1 /* 2 * Copyright (C) 2013 The Android Open Source Project 3 * 4 * Licensed under the Apache License, Version 2.0 (the "License"); 5 * you may not use this file except in compliance with the License. 6 * You may obtain a copy of the License at 7 * 8 * http://www.apache.org/licenses/LICENSE-2.0 9 * 10 * Unless required by applicable law or agreed to in writing, software 11 * distributed under the License is distributed on an "AS IS" BASIS, 12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 13 * See the License for the specific language governing permissions and 14 * limitations under the License. 15 */ 16 17 package com.android.inputmethod.latin.makedict; 18 19 import com.android.inputmethod.latin.makedict.BinaryDictDecoderUtils.CharEncoding; 20 import com.android.inputmethod.latin.makedict.FormatSpec.FormatOptions; 21 import com.android.inputmethod.latin.makedict.FusionDictionary.PtNode; 22 import com.android.inputmethod.latin.makedict.FusionDictionary.PtNodeArray; 23 24 import java.io.ByteArrayOutputStream; 25 import java.io.IOException; 26 import java.io.OutputStream; 27 import java.util.ArrayList; 28 import java.util.HashMap; 29 import java.util.Map.Entry; 30 31 /** 32 * Encodes binary files for a FusionDictionary. 33 * 34 * All the methods in this class are static. 35 * 36 * TODO: Rename this class to DictEncoderUtils. 37 */ 38 public class BinaryDictEncoderUtils { 39 40 private static final boolean DBG = MakedictLog.DBG; 41 42 private BinaryDictEncoderUtils() { 43 // This utility class is not publicly instantiable. 44 } 45 46 // Arbitrary limit to how much passes we consider address size compression should 47 // terminate in. At the time of this writing, our largest dictionary completes 48 // compression in five passes. 49 // If the number of passes exceeds this number, makedict bails with an exception on 50 // suspicion that a bug might be causing an infinite loop. 51 private static final int MAX_PASSES = 24; 52 53 /** 54 * Compute the binary size of the character array. 55 * 56 * If only one character, this is the size of this character. If many, it's the sum of their 57 * sizes + 1 byte for the terminator. 58 * 59 * @param characters the character array 60 * @return the size of the char array, including the terminator if any 61 */ 62 static int getPtNodeCharactersSize(final int[] characters, 63 final HashMap<Integer, Integer> codePointToOneByteCodeMap) { 64 int size = CharEncoding.getCharArraySize(characters, codePointToOneByteCodeMap); 65 if (characters.length > 1) size += FormatSpec.PTNODE_TERMINATOR_SIZE; 66 return size; 67 } 68 69 /** 70 * Compute the binary size of the character array in a PtNode 71 * 72 * If only one character, this is the size of this character. If many, it's the sum of their 73 * sizes + 1 byte for the terminator. 74 * 75 * @param ptNode the PtNode 76 * @return the size of the char array, including the terminator if any 77 */ 78 private static int getPtNodeCharactersSize(final PtNode ptNode, 79 final HashMap<Integer, Integer> codePointToOneByteCodeMap) { 80 return getPtNodeCharactersSize(ptNode.mChars, codePointToOneByteCodeMap); 81 } 82 83 /** 84 * Compute the binary size of the PtNode count for a node array. 85 * @param nodeArray the nodeArray 86 * @return the size of the PtNode count, either 1 or 2 bytes. 87 */ 88 private static int getPtNodeCountSize(final PtNodeArray nodeArray) { 89 return BinaryDictIOUtils.getPtNodeCountSize(nodeArray.mData.size()); 90 } 91 92 /** 93 * Compute the maximum size of a PtNode, assuming 3-byte addresses for everything. 94 * 95 * @param ptNode the PtNode to compute the size of. 96 * @return the maximum size of the PtNode. 97 */ 98 private static int getPtNodeMaximumSize(final PtNode ptNode, 99 final HashMap<Integer, Integer> codePointToOneByteCodeMap) { 100 int size = getNodeHeaderSize(ptNode, codePointToOneByteCodeMap); 101 if (ptNode.isTerminal()) { 102 // If terminal, one byte for the frequency. 103 size += FormatSpec.PTNODE_FREQUENCY_SIZE; 104 } 105 size += FormatSpec.PTNODE_MAX_ADDRESS_SIZE; // For children address 106 if (null != ptNode.mBigrams) { 107 size += (FormatSpec.PTNODE_ATTRIBUTE_FLAGS_SIZE 108 + FormatSpec.PTNODE_ATTRIBUTE_MAX_ADDRESS_SIZE) 109 * ptNode.mBigrams.size(); 110 } 111 return size; 112 } 113 114 /** 115 * Compute the maximum size of each PtNode of a PtNode array, assuming 3-byte addresses for 116 * everything, and caches it in the `mCachedSize' member of the nodes; deduce the size of 117 * the containing node array, and cache it it its 'mCachedSize' member. 