Home | History | Annotate | Download | only in src
      1 /*
      2 **
      3 ** Copyright 2010, The Android Open Source Project
      4 **
      5 ** Licensed under the Apache License, Version 2.0 (the "License");
      6 ** you may not use this file except in compliance with the License.
      7 ** You may obtain a copy of the License at
      8 **
      9 **     http://www.apache.org/licenses/LICENSE-2.0
     10 **
     11 ** Unless required by applicable law or agreed to in writing, software
     12 ** distributed under the License is distributed on an "AS IS" BASIS,
     13 ** WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
     14 ** See the License for the specific language governing permissions and
     15 ** limitations under the License.
     16 */
     17 
     18 #include <assert.h>
     19 #include <string.h>
     20 
     21 #define LOG_TAG "LatinIME: unigram_dictionary.cpp"
     22 
     23 #include "char_utils.h"
     24 #include "defines.h"
     25 #include "dictionary.h"
     26 #include "unigram_dictionary.h"
     27 
     28 #include "binary_format.h"
     29 #include "terminal_attributes.h"
     30 
     31 namespace latinime {
     32 
     33 const UnigramDictionary::digraph_t UnigramDictionary::GERMAN_UMLAUT_DIGRAPHS[] =
     34         { { 'a', 'e', 0x00E4 }, // U+00E4 : LATIN SMALL LETTER A WITH DIAERESIS
     35         { 'o', 'e', 0x00F6 }, // U+00F6 : LATIN SMALL LETTER O WITH DIAERESIS
     36         { 'u', 'e', 0x00FC } }; // U+00FC : LATIN SMALL LETTER U WITH DIAERESIS
     37 
     38 const UnigramDictionary::digraph_t UnigramDictionary::FRENCH_LIGATURES_DIGRAPHS[] =
     39         { { 'a', 'e', 0x00E6 }, // U+00E6 : LATIN SMALL LETTER AE
     40         { 'o', 'e', 0x0153 } }; // U+0153 : LATIN SMALL LIGATURE OE
     41 
     42 // TODO: check the header
     43 UnigramDictionary::UnigramDictionary(const uint8_t* const streamStart, int typedLetterMultiplier,
     44         int fullWordMultiplier, int maxWordLength, int maxWords, const unsigned int flags)
     45     : DICT_ROOT(streamStart), MAX_WORD_LENGTH(maxWordLength), MAX_WORDS(maxWords),
     46     TYPED_LETTER_MULTIPLIER(typedLetterMultiplier), FULL_WORD_MULTIPLIER(fullWordMultiplier),
     47       // TODO : remove this variable.
     48     ROOT_POS(0),
     49     BYTES_IN_ONE_CHAR(sizeof(int)),
     50     MAX_DIGRAPH_SEARCH_DEPTH(DEFAULT_MAX_DIGRAPH_SEARCH_DEPTH), FLAGS(flags) {
     51     if (DEBUG_DICT) {
     52         AKLOGI("UnigramDictionary - constructor");
     53     }
     54 }
     55 
     56 UnigramDictionary::~UnigramDictionary() {
     57 }
     58 
     59 static inline unsigned int getCodesBufferSize(const int *codes, const int codesSize) {
     60     return sizeof(*codes) * codesSize;
     61 }
     62 
     63 // TODO: This needs to take a const unsigned short* and not tinker with its contents
     64 static inline void addWord(
     65         unsigned short *word, int length, int frequency, WordsPriorityQueue *queue) {
     66     queue->push(frequency, word, length);
     67 }
     68 
     69 // Return the replacement code point for a digraph, or 0 if none.
     70 int UnigramDictionary::getDigraphReplacement(const int *codes, const int i, const int codesSize,
     71         const digraph_t* const digraphs, const unsigned int digraphsSize) const {
     72 
     73     // There can't be a digraph if we don't have at least 2 characters to examine
     74     if (i + 2 > codesSize) return false;
     75 
     76     // Search for the first char of some digraph
     77     int lastDigraphIndex = -1;
     78     const int thisChar = codes[i];
     79     for (lastDigraphIndex = digraphsSize - 1; lastDigraphIndex >= 0; --lastDigraphIndex) {
     80         if (thisChar == digraphs[lastDigraphIndex].first) break;
     81     }
     82     // No match: return early
     83     if (lastDigraphIndex < 0) return 0;
     84 
     85     // It's an interesting digraph if the second char matches too.
     86     if (digraphs[lastDigraphIndex].second == codes[i + 1]) {
     87         return digraphs[lastDigraphIndex].replacement;
     88     } else {
     89         return 0;
     90     }
     91 }
     92 
     93 // Mostly the same arguments as the non-recursive version, except:
     94 // codes is the original value. It points to the start of the work buffer, and gets passed as is.
     95 // codesSize is the size of the user input (thus, it is the size of codesSrc).
     96 // codesDest is the current point in the work buffer.
     97 // codesSrc is the current point in the user-input, original, content-unmodified buffer.
     98 // codesRemain is the remaining size in codesSrc.
     99 void UnigramDictionary::getWordWithDigraphSuggestionsRec(ProximityInfo *proximityInfo,
    100         const int *xcoordinates, const int *ycoordinates, const int *codesBuffer,
    101         int *xCoordinatesBuffer, int *yCoordinatesBuffer,
    102         const int codesBufferSize, const std::map<int, int> *bigramMap, const uint8_t *bigramFilter,
    103         const bool useFullEditDistance, const int *codesSrc,
    104         const int codesRemain, const int currentDepth, int *codesDest, Correction *correction,
    105         WordsPriorityQueuePool *queuePool,
    106         const digraph_t* const digraphs, const unsigned int digraphsSize) {
    107 
    108     const int startIndex = codesDest - codesBuffer;
    109     if (currentDepth < MAX_DIGRAPH_SEARCH_DEPTH) {
    110         for (int i = 0; i < codesRemain; ++i) {
    111             xCoordinatesBuffer[startIndex + i] = xcoordinates[codesBufferSize - codesRemain + i];
    112             yCoordinatesBuffer[startIndex + i] = ycoordinates[codesBufferSize - codesRemain + i];
    113             const int replacementCodePoint =
    114                     getDigraphReplacement(codesSrc, i, codesRemain, digraphs, digraphsSize);
    115             if (0 != replacementCodePoint) {
    116                 // Found a digraph. We will try both spellings. eg. the word is "pruefen"
    117 
    118                 // Copy the word up to the first char of the digraph, including proximity chars,
    119                 // and overwrite the primary code with the replacement code point. Then, continue
    120                 // processing on the remaining part of the word, skipping the second char of the
    121                 // digraph.
