The core protocol of WoopChain
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
woop/numeric/decimal.go

620 lines
15 KiB

package numeric
// Incorporated from cosmos-sdk
import (
"encoding/json"
"errors"
"fmt"
"math/big"
"strconv"
"strings"
)
// Dec represent a decimal. NOTE: never use new(Dec) or else we will panic unmarshalling into the
// nil embedded big.Int
type Dec struct {
*big.Int `json:"int"`
}
// number of decimal places
const (
Precision = 18
// bytes required to represent the above precision
// Ceiling[Log2[999 999 999 999 999 999]]
DecimalPrecisionBits = 60
)
var (
precisionReuse = new(big.Int).Exp(big.NewInt(10), big.NewInt(Precision), nil)
fivePrecision = new(big.Int).Quo(precisionReuse, big.NewInt(2))
precisionMultipliers []*big.Int
zeroInt = big.NewInt(0)
oneInt = big.NewInt(1)
tenInt = big.NewInt(10)
)
// Set precision multipliers
func init() {
precisionMultipliers = make([]*big.Int, Precision+1)
for i := 0; i <= Precision; i++ {
precisionMultipliers[i] = calcPrecisionMultiplier(int64(i))
}
}
func precisionInt() *big.Int {
return new(big.Int).Set(precisionReuse)
}
// ZeroDec ...
func ZeroDec() Dec { return Dec{new(big.Int).Set(zeroInt)} }
// OneDec ...
func OneDec() Dec { return Dec{precisionInt()} }
// SmallestDec ...
func SmallestDec() Dec { return Dec{new(big.Int).Set(oneInt)} }
// calculate the precision multiplier
func calcPrecisionMultiplier(prec int64) *big.Int {
if prec > Precision {
panic(fmt.Sprintf("too much precision, maximum %v, provided %v", Precision, prec))
}
zerosToAdd := Precision - prec
multiplier := new(big.Int).Exp(tenInt, big.NewInt(zerosToAdd), nil)
return multiplier
}
// get the precision multiplier, do not mutate result
func precisionMultiplier(prec int64) *big.Int {
if prec > Precision {
panic(fmt.Sprintf("too much precision, maximum %v, provided %v", Precision, prec))
}
return precisionMultipliers[prec]
}
//______________________________________________________________________________________________
// NewDec creates a new Dec from integer assuming whole number
func NewDec(i int64) Dec {
return NewDecWithPrec(i, 0)
}
// NewDecWithPrec creates a new Dec from integer with decimal place at prec
// CONTRACT: prec <= Precision
func NewDecWithPrec(i, prec int64) Dec {
return Dec{
new(big.Int).Mul(big.NewInt(i), precisionMultiplier(prec)),
}
}
// NewDecFromBigInt creates a new Dec from big integer assuming whole numbers
// CONTRACT: prec <= Precision
func NewDecFromBigInt(i *big.Int) Dec {
return NewDecFromBigIntWithPrec(i, 0)
}
// NewDecFromBigIntWithPrec creates a new Dec from big integer assuming whole numbers
// CONTRACT: prec <= Precision
func NewDecFromBigIntWithPrec(i *big.Int, prec int64) Dec {
return Dec{
new(big.Int).Mul(i, precisionMultiplier(prec)),
}
}
// NewDecFromInt creates a new Dec from big integer assuming whole numbers
// CONTRACT: prec <= Precision
func NewDecFromInt(i *big.Int) Dec {
return NewDecFromIntWithPrec(i, 0)
}
// NewDecFromIntWithPrec creates a new Dec from big integer with decimal place at prec
// CONTRACT: prec <= Precision
func NewDecFromIntWithPrec(i *big.Int, prec int64) Dec {
return Dec{
new(big.Int).Mul(i, precisionMultiplier(prec)),
}
}
// NewDecFromStr creates a decimal from an input decimal string.
// valid must come in the form:
//
// (-) whole integers (.) decimal integers
//
// examples of acceptable input include:
//
// -123.456
// 456.7890
// 345
// -456789
//
// NOTE - An error will return if more decimal places
// are provided in the string than the constant Precision.