118 * 119 * @param ptNodeArray the node array to compute the maximum size of. 120 */ 121 private static void calculatePtNodeArrayMaximumSize(final PtNodeArray ptNodeArray, 122 final HashMap<Integer, Integer> codePointToOneByteCodeMap) { 123 int size = getPtNodeCountSize(ptNodeArray); 124 for (PtNode node : ptNodeArray.mData) { 125 final int nodeSize = getPtNodeMaximumSize(node, codePointToOneByteCodeMap); 126 node.mCachedSize = nodeSize; 127 size += nodeSize; 128 } 129 ptNodeArray.mCachedSize = size; 130 } 131 132 /** 133 * Compute the size of the header (flag + [parent address] + characters size) of a PtNode. 134 * 135 * @param ptNode the PtNode of which to compute the size of the header 136 */ 137 private static int getNodeHeaderSize(final PtNode ptNode, 138 final HashMap<Integer, Integer> codePointToOneByteCodeMap) { 139 return FormatSpec.PTNODE_FLAGS_SIZE + getPtNodeCharactersSize(ptNode, 140 codePointToOneByteCodeMap); 141 } 142 143 /** 144 * Compute the size, in bytes, that an address will occupy. 145 * 146 * This can be used either for children addresses (which are always positive) or for 147 * attribute, which may be positive or negative but 148 * store their sign bit separately. 149 * 150 * @param address the address 151 * @return the byte size. 152 */ 153 static int getByteSize(final int address) { 154 assert(address <= FormatSpec.UINT24_MAX); 155 if (!BinaryDictIOUtils.hasChildrenAddress(address)) { 156 return 0; 157 } else if (Math.abs(address) <= FormatSpec.UINT8_MAX) { 158 return 1; 159 } else if (Math.abs(address) <= FormatSpec.UINT16_MAX) { 160 return 2; 161 } else { 162 return 3; 163 } 164 } 165 166 static int writeUIntToBuffer(final byte[] buffer, final int fromPosition, final int value, 167 final int size) { 168 int position = fromPosition; 169 switch(size) { 170 case 4: 171 buffer[position++] = (byte) ((value >> 24) & 0xFF); 172 /* fall through */ 173 case 3: 174 buffer[position++] = (byte) ((value >> 16) & 0xFF); 175 /* fall through */ 176 case 2: 177 buffer[position++] = (byte) ((value >> 8) & 0xFF); 178 /* fall through */ 179 case 1: 180 buffer[position++] = (byte) (value & 0xFF); 181 break; 182 default: 183 /* nop */ 184 } 185 return position; 186 } 187 188 static void writeUIntToStream(final OutputStream stream, final int value, final int size) 189 throws IOException { 190 switch(size) { 191 case 4: 192 stream.write((value >> 24) & 0xFF); 193 /* fall through */ 194 case 3: 195 stream.write((value >> 16) & 0xFF); 196 /* fall through */ 197 case 2: 198 stream.write((value >> 8) & 0xFF); 199 /* fall through */ 200 case 1: 201 stream.write(value & 0xFF); 202 break; 203 default: 204 /* nop */ 205 } 206 } 207 208 // End utility methods 209 210 // This method is responsible for finding a nice ordering of the nodes that favors run-time 211 // cache performance and dictionary size. 212 /* package for tests */ static ArrayList<PtNodeArray> flattenTree( 213 final PtNodeArray rootNodeArray) { 214 final int treeSize = FusionDictionary.countPtNodes(rootNodeArray); 215 MakedictLog.i("Counted nodes : " + treeSize); 216 final ArrayList<PtNodeArray> flatTree = new ArrayList<>(treeSize); 217 return flattenTreeInner(flatTree, rootNodeArray); 218 } 219 220 private static ArrayList<PtNodeArray> flattenTreeInner(final ArrayList<PtNodeArray> list, 221 final PtNodeArray ptNodeArray) { 222 // Removing the node is necessary if the tails are merged, because we would then 223 // add the same node several times when we only want it once. A number of places in 224 // the code also depends on any node being only once in the list. 225 // Merging tails can only be done if there are no attributes. Searching for attributes 226 // in LatinIME code depends on a total breadth-first ordering, which merging tails 227 // breaks. If there are no attributes, it should be fine (and reduce the file size) 228 // to merge tails, and removing the node from the list would be necessary. However, 229 // we don't merge tails because breaking the breadth-first ordering would result in 230 // extreme overhead at bigram lookup time (it would make the search function O(n) instead 231 // of the current O(log(n)), where n=number of nodes in the dictionary which is pretty 232 // high). 