mindoc/vendor/github.com/boombuler/barcode/aztec/state.go

265 lines
6.9 KiB
Go
Raw Normal View History

2017-05-30 19:09:50 +08:00
package aztec
import (
"fmt"
"github.com/boombuler/barcode/utils"
)
type encodingMode byte
const (
mode_upper encodingMode = iota // 5 bits
mode_lower // 5 bits
mode_digit // 4 bits
mode_mixed // 5 bits
mode_punct // 5 bits
)
var (
// The Latch Table shows, for each pair of Modes, the optimal method for
// getting from one mode to another. In the worst possible case, this can
// be up to 14 bits. In the best possible case, we are already there!
// The high half-word of each entry gives the number of bits.
// The low half-word of each entry are the actual bits necessary to change
latchTable = map[encodingMode]map[encodingMode]int{
mode_upper: {
mode_upper: 0,
mode_lower: (5 << 16) + 28,
mode_digit: (5 << 16) + 30,
mode_mixed: (5 << 16) + 29,
mode_punct: (10 << 16) + (29 << 5) + 30,
},
mode_lower: {
mode_upper: (9 << 16) + (30 << 4) + 14,
mode_lower: 0,
mode_digit: (5 << 16) + 30,
mode_mixed: (5 << 16) + 29,
mode_punct: (10 << 16) + (29 << 5) + 30,
},
mode_digit: {
mode_upper: (4 << 16) + 14,
mode_lower: (9 << 16) + (14 << 5) + 28,
mode_digit: 0,
mode_mixed: (9 << 16) + (14 << 5) + 29,
mode_punct: (14 << 16) + (14 << 10) + (29 << 5) + 30,
},
mode_mixed: {
mode_upper: (5 << 16) + 29,
mode_lower: (5 << 16) + 28,
mode_digit: (10 << 16) + (29 << 5) + 30,
mode_mixed: 0,
mode_punct: (5 << 16) + 30,
},
mode_punct: {
mode_upper: (5 << 16) + 31,
mode_lower: (10 << 16) + (31 << 5) + 28,
mode_digit: (10 << 16) + (31 << 5) + 30,
mode_mixed: (10 << 16) + (31 << 5) + 29,
mode_punct: 0,
},
}
// A map showing the available shift codes. (The shifts to BINARY are not shown)
shiftTable = map[encodingMode]map[encodingMode]int{
mode_upper: {
mode_punct: 0,
},
mode_lower: {
mode_punct: 0,
mode_upper: 28,
},
mode_mixed: {
mode_punct: 0,
},
mode_digit: {
mode_punct: 0,
mode_upper: 15,
},
}
charMap map[encodingMode][]int
)
type state struct {
mode encodingMode
tokens token
bShiftByteCount int
bitCount int
}
type stateSlice []*state
var initialState *state = &state{
mode: mode_upper,
tokens: nil,
bShiftByteCount: 0,
bitCount: 0,
}
func init() {
charMap = make(map[encodingMode][]int)
charMap[mode_upper] = make([]int, 256)
charMap[mode_lower] = make([]int, 256)
charMap[mode_digit] = make([]int, 256)
charMap[mode_mixed] = make([]int, 256)
charMap[mode_punct] = make([]int, 256)
charMap[mode_upper][' '] = 1
for c := 'A'; c <= 'Z'; c++ {
charMap[mode_upper][int(c)] = int(c - 'A' + 2)
}
charMap[mode_lower][' '] = 1
for c := 'a'; c <= 'z'; c++ {
charMap[mode_lower][c] = int(c - 'a' + 2)
}
charMap[mode_digit][' '] = 1
for c := '0'; c <= '9'; c++ {
charMap[mode_digit][c] = int(c - '0' + 2)
}
charMap[mode_digit][','] = 12
charMap[mode_digit]['.'] = 13
mixedTable := []int{
0, ' ', 1, 2, 3, 4, 5, 6, 7, 8, 9, 10,
11, 12, 13, 27, 28, 29, 30, 31, '@', '\\', '^',
'_', '`', '|', '~', 127,
}
for i, v := range mixedTable {
charMap[mode_mixed][v] = i
}
punctTable := []int{
0, '\r', 0, 0, 0, 0, '!', '\'', '#', '$', '%', '&', '\'',
'(', ')', '*', '+', ',', '-', '.', '/', ':', ';', '<', '=', '>', '?',
'[', ']', '{', '}',
}
for i, v := range punctTable {
if v > 0 {
charMap[mode_punct][v] = i
}
}
}
func (em encodingMode) BitCount() byte {
if em == mode_digit {
return 4
}
return 5
}
// Create a new state representing this state with a latch to a (not
// necessary different) mode, and then a code.
