registry/vendor/github.com/aws/aws-sdk-go/internal/ini/literal_tokens.go

325 lines
5.8 KiB
Go
Raw Normal View History

package ini
import (
"fmt"
"strconv"
"strings"
)
var (
runesTrue = []rune("true")
runesFalse = []rune("false")
)
var literalValues = [][]rune{
runesTrue,
runesFalse,
}
func isBoolValue(b []rune) bool {
for _, lv := range literalValues {
if isLitValue(lv, b) {
return true
}
}
return false
}
func isLitValue(want, have []rune) bool {
if len(have) < len(want) {
return false
}
for i := 0; i < len(want); i++ {
if want[i] != have[i] {
return false
}
}
return true
}
// isNumberValue will return whether not the leading characters in
// a byte slice is a number. A number is delimited by whitespace or
// the newline token.
//
// A number is defined to be in a binary, octal, decimal (int | float), hex format,
// or in scientific notation.
func isNumberValue(b []rune) bool {
negativeIndex := 0
helper := numberHelper{}
needDigit := false
for i := 0; i < len(b); i++ {
negativeIndex++
switch b[i] {
case '-':
if helper.IsNegative() || negativeIndex != 1 {
return false
}
helper.Determine(b[i])
needDigit = true
continue
case 'e', 'E':
if err := helper.Determine(b[i]); err != nil {
return false
}
negativeIndex = 0
needDigit = true
continue
case 'b':
if helper.numberFormat == hex {
break
}
fallthrough
case 'o', 'x':
needDigit = true
if i == 0 {
return false
}
fallthrough
case '.':
if err := helper.Determine(b[i]); err != nil {
return false
}
needDigit = true
continue
}
if i > 0 && (isNewline(b[i:]) || isWhitespace(b[i])) {
return !needDigit
}
if !helper.CorrectByte(b[i]) {
return false
}
needDigit = false
}
return !needDigit
}
func isValid(b []rune) (bool, int, error) {
if len(b) == 0 {
// TODO: should probably return an error
return false, 0, nil
}
return isValidRune(b[0]), 1, nil
}
func isValidRune(r rune) bool {
return r != ':' && r != '=' && r != '[' && r != ']' && r != ' ' && r != '\n'
}
// ValueType is an enum that will signify what type
// the Value is
type ValueType int
func (v ValueType) String() string {
switch v {
case NoneType:
return "NONE"
case DecimalType:
return "FLOAT"
case IntegerType:
return "INT"
case StringType:
return "STRING"
case BoolType:
return "BOOL"
}
return ""
}
// ValueType enums
const (
NoneType = ValueType(iota)
DecimalType
IntegerType
StringType
QuotedStringType
BoolType
)
// Value is a union container
type Value struct {
Type ValueType
raw []rune
integer int64
decimal float64
boolean bool
str string
}
func newValue(t ValueType, base int, raw []rune) (Value, error) {
v := Value{
Type: t,
raw: raw,
}
var err error
switch t {
case DecimalType:
v.decimal, err = strconv.ParseFloat(string(raw), 64)
case IntegerType:
if base != 10 {
raw = raw[2:]
}
v.integer, err = strconv.ParseInt(string(raw), base, 64)
case StringType:
v.str = string(raw)
case QuotedStringType:
v.str = string(raw[1 : len(raw)-1])
case BoolType:
v.boolean = runeCompare(v.raw, runesTrue)
}
// issue 2253
//
// if the value trying to be parsed is too large, then we will use
// the 'StringType' and raw value instead.
if nerr, ok := err.(*strconv.NumError); ok && nerr.Err == strconv.ErrRange {
v.Type = StringType
v.str = string(raw)
err = nil
}
return v, err
}
// Append will append values and change the type to a string
// type.
func (v *Value) Append(tok Token) {
r := tok.Raw()
if v.Type != QuotedStringType {
v.Type = StringType
r = tok.raw[1 : len(tok.raw)-1]
}
if tok.Type() != TokenLit {
v.raw = append(v.raw, tok.Raw()...)
} else {
v.raw = append(v.raw, r...)
}
}
func (v Value) String() string {
switch v.Type {
case DecimalType:
return fmt.Sprintf("decimal: %f", v.decimal)
case IntegerType:
return fmt.Sprintf("integer: %d", v.integer)
case StringType:
return fmt.Sprintf("string: %s", string(v.raw))
case QuotedStringType:
return fmt.Sprintf("quoted string: %s", string(v.raw))
case BoolType:
return fmt.Sprintf("bool: %t", v.boolean)
default:
return "union not set"
}
}
func newLitToken(b []rune) (Token, int, error) {
n := 0
var err error
token := Token{}
if b[0] == '"' {
n, err = getStringValue(b)
if err != nil {
return token, n, err
}
token = newToken(TokenLit, b[:n], QuotedStringType)
} else if isNumberValue(b) {
var base int
base, n, err = getNumericalValue(b)
if err != nil {
return token, 0, err
}
value := b[:n]
vType := IntegerType
if contains(value, '.') || hasExponent(value) {
vType = DecimalType
}
token = newToken(TokenLit, value, vType)
token.base = base
} else if isBoolValue(b) {
n, err = getBoolValue(b)
token = newToken(TokenLit, b[:n], BoolType)
} else {
n, err = getValue(b)
token = newToken(TokenLit, b[:n], StringType)
}
return token, n, err
}
// IntValue returns an integer value
func (v Value) IntValue() int64 {
return v.integer
}
// FloatValue returns a float value
func (v Value) FloatValue() float64 {
return v.decimal
}
// BoolValue returns a bool value
func (v Value) BoolValue() bool {
return v.boolean
}
func isTrimmable(r rune) bool {
switch r {
case '\n', ' ':
return true
}
return false
}
// StringValue returns the string value
func (v Value) StringValue() string {
switch v.Type {
case StringType:
return strings.TrimFunc(string(v.raw), isTrimmable)
case QuotedStringType:
// preserve all characters in the quotes
return string(removeEscapedCharacters(v.raw[1 : len(v.raw)-1]))
default:
return strings.TrimFunc(string(v.raw), isTrimmable)
}
}
func contains(runes []rune, c rune) bool {
for i := 0; i < len(runes); i++ {
if runes[i] == c {
return true
}
}
return false
}
func runeCompare(v1 []rune, v2 []rune) bool {
if len(v1) != len(v2) {
return false
}
for i := 0; i < len(v1); i++ {
if v1[i] != v2[i] {
return false
}
}
return true
}