233 // If no nodes are ever merged, we can't have the same node twice in the list, hence 234 // searching for duplicates in unnecessary. It is also very performance consuming, 235 // since `list' is an ArrayList so it's an O(n) operation that runs on all nodes, making 236 // this simple list.remove operation O(n*n) overall. On Android this overhead is very 237 // high. 238 // For future reference, the code to remove duplicate is a simple : list.remove(node); 239 list.add(ptNodeArray); 240 final ArrayList<PtNode> branches = ptNodeArray.mData; 241 for (PtNode ptNode : branches) { 242 if (null != ptNode.mChildren) flattenTreeInner(list, ptNode.mChildren); 243 } 244 return list; 245 } 246 247 /** 248 * Get the offset from a position inside a current node array to a target node array, during 249 * update. 250 * 251 * If the current node array is before the target node array, the target node array has not 252 * been updated yet, so we should return the offset from the old position of the current node 253 * array to the old position of the target node array. If on the other hand the target is 254 * before the current node array, it already has been updated, so we should return the offset 255 * from the new position in the current node array to the new position in the target node 256 * array. 257 * 258 * @param currentNodeArray node array containing the PtNode where the offset will be written 259 * @param offsetFromStartOfCurrentNodeArray offset, in bytes, from the start of currentNodeArray 260 * @param targetNodeArray the target node array to get the offset to 261 * @return the offset to the target node array 262 */ 263 private static int getOffsetToTargetNodeArrayDuringUpdate(final PtNodeArray currentNodeArray, 264 final int offsetFromStartOfCurrentNodeArray, final PtNodeArray targetNodeArray) { 265 final boolean isTargetBeforeCurrent = (targetNodeArray.mCachedAddressBeforeUpdate 266 < currentNodeArray.mCachedAddressBeforeUpdate); 267 if (isTargetBeforeCurrent) { 268 return targetNodeArray.mCachedAddressAfterUpdate 269 - (currentNodeArray.mCachedAddressAfterUpdate 270 + offsetFromStartOfCurrentNodeArray); 271 } 272 return targetNodeArray.mCachedAddressBeforeUpdate 273 - (currentNodeArray.mCachedAddressBeforeUpdate + offsetFromStartOfCurrentNodeArray); 274 } 275 276 /** 277 * Get the offset from a position inside a current node array to a target PtNode, during 278 * update. 279 * 280 * @param currentNodeArray node array containing the PtNode where the offset will be written 281 * @param offsetFromStartOfCurrentNodeArray offset, in bytes, from the start of currentNodeArray 282 * @param targetPtNode the target PtNode to get the offset to 283 * @return the offset to the target PtNode 284 */ 285 // TODO: is there any way to factorize this method with the one above? 286 private static int getOffsetToTargetPtNodeDuringUpdate(final PtNodeArray currentNodeArray, 287 final int offsetFromStartOfCurrentNodeArray, final PtNode targetPtNode) { 288 final int oldOffsetBasePoint = currentNodeArray.mCachedAddressBeforeUpdate 289 + offsetFromStartOfCurrentNodeArray; 290 final boolean isTargetBeforeCurrent = (targetPtNode.mCachedAddressBeforeUpdate 291 < oldOffsetBasePoint); 292 // If the target is before the current node array, then its address has already been 293 // updated. We can use the AfterUpdate member, and compare it to our own member after 294 // update. Otherwise, the AfterUpdate member is not updated yet, so we need to use the 295 // BeforeUpdate member, and of course we have to compare this to our own address before 296 // update. 297 if (isTargetBeforeCurrent) { 298 final int newOffsetBasePoint = currentNodeArray.mCachedAddressAfterUpdate 299 + offsetFromStartOfCurrentNodeArray; 300 return targetPtNode.mCachedAddressAfterUpdate - newOffsetBasePoint; 301 } 302 return targetPtNode.mCachedAddressBeforeUpdate - oldOffsetBasePoint; 303 } 304 305 /** 306 * Computes the actual node array size, based on the cached addresses of the children nodes. 307 * 308 * Each node array stores its tentative address. During dictionary address computing, these 309 * are not final, but they can be used to compute the node array size (the node array size 310 * depends on the address of the children because the number of bytes necessary to store an 311 * address depends on its numeric value. The return value indicates whether the node array 312 * contents (as in, any of the addresses stored in the cache fields) have changed with 313 * respect to their previous value. 314 * 315 * @param ptNodeArray the node array to compute the size of. 316 * @param dict the dictionary in which the word/attributes are to be found. 317 * @return false if none of the cached addresses inside the node array changed, true otherwise. 