moedinha

Fixed-precision decimal numbers in Go, aiming to represent currency values.

Example

package main

import (
	"fmt"

	"github.com/mqzabin/moedinha"
)

func main() {
	a, _ := moedinha.NewFromString("99999999999999999999999999999999999.999999999999999999")
	b, _ := moedinha.NewFromString("888888888888888888.888888888888888888")

	fmt.Println("a + b =", a.Add(b).String())
	// a + b = 100000000000000000888888888888888888.888888888888888887
	fmt.Println("a - b =", a.Sub(b).String())
	// a - b = 99999999999999999111111111111111111.111111111111111111
	fmt.Println("b - a =", b.Sub(a).String())
	// b - a = -99999999999999999111111111111111111.111111111111111111
	fmt.Println("a * b =", a.Mul(b).String())
	// a * b = 88888888888888888888888888888888888799999999999999999.111111111111111111
}

How it works?

moedinha uses an array of uint64 to represent decimal numbers. Each uint64 represents up to 18-digits.

You can set how many uint64 you want to use, and how many of those you want to use as decimal digits. Those settings are set through editing the settings.go file.

The default setting is to use 4 uint64 and 1 of those as decimal digits. This settings can represent numbers up to:

999999999999999999999999999999999999999999999999999999.999999999999999999,

i.e. 54 integer digits and 18 decimal digits.

Since the precision is fixed, overflows during arithmetic operations can happen and the package will call a panic.

Motivation

The shopspring/decimal solve the problem of arbitrary precision decimals in Go, wrapping the math/big structure with an easy-to-use API.

The API translation partially loses the math/big's memory allocation optimizations, making many arithmetic operations reallocate unnecessary memory. For low throughput scenarios, the garbage collector will handle this and the easy API + not worrying about precision/value limits will pay off.

However, in scenarios where there are high throughput and many arithmetical operations per request, the garbage collector will start to be a performance bottleneck.

These problems arise from the "arbitrary precision" hypothesis, which is not required in some real-world scenarios. moedinha makes this hypothesis false to reduce allocations to zero, using fixed-size arrays to represent integer values.

Roadmap

Benchmarks

goos: linux
goarch: amd64
pkg: github.com/mqzabin/moedinha
cpu: AMD Ryzen 7 5700G with Radeon Graphics         
BenchmarkNewFromString/moedinha-16               1643692               735.1 ns/op             0 B/op          0 allocs/op
BenchmarkNewFromString/shopspring-16             1362316               844.2 ns/op           184 B/op          5 allocs/op
BenchmarkString/moedinha-16                      4755722               239.9 ns/op            64 B/op          1 allocs/op
BenchmarkString/shopspring-16                    2683314               482.7 ns/op           320 B/op          5 allocs/op
BenchmarkAdd/moedinha-16                        29274129                39.17 ns/op            0 B/op          0 allocs/op
BenchmarkAdd/shopspring-16                       2546943               426.8 ns/op           304 B/op          8 allocs/op
BenchmarkSub/moedinha-16                         3190180               371.5 ns/op             0 B/op          0 allocs/op
BenchmarkSub/shopspring-16                      13402520               132.8 ns/op            96 B/op          2 allocs/op
BenchmarkMul/moedinha-16                        17169872                63.04 ns/op            0 B/op          0 allocs/op
BenchmarkMul/shopspring-16                       3261558               438.9 ns/op           304 B/op          8 allocs/op

mqzabin / moedinha