| 1 |
747 |
jeremybenn |
// Copyright 2011 The Go Authors. All rights reserved.
|
| 2 |
|
|
// Use of this source code is governed by a BSD-style
|
| 3 |
|
|
// license that can be found in the LICENSE file.
|
| 4 |
|
|
|
| 5 |
|
|
// Trie table generator.
|
| 6 |
|
|
// Used by make*tables tools to generate a go file with trie data structures
|
| 7 |
|
|
// for mapping UTF-8 to a 16-bit value. All but the last byte in a UTF-8 byte
|
| 8 |
|
|
// sequence are used to lookup offsets in the index table to be used for the
|
| 9 |
|
|
// next byte. The last byte is used to index into a table with 16-bit values.
|
| 10 |
|
|
|
| 11 |
|
|
package main
|
| 12 |
|
|
|
| 13 |
|
|
import (
|
| 14 |
|
|
"fmt"
|
| 15 |
|
|
"hash/crc32"
|
| 16 |
|
|
"log"
|
| 17 |
|
|
"unicode/utf8"
|
| 18 |
|
|
)
|
| 19 |
|
|
|
| 20 |
|
|
const blockSize = 64
|
| 21 |
|
|
const maxSparseEntries = 16
|
| 22 |
|
|
|
| 23 |
|
|
// Intermediate trie structure
|
| 24 |
|
|
type trieNode struct {
|
| 25 |
|
|
table [256]*trieNode
|
| 26 |
|
|
value int
|
| 27 |
|
|
b byte
|
| 28 |
|
|
leaf bool
|
| 29 |
|
|
}
|
| 30 |
|
|
|
| 31 |
|
|
func newNode() *trieNode {
|
| 32 |
|
|
return new(trieNode)
|
| 33 |
|
|
}
|
| 34 |
|
|
|
| 35 |
|
|
func (n trieNode) String() string {
|
| 36 |
|
|
s := fmt.Sprint("trieNode{table: { non-nil at index: ")
|
| 37 |
|
|
for i, v := range n.table {
|
| 38 |
|
|
if v != nil {
|
| 39 |
|
|
s += fmt.Sprintf("%d, ", i)
|
| 40 |
|
|
}
|
| 41 |
|
|
}
|
| 42 |
|
|
s += fmt.Sprintf("}, value:%#x, b:%#x leaf:%v}", n.value, n.b, n.leaf)
|
| 43 |
|
|
return s
|
| 44 |
|
|
}
|
| 45 |
|
|
|
| 46 |
|
|
func (n trieNode) isInternal() bool {
|
| 47 |
|
|
internal := true
|
| 48 |
|
|
for i := 0; i < 256; i++ {
|
| 49 |
|
|
if nn := n.table[i]; nn != nil {
|
| 50 |
|
|
if !internal && !nn.leaf {
|
| 51 |
|
|
log.Fatalf("triegen: isInternal: node contains both leaf and non-leaf children (%v)", n)
|
| 52 |
|
|
}
|
| 53 |
|
|
internal = internal && !nn.leaf
|
| 54 |
|
|
}
|
| 55 |
|
|
}
|
| 56 |
|
|
return internal
|
| 57 |
|
|
}
|
| 58 |
|
|
|
| 59 |
|
|
func (n trieNode) mostFrequentStride() int {
|
| 60 |
|
|
counts := make(map[int]int)
|
| 61 |
|
|
v := 0
|
| 62 |
|
|
for _, t := range n.table[0x80 : 0x80+blockSize] {
|
| 63 |
|
|
if t != nil {
|
| 64 |
|
|
if stride := t.value - v; v != 0 && stride >= 0 {
|
| 65 |
|
|
counts[stride]++
|
| 66 |
|
|
}
|
| 67 |
|
|
v = t.value
|
| 68 |
|
|
} else {
|
| 69 |
|
|
v = 0
|
| 70 |
|
|
}
|