318 */ 319 private static boolean computeActualPtNodeArraySize(final PtNodeArray ptNodeArray, 320 final FusionDictionary dict, 321 final HashMap<Integer, Integer> codePointToOneByteCodeMap) { 322 boolean changed = false; 323 int size = getPtNodeCountSize(ptNodeArray); 324 for (PtNode ptNode : ptNodeArray.mData) { 325 ptNode.mCachedAddressAfterUpdate = ptNodeArray.mCachedAddressAfterUpdate + size; 326 if (ptNode.mCachedAddressAfterUpdate != ptNode.mCachedAddressBeforeUpdate) { 327 changed = true; 328 } 329 int nodeSize = getNodeHeaderSize(ptNode, codePointToOneByteCodeMap); 330 if (ptNode.isTerminal()) { 331 nodeSize += FormatSpec.PTNODE_FREQUENCY_SIZE; 332 } 333 if (null != ptNode.mChildren) { 334 nodeSize += getByteSize(getOffsetToTargetNodeArrayDuringUpdate(ptNodeArray, 335 nodeSize + size, ptNode.mChildren)); 336 } 337 if (null != ptNode.mBigrams) { 338 for (WeightedString bigram : ptNode.mBigrams) { 339 final int offset = getOffsetToTargetPtNodeDuringUpdate(ptNodeArray, 340 nodeSize + size + FormatSpec.PTNODE_ATTRIBUTE_FLAGS_SIZE, 341 FusionDictionary.findWordInTree(dict.mRootNodeArray, bigram.mWord)); 342 nodeSize += getByteSize(offset) + FormatSpec.PTNODE_ATTRIBUTE_FLAGS_SIZE; 343 } 344 } 345 ptNode.mCachedSize = nodeSize; 346 size += nodeSize; 347 } 348 if (ptNodeArray.mCachedSize != size) { 349 ptNodeArray.mCachedSize = size; 350 changed = true; 351 } 352 return changed; 353 } 354 355 /** 356 * Initializes the cached addresses of node arrays and their containing nodes from their size. 357 * 358 * @param flatNodes the list of node arrays. 359 * @return the byte size of the entire stack. 360 */ 361 private static int initializePtNodeArraysCachedAddresses( 362 final ArrayList<PtNodeArray> flatNodes) { 363 int nodeArrayOffset = 0; 364 for (final PtNodeArray nodeArray : flatNodes) { 365 nodeArray.mCachedAddressBeforeUpdate = nodeArrayOffset; 366 int nodeCountSize = getPtNodeCountSize(nodeArray); 367 int nodeffset = 0; 368 for (final PtNode ptNode : nodeArray.mData) { 369 ptNode.mCachedAddressBeforeUpdate = ptNode.mCachedAddressAfterUpdate = 370 nodeCountSize + nodeArrayOffset + nodeffset; 371 nodeffset += ptNode.mCachedSize; 372 } 373 nodeArrayOffset += nodeArray.mCachedSize; 374 } 375 return nodeArrayOffset; 376 } 377 378 /** 379 * Updates the cached addresses of node arrays after recomputing their new positions. 380 * 381 * @param flatNodes the list of node arrays. 382 */ 383 private static void updatePtNodeArraysCachedAddresses(final ArrayList<PtNodeArray> flatNodes) { 384 for (final PtNodeArray nodeArray : flatNodes) { 385 nodeArray.mCachedAddressBeforeUpdate = nodeArray.mCachedAddressAfterUpdate; 386 for (final PtNode ptNode : nodeArray.mData) { 387 ptNode.mCachedAddressBeforeUpdate = ptNode.mCachedAddressAfterUpdate; 388 } 389 } 390 } 391 392 /** 393 * Compute the addresses and sizes of an ordered list of PtNode arrays. 394 * 395 * This method takes a list of PtNode arrays and will update their cached address and size 396 * values so that they can be written into a file. It determines the smallest size each of the 397 * PtNode arrays can be given the addresses of its children and attributes, and store that into 398 * each PtNode. 399 * The order of the PtNode is given by the order of the array. This method makes no effort 400 * to find a good order; it only mechanically computes the size this order results in. 401 * 402 * @param dict the dictionary 403 * @param flatNodes the ordered list of PtNode arrays 404 * @return the same array it was passed. The nodes have been updated for address and size. 405 */ 406 /* package */ static ArrayList<PtNodeArray> computeAddresses(final FusionDictionary dict, 407 final ArrayList<PtNodeArray> flatNodes, 408 final HashMap<Integer, Integer> codePointToOneByteCodeMap) { 409 // First get the worst possible sizes and offsets 410 for (final PtNodeArray n : flatNodes) { 411 calculatePtNodeArrayMaximumSize(n, codePointToOneByteCodeMap); 412 } 413 final int offset = initializePtNodeArraysCachedAddresses(flatNodes); 414 415 MakedictLog.i("Compressing the array addresses. Original size : " + offset); 416 MakedictLog.i("(Recursively seen size : " + offset + ")"); 417 418 int passes = 0; 419 boolean changesDone = false; 420 do { 421 changesDone = false; 422 int ptNodeArrayStartOffset = 0; 423 for (final PtNodeArray ptNodeArray : flatNodes) { 424 ptNodeArray.mCachedAddressAfterUpdate = ptNodeArrayStartOffset; 425 final int oldNodeArraySize = ptNodeArray.mCachedSize; 426 final boolean changed = computeActualPtNodeArraySize(ptNodeArray, dict, 427 codePointToOneByteCodeMap); 428 final int newNodeArraySize = ptNodeArray.mCachedSize; 429 