func (s *state) latchAndAppend(mode encodingMode, value int) *state {
bitCount := s.bitCount
tokens := s.tokens
if mode != s.mode {
latch := latchTable[s.mode][mode]
tokens = newSimpleToken(tokens, latch&0xFFFF, byte(latch>>16))
bitCount += latch >> 16
}
tokens = newSimpleToken(tokens, value, mode.BitCount())
return &state{
mode: mode,
tokens: tokens,
bShiftByteCount: 0,
bitCount: bitCount + int(mode.BitCount()),
}
}
// Create a new state representing this state, with a temporary shift
// to a different mode to output a single value.
func (s *state) shiftAndAppend(mode encodingMode, value int) *state {
tokens := s.tokens
// Shifts exist only to UPPER and PUNCT, both with tokens size 5.
tokens = newSimpleToken(tokens, shiftTable[s.mode][mode], s.mode.BitCount())
tokens = newSimpleToken(tokens, value, 5)
return &state{
mode: s.mode,
tokens: tokens,
bShiftByteCount: 0,
bitCount: s.bitCount + int(s.mode.BitCount()) + 5,
}
}
// Create a new state representing this state, but an additional character
// output in Binary Shift mode.
func (s *state) addBinaryShiftChar(index int) *state {
tokens := s.tokens
mode := s.mode
bitCnt := s.bitCount
if s.mode == mode_punct || s.mode == mode_digit {
latch := latchTable[s.mode][mode_upper]
tokens = newSimpleToken(tokens, latch&0xFFFF, byte(latch>>16))
bitCnt += latch >> 16
mode = mode_upper
}
deltaBitCount := 8
if s.bShiftByteCount == 0 || s.bShiftByteCount == 31 {
deltaBitCount = 18
} else if s.bShiftByteCount == 62 {
deltaBitCount = 9
}
result := &state{
mode: mode,
tokens: tokens,
bShiftByteCount: s.bShiftByteCount + 1,
bitCount: bitCnt + deltaBitCount,
}
if result.bShiftByteCount == 2047+31 {
// The string is as long as it's allowed to be. We should end it.
result = result.endBinaryShift(index + 1)
}
return result
}
// Create the state identical to this one, but we are no longer in
// Binary Shift mode.
func (s *state) endBinaryShift(index int) *state {
if s.bShiftByteCount == 0 {
return s
}
tokens := newShiftToken(s.tokens, index-s.bShiftByteCount, s.bShiftByteCount)
return &state{
mode: s.mode,
tokens: tokens,
bShiftByteCount: 0,
bitCount: s.bitCount,
}
}
// Returns true if "this" state is better (or equal) to be in than "that"
// state under all possible circumstances.
func (this *state) isBetterThanOrEqualTo(other *state) bool {
mySize := this.bitCount + (latchTable[this.mode][other.mode] >> 16)
if other.bShiftByteCount > 0 && (this.bShiftByteCount == 0 || this.bShiftByteCount > other.bShiftByteCount) {
mySize += 10 // Cost of entering Binary Shift mode.
}
return mySize <= other.bitCount
}
func (s *state) toBitList(text []byte) *utils.BitList {
tokens := make([]token, 0)
se := s.endBinaryShift(len(text))
for t := se.tokens; t != nil; t = t.prev() {
tokens = append(tokens, t)
}
res := new(utils.BitList)
for i := len(tokens) - 1; i >= 0; i-- {
tokens[i].appendTo(res, text)
}
return res
}
func (s *state) String() string {
tokens := make([]token, 0)
for t := s.tokens; t != nil; t = t.prev() {
tokens = append([]token{t}, tokens...)
}
return fmt.Sprintf("M:%d bits=%d bytes=%d: %v", s.mode, s.bitCount, s.bShiftByteCount, tokens)
}