    122                 // In our example, copy "pru", replace "u" with the version with the diaeresis and
    123                 // continue running on "fen".
    124                 // Make i the index of the second char of the digraph for simplicity. Forgetting
    125                 // to do that results in an infinite recursion so take care!
    126                 ++i;
    127                 memcpy(codesDest, codesSrc, i * BYTES_IN_ONE_CHAR);
    128                 codesDest[(i - 1) * (BYTES_IN_ONE_CHAR / sizeof(codesDest[0]))] =
    129                         replacementCodePoint;
    130                 getWordWithDigraphSuggestionsRec(proximityInfo, xcoordinates, ycoordinates,
    131                         codesBuffer, xCoordinatesBuffer, yCoordinatesBuffer, codesBufferSize,
    132                         bigramMap, bigramFilter, useFullEditDistance, codesSrc + i + 1,
    133                         codesRemain - i - 1, currentDepth + 1, codesDest + i, correction,
    134                         queuePool, digraphs, digraphsSize);
    135 
    136                 // Copy the second char of the digraph in place, then continue processing on
    137                 // the remaining part of the word.
    138                 // In our example, after "pru" in the buffer copy the "e", and continue on "fen"
    139                 memcpy(codesDest + i, codesSrc + i, BYTES_IN_ONE_CHAR);
    140                 getWordWithDigraphSuggestionsRec(proximityInfo, xcoordinates, ycoordinates,
    141                         codesBuffer, xCoordinatesBuffer, yCoordinatesBuffer, codesBufferSize,
    142                         bigramMap, bigramFilter, useFullEditDistance, codesSrc + i, codesRemain - i,
    143                         currentDepth + 1, codesDest + i, correction, queuePool, digraphs,
    144                         digraphsSize);
    145                 return;
    146             }
    147         }
    148     }
    149 
    150     // If we come here, we hit the end of the word: let's check it against the dictionary.
    151     // In our example, we'll come here once for "prufen" and then once for "pruefen".
    152     // If the word contains several digraphs, we'll come it for the product of them.
    153     // eg. if the word is "ueberpruefen" we'll test, in order, against
    154     // "uberprufen", "uberpruefen", "ueberprufen", "ueberpruefen".
    155     const unsigned int remainingBytes = BYTES_IN_ONE_CHAR * codesRemain;
    156     if (0 != remainingBytes) {
    157         memcpy(codesDest, codesSrc, remainingBytes);
    158         memcpy(&xCoordinatesBuffer[startIndex], &xcoordinates[codesBufferSize - codesRemain],
    159                 sizeof(int) * codesRemain);
    160         memcpy(&yCoordinatesBuffer[startIndex], &ycoordinates[codesBufferSize - codesRemain],
    161                 sizeof(int) * codesRemain);
    162     }
    163 
    164     getWordSuggestions(proximityInfo, xCoordinatesBuffer, yCoordinatesBuffer, codesBuffer,
    165             startIndex + codesRemain, bigramMap, bigramFilter, useFullEditDistance, correction,
    166             queuePool);
    167 }
    168 
    169 // bigramMap contains the association <bigram address> -> <bigram frequency>
    170 // bigramFilter is a bloom filter for fast rejection: see functions setInFilter and isInFilter
    171 // in bigram_dictionary.cpp
    172 int UnigramDictionary::getSuggestions(ProximityInfo *proximityInfo,
    173         WordsPriorityQueuePool *queuePool, Correction *correction, const int *xcoordinates,
    174         const int *ycoordinates, const int *codes, const int codesSize,
    175         const std::map<int, int> *bigramMap, const uint8_t *bigramFilter,
    176         const bool useFullEditDistance, unsigned short *outWords, int *frequencies) {
    177 
    178     queuePool->clearAll();
    179     Correction* masterCorrection = correction;
    180     correction->resetCorrection();
    181     if (BinaryFormat::REQUIRES_GERMAN_UMLAUT_PROCESSING & FLAGS)
    182     { // Incrementally tune the word and try all possibilities
    183         int codesBuffer[getCodesBufferSize(codes, codesSize)];
    184         int xCoordinatesBuffer[codesSize];
    185         int yCoordinatesBuffer[codesSize];
    186         getWordWithDigraphSuggestionsRec(proximityInfo, xcoordinates, ycoordinates, codesBuffer,
    187                 xCoordinatesBuffer, yCoordinatesBuffer, codesSize, bigramMap, bigramFilter,
    188                 useFullEditDistance, codes, codesSize, 0, codesBuffer, masterCorrection,
    189                 queuePool, GERMAN_UMLAUT_DIGRAPHS,
    190                 sizeof(GERMAN_UMLAUT_DIGRAPHS) / sizeof(GERMAN_UMLAUT_DIGRAPHS[0]));
    191     } else if (BinaryFormat::REQUIRES_FRENCH_LIGATURES_PROCESSING & FLAGS) {
    192         int codesBuffer[getCodesBufferSize(codes, codesSize)];
    193         int xCoordinatesBuffer[codesSize];
    194         int yCoordinatesBuffer[codesSize];
    195         getWordWithDigraphSuggestionsRec(proximityInfo, xcoordinates, ycoordinates, codesBuffer,
    196                 xCoordinatesBuffer, yCoordinatesBuffer, codesSize, bigramMap, bigramFilter,
    197                 useFullEditDistance, codes, codesSize, 0, codesBuffer, masterCorrection,
    198                 queuePool, FRENCH_LIGATURES_DIGRAPHS,
    199                 sizeof(FRENCH_LIGATURES_DIGRAPHS) / sizeof(FRENCH_LIGATURES_DIGRAPHS[0]));
    200     } else { // Normal processing
    201         getWordSuggestions(proximityInfo, xcoordinates, ycoordinates, codes, codesSize,
    202                 bigramMap, bigramFilter, useFullEditDistance, masterCorrection, queuePool);
    203     }
    204 
    205     PROF_START(20);
    206     if (DEBUG_DICT) {
    207         float ns = queuePool->getMasterQueue()->getHighestNormalizedScore(
    208                 proximityInfo->getPrimaryInputWord(), codesSize, 0, 0, 0);
    209         ns += 0;
    210         AKLOGI("Max normalized score = %f", ns);
    211     }
    212     const int suggestedWordsCount =
    213             queuePool->getMasterQueue()->outputSuggestions(
    214                     proximityInfo->getPrimaryInputWord(), codesSize, frequencies, outWords);
    215 
    216     if (DEBUG_DICT) {
    217         float ns = queuePool->getMasterQueue()->getHighestNormalizedScore(
    218                 proximityInfo->getPrimaryInputWord(), codesSize, 0, 0, 0);
    219         ns += 0;