//
// CONTRACT - This function does not mutate the input str.
func NewDecFromStr(str string) (d Dec, err error) {
if len(str) == 0 {
return d, errors.New("decimal string is empty")
}
// first extract any negative symbol
neg := false
if str[0] == '-' {
neg = true
str = str[1:]
}
if len(str) == 0 {
return d, errors.New("decimal string is empty")
}
strs := strings.Split(str, ".")
lenDecs := 0
combinedStr := strs[0]
if len(strs) == 2 { // has a decimal place
lenDecs = len(strs[1])
if lenDecs == 0 || len(combinedStr) == 0 {
return d, errors.New("bad decimal length")
}
combinedStr += strs[1]
} else if len(strs) > 2 {
return d, errors.New("too many periods to be a decimal string")
}
if lenDecs > Precision {
return d, fmt.Errorf("too much precision, maximum %v, len decimal %v", Precision, lenDecs)
}
// add some extra zero's to correct to the Precision factor
zerosToAdd := Precision - lenDecs
zeros := fmt.Sprintf(`%0`+strconv.Itoa(zerosToAdd)+`s`, "")
combinedStr += zeros
combined, ok := new(big.Int).SetString(combinedStr, 10) // base 10
if !ok {
return d, fmt.Errorf("bad string to integer conversion, combinedStr: %v", combinedStr)
}
if neg {
combined = new(big.Int).Neg(combined)
}
return Dec{combined}, nil
}
// MustNewDecFromStr Decimal from string, panic on error
func MustNewDecFromStr(s string) Dec {
dec, err := NewDecFromStr(s)
if err != nil {
panic(err)
}
return dec
}
// Copy makes a deep copy of the dec
func (d Dec) Copy() Dec {
if d.IsNil() {
return Dec{}
}
return Dec{
new(big.Int).Set(d.Int),
}
}
// IsNil ...
func (d Dec) IsNil() bool { return d.Int == nil } // is decimal nil
// IsZero ...
func (d Dec) IsZero() bool { return (d.Int).Sign() == 0 } // is equal to zero
// IsNegative ...
func (d Dec) IsNegative() bool { return (d.Int).Sign() == -1 } // is negative
// IsPositive ...
func (d Dec) IsPositive() bool { return (d.Int).Sign() == 1 } // is positive
// Equal ...
func (d Dec) Equal(d2 Dec) bool { return (d.Int).Cmp(d2.Int) == 0 } // equal decimals
// GT ...
func (d Dec) GT(d2 Dec) bool { return (d.Int).Cmp(d2.Int) > 0 } // greater than
// GTE ...
func (d Dec) GTE(d2 Dec) bool { return (d.Int).Cmp(d2.Int) >= 0 } // greater than or equal
// LT ...
func (d Dec) LT(d2 Dec) bool { return (d.Int).Cmp(d2.Int) < 0 } // less than
// LTE ...
func (d Dec) LTE(d2 Dec) bool { return (d.Int).Cmp(d2.Int) <= 0 } // less than or equal
// Neg ...
func (d Dec) Neg() Dec { return Dec{new(big.Int).Neg(d.Int)} } // reverse the decimal sign
// Abs ...
func (d Dec) Abs() Dec { return Dec{new(big.Int).Abs(d.Int)} } // absolute value
// Add addition
func (d Dec) Add(d2 Dec) Dec {
res := new(big.Int).Add(d.Int, d2.Int)
if res.BitLen() > 255+DecimalPrecisionBits {
panic("Int overflow")
}
return Dec{res}
}
// Sub subtraction
func (d Dec) Sub(d2 Dec) Dec {
res := new(big.Int).Sub(d.Int, d2.Int)
if res.BitLen() > 255+DecimalPrecisionBits {
panic("Int overflow")
}
return Dec{res}
}
// Mul multiplication
func (d Dec) Mul(d2 Dec) Dec {
mul := new(big.Int).Mul(d.Int, d2.Int)
chopped := chopPrecisionAndRound(mul)
if chopped.BitLen() > 255+DecimalPrecisionBits {
panic("Int overflow")
}
return Dec{chopped}
}
// MulTruncate multiplication truncate
func (d Dec) MulTruncate(d2 Dec) Dec {
mul := new(big.Int).Mul(d.Int, d2.Int)
chopped := chopPrecisionAndTruncate(mul)
if chopped.BitLen() > 255+DecimalPrecisionBits {
panic("Int overflow")
}
return Dec{chopped}
}
// MulInt multiplication
func (d Dec) MulInt(i *big.Int) Dec {
mul := new(big.Int).Mul(d.Int, i)
if mul.BitLen() > 255+DecimalPrecisionBits {
panic("Int overflow")
}
return Dec{mul}
}
// MulInt64 - multiplication with int64
func (d Dec) MulInt64(i int64) Dec {
mul := new(big.Int).Mul(d.Int, big.NewInt(i))
if mul.BitLen() > 255+DecimalPrecisionBits {
panic("Int overflow")
}
return Dec{mul}
}