if (oldNodeArraySize < newNodeArraySize) { 430 throw new RuntimeException("Increased size ?!"); 431 } 432 ptNodeArrayStartOffset += newNodeArraySize; 433 changesDone |= changed; 434 } 435 updatePtNodeArraysCachedAddresses(flatNodes); 436 ++passes; 437 if (passes > MAX_PASSES) throw new RuntimeException("Too many passes - probably a bug"); 438 } while (changesDone); 439 440 final PtNodeArray lastPtNodeArray = flatNodes.get(flatNodes.size() - 1); 441 MakedictLog.i("Compression complete in " + passes + " passes."); 442 MakedictLog.i("After address compression : " 443 + (lastPtNodeArray.mCachedAddressAfterUpdate + lastPtNodeArray.mCachedSize)); 444 445 return flatNodes; 446 } 447 448 /** 449 * Sanity-checking method. 450 * 451 * This method checks a list of PtNode arrays for juxtaposition, that is, it will do 452 * nothing if each node array's cached address is actually the previous node array's address 453 * plus the previous node's size. 454 * If this is not the case, it will throw an exception. 455 * 456 * @param arrays the list of node arrays to check 457 */ 458 /* package */ static void checkFlatPtNodeArrayList(final ArrayList<PtNodeArray> arrays) { 459 int offset = 0; 460 int index = 0; 461 for (final PtNodeArray ptNodeArray : arrays) { 462 // BeforeUpdate and AfterUpdate addresses are the same here, so it does not matter 463 // which we use. 464 if (ptNodeArray.mCachedAddressAfterUpdate != offset) { 465 throw new RuntimeException("Wrong address for node " + index 466 + " : expected " + offset + ", got " + 467 ptNodeArray.mCachedAddressAfterUpdate); 468 } 469 ++index; 470 offset += ptNodeArray.mCachedSize; 471 } 472 } 473 474 /** 475 * Helper method to write a children position to a file. 476 * 477 * @param buffer the buffer to write to. 478 * @param fromIndex the index in the buffer to write the address to. 479 * @param position the position to write. 480 * @return the size in bytes the address actually took. 481 */ 482 /* package */ static int writeChildrenPosition(final byte[] buffer, final int fromIndex, 483 final int position) { 484 int index = fromIndex; 485 switch (getByteSize(position)) { 486 case 1: 487 buffer[index++] = (byte)position; 488 return 1; 489 case 2: 490 buffer[index++] = (byte)(0xFF & (position >> 8)); 491 buffer[index++] = (byte)(0xFF & position); 492 return 2; 493 case 3: 494 buffer[index++] = (byte)(0xFF & (position >> 16)); 495 buffer[index++] = (byte)(0xFF & (position >> 8)); 496 buffer[index++] = (byte)(0xFF & position); 497 return 3; 498 case 0: 499 return 0; 500 default: 501 throw new RuntimeException("Position " + position + " has a strange size"); 502 } 503 } 504 505 /** 506 * Makes the flag value for a PtNode. 507 * 508 * @param hasMultipleChars whether the PtNode has multiple chars. 509 * @param isTerminal whether the PtNode is terminal. 510 * @param childrenAddressSize the size of a children address. 511 * @param hasBigrams whether the PtNode has bigrams. 512 * @param isNotAWord whether the PtNode is not a word. 513 * @param isPossiblyOffensive whether the PtNode is a possibly offensive entry. 514 * @return the flags 515 */ 516 static int makePtNodeFlags(final boolean hasMultipleChars, final boolean isTerminal, 517 final int childrenAddressSize, final boolean hasBigrams, 518 final boolean isNotAWord, final boolean isPossiblyOffensive) { 519 byte flags = 0; 520 if (hasMultipleChars) flags |= FormatSpec.FLAG_HAS_MULTIPLE_CHARS; 521 if (isTerminal) flags |= FormatSpec.FLAG_IS_TERMINAL; 522 switch (childrenAddressSize) { 523 case 1: 524 flags |= FormatSpec.FLAG_CHILDREN_ADDRESS_TYPE_ONEBYTE; 525 break; 526 case 2: 527 flags |= FormatSpec.FLAG_CHILDREN_ADDRESS_TYPE_TWOBYTES; 528 break; 529 case 3: 530 flags |= FormatSpec.FLAG_CHILDREN_ADDRESS_TYPE_THREEBYTES; 531 break; 532 case 0: 533 flags |= FormatSpec.FLAG_CHILDREN_ADDRESS_TYPE_NOADDRESS; 534 break; 535 default: 536 throw new RuntimeException("Node with a strange address"); 537 } 538 if (hasBigrams) flags |= FormatSpec.FLAG_HAS_BIGRAMS; 539 if (isNotAWord) flags |= FormatSpec.FLAG_IS_NOT_A_WORD; 540 if (isPossiblyOffensive) flags |= FormatSpec.FLAG_IS_POSSIBLY_OFFENSIVE; 541 return flags; 542 } 543 544 /* package */ static byte makePtNodeFlags(final PtNode node, final int childrenOffset) { 545 return (byte) makePtNodeFlags(node.mChars.length > 1, node.isTerminal(), 546 getByteSize(childrenOffset), 547 node.mBigrams != null && !node.mBigrams.isEmpty(), 548 node.mIsNotAWord, node.mIsPossiblyOffensive); 549 } 550 551 /** 552 * Makes the flag value for a bigram. 