    220         AKLOGI("Returning %d words", suggestedWordsCount);
    221         /// Print the returned words
    222         for (int j = 0; j < suggestedWordsCount; ++j) {
    223             short unsigned int* w = outWords + j * MAX_WORD_LENGTH;
    224             char s[MAX_WORD_LENGTH];
    225             for (int i = 0; i <= MAX_WORD_LENGTH; i++) s[i] = w[i];
    226             AKLOGI("%s %i", s, frequencies[j]);
    227         }
    228     }
    229     PROF_END(20);
    230     PROF_CLOSE;
    231     return suggestedWordsCount;
    232 }
    233 
    234 void UnigramDictionary::getWordSuggestions(ProximityInfo *proximityInfo,
    235         const int *xcoordinates, const int *ycoordinates, const int *codes,
    236         const int inputLength, const std::map<int, int> *bigramMap, const uint8_t *bigramFilter,
    237         const bool useFullEditDistance, Correction *correction, WordsPriorityQueuePool *queuePool) {
    238 
    239     PROF_OPEN;
    240     PROF_START(0);
    241     PROF_END(0);
    242 
    243     PROF_START(1);
    244     getOneWordSuggestions(proximityInfo, xcoordinates, ycoordinates, codes, bigramMap, bigramFilter,
    245             useFullEditDistance, inputLength, correction, queuePool);
    246     PROF_END(1);
    247 
    248     PROF_START(2);
    249     // Note: This line is intentionally left blank
    250     PROF_END(2);
    251 
    252     PROF_START(3);
    253     // Note: This line is intentionally left blank
    254     PROF_END(3);
    255 
    256     PROF_START(4);
    257     bool hasAutoCorrectionCandidate = false;
    258     WordsPriorityQueue* masterQueue = queuePool->getMasterQueue();
    259     if (masterQueue->size() > 0) {
    260         float nsForMaster = masterQueue->getHighestNormalizedScore(
    261                 proximityInfo->getPrimaryInputWord(), inputLength, 0, 0, 0);
    262         hasAutoCorrectionCandidate = (nsForMaster > START_TWO_WORDS_CORRECTION_THRESHOLD);
    263     }
    264     PROF_END(4);
    265 
    266     PROF_START(5);
    267     // Multiple word suggestions
    268     if (SUGGEST_MULTIPLE_WORDS
    269             && inputLength >= MIN_USER_TYPED_LENGTH_FOR_MULTIPLE_WORD_SUGGESTION) {
    270         getSplitMultipleWordsSuggestions(proximityInfo, xcoordinates, ycoordinates, codes,
    271                 useFullEditDistance, inputLength, correction, queuePool,
    272                 hasAutoCorrectionCandidate);
    273     }
    274     PROF_END(5);
    275 
    276     PROF_START(6);
    277     // Note: This line is intentionally left blank
    278     PROF_END(6);
    279 
    280     if (DEBUG_DICT) {
    281         queuePool->dumpSubQueue1TopSuggestions();
    282         for (int i = 0; i < SUB_QUEUE_MAX_COUNT; ++i) {
    283             WordsPriorityQueue* queue = queuePool->getSubQueue(FIRST_WORD_INDEX, i);
    284             if (queue->size() > 0) {
    285                 WordsPriorityQueue::SuggestedWord* sw = queue->top();
    286                 const int score = sw->mScore;
    287                 const unsigned short* word = sw->mWord;
    288                 const int wordLength = sw->mWordLength;
    289                 float ns = Correction::RankingAlgorithm::calcNormalizedScore(
    290                         proximityInfo->getPrimaryInputWord(), i, word, wordLength, score);
    291                 ns += 0;
    292                 AKLOGI("--- TOP SUB WORDS for %d --- %d %f [%d]", i, score, ns,
    293                         (ns > TWO_WORDS_CORRECTION_WITH_OTHER_ERROR_THRESHOLD));
    294                 DUMP_WORD(proximityInfo->getPrimaryInputWord(), i);
    295                 DUMP_WORD(word, wordLength);
    296             }
    297         }
    298     }
    299 }
    300 
    301 void UnigramDictionary::initSuggestions(ProximityInfo *proximityInfo, const int *xCoordinates,
    302         const int *yCoordinates, const int *codes, const int inputLength, Correction *correction) {
    303     if (DEBUG_DICT) {
    304         AKLOGI("initSuggest");
    305         DUMP_WORD_INT(codes, inputLength);
    306     }
    307     proximityInfo->setInputParams(codes, inputLength, xCoordinates, yCoordinates);
    308     const int maxDepth = min(inputLength * MAX_DEPTH_MULTIPLIER, MAX_WORD_LENGTH);
    309     correction->initCorrection(proximityInfo, inputLength, maxDepth);
    310 }
    311 
    312 static const char QUOTE = '\'';
    313 static const char SPACE = ' ';
    314 
    315 void UnigramDictionary::getOneWordSuggestions(ProximityInfo *proximityInfo,
    316         const int *xcoordinates, const int *ycoordinates, const int *codes,
    317         const std::map<int, int> *bigramMap, const uint8_t *bigramFilter,
    318         const bool useFullEditDistance, const int inputLength,
    319         Correction *correction, WordsPriorityQueuePool *queuePool) {
    320     initSuggestions(proximityInfo, xcoordinates, ycoordinates, codes, inputLength, correction);
    321     getSuggestionCandidates(useFullEditDistance, inputLength, bigramMap, bigramFilter, correction,
    322             queuePool, true /* doAutoCompletion */, DEFAULT_MAX_ERRORS, FIRST_WORD_INDEX);
    323 }
    324 
    325 void UnigramDictionary::getSuggestionCandidates(const bool useFullEditDistance,
    326         const int inputLength, const std::map<int, int> *bigramMap, const uint8_t *bigramFilter,
    327         Correction *correction, WordsPriorityQueuePool *queuePool,
    328         const bool doAutoCompletion, const int maxErrors, const int currentWordIndex) {
    329     uint8_t totalTraverseCount = correction->pushAndGetTotalTraverseCount();
    330     if (DEBUG_DICT) {
    331         AKLOGI("Traverse count %d", totalTraverseCount);
    332     }
    333     if (totalTraverseCount > MULTIPLE_WORDS_SUGGESTION_MAX_TOTAL_TRAVERSE_COUNT) {
    334         if (DEBUG_DICT) {
    335             AKLOGI("Abort traversing %d", totalTraverseCount);
    336         }
    337         return;
    338     }
    339     // TODO: Remove setCorrectionParams
    340     correction->setCorrectionParams(0, 0, 0,
    341             -1 /* spaceProximityPos */, -1 /* missingSpacePos */, useFullEditDistance,
    342             doAutoCompletion, maxErrors);
    343     int rootPosition = ROOT_POS;
    344     // Get the number of children of root, then increment the position
    345     int childCount = BinaryFormat::getGroupCountAndForwardPointer(DICT_ROOT, &rootPosition);
    346     int outputIndex = 0;