// Quo quotient
func (d Dec) Quo(d2 Dec) Dec {
// multiply precision twice
mul := new(big.Int).Mul(d.Int, precisionReuse)
mul.Mul(mul, precisionReuse)
quo := new(big.Int).Quo(mul, d2.Int)
chopped := chopPrecisionAndRound(quo)
if chopped.BitLen() > 255+DecimalPrecisionBits {
panic("Int overflow")
}
return Dec{chopped}
}
// QuoTruncate quotient truncate
func (d Dec) QuoTruncate(d2 Dec) Dec {
// multiply precision twice
mul := new(big.Int).Mul(d.Int, precisionReuse)
mul.Mul(mul, precisionReuse)
quo := new(big.Int).Quo(mul, d2.Int)
chopped := chopPrecisionAndTruncate(quo)
if chopped.BitLen() > 255+DecimalPrecisionBits {
panic("Int overflow")
}
return Dec{chopped}
}
// QuoRoundUp quotient, round up
func (d Dec) QuoRoundUp(d2 Dec) Dec {
// multiply precision twice
mul := new(big.Int).Mul(d.Int, precisionReuse)
mul.Mul(mul, precisionReuse)
quo := new(big.Int).Quo(mul, d2.Int)
chopped := chopPrecisionAndRoundUp(quo)
if chopped.BitLen() > 255+DecimalPrecisionBits {
panic("Int overflow")
}
return Dec{chopped}
}
// QuoInt quotient
func (d Dec) QuoInt(i *big.Int) Dec {
mul := new(big.Int).Quo(d.Int, i)
return Dec{mul}
}
// QuoInt64 - quotient with int64
func (d Dec) QuoInt64(i int64) Dec {
mul := new(big.Int).Quo(d.Int, big.NewInt(i))
return Dec{mul}
}
// IsInteger is integer, e.g. decimals are zero
func (d Dec) IsInteger() bool {
return new(big.Int).Rem(d.Int, precisionReuse).Sign() == 0
}
// Format decimal state
func (d Dec) Format(s fmt.State, verb rune) {
_, err := s.Write([]byte(d.String()))
if err != nil {
panic(err)
}
}
func (d Dec) String() string {
if d.Int == nil {
return d.Int.String()
}
isNeg := d.IsNegative()
if d.IsNegative() {
d = d.Neg()
}
bzInt, err := d.Int.MarshalText()
if err != nil {
return ""
}
inputSize := len(bzInt)
var bzStr []byte
// TODO: Remove trailing zeros
// case 1, purely decimal
if inputSize <= Precision {
bzStr = make([]byte, Precision+2)
// 0. prefix
bzStr[0] = byte('0')
bzStr[1] = byte('.')
// set relevant digits to 0
for i := 0; i < Precision-inputSize; i++ {
bzStr[i+2] = byte('0')
}
// set final digits
copy(bzStr[2+(Precision-inputSize):], bzInt)
} else {
// inputSize + 1 to account for the decimal point that is being added
bzStr = make([]byte, inputSize+1)
decPointPlace := inputSize - Precision
copy(bzStr, bzInt[:decPointPlace]) // pre-decimal digits
bzStr[decPointPlace] = byte('.') // decimal point
copy(bzStr[decPointPlace+1:], bzInt[decPointPlace:]) // post-decimal digits
}
if isNeg {
return "-" + string(bzStr)
}
return string(bzStr)
}
// ____
// __| |__ "chop 'em
// ` \ round!"
// ___|| ~ _ -bankers
// | | __
// | | | __|__|__
// |_____: / | $$$ |
// |________|
// nolint - go-cyclo
// Remove a Precision amount of rightmost digits and perform bankers rounding
// on the remainder (gaussian rounding) on the digits which have been removed.
//
// Mutates the input. Use the non-mutative version if that is undesired
func chopPrecisionAndRound(d *big.Int) *big.Int {
// remove the negative and add it back when returning
if d.Sign() == -1 {
// make d positive, compute chopped value, and then un-mutate d
d = d.Neg(d)
d = chopPrecisionAndRound(d)
d = d.Neg(d)
return d
}
// get the truncated quotient and remainder
quo, rem := d, big.NewInt(0)
quo, rem = quo.QuoRem(d, precisionReuse, rem)
if rem.Sign() == 0 { // remainder is zero
return quo
}
switch rem.Cmp(fivePrecision) {
case -1:
return quo
case 1:
return quo.Add(quo, oneInt)
default: // bankers rounding must take place
// always round to an even number
if quo.Bit(0) == 0 {
return quo
}
return quo.Add(quo, oneInt)
}
}
func chopPrecisionAndRoundUp(d *big.Int) *big.Int {
// remove the negative and add it back when returning
if d.Sign() == -1 {
// make d positive, compute chopped value, and then un-mutate d
d = d.Neg(d)
// truncate since d is negative...
