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 
BinaryDictEncoderUtils()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      */
getPtNodeCharactersSize(final int[] characters, final HashMap<Integer, Integer> codePointToOneByteCodeMap)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      */
getPtNodeCharactersSize(final PtNode ptNode, final HashMap<Integer, Integer> codePointToOneByteCodeMap)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      */
getPtNodeCountSize(final PtNodeArray nodeArray)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      */
getPtNodeMaximumSize(final PtNode ptNode, final HashMap<Integer, Integer> codePointToOneByteCodeMap)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      */
calculatePtNodeArrayMaximumSize(final PtNodeArray ptNodeArray, final HashMap<Integer, Integer> codePointToOneByteCodeMap)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      */
getNodeHeaderSize(final PtNode ptNode, final HashMap<Integer, Integer> codePointToOneByteCodeMap)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      */
getByteSize(final int address)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 
writeUIntToBuffer(final byte[] buffer, final int fromPosition, final int value, final int size)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 
writeUIntToStream(final OutputStream stream, final int value, final int size)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.
flattenTree( final PtNodeArray rootNodeArray)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 
flattenTreeInner(final ArrayList<PtNodeArray> list, final PtNodeArray ptNodeArray)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      */
getOffsetToTargetNodeArrayDuringUpdate(final PtNodeArray currentNodeArray, final int offsetFromStartOfCurrentNodeArray, final PtNodeArray targetNodeArray)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?
getOffsetToTargetPtNodeDuringUpdate(final PtNodeArray currentNodeArray, final int offsetFromStartOfCurrentNodeArray, final PtNode targetPtNode)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      */
computeActualPtNodeArraySize(final PtNodeArray ptNodeArray, final FusionDictionary dict, final HashMap<Integer, Integer> codePointToOneByteCodeMap)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      */
initializePtNodeArraysCachedAddresses( final ArrayList<PtNodeArray> flatNodes)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      */
updatePtNodeArraysCachedAddresses(final ArrayList<PtNodeArray> flatNodes)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      */
computeAddresses(final FusionDictionary dict, final ArrayList<PtNodeArray> flatNodes, final HashMap<Integer, Integer> codePointToOneByteCodeMap)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      * Validity-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      */
checkFlatPtNodeArrayList(final ArrayList<PtNodeArray> arrays)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      */
writeChildrenPosition(final byte[] buffer, final int fromIndex, final int position)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      */
makePtNodeFlags(final boolean hasMultipleChars, final boolean isTerminal, final int childrenAddressSize, final boolean hasBigrams, final boolean isNotAWord, final boolean isPossiblyOffensive)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 
makePtNodeFlags(final PtNode node, final int childrenOffset)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      */
makeBigramFlags(final boolean more, final int offset, final int bigramFrequency, final int unigramFrequency, final String word)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 
getBigramFrequencyDiff(final int unigramFrequency, final int bigramFrequency)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 
getChildrenPosition(final PtNode ptNode, final HashMap<Integer, Integer> codePointToOneByteCodeMap)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      */
writePlacedPtNodeArray(final FusionDictionary dict, final DictEncoder dictEncoder, final PtNodeArray ptNodeArray, final HashMap<Integer, Integer> codePointToOneByteCodeMap)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             // Validity 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      */
showStatistics(ArrayList<PtNodeArray> ptNodeArrays)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      */
writeDictionaryHeader(final OutputStream destination, final FusionDictionary dict, final FormatOptions formatOptions, final ArrayList<Entry<Integer, Integer>> codePointOccurrenceArray)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
CodePointTable()814         CodePointTable() {
815             mCodePointToOneByteCodeMap = null;
816             mCodePointOccurrenceArray = null;
817         }
818 
CodePointTable(final HashMap<Integer, Integer> codePointToOneByteCodeMap, final ArrayList<Entry<Integer, Integer>> codePointOccurrenceArray)819         CodePointTable(final HashMap<Integer, Integer> codePointToOneByteCodeMap,
820                 final ArrayList<Entry<Integer, Integer>> codePointOccurrenceArray) {
821             mCodePointToOneByteCodeMap = codePointToOneByteCodeMap;
822             mCodePointOccurrenceArray = codePointOccurrenceArray;
823         }
824     }
825 
encodeCodePointTable( final ArrayList<Entry<Integer, Integer>> codePointOccurrenceArray)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