| 71 |
|
|
}
|
| 72 |
|
|
var maxs, maxc int
|
| 73 |
|
|
for stride, cnt := range counts {
|
| 74 |
|
|
if cnt > maxc || (cnt == maxc && stride < maxs) {
|
| 75 |
|
|
maxs, maxc = stride, cnt
|
| 76 |
|
|
}
|
| 77 |
|
|
}
|
| 78 |
|
|
return maxs
|
| 79 |
|
|
}
|
| 80 |
|
|
|
| 81 |
|
|
func (n trieNode) countSparseEntries() int {
|
| 82 |
|
|
stride := n.mostFrequentStride()
|
| 83 |
|
|
var count, v int
|
| 84 |
|
|
for _, t := range n.table[0x80 : 0x80+blockSize] {
|
| 85 |
|
|
tv := 0
|
| 86 |
|
|
if t != nil {
|
| 87 |
|
|
tv = t.value
|
| 88 |
|
|
}
|
| 89 |
|
|
if tv-v != stride {
|
| 90 |
|
|
if tv != 0 {
|
| 91 |
|
|
count++
|
| 92 |
|
|
}
|
| 93 |
|
|
}
|
| 94 |
|
|
v = tv
|
| 95 |
|
|
}
|
| 96 |
|
|
return count
|
| 97 |
|
|
}
|
| 98 |
|
|
|
| 99 |
|
|
func (n *trieNode) insert(r rune, value uint16) {
|
| 100 |
|
|
var p [utf8.UTFMax]byte
|
| 101 |
|
|
sz := utf8.EncodeRune(p[:], r)
|
| 102 |
|
|
|
| 103 |
|
|
for i := 0; i < sz; i++ {
|
| 104 |
|
|
if n.leaf {
|
| 105 |
|
|
log.Fatalf("triegen: insert: node (%#v) should not be a leaf", n)
|
| 106 |
|
|
}
|
| 107 |
|
|
nn := n.table[p[i]]
|
| 108 |
|
|
if nn == nil {
|
| 109 |
|
|
nn = newNode()
|
| 110 |
|
|
nn.b = p[i]
|
| 111 |
|
|
n.table[p[i]] = nn
|
| 112 |
|
|
}
|
| 113 |
|
|
n = nn
|
| 114 |
|
|
}
|
| 115 |
|
|
n.value = int(value)
|
| 116 |
|
|
n.leaf = true
|
| 117 |
|
|
}
|
| 118 |
|
|
|
| 119 |
|
|
type nodeIndex struct {
|
| 120 |
|
|
lookupBlocks []*trieNode
|
| 121 |
|
|
valueBlocks []*trieNode
|
| 122 |
|
|
sparseBlocks []*trieNode
|
| 123 |
|
|
sparseOffset []uint16
|
| 124 |
|
|
sparseCount int
|
| 125 |
|
|
|
| 126 |
|
|
lookupBlockIdx map[uint32]int
|
| 127 |
|
|
valueBlockIdx map[uint32]int
|
| 128 |
|
|
}
|
| 129 |
|
|
|
| 130 |
|
|
func newIndex() *nodeIndex {
|
| 131 |
|
|
index := &nodeIndex{}
|
| 132 |
|
|
index.lookupBlocks = make([]*trieNode, 0)
|
| 133 |
|
|
index.valueBlocks = make([]*trieNode, 0)
|
| 134 |
|
|
index.sparseBlocks = make([]*trieNode, 0)
|
| 135 |
|
|
index.sparseOffset = make([]uint16, 1)
|
| 136 |
|
|
index.lookupBlockIdx = make(map[uint32]int)
|
| 137 |
|
|
index.valueBlockIdx = make(map[uint32]int)
|
| 138 |
|
|
return index
|
| 139 |
|
|
}
|
| 140 |
|
|
|
| 141 |
|
|
func computeOffsets(index *nodeIndex, n *trieNode) int {
|
| 142 |
|
|
if n.leaf {
|
| 143 |
|
|
return n.value
|
| 144 |
|
|
}
|
| 145 |
|
|
hasher := crc32.New(crc32.MakeTable(crc32.IEEE))
|
| 146 |
|
|
// We only index continuation bytes.