553 * 554 * @param more whether there are more bigrams after this one. 555 * @param offset the offset of the bigram. 556 * @param bigramFrequency the frequency of the bigram, 0..255. 557 * @param unigramFrequency the unigram frequency of the same word, 0..255. 558 * @param word the second bigram, for debugging purposes 559 * @return the flags 560 */ 561 /* package */ static int makeBigramFlags(final boolean more, final int offset, 562 final int bigramFrequency, final int unigramFrequency, final String word) { 563 int bigramFlags = (more ? FormatSpec.FLAG_BIGRAM_SHORTCUT_ATTR_HAS_NEXT : 0) 564 + (offset < 0 ? FormatSpec.FLAG_BIGRAM_ATTR_OFFSET_NEGATIVE : 0); 565 switch (getByteSize(offset)) { 566 case 1: 567 bigramFlags |= FormatSpec.FLAG_BIGRAM_ATTR_ADDRESS_TYPE_ONEBYTE; 568 break; 569 case 2: 570 bigramFlags |= FormatSpec.FLAG_BIGRAM_ATTR_ADDRESS_TYPE_TWOBYTES; 571 break; 572 case 3: 573 bigramFlags |= FormatSpec.FLAG_BIGRAM_ATTR_ADDRESS_TYPE_THREEBYTES; 574 break; 575 default: 576 throw new RuntimeException("Strange offset size"); 577 } 578 final int frequency; 579 if (unigramFrequency > bigramFrequency) { 580 MakedictLog.e("Unigram freq is superior to bigram freq for \"" + word 581 + "\". Bigram freq is " + bigramFrequency + ", unigram freq for " 582 + word + " is " + unigramFrequency); 583 frequency = unigramFrequency; 584 } else { 585 frequency = bigramFrequency; 586 } 587 bigramFlags += getBigramFrequencyDiff(unigramFrequency, frequency) 588 & FormatSpec.FLAG_BIGRAM_SHORTCUT_ATTR_FREQUENCY; 589 return bigramFlags; 590 } 591 592 public static int getBigramFrequencyDiff(final int unigramFrequency, 593 final int bigramFrequency) { 594 // We compute the difference between 255 (which means probability = 1) and the 595 // unigram score. We split this into a number of discrete steps. 596 // Now, the steps are numbered 0~15; 0 represents an increase of 1 step while 15 597 // represents an increase of 16 steps: a value of 15 will be interpreted as the median 598 // value of the 16th step. In all justice, if the bigram frequency is low enough to be 599 // rounded below the first step (which means it is less than half a step higher than the 600 // unigram frequency) then the unigram frequency itself is the best approximation of the 601 // bigram freq that we could possibly supply, hence we should *not* include this bigram 602 // in the file at all. 603 // until this is done, we'll write 0 and slightly overestimate this case. 604 // In other words, 0 means "between 0.5 step and 1.5 step", 1 means "between 1.5 step 605 // and 2.5 steps", and 15 means "between 15.5 steps and 16.5 steps". So we want to 606 // divide our range [unigramFreq..MAX_TERMINAL_FREQUENCY] in 16.5 steps to get the 607 // step size. Then we compute the start of the first step (the one where value 0 starts) 608 // by adding half-a-step to the unigramFrequency. From there, we compute the integer 609 // number of steps to the bigramFrequency. One last thing: we want our steps to include 610 // their lower bound and exclude their higher bound so we need to have the first step 611 // start at exactly 1 unit higher than floor(unigramFreq + half a step). 612 // Note : to reconstruct the score, the dictionary reader will need to divide 613 // MAX_TERMINAL_FREQUENCY - unigramFreq by 16.5 likewise to get the value of the step, 614 // and add (discretizedFrequency + 0.5 + 0.5) times this value to get the best 615 // approximation. (0.5 to get the first step start, and 0.5 to get the middle of the 616 // step pointed by the discretized frequency. 617 final float stepSize = 618 (FormatSpec.MAX_TERMINAL_FREQUENCY - unigramFrequency) 619 / (1.5f + FormatSpec.MAX_BIGRAM_FREQUENCY); 620 final float firstStepStart = 1 + unigramFrequency + (stepSize / 2.0f); 621 final int discretizedFrequency = (int)((bigramFrequency - firstStepStart) / stepSize); 622 // If the bigram freq is less than half-a-step higher than the unigram freq, we get -1 623 // here. The best approximation would be the unigram freq itself, so we should not 624 // include this bigram in the dictionary. For now, register as 0, and live with the 625 // small over-estimation that we get in this case. TODO: actually remove this bigram 626 // if discretizedFrequency < 0. 627 return discretizedFrequency > 0 ? discretizedFrequency : 0; 628 } 629 630 /* package */ static int getChildrenPosition(final PtNode ptNode, 631 final HashMap<Integer, Integer> codePointToOneByteCodeMap) { 632 int positionOfChildrenPosField = ptNode.mCachedAddressAfterUpdate 633 + getNodeHeaderSize(ptNode, codePointToOneByteCodeMap); 634 if (ptNode.isTerminal()) { 635 // A terminal node has the frequency. 