    347 
    348     correction->initCorrectionState(rootPosition, childCount, (inputLength <= 0));
    349 
    350     // Depth first search
    351     while (outputIndex >= 0) {
    352         if (correction->initProcessState(outputIndex)) {
    353             int siblingPos = correction->getTreeSiblingPos(outputIndex);
    354             int firstChildPos;
    355 
    356             const bool needsToTraverseChildrenNodes = processCurrentNode(siblingPos,
    357                     bigramMap, bigramFilter, correction, &childCount, &firstChildPos, &siblingPos,
    358                     queuePool, currentWordIndex);
    359             // Update next sibling pos
    360             correction->setTreeSiblingPos(outputIndex, siblingPos);
    361 
    362             if (needsToTraverseChildrenNodes) {
    363                 // Goes to child node
    364                 outputIndex = correction->goDownTree(outputIndex, childCount, firstChildPos);
    365             }
    366         } else {
    367             // Goes to parent sibling node
    368             outputIndex = correction->getTreeParentIndex(outputIndex);
    369         }
    370     }
    371 }
    372 
    373 inline void UnigramDictionary::onTerminal(const int probability,
    374         const TerminalAttributes& terminalAttributes, Correction *correction,
    375         WordsPriorityQueuePool *queuePool, const bool addToMasterQueue,
    376         const int currentWordIndex) {
    377     const int inputIndex = correction->getInputIndex();
    378     const bool addToSubQueue = inputIndex < SUB_QUEUE_MAX_COUNT;
    379 
    380     int wordLength;
    381     unsigned short* wordPointer;
    382 
    383     if ((currentWordIndex == FIRST_WORD_INDEX) && addToMasterQueue) {
    384         WordsPriorityQueue *masterQueue = queuePool->getMasterQueue();
    385         const int finalProbability =
    386                 correction->getFinalProbability(probability, &wordPointer, &wordLength);
    387         if (finalProbability != NOT_A_PROBABILITY) {
    388             addWord(wordPointer, wordLength, finalProbability, masterQueue);
    389 
    390             const int shortcutProbability = finalProbability > 0 ? finalProbability - 1 : 0;
    391             // Please note that the shortcut candidates will be added to the master queue only.
    392             TerminalAttributes::ShortcutIterator iterator =
    393                     terminalAttributes.getShortcutIterator();
    394             while (iterator.hasNextShortcutTarget()) {
    395                 // TODO: addWord only supports weak ordering, meaning we have no means
    396                 // to control the order of the shortcuts relative to one another or to the word.
    397                 // We need to either modulate the probability of each shortcut according
    398                 // to its own shortcut probability or to make the queue
    399                 // so that the insert order is protected inside the queue for words
    400                 // with the same score. For the moment we use -1 to make sure the shortcut will
    401                 // never be in front of the word.
    402                 uint16_t shortcutTarget[MAX_WORD_LENGTH_INTERNAL];
    403                 const int shortcutTargetStringLength = iterator.getNextShortcutTarget(
    404                         MAX_WORD_LENGTH_INTERNAL, shortcutTarget);
    405                 addWord(shortcutTarget, shortcutTargetStringLength, shortcutProbability,
    406                         masterQueue);
    407             }
    408         }
    409     }
    410 
    411     // We only allow two words + other error correction for words with SUB_QUEUE_MIN_WORD_LENGTH
    412     // or more length.
    413     if (inputIndex >= SUB_QUEUE_MIN_WORD_LENGTH && addToSubQueue) {
    414         WordsPriorityQueue *subQueue;
    415         subQueue = queuePool->getSubQueue(currentWordIndex, inputIndex);
    416         if (!subQueue) {
    417             return;
    418         }
    419         const int finalProbability = correction->getFinalProbabilityForSubQueue(
    420                 probability, &wordPointer, &wordLength, inputIndex);
    421         addWord(wordPointer, wordLength, finalProbability, subQueue);
    422     }
    423 }
    424 
    425 int UnigramDictionary::getSubStringSuggestion(
    426         ProximityInfo *proximityInfo, const int *xcoordinates, const int *ycoordinates,
    427         const int *codes, const bool useFullEditDistance, Correction *correction,
    428         WordsPriorityQueuePool* queuePool, const int inputLength,
    429         const bool hasAutoCorrectionCandidate, const int currentWordIndex,
    430         const int inputWordStartPos, const int inputWordLength,
    431         const int outputWordStartPos, const bool isSpaceProximity, int *freqArray,
    432         int*wordLengthArray, unsigned short* outputWord, int *outputWordLength) {
    433     if (inputWordLength > MULTIPLE_WORDS_SUGGESTION_MAX_WORD_LENGTH) {
    434         return FLAG_MULTIPLE_SUGGEST_ABORT;
    435     }
    436 
    437     /////////////////////////////////////////////
    438     // safety net for multiple word suggestion //
    439     // TODO: Remove this safety net            //
    440     /////////////////////////////////////////////
    441     int smallWordCount = 0;
    442     int singleLetterWordCount = 0;
    443     if (inputWordLength == 1) {
    444         ++singleLetterWordCount;
    445     }
    446     if (inputWordLength <= 2) {
    447         // small word == single letter or 2-letter word
    448         ++smallWordCount;
    449     }
    450     for (int i = 0; i < currentWordIndex; ++i) {
    451         const int length = wordLengthArray[i];
    452         if (length == 1) {
    453             ++singleLetterWordCount;
    454             // Safety net to avoid suggesting sequential single letter words
    455             if (i < (currentWordIndex - 1)) {
    456                 if (wordLengthArray[i + 1] == 1) {
    457                     return FLAG_MULTIPLE_SUGGEST_ABORT;
    458                 }
    459             } else if (inputWordLength == 1) {
    460                 return FLAG_MULTIPLE_SUGGEST_ABORT;
    461             }
    462         }
    463         if (length <= 2) {
    464             ++smallWordCount;
    465         }
    466         // Safety net to avoid suggesting multiple words with many (4 or more, for now) small words
