Decimal, ParseDecimalError};
# use std::str::FromStr;
let d = Decimal::<4>::from_str("38.207")?;
assert_eq!(d.to_string(), "38.2070");
# Ok::<(), ParseDecimalError>(())

The sign of a Decimal can be inverted using the unary minus operator and a Decimal instance can be compared to other instances of type Decimal or all basic types of integers (besides u128 and atm besides u8, which causes a compiler error):

# #![allow(incomplete_features)]
# #![feature(generic_const_exprs)]
# use rust_fixed_point_decimal::{Dec, Decimal};
let x = Dec!(129.24);
let y = -x;
assert_eq!(y.to_string(), "-129.24");
assert!(-129_i64 > y);
let z = -y;
assert_eq!(x, z);
let z = Dec!(0.00097);
assert!(x > z);
assert!(y <= z);
assert!(z != 7_u32);
assert!(7_u32 == Dec!(7.00));

Decimal supports all five binary numerical operators +, -, *, /, and %, with two Decimals or with a Decimal and a basic integer (besides u128):

# #![allow(incomplete_features)]
# #![feature(generic_const_exprs)]
# use rust_fixed_point_decimal::{Dec, Decimal};
let x = Dec!(17.5);
let y = Dec!(6.40);
let z = x + y;
assert_eq!(z.to_string(), "23.90");
let z = x - y;
assert_eq!(z.to_string(), "11.10");
let z = x * y;
assert_eq!(z.to_string(), "112.000");
let z = x / y;
assert_eq!(z.to_string(), "2.734375000");
let z = x % y;
assert_eq!(z.to_string(), "4.70");
# #![allow(incomplete_features)]
# #![feature(generic_const_exprs)]
# use rust_fixed_point_decimal::{Dec, Decimal};
let x = Dec!(17.5);
let y = -5_i64;
let z = x + y;
assert_eq!(z.to_string(), "12.5");
let z = x - y;
assert_eq!(z.to_string(), "22.5");
let z = y * x;
assert_eq!(z.to_string(), "-87.5");
let z = x / y;
assert_eq!(z.to_string(), "-3.500000000");
let z = x % y;
assert_eq!(z.to_string(), "2.5");

All these binary numeric operators panic if the result is not representable as a Decimal according to the constraints stated above. In addition there are functions implementing "checked" variants of the operators which return Option::None instead of panicking.

For Multiplication and Division there are also functions which return a result rounded to a number of fractional digits determined by the target type:

# #![allow(incomplete_features)]
# #![feature(generic_const_exprs)]
# use rust_fixed_point_decimal::{Dec, Decimal, DivRounded, MulRounded};
let x = Dec!(17.5);
let y = Dec!(6.47);
let z: Decimal<1> = x.mul_rounded(y);
assert_eq!(z.to_string(), "113.2");
let z: Decimal<3> = x.div_rounded(y);
assert_eq!(z.to_string(), "2.705");

No runtime deps