|
| 147 |
|
|
for i := 0; i < blockSize; i++ {
|
| 148 |
|
|
v := 0
|
| 149 |
|
|
if nn := n.table[0x80+i]; nn != nil {
|
| 150 |
|
|
v = computeOffsets(index, nn)
|
| 151 |
|
|
}
|
| 152 |
|
|
hasher.Write([]byte{uint8(v >> 8), uint8(v)})
|
| 153 |
|
|
}
|
| 154 |
|
|
h := hasher.Sum32()
|
| 155 |
|
|
if n.isInternal() {
|
| 156 |
|
|
v, ok := index.lookupBlockIdx[h]
|
| 157 |
|
|
if !ok {
|
| 158 |
|
|
v = len(index.lookupBlocks)
|
| 159 |
|
|
index.lookupBlocks = append(index.lookupBlocks, n)
|
| 160 |
|
|
index.lookupBlockIdx[h] = v
|
| 161 |
|
|
}
|
| 162 |
|
|
n.value = v
|
| 163 |
|
|
} else {
|
| 164 |
|
|
v, ok := index.valueBlockIdx[h]
|
| 165 |
|
|
if !ok {
|
| 166 |
|
|
if c := n.countSparseEntries(); c > maxSparseEntries {
|
| 167 |
|
|
v = len(index.valueBlocks)
|
| 168 |
|
|
index.valueBlocks = append(index.valueBlocks, n)
|
| 169 |
|
|
index.valueBlockIdx[h] = v
|
| 170 |
|
|
} else {
|
| 171 |
|
|
v = -len(index.sparseOffset)
|
| 172 |
|
|
index.sparseBlocks = append(index.sparseBlocks, n)
|
| 173 |
|
|
index.sparseOffset = append(index.sparseOffset, uint16(index.sparseCount))
|
| 174 |
|
|
index.sparseCount += c + 1
|
| 175 |
|
|
index.valueBlockIdx[h] = v
|
| 176 |
|
|
}
|
| 177 |
|
|
}
|
| 178 |
|
|
n.value = v
|
| 179 |
|
|
}
|
| 180 |
|
|
return n.value
|
| 181 |
|
|
}
|
| 182 |
|
|
|
| 183 |
|
|
func printValueBlock(nr int, n *trieNode, offset int) {
|
| 184 |
|
|
boff := nr * blockSize
|
| 185 |
|
|
fmt.Printf("\n// Block %#x, offset %#x", nr, boff)
|
| 186 |
|
|
var printnewline bool
|
| 187 |
|
|
for i := 0; i < blockSize; i++ {
|
| 188 |
|
|
if i%6 == 0 {
|
| 189 |
|
|
printnewline = true
|
| 190 |
|
|
}
|
| 191 |
|
|
v := 0
|
| 192 |
|
|
if nn := n.table[i+offset]; nn != nil {
|
| 193 |
|
|
v = nn.value
|
| 194 |
|
|
}
|
| 195 |
|
|
if v != 0 {
|
| 196 |
|
|
if printnewline {
|
| 197 |
|
|
fmt.Printf("\n")
|
| 198 |
|
|
printnewline = false
|
| 199 |
|
|
}
|
| 200 |
|
|
fmt.Printf("%#04x:%#04x, ", boff+i, v)
|
| 201 |
|
|
}
|
| 202 |
|
|
}
|
| 203 |
|
|
}
|
| 204 |
|
|
|
| 205 |
|
|
func printSparseBlock(nr int, n *trieNode) {
|
| 206 |
|
|
boff := -n.value
|
| 207 |
|
|
fmt.Printf("\n// Block %#x, offset %#x", nr, boff)
|
| 208 |
|
|
v := 0
|
| 209 |
|
|
//stride := f(n)
|
| 210 |
|
|
stride := n.mostFrequentStride()
|
| 211 |
|
|
c := n.countSparseEntries()
|
| 212 |
|
|
fmt.Printf("\n{value:%#04x,lo:%#02x},", stride, uint8(c))
|
| 213 |
|
|
for i, nn := range n.table[0x80 : 0x80+blockSize] {
|
| 214 |
|
|
nv := 0
|
| 215 |
|
|
if nn != nil {
|
| 216 |
|
|
nv = nn.value
|
| 217 |
|
|
}
|
| 218 |
|
|
if nv-v != stride {
|
| 219 |
|
|
if v != 0 {
|
| 220 |
|
|
fmt.Printf(",hi:%#02x},", 0x80+i-1)
|
| 221 |
|
|
}
|
| 222 |
|
|
if nv != 0 {
|
| 223 |
|
|
fmt.Printf("\n{value:%#04x,lo:%#02x", nv, nn.b)
|
| 224 |
|
|
}
|
| 225 |
|
|
}
|
| 226 |
|
|
v = nv
|
| 227 |
|
|
}
|
| 228 |
|
|
if v != 0 {
|
| 229 |
|
|
fmt.Printf(",hi:%#02x},", 0x80+blockSize-1)
|
| 230 |
|
|
}
|
| 231 |
|
|
}
|
| 232 |
|
|
|
| 233 |
|
|
func printLookupBlock(nr int, n *trieNode, offset, cutoff int) {
|
| 234 |
|
|
boff := nr * blockSize
|
| 235 |
|
|
fmt.Printf("\n// Block %#x, offset %#x", nr, boff)
|
| 236 |
|
|
var printnewline bool
|
| 237 |
|
|
for i := 0; i < blockSize; i++ {
|
| 238 |
|
|
if i%8 == 0 {
|
| 239 |
|
|
printnewline = true
|
| 240 |
|
|
}
|
| 241 |
|
|
v := 0
|
| 242 |
|
|
if nn := n.table[i+offset]; nn != nil {
|
| 243 |
|
|
v = nn.value
|
| 244 |
|
|
}
|
| 245 |
|
|
if v != 0 {
|
| 246 |
|
|
if v < 0 {
|
| 247 |
|
|
v = -v - 1 + cutoff
|
| 248 |
|
|
}
|
| 249 |
|
|
if printnewline {
|
| 250 |
|
|
fmt.Printf("\n")
|
| 251 |
|
|
printnewline = false
|
| 252 |
|
|
}
|
| 253 |
|
|
fmt.Printf("%#03x:%#02x, ", boff+i, v)
|
| 254 |
|
|
}
|
| 255 |
|
|
}
|
| 256 |
|
|
}
|
| 257 |
|
|
|
| 258 |
|
|
// printTables returns the size in bytes of the generated tables.