    467         if (singleLetterWordCount >= 3 || smallWordCount >= 4) {
    468             return FLAG_MULTIPLE_SUGGEST_ABORT;
    469         }
    470     }
    471     //////////////////////////////////////////////
    472     // TODO: Remove the safety net above        //
    473     //////////////////////////////////////////////
    474 
    475     unsigned short* tempOutputWord = 0;
    476     int nextWordLength = 0;
    477     // TODO: Optimize init suggestion
    478     initSuggestions(proximityInfo, xcoordinates, ycoordinates, codes,
    479             inputLength, correction);
    480 
    481     int freq = getMostFrequentWordLike(
    482             inputWordStartPos, inputWordLength, proximityInfo, mWord);
    483     if (freq > 0) {
    484         nextWordLength = inputWordLength;
    485         tempOutputWord = mWord;
    486     } else if (!hasAutoCorrectionCandidate) {
    487         if (inputWordStartPos > 0) {
    488             const int offset = inputWordStartPos;
    489             initSuggestions(proximityInfo, &xcoordinates[offset], &ycoordinates[offset],
    490                     codes + offset, inputWordLength, correction);
    491             queuePool->clearSubQueue(currentWordIndex);
    492             // TODO: pass the bigram list for substring suggestion
    493             getSuggestionCandidates(useFullEditDistance, inputWordLength,
    494                     0 /* bigramMap */, 0 /* bigramFilter */, correction, queuePool,
    495                     false /* doAutoCompletion */, MAX_ERRORS_FOR_TWO_WORDS, currentWordIndex);
    496             if (DEBUG_DICT) {
    497                 if (currentWordIndex < MULTIPLE_WORDS_SUGGESTION_MAX_WORDS) {
    498                     AKLOGI("Dump word candidates(%d) %d", currentWordIndex, inputWordLength);
    499                     for (int i = 0; i < SUB_QUEUE_MAX_COUNT; ++i) {
    500                         queuePool->getSubQueue(currentWordIndex, i)->dumpTopWord();
    501                     }
    502                 }
    503             }
    504         }
    505         WordsPriorityQueue* queue = queuePool->getSubQueue(currentWordIndex, inputWordLength);
    506         // TODO: Return the correct value depending on doAutoCompletion
    507         if (!queue || queue->size() <= 0) {
    508             return FLAG_MULTIPLE_SUGGEST_ABORT;
    509         }
    510         int score = 0;
    511         const float ns = queue->getHighestNormalizedScore(
    512                 proximityInfo->getPrimaryInputWord(), inputWordLength,
    513                 &tempOutputWord, &score, &nextWordLength);
    514         if (DEBUG_DICT) {
    515             AKLOGI("NS(%d) = %f, Score = %d", currentWordIndex, ns, score);
    516         }
    517         // Two words correction won't be done if the score of the first word doesn't exceed the
    518         // threshold.
    519         if (ns < TWO_WORDS_CORRECTION_WITH_OTHER_ERROR_THRESHOLD
    520                 || nextWordLength < SUB_QUEUE_MIN_WORD_LENGTH) {
    521             return FLAG_MULTIPLE_SUGGEST_SKIP;
    522         }
    523         freq = score >> (nextWordLength + TWO_WORDS_PLUS_OTHER_ERROR_CORRECTION_DEMOTION_DIVIDER);
    524     }
    525     if (DEBUG_DICT) {
    526         AKLOGI("Freq(%d): %d, length: %d, input length: %d, input start: %d (%d)"
    527                 , currentWordIndex, freq, nextWordLength, inputWordLength, inputWordStartPos,
    528                 wordLengthArray[0]);
    529     }
    530     if (freq <= 0 || nextWordLength <= 0
    531             || MAX_WORD_LENGTH <= (outputWordStartPos + nextWordLength)) {
    532         return FLAG_MULTIPLE_SUGGEST_SKIP;
    533     }
    534     for (int i = 0; i < nextWordLength; ++i) {
    535         outputWord[outputWordStartPos + i] = tempOutputWord[i];
    536     }
    537 
    538     // Put output values
    539     freqArray[currentWordIndex] = freq;
    540     // TODO: put output length instead of input length
    541     wordLengthArray[currentWordIndex] = inputWordLength;
    542     const int tempOutputWordLength = outputWordStartPos + nextWordLength;
    543     if (outputWordLength) {
    544         *outputWordLength = tempOutputWordLength;
    545     }
    546 
    547     if ((inputWordStartPos + inputWordLength) < inputLength) {
    548         if (outputWordStartPos + nextWordLength >= MAX_WORD_LENGTH) {
    549             return FLAG_MULTIPLE_SUGGEST_SKIP;
    550         }
    551         outputWord[tempOutputWordLength] = SPACE;
    552         if (outputWordLength) {
    553             ++*outputWordLength;
    554         }
    555     } else if (currentWordIndex >= 1) {
    556         // TODO: Handle 3 or more words
    557         const int pairFreq = correction->getFreqForSplitMultipleWords(
    558                 freqArray, wordLengthArray, currentWordIndex + 1, isSpaceProximity, outputWord);
    559         if (DEBUG_DICT) {
    560             DUMP_WORD(outputWord, tempOutputWordLength);
    561             for (int i = 0; i < currentWordIndex + 1; ++i) {
    562                 AKLOGI("Split %d,%d words: freq = %d, length = %d", i, currentWordIndex + 1,
    563                         freqArray[i], wordLengthArray[i]);
    564             }
    565             AKLOGI("Split two words: freq = %d, length = %d, %d, isSpace ? %d", pairFreq,
    566                     inputLength, tempOutputWordLength, isSpaceProximity);
    567         }
    568         addWord(outputWord, tempOutputWordLength, pairFreq, queuePool->getMasterQueue());
    569     }
    570     return FLAG_MULTIPLE_SUGGEST_CONTINUE;
    571 }
    572 
    573 void UnigramDictionary::getMultiWordsSuggestionRec(ProximityInfo *proximityInfo,
    574         const int *xcoordinates, const int *ycoordinates, const int *codes,
    575         const bool useFullEditDistance, const int inputLength,
    576         Correction *correction, WordsPriorityQueuePool* queuePool,
    577         const bool hasAutoCorrectionCandidate, const int startInputPos, const int startWordIndex,
    578         const int outputWordLength, int *freqArray, int* wordLengthArray,
    579         unsigned short* outputWord) {
    580     if (startWordIndex >= (MULTIPLE_WORDS_SUGGESTION_MAX_WORDS - 1)) {
    581         // Return if the last word index
    582         return;
    583     }
    584     if (startWordIndex >= 1
