355 lines
7.3 KiB
Go
355 lines
7.3 KiB
Go
|
package pdf417
|
||
|
|
||
|
import (
|
||
|
"errors"
|
||
|
"math/big"
|
||
|
|
||
|
"github.com/boombuler/barcode/utils"
|
||
|
)
|
||
|
|
||
|
type encodingMode byte
|
||
|
|
||
|
type subMode byte
|
||
|
|
||
|
const (
|
||
|
encText encodingMode = iota
|
||
|
encNumeric
|
||
|
encBinary
|
||
|
|
||
|
subUpper subMode = iota
|
||
|
subLower
|
||
|
subMixed
|
||
|
subPunct
|
||
|
|
||
|
latch_to_text = 900
|
||
|
latch_to_byte_padded = 901
|
||
|
latch_to_numeric = 902
|
||
|
latch_to_byte = 924
|
||
|
shift_to_byte = 913
|
||
|
|
||
|
min_numeric_count = 13
|
||
|
)
|
||
|
|
||
|
var (
|
||
|
mixedMap map[rune]int
|
||
|
punctMap map[rune]int
|
||
|
)
|
||
|
|
||
|
func init() {
|
||
|
mixedMap = make(map[rune]int)
|
||
|
mixedRaw := []rune{
|
||
|
48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 38, 13, 9, 44, 58,
|
||
|
35, 45, 46, 36, 47, 43, 37, 42, 61, 94, 0, 32, 0, 0, 0,
|
||
|
}
|
||
|
for idx, ch := range mixedRaw {
|
||
|
if ch > 0 {
|
||
|
mixedMap[ch] = idx
|
||
|
}
|
||
|
}
|
||
|
|
||
|
punctMap = make(map[rune]int)
|
||
|
punctRaw := []rune{
|
||
|
59, 60, 62, 64, 91, 92, 93, 95, 96, 126, 33, 13, 9, 44, 58,
|
||
|
10, 45, 46, 36, 47, 34, 124, 42, 40, 41, 63, 123, 125, 39, 0,
|
||
|
}
|
||
|
for idx, ch := range punctRaw {
|
||
|
if ch > 0 {
|
||
|
punctMap[ch] = idx
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
|
||
|
func determineConsecutiveDigitCount(data []rune) int {
|
||
|
cnt := 0
|
||
|
for _, r := range data {
|
||
|
if utils.RuneToInt(r) == -1 {
|
||
|
break
|
||
|
}
|
||
|
cnt++
|
||
|
}
|
||
|
return cnt
|
||
|
}
|
||
|
|
||
|
func encodeNumeric(digits []rune) ([]int, error) {
|
||
|
digitCount := len(digits)
|
||
|
chunkCount := digitCount / 44
|
||
|
if digitCount%44 != 0 {
|
||
|
chunkCount++
|
||
|
}
|
||
|
|
||
|
codeWords := []int{}
|
||
|
|
||
|
for i := 0; i < chunkCount; i++ {
|
||
|
start := i * 44
|
||
|
end := start + 44
|
||
|
if end > digitCount {
|
||
|
end = digitCount
|
||
|
}
|
||
|
chunk := digits[start:end]
|
||
|
|
||
|
chunkNum := big.NewInt(0)
|
||
|
_, ok := chunkNum.SetString("1"+string(chunk), 10)
|
||
|
|
||
|
if !ok {
|
||
|
return nil, errors.New("Failed converting: " + string(chunk))
|
||
|
}
|
||
|
|
||
|
cws := []int{}
|
||
|
|
||
|
for chunkNum.Cmp(big.NewInt(0)) > 0 {
|
||
|
newChunk, cw := chunkNum.DivMod(chunkNum, big.NewInt(900), big.NewInt(0))
|
||
|
chunkNum = newChunk
|
||
|
cws = append([]int{int(cw.Int64())}, cws...)
|
||
|
}
|
||
|
|
||
|
codeWords = append(codeWords, cws...)