|
| 259 |
|
|
func (t *trieNode) printTables(name string) int {
|
| 260 |
|
|
index := newIndex()
|
| 261 |
|
|
// Values for 7-bit ASCII are stored in first two block, followed by nil block.
|
| 262 |
|
|
index.valueBlocks = append(index.valueBlocks, nil, nil, nil)
|
| 263 |
|
|
// First byte of multi-byte UTF-8 codepoints are indexed in 4th block.
|
| 264 |
|
|
index.lookupBlocks = append(index.lookupBlocks, nil, nil, nil, nil)
|
| 265 |
|
|
// Index starter bytes of multi-byte UTF-8.
|
| 266 |
|
|
for i := 0xC0; i < 0x100; i++ {
|
| 267 |
|
|
if t.table[i] != nil {
|
| 268 |
|
|
computeOffsets(index, t.table[i])
|
| 269 |
|
|
}
|
| 270 |
|
|
}
|
| 271 |
|
|
|
| 272 |
|
|
nv := len(index.valueBlocks) * blockSize
|
| 273 |
|
|
fmt.Printf("// %sValues: %d entries, %d bytes\n", name, nv, nv*2)
|
| 274 |
|
|
fmt.Printf("// Block 2 is the null block.\n")
|
| 275 |
|
|
fmt.Printf("var %sValues = [%d]uint16 {", name, nv)
|
| 276 |
|
|
printValueBlock(0, t, 0)
|
| 277 |
|
|
printValueBlock(1, t, 64)
|
| 278 |
|
|
printValueBlock(2, newNode(), 0)
|
| 279 |
|
|
for i := 3; i < len(index.valueBlocks); i++ {
|
| 280 |
|
|
printValueBlock(i, index.valueBlocks[i], 0x80)
|
| 281 |
|
|
}
|
| 282 |
|
|
fmt.Print("\n}\n\n")
|
| 283 |
|
|
|
| 284 |
|
|
ls := len(index.sparseBlocks)
|
| 285 |
|
|
fmt.Printf("// %sSparseOffset: %d entries, %d bytes\n", name, ls, ls*2)
|
| 286 |
|
|
fmt.Printf("var %sSparseOffset = %#v\n\n", name, index.sparseOffset[1:])
|
| 287 |
|
|
|
| 288 |
|
|
ns := index.sparseCount
|
| 289 |
|
|
fmt.Printf("// %sSparseValues: %d entries, %d bytes\n", name, ns, ns*4)
|
| 290 |
|
|
fmt.Printf("var %sSparseValues = [%d]valueRange {", name, ns)
|
| 291 |
|
|
for i, n := range index.sparseBlocks {
|
| 292 |
|
|
printSparseBlock(i, n)
|
| 293 |
|
|
}
|
| 294 |
|
|
fmt.Print("\n}\n\n")
|
| 295 |
|
|
|
| 296 |
|
|
cutoff := len(index.valueBlocks)
|
| 297 |
|
|
ni := len(index.lookupBlocks) * blockSize
|
| 298 |
|
|
fmt.Printf("// %sLookup: %d bytes\n", name, ni)
|
| 299 |
|
|
fmt.Printf("// Block 0 is the null block.\n")
|
| 300 |
|
|
fmt.Printf("var %sLookup = [%d]uint8 {", name, ni)
|
| 301 |
|
|
printLookupBlock(0, newNode(), 0, cutoff)
|
| 302 |
|
|
printLookupBlock(1, newNode(), 0, cutoff)
|
| 303 |
|
|
printLookupBlock(2, newNode(), 0, cutoff)
|
| 304 |
|
|
printLookupBlock(3, t, 0xC0, cutoff)
|
| 305 |
|
|
for i := 4; i < len(index.lookupBlocks); i++ {
|
| 306 |
|
|
printLookupBlock(i, index.lookupBlocks[i], 0x80, cutoff)
|
| 307 |
|
|
}
|
| 308 |
|
|
fmt.Print("\n}\n\n")
|
| 309 |
|
|
fmt.Printf("var %sTrie = trie{ %sLookup[:], %sValues[:], %sSparseValues[:], %sSparseOffset[:], %d}\n\n",
|
| 310 |
|
|
name, name, name, name, name, cutoff)
|
| 311 |
|
|
return nv*2 + ns*4 + ni + ls*2
|
| 312 |
|
|
}
|