    585             && (hasAutoCorrectionCandidate
    586                     || inputLength < MIN_INPUT_LENGTH_FOR_THREE_OR_MORE_WORDS_CORRECTION)) {
    587         // Do not suggest 3+ words if already has auto correction candidate
    588         return;
    589     }
    590     for (int i = startInputPos + 1; i < inputLength; ++i) {
    591         if (DEBUG_CORRECTION_FREQ) {
    592             AKLOGI("Multi words(%d), start in %d sep %d start out %d",
    593                     startWordIndex, startInputPos, i, outputWordLength);
    594             DUMP_WORD(outputWord, outputWordLength);
    595         }
    596         int tempOutputWordLength = 0;
    597         // Current word
    598         int inputWordStartPos = startInputPos;
    599         int inputWordLength = i - startInputPos;
    600         const int suggestionFlag = getSubStringSuggestion(proximityInfo, xcoordinates, ycoordinates,
    601                 codes, useFullEditDistance, correction, queuePool, inputLength,
    602                 hasAutoCorrectionCandidate, startWordIndex, inputWordStartPos, inputWordLength,
    603                 outputWordLength, true /* not used */, freqArray, wordLengthArray, outputWord,
    604                 &tempOutputWordLength);
    605         if (suggestionFlag == FLAG_MULTIPLE_SUGGEST_ABORT) {
    606             // TODO: break here
    607             continue;
    608         } else if (suggestionFlag == FLAG_MULTIPLE_SUGGEST_SKIP) {
    609             continue;
    610         }
    611 
    612         if (DEBUG_CORRECTION_FREQ) {
    613             AKLOGI("Do missing space correction");
    614         }
    615         // Next word
    616         // Missing space
    617         inputWordStartPos = i;
    618         inputWordLength = inputLength - i;
    619         if(getSubStringSuggestion(proximityInfo, xcoordinates, ycoordinates, codes,
    620                 useFullEditDistance, correction, queuePool, inputLength, hasAutoCorrectionCandidate,
    621                 startWordIndex + 1, inputWordStartPos, inputWordLength, tempOutputWordLength,
    622                 false /* missing space */, freqArray, wordLengthArray, outputWord, 0)
    623                         != FLAG_MULTIPLE_SUGGEST_CONTINUE) {
    624             getMultiWordsSuggestionRec(proximityInfo, xcoordinates, ycoordinates, codes,
    625                     useFullEditDistance, inputLength, correction, queuePool,
    626                     hasAutoCorrectionCandidate, inputWordStartPos, startWordIndex + 1,
    627                     tempOutputWordLength, freqArray, wordLengthArray, outputWord);
    628         }
    629 
    630         // Mistyped space
    631         ++inputWordStartPos;
    632         --inputWordLength;
    633 
    634         if (inputWordLength <= 0) {
    635             continue;
    636         }
    637 
    638         const int x = xcoordinates[inputWordStartPos - 1];
    639         const int y = ycoordinates[inputWordStartPos - 1];
    640         if (!proximityInfo->hasSpaceProximity(x, y)) {
    641             continue;
    642         }
    643 
    644         if (DEBUG_CORRECTION_FREQ) {
    645             AKLOGI("Do mistyped space correction");
    646         }
    647         getSubStringSuggestion(proximityInfo, xcoordinates, ycoordinates, codes,
    648                 useFullEditDistance, correction, queuePool, inputLength, hasAutoCorrectionCandidate,
    649                 startWordIndex + 1, inputWordStartPos, inputWordLength, tempOutputWordLength,
    650                 true /* mistyped space */, freqArray, wordLengthArray, outputWord, 0);
    651     }
    652 }
    653 
    654 void UnigramDictionary::getSplitMultipleWordsSuggestions(ProximityInfo *proximityInfo,
    655         const int *xcoordinates, const int *ycoordinates, const int *codes,
    656         const bool useFullEditDistance, const int inputLength,
    657         Correction *correction, WordsPriorityQueuePool* queuePool,
    658         const bool hasAutoCorrectionCandidate) {
    659     if (inputLength >= MAX_WORD_LENGTH) return;
    660     if (DEBUG_DICT) {
    661         AKLOGI("--- Suggest multiple words");
    662     }
    663 
    664     // Allocating fixed length array on stack
    665     unsigned short outputWord[MAX_WORD_LENGTH];
    666     int freqArray[MULTIPLE_WORDS_SUGGESTION_MAX_WORDS];
    667     int wordLengthArray[MULTIPLE_WORDS_SUGGESTION_MAX_WORDS];
    668     const int outputWordLength = 0;
    669     const int startInputPos = 0;
    670     const int startWordIndex = 0;
    671     getMultiWordsSuggestionRec(proximityInfo, xcoordinates, ycoordinates, codes,
    672             useFullEditDistance, inputLength, correction, queuePool, hasAutoCorrectionCandidate,
    673             startInputPos, startWordIndex, outputWordLength, freqArray, wordLengthArray,
    674             outputWord);
    675 }
    676 
    677 // Wrapper for getMostFrequentWordLikeInner, which matches it to the previous
    678 // interface.
    679 inline int UnigramDictionary::getMostFrequentWordLike(const int startInputIndex,
    680         const int inputLength, ProximityInfo *proximityInfo, unsigned short *word) {
    681     uint16_t inWord[inputLength];
    682 
    683     for (int i = 0; i < inputLength; ++i) {
    684         inWord[i] = (uint16_t)proximityInfo->getPrimaryCharAt(startInputIndex + i);
    685     }
    686     return getMostFrequentWordLikeInner(inWord, inputLength, word);
    687 }
    688 
    689 // This function will take the position of a character array within a CharGroup,
    690 // and check it actually like-matches the word in inWord starting at startInputIndex,
    691 // that is, it matches it with case and accents squashed.
    692 // The function returns true if there was a full match, false otherwise.
    693 // The function will copy on-the-fly the characters in the CharGroup to outNewWord.
    694 // It will also place the end position of the array in outPos; in outInputIndex,
    695 // it will place the index of the first char AFTER the match if there was a match,
    696 // and the initial position if there was not. It makes sense because if there was
    697 // a match we want to continue searching, but if there was not, we want to go to
    698 // the next CharGroup.
    699 // In and out parameters may point to the same location. This function takes care
    700 // not to use any input parameters after it wrote into its outputs.
