khenaidoo | ac63710 | 2019-01-14 15:44:34 -0500 | [diff] [blame] | 1 | // Copyright 2012 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 | // The trie in this file is used to associate the first full character |
| 6 | // in a UTF-8 string to a collation element. |
| 7 | // All but the last byte in a UTF-8 byte sequence are |
| 8 | // used to look up offsets in the index table to be used for the next byte. |
| 9 | // The last byte is used to index into a table of collation elements. |
| 10 | // This file contains the code for the generation of the trie. |
| 11 | |
| 12 | package build |
| 13 | |
| 14 | import ( |
| 15 | "fmt" |
| 16 | "hash/fnv" |
| 17 | "io" |
| 18 | "reflect" |
| 19 | ) |
| 20 | |
| 21 | const ( |
| 22 | blockSize = 64 |
| 23 | blockOffset = 2 // Subtract 2 blocks to compensate for the 0x80 added to continuation bytes. |
| 24 | ) |
| 25 | |
| 26 | type trieHandle struct { |
| 27 | lookupStart uint16 // offset in table for first byte |
| 28 | valueStart uint16 // offset in table for first byte |
| 29 | } |
| 30 | |
| 31 | type trie struct { |
| 32 | index []uint16 |
| 33 | values []uint32 |
| 34 | } |
| 35 | |
| 36 | // trieNode is the intermediate trie structure used for generating a trie. |
| 37 | type trieNode struct { |
| 38 | index []*trieNode |
| 39 | value []uint32 |
| 40 | b byte |
| 41 | refValue uint16 |
| 42 | refIndex uint16 |
| 43 | } |
| 44 | |
| 45 | func newNode() *trieNode { |
| 46 | return &trieNode{ |
| 47 | index: make([]*trieNode, 64), |
| 48 | value: make([]uint32, 128), // root node size is 128 instead of 64 |
| 49 | } |
| 50 | } |
| 51 | |
| 52 | func (n *trieNode) isInternal() bool { |
| 53 | return n.value != nil |
| 54 | } |
| 55 | |
| 56 | func (n *trieNode) insert(r rune, value uint32) { |
| 57 | const maskx = 0x3F // mask out two most-significant bits |
| 58 | str := string(r) |
| 59 | if len(str) == 1 { |
| 60 | n.value[str[0]] = value |
| 61 | return |
| 62 | } |
| 63 | for i := 0; i < len(str)-1; i++ { |
| 64 | b := str[i] & maskx |
| 65 | if n.index == nil { |
| 66 | n.index = make([]*trieNode, blockSize) |
| 67 | } |
| 68 | nn := n.index[b] |
| 69 | if nn == nil { |
| 70 | nn = &trieNode{} |
| 71 | nn.b = b |
| 72 | n.index[b] = nn |
| 73 | } |
| 74 | n = nn |
| 75 | } |
| 76 | if n.value == nil { |
| 77 | n.value = make([]uint32, blockSize) |
| 78 | } |
| 79 | b := str[len(str)-1] & maskx |
| 80 | n.value[b] = value |
| 81 | } |
| 82 | |
| 83 | type trieBuilder struct { |
| 84 | t *trie |
| 85 | |
| 86 | roots []*trieHandle |
| 87 | |
| 88 | lookupBlocks []*trieNode |
| 89 | valueBlocks []*trieNode |
| 90 | |
| 91 | lookupBlockIdx map[uint32]*trieNode |
| 92 | valueBlockIdx map[uint32]*trieNode |
| 93 | } |
| 94 | |
| 95 | func newTrieBuilder() *trieBuilder { |
| 96 | index := &trieBuilder{} |
| 97 | index.lookupBlocks = make([]*trieNode, 0) |
| 98 | index.valueBlocks = make([]*trieNode, 0) |
| 99 | index.lookupBlockIdx = make(map[uint32]*trieNode) |
| 100 | index.valueBlockIdx = make(map[uint32]*trieNode) |
| 101 | // The third nil is the default null block. The other two blocks |
| 102 | // are used to guarantee an offset of at least 3 for each block. |