|
||
|
}
|
||
|
|
||
|
return codeWords, nil
|
||
|
}
|
||
|
|
||
|
func determineConsecutiveTextCount(msg []rune) int {
|
||
|
result := 0
|
||
|
|
||
|
isText := func(ch rune) bool {
|
||
|
return ch == '\t' || ch == '\n' || ch == '\r' || (ch >= 32 && ch <= 126)
|
||
|
}
|
||
|
|
||
|
for i, ch := range msg {
|
||
|
numericCount := determineConsecutiveDigitCount(msg[i:])
|
||
|
if numericCount >= min_numeric_count || (numericCount == 0 && !isText(ch)) {
|
||
|
break
|
||
|
}
|
||
|
|
||
|
result++
|
||
|
}
|
||
|
return result
|
||
|
}
|
||
|
|
||
|
func encodeText(text []rune, submode subMode) (subMode, []int) {
|
||
|
isAlphaUpper := func(ch rune) bool {
|
||
|
return ch == ' ' || (ch >= 'A' && ch <= 'Z')
|
||
|
}
|
||
|
isAlphaLower := func(ch rune) bool {
|
||
|
return ch == ' ' || (ch >= 'a' && ch <= 'z')
|
||
|
}
|
||
|
isMixed := func(ch rune) bool {
|
||
|
_, ok := mixedMap[ch]
|
||
|
return ok
|
||
|
}
|
||
|
isPunctuation := func(ch rune) bool {
|
||
|
_, ok := punctMap[ch]
|
||
|
return ok
|
||
|
}
|
||
|
|
||
|
idx := 0
|
||
|
var tmp []int
|
||
|
for idx < len(text) {
|
||
|
ch := text[idx]
|
||
|
switch submode {
|
||
|
case subUpper:
|
||
|
if isAlphaUpper(ch) {
|
||
|
if ch == ' ' {
|
||
|
tmp = append(tmp, 26) //space
|
||
|
} else {
|
||
|
tmp = append(tmp, int(ch-'A'))
|
||
|
}
|
||
|
} else {
|
||
|
if isAlphaLower(ch) {
|
||
|
submode = subLower
|
||
|
tmp = append(tmp, 27) // lower latch
|
||
|
continue
|
||
|
} else if isMixed(ch) {
|
||
|
submode = subMixed
|
||
|
tmp = append(tmp, 28) // mixed latch
|
||
|
continue
|
||
|
} else {
|
||
|
tmp = append(tmp, 29) // punctuation switch
|
||
|
tmp = append(tmp, punctMap[ch])
|
||
|
break
|
||
|
}
|
||
|
}
|
||
|
break
|
||
|
case subLower:
|
||
|
if isAlphaLower(ch) {
|
||
|
if ch == ' ' {
|
||
|
tmp = append(tmp, 26) //space
|
||
|
} else {
|
||
|
tmp = append(tmp, int(ch-'a'))
|
||
|
}
|
||
|
} else {
|
||
|
if isAlphaUpper(ch) {
|
||
|
tmp = append(tmp, 27) //upper switch
|
||
|
tmp = append(tmp, int(ch-'A'))
|
||
|
break
|
||
|
} else if isMixed(ch) {
|
||
|
submode = subMixed
|
||
|
tmp = append(tmp, 28) //mixed latch
|
||
|
continue
|
||
|
} else {
|
||
|
tmp = append(tmp, 29) //punctuation switch
|
||
|
tmp = append(tmp, punctMap[ch])
|
||
|
break
|
||
|
}
|
||
|
}
|
||
|
break
|
||
|
case subMixed:
|
||
|
if isMixed(ch) {
|
||
|
tmp = append(tmp, mixedMap[ch])
|
||
|
} else {
|
||
|
if isAlphaUpper(ch) {
|
||
|
submode = subUpper
|
||
|
tmp = append(tmp, 28) //upper latch
|
||
|
continue
|
||
|
} else if isAlphaLower(ch) {
|
||
|
submode = subLower
|
||
|
tmp = append(tmp, 27) //lower latch
|
||
|
continue
|
||
|
} else {
|
||
|
if idx+1 < len(text) {
|
||
|
next := text[idx+1]
|
||
|
if isPunctuation(next) {
|
||
|
submode = subPunct
|
||
|
tmp = append(tmp, 25) //punctuation latch
|
||
|
continue
|
||
|
}
|
||
|
}
|
||
|
tmp = append(tmp, 29) //punctuation switch
|
||
|
tmp = append(tmp, punctMap[ch])
|
||
|
}
|
||
|
}
|
||
|
break
|
||
|
default: //subPunct
|
||
|
if isPunctuation(ch) {
|
||
|
tmp = append(tmp, punctMap[ch])
|
||
|
} else {
|
||
|
submode = subUpper
|
||
|
tmp = append(tmp, 29) //upper latch
|
||
|
continue
|
||
|
}
|
||
|
}
|
||
|
idx++
|
||
|
}
|
||
|
|
||
|
h := 0
|
||
|
result := []int{}
|
||
|
for i, val := range tmp {
|
||
|
if i%2 != 0 {
|
||
|
h = (h * 30) + val
|