636 // If positionOfChildrenPosField is incorrect, we may crash when jumping to the children 637 // position. 638 positionOfChildrenPosField += FormatSpec.PTNODE_FREQUENCY_SIZE; 639 } 640 return null == ptNode.mChildren ? FormatSpec.NO_CHILDREN_ADDRESS 641 : ptNode.mChildren.mCachedAddressAfterUpdate - positionOfChildrenPosField; 642 } 643 644 /** 645 * Write a PtNodeArray. The PtNodeArray is expected to have its final position cached. 646 * 647 * @param dict the dictionary the node array is a part of (for relative offsets). 648 * @param dictEncoder the dictionary encoder. 649 * @param ptNodeArray the node array to write. 650 * @param codePointToOneByteCodeMap the map to convert the code points. 651 */ 652 /* package */ static void writePlacedPtNodeArray(final FusionDictionary dict, 653 final DictEncoder dictEncoder, final PtNodeArray ptNodeArray, 654 final HashMap<Integer, Integer> codePointToOneByteCodeMap) { 655 // TODO: Make the code in common with BinaryDictIOUtils#writePtNode 656 dictEncoder.setPosition(ptNodeArray.mCachedAddressAfterUpdate); 657 658 final int ptNodeCount = ptNodeArray.mData.size(); 659 dictEncoder.writePtNodeCount(ptNodeCount); 660 for (int i = 0; i < ptNodeCount; ++i) { 661 final PtNode ptNode = ptNodeArray.mData.get(i); 662 if (dictEncoder.getPosition() != ptNode.mCachedAddressAfterUpdate) { 663 throw new RuntimeException("Bug: write index is not the same as the cached address " 664 + "of the node : " + dictEncoder.getPosition() + " <> " 665 + ptNode.mCachedAddressAfterUpdate); 666 } 667 // Sanity checks. 668 if (DBG && ptNode.getProbability() > FormatSpec.MAX_TERMINAL_FREQUENCY) { 669 throw new RuntimeException("A node has a frequency > " 670 + FormatSpec.MAX_TERMINAL_FREQUENCY 671 + " : " + ptNode.mProbabilityInfo.toString()); 672 } 673 dictEncoder.writePtNode(ptNode, dict, codePointToOneByteCodeMap); 674 } 675 if (dictEncoder.getPosition() != ptNodeArray.mCachedAddressAfterUpdate 676 + ptNodeArray.mCachedSize) { 677 throw new RuntimeException("Not the same size : written " 678 + (dictEncoder.getPosition() - ptNodeArray.mCachedAddressAfterUpdate) 679 + " bytes from a node that should have " + ptNodeArray.mCachedSize + " bytes"); 680 } 681 } 682 683 /** 684 * Dumps a collection of useful statistics about a list of PtNode arrays. 685 * 686 * This prints purely informative stuff, like the total estimated file size, the 687 * number of PtNode arrays, of PtNodes, the repartition of each address size, etc 688 * 689 * @param ptNodeArrays the list of PtNode arrays. 690 */ 691 /* package */ static void showStatistics(ArrayList<PtNodeArray> ptNodeArrays) { 692 int firstTerminalAddress = Integer.MAX_VALUE; 693 int lastTerminalAddress = Integer.MIN_VALUE; 694 int size = 0; 695 int ptNodes = 0; 696 int maxNodes = 0; 697 int maxRuns = 0; 698 for (final PtNodeArray ptNodeArray : ptNodeArrays) { 699 if (maxNodes < ptNodeArray.mData.size()) maxNodes = ptNodeArray.mData.size(); 700 for (final PtNode ptNode : ptNodeArray.mData) { 701 ++ptNodes; 702 if (ptNode.mChars.length > maxRuns) maxRuns = ptNode.mChars.length; 703 if (ptNode.isTerminal()) { 704 if (ptNodeArray.mCachedAddressAfterUpdate < firstTerminalAddress) 705 firstTerminalAddress = ptNodeArray.mCachedAddressAfterUpdate; 706 if (ptNodeArray.mCachedAddressAfterUpdate > lastTerminalAddress) 707 lastTerminalAddress = ptNodeArray.mCachedAddressAfterUpdate; 708 } 709 } 710 if (ptNodeArray.mCachedAddressAfterUpdate + ptNodeArray.mCachedSize > size) { 711 size = ptNodeArray.mCachedAddressAfterUpdate + ptNodeArray.mCachedSize; 712 } 713 } 714 final int[] ptNodeCounts = new int[maxNodes + 1]; 715 final int[] runCounts = new int[maxRuns + 1]; 716 for (final PtNodeArray ptNodeArray : ptNodeArrays) { 717 ++ptNodeCounts[ptNodeArray.mData.size()]; 718 for (final PtNode ptNode : ptNodeArray.mData) { 719 ++runCounts[ptNode.mChars.length]; 720 } 721 } 722 723 MakedictLog.i("Statistics:\n" 724 + " Total file size " + size + "\n" 725 + " " + ptNodeArrays.size() + " node arrays\n" 726 + " " + ptNodes + " PtNodes (" + ((float)ptNodes / ptNodeArrays.size()) 727 + " PtNodes per node)\n" 728 + " First terminal at " + firstTerminalAddress + "\n" 729 + " Last terminal at " + lastTerminalAddress + "\n" 730 + " PtNode stats : max = " + maxNodes); 731 } 732 733 /** 734 * Writes a file header to an output stream. 735 * 736 * @param destination the stream to write the file header to. 737 * @param dict the dictionary to write. 