githaven/vendor/github.com/denisenkom/go-mssqldb/internal/decimal/decimal.go

253 lines
5.5 KiB
Go
Vendored

package decimal
import (
"encoding/binary"
"errors"
"fmt"
"math"
"math/big"
"strings"
)
// Decimal represents decimal type in the Microsoft Open Specifications: http://msdn.microsoft.com/en-us/library/ee780893.aspx
type Decimal struct {
integer [4]uint32 // Little-endian
positive bool
prec uint8
scale uint8
}
var (
scaletblflt64 [39]float64
int10 big.Int
int1e5 big.Int
)
func init() {
var acc float64 = 1
for i := 0; i <= 38; i++ {
scaletblflt64[i] = acc
acc *= 10
}
int10.SetInt64(10)
int1e5.SetInt64(1e5)
}
const autoScale = 100
// SetInteger sets the ind'th element in the integer array
func (d *Decimal) SetInteger(integer uint32, ind uint8) {
d.integer[ind] = integer
}
// SetPositive sets the positive member
func (d *Decimal) SetPositive(positive bool) {
d.positive = positive
}
// SetPrec sets the prec member
func (d *Decimal) SetPrec(prec uint8) {
d.prec = prec
}
// SetScale sets the scale member
func (d *Decimal) SetScale(scale uint8) {
d.scale = scale
}
// IsPositive returns true if the Decimal is positive
func (d *Decimal) IsPositive() bool {
return d.positive
}
// ToFloat64 converts decimal to a float64
func (d Decimal) ToFloat64() float64 {
val := float64(0)
for i := 3; i >= 0; i-- {
val *= 0x100000000
val += float64(d.integer[i])
}
if !d.positive {
val = -val
}
if d.scale != 0 {
val /= scaletblflt64[d.scale]
}
return val
}
// BigInt converts decimal to a bigint
func (d Decimal) BigInt() big.Int {
bytes := make([]byte, 16)
binary.BigEndian.PutUint32(bytes[0:4], d.integer[3])
binary.BigEndian.PutUint32(bytes[4:8], d.integer[2])
binary.BigEndian.PutUint32(bytes[8:12], d.integer[1])
binary.BigEndian.PutUint32(bytes[12:16], d.integer[0])
var x big.Int
x.SetBytes(bytes)
if !d.positive {
x.Neg(&x)
}
return x
}
// Bytes converts decimal to a scaled byte slice
func (d Decimal) Bytes() []byte {
x := d.BigInt()
return ScaleBytes(x.String(), d.scale)
}
// UnscaledBytes converts decimal to a unscaled byte slice
func (d Decimal) UnscaledBytes() []byte {
x := d.BigInt()
return x.Bytes()
}
// String converts decimal to a string
func (d Decimal) String() string {
return string(d.Bytes())
}
// Float64ToDecimal converts float64 to decimal
func Float64ToDecimal(f float64) (Decimal, error) {
return Float64ToDecimalScale(f, autoScale)
}
// Float64ToDecimalScale converts float64 to decimal; user can specify the scale
func Float64ToDecimalScale(f float64, scale uint8) (Decimal, error) {
var dec Decimal
if math.IsNaN(f) {
return dec, errors.New("NaN")
}
if math.IsInf(f, 0) {
return dec, errors.New("Infinity can't be converted to decimal")
}
dec.positive = f >= 0
if !dec.positive {
f = math.Abs(f)
}
if f > 3.402823669209385e+38 {
return dec, errors.New("Float value is out of range")
}
dec.prec = 20
var integer float64
for dec.scale = 0; dec.scale <= scale; dec.scale++ {
integer = f * scaletblflt64[dec.scale]
_, frac := math.Modf(integer)
if frac == 0 && scale == autoScale {
break
}
}
for i := 0; i < 4; i++ {
mod := math.Mod(integer, 0x100000000)
integer -= mod
integer /= 0x100000000
dec.integer[i] = uint32(mod)
if mod-math.Trunc(mod) >= 0.5 {
dec.integer[i] = uint32(mod) + 1
}
}
return dec, nil
}
// Int64ToDecimalScale converts float64 to decimal; user can specify the scale
func Int64ToDecimalScale(v int64, scale uint8) Decimal {
positive := v >= 0
if !positive {
if v == math.MinInt64 {
// Special case - can't negate
return Decimal{
integer: [4]uint32{0, 0x80000000, 0, 0},
positive: false,
prec: 20,
scale: 0,
}
}
v = -v
}
return Decimal{
integer: [4]uint32{uint32(v), uint32(v >> 32), 0, 0},
positive: positive,
prec: 20,
scale: scale,
}
}
// StringToDecimalScale converts string to decimal
func StringToDecimalScale(v string, outScale uint8) (Decimal, error) {
var r big.Int
var unscaled string
var inScale int
point := strings.LastIndexByte(v, '.')
if point == -1 {
inScale = 0
unscaled = v
} else {
inScale = len(v) - point - 1
unscaled = v[:point] + v[point+1:]
}
if inScale > math.MaxUint8 {
return Decimal{}, fmt.Errorf("can't parse %q as a decimal number: scale too large", v)
}
_, ok := r.SetString(unscaled, 10)
if !ok {
return Decimal{}, fmt.Errorf("can't parse %q as a decimal number", v)
}
if inScale > int(outScale) {
return Decimal{}, fmt.Errorf("can't parse %q as a decimal number: scale %d is larger than the scale %d of the target column", v, inScale, outScale)
}
for inScale < int(outScale) {
if int(outScale)-inScale >= 5 {
r.Mul(&r, &int1e5)
inScale += 5
} else {
r.Mul(&r, &int10)
inScale++
}
}
bytes := r.Bytes()
if len(bytes) > 16 {
return Decimal{}, fmt.Errorf("can't parse %q as a decimal number: precision too large", v)
}
var out [4]uint32
for i, b := range bytes {
pos := len(bytes) - i - 1
out[pos/4] += uint32(b) << uint(pos%4*8)
}
return Decimal{
integer: out,
positive: r.Sign() >= 0,
prec: 20,
scale: uint8(inScale),
}, nil
}
// ScaleBytes converts a stringified decimal to a scaled byte slice
func ScaleBytes(s string, scale uint8) []byte {
z := make([]byte, 0, len(s)+1)
if s[0] == '-' || s[0] == '+' {
z = append(z, byte(s[0]))
s = s[1:]
}
pos := len(s) - int(scale)
if pos <= 0 {
z = append(z, byte('0'))
} else if pos > 0 {
z = append(z, []byte(s[:pos])...)
}
if scale > 0 {
z = append(z, byte('.'))
for pos < 0 {
z = append(z, byte('0'))
pos++
}
z = append(z, []byte(s[pos:])...)
}
return z
}