||
|
result = append(result, h)
|
||
|
} else {
|
||
|
h = val
|
||
|
}
|
||
|
}
|
||
|
if len(tmp)%2 != 0 {
|
||
|
result = append(result, (h*30)+29)
|
||
|
}
|
||
|
return submode, result
|
||
|
}
|
||
|
|
||
|
func determineConsecutiveBinaryCount(msg []byte) int {
|
||
|
result := 0
|
||
|
|
||
|
for i, _ := range msg {
|
||
|
numericCount := determineConsecutiveDigitCount([]rune(string(msg[i:])))
|
||
|
if numericCount >= min_numeric_count {
|
||
|
break
|
||
|
}
|
||
|
textCount := determineConsecutiveTextCount([]rune(string(msg[i:])))
|
||
|
if textCount > 5 {
|
||
|
break
|
||
|
}
|
||
|
result++
|
||
|
}
|
||
|
return result
|
||
|
}
|
||
|
|
||
|
func encodeBinary(data []byte, startmode encodingMode) []int {
|
||
|
result := []int{}
|
||
|
|
||
|
count := len(data)
|
||
|
if count == 1 && startmode == encText {
|
||
|
result = append(result, shift_to_byte)
|
||
|
} else if (count % 6) == 0 {
|
||
|
result = append(result, latch_to_byte)
|
||
|
} else {
|
||
|
result = append(result, latch_to_byte_padded)
|
||
|
}
|
||
|
|
||
|
idx := 0
|
||
|
// Encode sixpacks
|
||
|
if count >= 6 {
|
||
|
words := make([]int, 5)
|
||
|
for (count - idx) >= 6 {
|
||
|
var t int64 = 0
|
||
|
for i := 0; i < 6; i++ {
|
||
|
t = t << 8
|
||
|
t += int64(data[idx+i])
|
||
|
}
|
||
|
for i := 0; i < 5; i++ {
|
||
|
words[4-i] = int(t % 900)
|
||
|
t = t / 900
|
||
|
}
|
||
|
result = append(result, words...)
|
||
|
idx += 6
|
||
|
}
|
||
|
}
|
||
|
//Encode rest (remaining n<5 bytes if any)
|
||
|
for i := idx; i < count; i++ {
|
||
|
result = append(result, int(data[i]&0xff))
|
||
|
}
|
||
|
return result
|
||
|
}
|
||
|
|
||
|
func highlevelEncode(dataStr string) ([]int, error) {
|
||
|
encodingMode := encText
|
||
|
textSubMode := subUpper
|
||
|
|
||
|
result := []int{}
|
||
|
|
||
|
data := []byte(dataStr)
|
||
|
|
||
|
for len(data) > 0 {
|
||
|
numericCount := determineConsecutiveDigitCount([]rune(string(data)))
|
||
|
if numericCount >= min_numeric_count || numericCount == len(data) {
|
||
|
result = append(result, latch_to_numeric)
|
||
|
encodingMode = encNumeric
|
||
|
textSubMode = subUpper
|
||
|
numData, err := encodeNumeric([]rune(string(data[:numericCount])))
|
||
|
if err != nil {
|
||
|
return nil, err
|
||
|
}
|
||
|
result = append(result, numData...)
|
||
|
data = data[numericCount:]
|
||
|
} else {
|
||
|
textCount := determineConsecutiveTextCount([]rune(string(data)))
|
||
|
if textCount >= 5 || textCount == len(data) {
|
||
|
if encodingMode != encText {
|
||
|
result = append(result, latch_to_text)
|
||
|
encodingMode = encText
|
||
|
textSubMode = subUpper
|
||
|
}
|
||
|
var txtData []int
|
||
|
textSubMode, txtData = encodeText([]rune(string(data[:textCount])), textSubMode)
|
||
|
result = append(result, txtData...)
|
||
|
data = data[textCount:]
|
||
|
} else {
|
||
|
binaryCount := determineConsecutiveBinaryCount(data)
|
||
|
if binaryCount == 0 {
|
||
|
binaryCount = 1
|
||
|
}
|
||
|
bytes := data[:binaryCount]
|
||
|
if len(bytes) != 1 || encodingMode != encText {
|
||
|
encodingMode = encBinary
|
||
|
textSubMode = subUpper
|
||
|
}
|
||
|
byteData := encodeBinary(bytes, encodingMode)
|
||
|
result = append(result, byteData...)
|
||
|
data = data[binaryCount:]
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
|
||
|
return result, nil
|
||
|
}
|