738 * @param formatOptions file format options. 739 * @param codePointOccurrenceArray code points ordered by occurrence count. 740 * @return the size of the header. 741 */ 742 /* package */ static int writeDictionaryHeader(final OutputStream destination, 743 final FusionDictionary dict, final FormatOptions formatOptions, 744 final ArrayList<Entry<Integer, Integer>> codePointOccurrenceArray) 745 throws IOException, UnsupportedFormatException { 746 final int version = formatOptions.mVersion; 747 if ((version >= FormatSpec.MINIMUM_SUPPORTED_STATIC_VERSION && 748 version <= FormatSpec.MAXIMUM_SUPPORTED_STATIC_VERSION) || ( 749 version >= FormatSpec.MINIMUM_SUPPORTED_DYNAMIC_VERSION && 750 version <= FormatSpec.MAXIMUM_SUPPORTED_DYNAMIC_VERSION)) { 751 // Dictionary is valid 752 } else { 753 throw new UnsupportedFormatException("Requested file format version " + version 754 + ", but this implementation only supports static versions " 755 + FormatSpec.MINIMUM_SUPPORTED_STATIC_VERSION + " through " 756 + FormatSpec.MAXIMUM_SUPPORTED_STATIC_VERSION + " and dynamic versions " 757 + FormatSpec.MINIMUM_SUPPORTED_DYNAMIC_VERSION + " through " 758 + FormatSpec.MAXIMUM_SUPPORTED_DYNAMIC_VERSION); 759 } 760 761 ByteArrayOutputStream headerBuffer = new ByteArrayOutputStream(256); 762 763 // The magic number in big-endian order. 764 // Magic number for all versions. 765 headerBuffer.write((byte) (0xFF & (FormatSpec.MAGIC_NUMBER >> 24))); 766 headerBuffer.write((byte) (0xFF & (FormatSpec.MAGIC_NUMBER >> 16))); 767 headerBuffer.write((byte) (0xFF & (FormatSpec.MAGIC_NUMBER >> 8))); 768 headerBuffer.write((byte) (0xFF & FormatSpec.MAGIC_NUMBER)); 769 // Dictionary version. 770 headerBuffer.write((byte) (0xFF & (version >> 8))); 771 headerBuffer.write((byte) (0xFF & version)); 772 773 // Options flags 774 // TODO: Remove this field. 775 final int options = 0; 776 headerBuffer.write((byte) (0xFF & (options >> 8))); 777 headerBuffer.write((byte) (0xFF & options)); 778 final int headerSizeOffset = headerBuffer.size(); 779 // Placeholder to be written later with header size. 780 for (int i = 0; i < 4; ++i) { 781 headerBuffer.write(0); 782 } 783 // Write out the options. 784 for (final String key : dict.mOptions.mAttributes.keySet()) { 785 final String value = dict.mOptions.mAttributes.get(key); 786 CharEncoding.writeString(headerBuffer, key, null); 787 CharEncoding.writeString(headerBuffer, value, null); 788 } 789 // Write out the codePointTable if there is codePointOccurrenceArray. 790 if (codePointOccurrenceArray != null) { 791 final String codePointTableString = 792 encodeCodePointTable(codePointOccurrenceArray); 793 CharEncoding.writeString(headerBuffer, DictionaryHeader.CODE_POINT_TABLE_KEY, null); 794 CharEncoding.writeString(headerBuffer, codePointTableString, null); 795 } 796 final int size = headerBuffer.size(); 797 final byte[] bytes = headerBuffer.toByteArray(); 798 // Write out the header size. 799 bytes[headerSizeOffset] = (byte) (0xFF & (size >> 24)); 800 bytes[headerSizeOffset + 1] = (byte) (0xFF & (size >> 16)); 801 bytes[headerSizeOffset + 2] = (byte) (0xFF & (size >> 8)); 802 bytes[headerSizeOffset + 3] = (byte) (0xFF & (size >> 0)); 803 destination.write(bytes); 804 805 headerBuffer.close(); 806 return size; 807 } 808 809 static final class CodePointTable { 810 final HashMap<Integer, Integer> mCodePointToOneByteCodeMap; 811 final ArrayList<Entry<Integer, Integer>> mCodePointOccurrenceArray; 812 813 // Let code point table empty for version 200 dictionary which used in test 814 CodePointTable() { 815 mCodePointToOneByteCodeMap = null; 816 mCodePointOccurrenceArray = null; 817 } 818 819 CodePointTable(final HashMap<Integer, Integer> codePointToOneByteCodeMap, 820 final ArrayList<Entry<Integer, Integer>> codePointOccurrenceArray) { 821 mCodePointToOneByteCodeMap = codePointToOneByteCodeMap; 822 mCodePointOccurrenceArray = codePointOccurrenceArray; 823 } 824 } 825 826 private static String encodeCodePointTable( 827 final ArrayList<Entry<Integer, Integer>> codePointOccurrenceArray) { 828 final StringBuilder codePointTableString = new StringBuilder(); 829 int currentCodePointTableIndex = FormatSpec.MINIMAL_ONE_BYTE_CHARACTER_VALUE; 830 for (final Entry<Integer, Integer> entry : codePointOccurrenceArray) { 831 // Native reads the table as a string 832 codePointTableString.appendCodePoint(entry.getKey()); 833 if (FormatSpec.MAXIMAL_ONE_BYTE_CHARACTER_VALUE < ++currentCodePointTableIndex) { 834 break; 835 } 836 } 837 return codePointTableString.toString(); 838 } 839 } 840