d = chopPrecisionAndTruncate(d)
d = d.Neg(d)
return d
}
// get the truncated quotient and remainder
quo, rem := d, big.NewInt(0)
quo, rem = quo.QuoRem(d, precisionReuse, rem)
if rem.Sign() == 0 { // remainder is zero
return quo
}
return quo.Add(quo, oneInt)
}
func chopPrecisionAndRoundNonMutative(d *big.Int) *big.Int {
tmp := new(big.Int).Set(d)
return chopPrecisionAndRound(tmp)
}
// RoundInt64 rounds the decimal using bankers rounding
func (d Dec) RoundInt64() int64 {
chopped := chopPrecisionAndRoundNonMutative(d.Int)
if !chopped.IsInt64() {
panic("Int64() out of bound")
}
return chopped.Int64()
}
// RoundInt round the decimal using bankers rounding
func (d Dec) RoundInt() *big.Int {
return chopPrecisionAndRoundNonMutative(d.Int)
}
//___________________________________________________________________________________
// similar to chopPrecisionAndRound, but always rounds down
func chopPrecisionAndTruncate(d *big.Int) *big.Int {
return d.Quo(d, precisionReuse)
}
func chopPrecisionAndTruncateNonMutative(d *big.Int) *big.Int {
tmp := new(big.Int).Set(d)
return chopPrecisionAndTruncate(tmp)
}
// TruncateInt64 truncates the decimals from the number and returns an int64
func (d Dec) TruncateInt64() int64 {
chopped := chopPrecisionAndTruncateNonMutative(d.Int)
if !chopped.IsInt64() {
panic("Int64() out of bound")
}
return chopped.Int64()
}
// TruncateInt truncates the decimals from the number and returns an Int
func (d Dec) TruncateInt() *big.Int {
return chopPrecisionAndTruncateNonMutative(d.Int)
}
// TruncateDec truncates the decimals from the number and returns a Dec
func (d Dec) TruncateDec() Dec {
return NewDecFromBigInt(chopPrecisionAndTruncateNonMutative(d.Int))
}
// Ceil returns the smallest interger value (as a decimal) that is greater than
// or equal to the given decimal.
func (d Dec) Ceil() Dec {
tmp := new(big.Int).Set(d.Int)
quo, rem := tmp, big.NewInt(0)
quo, rem = quo.QuoRem(tmp, precisionReuse, rem)
// no need to round with a zero remainder regardless of sign
if rem.Cmp(zeroInt) == 0 {
return NewDecFromBigInt(quo)
}
if rem.Sign() == -1 {
return NewDecFromBigInt(quo)
}
return NewDecFromBigInt(quo.Add(quo, oneInt))
}
//___________________________________________________________________________________
// MarshalJSON marshals the decimal
func (d Dec) MarshalJSON() ([]byte, error) {
if d.Int == nil {
return []byte{}, nil
}
return json.Marshal(d.String())
}
// UnmarshalJSON defines custom decoding scheme
func (d *Dec) UnmarshalJSON(bz []byte) error {
if d.Int == nil {
d.Int = new(big.Int)
}
var text string
err := json.Unmarshal(bz, &text)
if err != nil {
return err
}
// TODO: Reuse dec allocation
newDec, err := NewDecFromStr(text)
if err != nil {
return err
}
d.Int = newDec.Int
return nil
}
// MarshalYAML returns Ythe AML representation.
func (d Dec) MarshalYAML() (interface{}, error) { return d.String(), nil }
//___________________________________________________________________________________
// helpers
// DecsEqual test if two decimal arrays are equal
func DecsEqual(d1s, d2s []Dec) bool {
if len(d1s) != len(d2s) {
return false
}
for i, d1 := range d1s {
if !d1.Equal(d2s[i]) {
return false
}
}
return true
}
// MinDec minimum decimal between two
func MinDec(d1, d2 Dec) Dec {
if d1.LT(d2) {
return d1
}
return d2
}
// MaxDec maximum decimal between two
func MaxDec(d1, d2 Dec) Dec {
if d1.LT(d2) {
return d2
}
return d1
}