    701 static inline bool testCharGroupForContinuedLikeness(const uint8_t flags,
    702         const uint8_t* const root, const int startPos,
    703         const uint16_t* const inWord, const int startInputIndex,
    704         int32_t* outNewWord, int* outInputIndex, int* outPos) {
    705     const bool hasMultipleChars = (0 != (UnigramDictionary::FLAG_HAS_MULTIPLE_CHARS & flags));
    706     int pos = startPos;
    707     int32_t character = BinaryFormat::getCharCodeAndForwardPointer(root, &pos);
    708     int32_t baseChar = toBaseLowerCase(character);
    709     const uint16_t wChar = toBaseLowerCase(inWord[startInputIndex]);
    710 
    711     if (baseChar != wChar) {
    712         *outPos = hasMultipleChars ? BinaryFormat::skipOtherCharacters(root, pos) : pos;
    713         *outInputIndex = startInputIndex;
    714         return false;
    715     }
    716     int inputIndex = startInputIndex;
    717     outNewWord[inputIndex] = character;
    718     if (hasMultipleChars) {
    719         character = BinaryFormat::getCharCodeAndForwardPointer(root, &pos);
    720         while (NOT_A_CHARACTER != character) {
    721             baseChar = toBaseLowerCase(character);
    722             if (toBaseLowerCase(inWord[++inputIndex]) != baseChar) {
    723                 *outPos = BinaryFormat::skipOtherCharacters(root, pos);
    724                 *outInputIndex = startInputIndex;
    725                 return false;
    726             }
    727             outNewWord[inputIndex] = character;
    728             character = BinaryFormat::getCharCodeAndForwardPointer(root, &pos);
    729         }
    730     }
    731     *outInputIndex = inputIndex + 1;
    732     *outPos = pos;
    733     return true;
    734 }
    735 
    736 // This function is invoked when a word like the word searched for is found.
    737 // It will compare the frequency to the max frequency, and if greater, will
    738 // copy the word into the output buffer. In output value maxFreq, it will
    739 // write the new maximum frequency if it changed.
    740 static inline void onTerminalWordLike(const int freq, int32_t* newWord, const int length,
    741         short unsigned int* outWord, int* maxFreq) {
    742     if (freq > *maxFreq) {
    743         for (int q = 0; q < length; ++q)
    744             outWord[q] = newWord[q];
    745         outWord[length] = 0;
    746         *maxFreq = freq;
    747     }
    748 }
    749 
    750 // Will find the highest frequency of the words like the one passed as an argument,
    751 // that is, everything that only differs by case/accents.
    752 int UnigramDictionary::getMostFrequentWordLikeInner(const uint16_t * const inWord,
    753         const int length, short unsigned int* outWord) {
    754     int32_t newWord[MAX_WORD_LENGTH_INTERNAL];
    755     int depth = 0;
    756     int maxFreq = -1;
    757     const uint8_t* const root = DICT_ROOT;
    758 
    759     int startPos = 0;
    760     mStackChildCount[0] = BinaryFormat::getGroupCountAndForwardPointer(root, &startPos);
    761     mStackInputIndex[0] = 0;
    762     mStackSiblingPos[0] = startPos;
    763     while (depth >= 0) {
    764         const int charGroupCount = mStackChildCount[depth];
    765         int pos = mStackSiblingPos[depth];
    766         for (int charGroupIndex = charGroupCount - 1; charGroupIndex >= 0; --charGroupIndex) {
    767             int inputIndex = mStackInputIndex[depth];
    768             const uint8_t flags = BinaryFormat::getFlagsAndForwardPointer(root, &pos);
    769             // Test whether all chars in this group match with the word we are searching for. If so,
    770             // we want to traverse its children (or if the length match, evaluate its frequency).
    771             // Note that this function will output the position regardless, but will only write
    772             // into inputIndex if there is a match.
    773             const bool isAlike = testCharGroupForContinuedLikeness(flags, root, pos, inWord,
    774                     inputIndex, newWord, &inputIndex, &pos);
    775             if (isAlike && (FLAG_IS_TERMINAL & flags) && (inputIndex == length)) {
    776                 const int frequency = BinaryFormat::readFrequencyWithoutMovingPointer(root, pos);
    777                 onTerminalWordLike(frequency, newWord, inputIndex, outWord, &maxFreq);
    778             }
    779             pos = BinaryFormat::skipFrequency(flags, pos);
    780             const int siblingPos = BinaryFormat::skipChildrenPosAndAttributes(root, flags, pos);
    781             const int childrenNodePos = BinaryFormat::readChildrenPosition(root, flags, pos);
    782             // If we had a match and the word has children, we want to traverse them. We don't have
    783             // to traverse words longer than the one we are searching for, since they will not match
    784             // anyway, so don't traverse unless inputIndex < length.
    785             if (isAlike && (-1 != childrenNodePos) && (inputIndex < length)) {
    786                 // Save position for this depth, to get back to this once children are done
    787                 mStackChildCount[depth] = charGroupIndex;
    788                 mStackSiblingPos[depth] = siblingPos;
    789                 // Prepare stack values for next depth
    790                 ++depth;
    791                 int childrenPos = childrenNodePos;
    792                 mStackChildCount[depth] =
    793                         BinaryFormat::getGroupCountAndForwardPointer(root, &childrenPos);
    794                 mStackSiblingPos[depth] = childrenPos;
    795                 mStackInputIndex[depth] = inputIndex;
    796                 pos = childrenPos;
    797                 // Go to the next depth level.
    798                 ++depth;
    799                 break;
    800             } else {
    801                 // No match, or no children, or word too long to ever match: go the next sibling.
    802                 pos = siblingPos;
    803             }
    804         }
    805         --depth;
    806     }
    807     return maxFreq;
    808 }
    809 
    810 int UnigramDictionary::getFrequency(const int32_t* const inWord, const int length) const {
    811     const uint8_t* const root = DICT_ROOT;
    812     int pos = BinaryFormat::getTerminalPosition(root, inWord, length);
    813     if (NOT_VALID_WORD == pos) {
    814         return NOT_A_PROBABILITY;
    815     }
    816     const uint8_t flags = BinaryFormat::getFlagsAndForwardPointer(root, &pos);
    817     const bool hasMultipleChars = (0 != (FLAG_HAS_MULTIPLE_CHARS & flags));
    818     if (hasMultipleChars) {
    819         pos = BinaryFormat::skipOtherCharacters(root, pos);
    820     } else {
    821         BinaryFormat::getCharCodeAndForwardPointer(DICT_ROOT, &pos);
    822     }
    823     const int unigramFreq = BinaryFormat::readFrequencyWithoutMovingPointer(root, pos);
    824     return unigramFreq;
    825 }
    826 
    827 // TODO: remove this function.