| 103 | index.lookupBlocks = append(index.lookupBlocks, nil, nil, nil) |
| 104 | index.t = &trie{} |
| 105 | return index |
| 106 | } |
| 107 | |
| 108 | func (b *trieBuilder) computeOffsets(n *trieNode) *trieNode { |
| 109 | hasher := fnv.New32() |
| 110 | if n.index != nil { |
| 111 | for i, nn := range n.index { |
| 112 | var vi, vv uint16 |
| 113 | if nn != nil { |
| 114 | nn = b.computeOffsets(nn) |
| 115 | n.index[i] = nn |
| 116 | vi = nn.refIndex |
| 117 | vv = nn.refValue |
| 118 | } |
| 119 | hasher.Write([]byte{byte(vi >> 8), byte(vi)}) |
| 120 | hasher.Write([]byte{byte(vv >> 8), byte(vv)}) |
| 121 | } |
| 122 | h := hasher.Sum32() |
| 123 | nn, ok := b.lookupBlockIdx[h] |
| 124 | if !ok { |
| 125 | n.refIndex = uint16(len(b.lookupBlocks)) - blockOffset |
| 126 | b.lookupBlocks = append(b.lookupBlocks, n) |
| 127 | b.lookupBlockIdx[h] = n |
| 128 | } else { |
| 129 | n = nn |
| 130 | } |
| 131 | } else { |
| 132 | for _, v := range n.value { |
| 133 | hasher.Write([]byte{byte(v >> 24), byte(v >> 16), byte(v >> 8), byte(v)}) |
| 134 | } |
| 135 | h := hasher.Sum32() |
| 136 | nn, ok := b.valueBlockIdx[h] |
| 137 | if !ok { |
| 138 | n.refValue = uint16(len(b.valueBlocks)) - blockOffset |
| 139 | n.refIndex = n.refValue |
| 140 | b.valueBlocks = append(b.valueBlocks, n) |
| 141 | b.valueBlockIdx[h] = n |
| 142 | } else { |
| 143 | n = nn |
| 144 | } |
| 145 | } |
| 146 | return n |
| 147 | } |
| 148 | |
| 149 | func (b *trieBuilder) addStartValueBlock(n *trieNode) uint16 { |
| 150 | hasher := fnv.New32() |
| 151 | for _, v := range n.value[:2*blockSize] { |
| 152 | hasher.Write([]byte{byte(v >> 24), byte(v >> 16), byte(v >> 8), byte(v)}) |
| 153 | } |
| 154 | h := hasher.Sum32() |
| 155 | nn, ok := b.valueBlockIdx[h] |
| 156 | if !ok { |
| 157 | n.refValue = uint16(len(b.valueBlocks)) |
| 158 | n.refIndex = n.refValue |
| 159 | b.valueBlocks = append(b.valueBlocks, n) |
| 160 | // Add a dummy block to accommodate the double block size. |
| 161 | b.valueBlocks = append(b.valueBlocks, nil) |
| 162 | b.valueBlockIdx[h] = n |
| 163 | } else { |
| 164 | n = nn |
| 165 | } |
| 166 | return n.refValue |
| 167 | } |
| 168 | |
| 169 | func genValueBlock(t *trie, n *trieNode) { |
| 170 | if n != nil { |
| 171 | for _, v := range n.value { |
| 172 | t.values = append(t.values, v) |
| 173 | } |
| 174 | } |
| 175 | } |
| 176 | |
| 177 | func genLookupBlock(t *trie, n *trieNode) { |
| 178 | for _, nn := range n.index { |
| 179 | v := uint16(0) |
| 180 | if nn != nil { |
| 181 | if n.index != nil { |
| 182 | v = nn.refIndex |
| 183 | } else { |
| 184 | v = nn.refValue |
| 185 | } |
| 186 | } |
| 187 | t.index = append(t.index, v) |
| 188 | } |
| 189 | } |
| 190 | |
| 191 | func (b *trieBuilder) addTrie(n *trieNode) *trieHandle { |
| 192 | h := &trieHandle{} |
| 193 | b.roots = append(b.roots, h) |
| 194 | h.valueStart = b.addStartValueBlock(n) |
| 195 | if len(b.roots) == 1 { |
| 196 | // We insert a null block after the first start value block. |
| 197 | // This ensures that continuation bytes UTF-8 sequences of length |
| 198 | // greater than 2 will automatically hit a null block if there |
| 199 | // was an undefined entry. |
| 200 | b.valueBlocks = append(b.valueBlocks, nil) |
| 201 | } |
| 202 | n = b.computeOffsets(n) |
| 203 | // Offset by one extra block as the first byte starts at 0xC0 instead of 0x80. |
| 204 | h.lookupStart = n.refIndex - 1 |
| 205 | return h |
| 206 | } |
| 207 | |
| 208 | // generate generates and returns the trie for n. |
| 209 | func (b *trieBuilder) generate() (t *trie, err error) { |
| 210 | t = b.t |
| 211 | if len(b.valueBlocks) >= 1<<16 { |
| 212 | return nil, fmt.Errorf("maximum number of value blocks exceeded (%d > %d)", len(b.valueBlocks), 1<<16) |
| 213 | } |
| 214 | if len(b.lookupBlocks) >= 1<<16 { |
| 215 | return nil, fmt.Errorf("maximum number of lookup blocks exceeded (%d > %d)", len(b.lookupBlocks), 1<<16) |
| 216 | } |
| 217 | genValueBlock(t, b.valueBlocks[0]) |
| 218 | genValueBlock(t, &trieNode{value: make([]uint32, 64)}) |
| 219 | for i := 2; i < len(b.valueBlocks); i++ { |
| 220 | genValueBlock(t, b.valueBlocks[i]) |
| 221 | } |
| 222 | n := &trieNode{index: make([]*trieNode, 64)} |
| 223 | genLookupBlock(t, n) |
| 224 | genLookupBlock(t, n) |
| 225 | genLookupBlock(t, n) |
| 226 | for i := 3; i < len(b.lookupBlocks); i++ { |
| 227 | genLookupBlock(t, b.lookupBlocks[i]) |
| 228 | } |
| 229 | return b.t, nil |
| 230 | } |
| 231 | |
| 232 | func (t *trie) printArrays(w io.Writer, name string) (n, size int, err error) { |
| 233 | p := func(f string, a ...interface{}) { |
| 234 | nn, e := fmt.Fprintf(w, f, a...) |
| 235 | n += nn |
| 236 | if err == nil { |
| 237 | err = e |
| 238 | } |
| 239 | } |
| 240 | nv := len(t.values) |
| 241 | p("// %sValues: %d entries, %d bytes\n", name, nv, nv*4) |
| 242 | p("// Block 2 is the null block.\n") |
| 243 | p("var %sValues = [%d]uint32 {", name, nv) |
| 244 | var printnewline bool |
| 245 | for i, v := range t.values { |
| 246 | if i%blockSize == 0 { |
| 247 | p("\n\t// Block %#x, offset %#x", i/blockSize, i) |
| 248 | } |
| 249 | if i%4 == 0 { |
| 250 | printnewline = true |
| 251 | } |
| 252 | if v != 0 { |
| 253 | if printnewline { |
| 254 | p("\n\t") |
| 255 | printnewline = false |
| 256 | } |
| 257 | p("%#04x:%#08x, ", i, v) |
| 258 | } |
| 259 | } |
| 260 | p("\n}\n\n") |
| 261 | ni := len(t.index) |
| 262 | p("// %sLookup: %d entries, %d bytes\n", name, ni, ni*2) |
| 263 | p("// Block 0 is the null block.\n") |
| 264 | p("var %sLookup = [%d]uint16 {", name, ni) |
| 265 | printnewline = false |
| 266 | for i, v := range t.index { |
| 267 | if i%blockSize == 0 { |
| 268 | p("\n\t// Block %#x, offset %#x", i/blockSize, i) |
| 269 | } |
| 270 | if i%8 == 0 { |
| 271 | printnewline = true |
| 272 | } |
| 273 | if v != 0 { |
| 274 | if printnewline { |
| 275 | p("\n\t") |
| 276 | printnewline = false |
| 277 | } |
| 278 | p("%#03x:%#02x, ", i, v) |
| 279 | } |
| 280 | } |
| 281 | p("\n}\n\n") |
| 282 | return n, nv*4 + ni*2, err |
| 283 | } |
| 284 | |
| 285 | func (t *trie) printStruct(w io.Writer, handle *trieHandle, name string) (n, sz int, err error) { |
| 286 | const msg = "trie{ %sLookup[%d:], %sValues[%d:], %sLookup[:], %sValues[:]}" |
| 287 | n, err = fmt.Fprintf(w, msg, name, handle.lookupStart*blockSize, name, handle.valueStart*blockSize, name, name) |
| 288 | sz += int(reflect.TypeOf(trie{}).Size()) |
| 289 | return |
| 290 | } |