    828 int UnigramDictionary::getBigramPosition(int pos, unsigned short *word, int offset,
    829         int length) const {
    830     return -1;
    831 }
    832 
    833 // ProcessCurrentNode returns a boolean telling whether to traverse children nodes or not.
    834 // If the return value is false, then the caller should read in the output "nextSiblingPosition"
    835 // to find out the address of the next sibling node and pass it to a new call of processCurrentNode.
    836 // It is worthy to note that when false is returned, the output values other than
    837 // nextSiblingPosition are undefined.
    838 // If the return value is true, then the caller must proceed to traverse the children of this
    839 // node. processCurrentNode will output the information about the children: their count in
    840 // newCount, their position in newChildrenPosition, the traverseAllNodes flag in
    841 // newTraverseAllNodes, the match weight into newMatchRate, the input index into newInputIndex, the
    842 // diffs into newDiffs, the sibling position in nextSiblingPosition, and the output index into
    843 // newOutputIndex. Please also note the following caveat: processCurrentNode does not know when
    844 // there aren't any more nodes at this level, it merely returns the address of the first byte after
    845 // the current node in nextSiblingPosition. Thus, the caller must keep count of the nodes at any
    846 // given level, as output into newCount when traversing this level's parent.
    847 inline bool UnigramDictionary::processCurrentNode(const int initialPos,
    848         const std::map<int, int> *bigramMap, const uint8_t *bigramFilter, Correction *correction,
    849         int *newCount, int *newChildrenPosition, int *nextSiblingPosition,
    850         WordsPriorityQueuePool *queuePool, const int currentWordIndex) {
    851     if (DEBUG_DICT) {
    852         correction->checkState();
    853     }
    854     int pos = initialPos;
    855 
    856     // Flags contain the following information:
    857     // - Address type (MASK_GROUP_ADDRESS_TYPE) on two bits:
    858     //   - FLAG_GROUP_ADDRESS_TYPE_{ONE,TWO,THREE}_BYTES means there are children and their address
    859     //     is on the specified number of bytes.
    860     //   - FLAG_GROUP_ADDRESS_TYPE_NOADDRESS means there are no children, and therefore no address.
    861     // - FLAG_HAS_MULTIPLE_CHARS: whether this node has multiple char or not.
    862     // - FLAG_IS_TERMINAL: whether this node is a terminal or not (it may still have children)
    863     // - FLAG_HAS_BIGRAMS: whether this node has bigrams or not
    864     const uint8_t flags = BinaryFormat::getFlagsAndForwardPointer(DICT_ROOT, &pos);
    865     const bool hasMultipleChars = (0 != (FLAG_HAS_MULTIPLE_CHARS & flags));
    866     const bool isTerminalNode = (0 != (FLAG_IS_TERMINAL & flags));
    867 
    868     bool needsToInvokeOnTerminal = false;
    869 
    870     // This gets only ONE character from the stream. Next there will be:
    871     // if FLAG_HAS_MULTIPLE CHARS: the other characters of the same node
    872     // else if FLAG_IS_TERMINAL: the frequency
    873     // else if MASK_GROUP_ADDRESS_TYPE is not NONE: the children address
    874     // Note that you can't have a node that both is not a terminal and has no children.
    875     int32_t c = BinaryFormat::getCharCodeAndForwardPointer(DICT_ROOT, &pos);
    876     assert(NOT_A_CHARACTER != c);
    877 
    878     // We are going to loop through each character and make it look like it's a different
    879     // node each time. To do that, we will process characters in this node in order until
    880     // we find the character terminator. This is signalled by getCharCode* returning
    881     // NOT_A_CHARACTER.
    882     // As a special case, if there is only one character in this node, we must not read the
    883     // next bytes so we will simulate the NOT_A_CHARACTER return by testing the flags.
    884     // This way, each loop run will look like a "virtual node".
    885     do {
    886         // We prefetch the next char. If 'c' is the last char of this node, we will have
    887         // NOT_A_CHARACTER in the next char. From this we can decide whether this virtual node
    888         // should behave as a terminal or not and whether we have children.
    889         const int32_t nextc = hasMultipleChars
    890                 ? BinaryFormat::getCharCodeAndForwardPointer(DICT_ROOT, &pos) : NOT_A_CHARACTER;
    891         const bool isLastChar = (NOT_A_CHARACTER == nextc);
    892         // If there are more chars in this nodes, then this virtual node is not a terminal.
    893         // If we are on the last char, this virtual node is a terminal if this node is.
    894         const bool isTerminal = isLastChar && isTerminalNode;
    895 
    896         Correction::CorrectionType stateType = correction->processCharAndCalcState(
    897                 c, isTerminal);
    898         if (stateType == Correction::TRAVERSE_ALL_ON_TERMINAL
    899                 || stateType == Correction::ON_TERMINAL) {
    900             needsToInvokeOnTerminal = true;
    901         } else if (stateType == Correction::UNRELATED || correction->needsToPrune()) {
    902             // We found that this is an unrelated character, so we should give up traversing
    903             // this node and its children entirely.
    904             // However we may not be on the last virtual node yet so we skip the remaining
    905             // characters in this node, the frequency if it's there, read the next sibling
    906             // position to output it, then return false.
    907             // We don't have to output other values because we return false, as in
    908             // "don't traverse children".
    909             if (!isLastChar) {
    910                 pos = BinaryFormat::skipOtherCharacters(DICT_ROOT, pos);
    911             }
    912             pos = BinaryFormat::skipFrequency(flags, pos);
    913             *nextSiblingPosition =
    914                     BinaryFormat::skipChildrenPosAndAttributes(DICT_ROOT, flags, pos);
    915             return false;
    916         }
    917 
    918         // Prepare for the next character. Promote the prefetched char to current char - the loop
    919         // will take care of prefetching the next. If we finally found our last char, nextc will
    920         // contain NOT_A_CHARACTER.
    921         c = nextc;
    922     } while (NOT_A_CHARACTER != c);
    923 
    924     if (isTerminalNode) {
    925         // The frequency should be here, because we come here only if this is actually
    926         // a terminal node, and we are on its last char.
    927         const int unigramFreq = BinaryFormat::readFrequencyWithoutMovingPointer(DICT_ROOT, pos);
    928         const int childrenAddressPos = BinaryFormat::skipFrequency(flags, pos);
    929         const int attributesPos = BinaryFormat::skipChildrenPosition(flags, childrenAddressPos);
    930         TerminalAttributes terminalAttributes(DICT_ROOT, flags, attributesPos);
    931         // bigramMap contains the bigram frequencies indexed by addresses for fast lookup.
    932         // bigramFilter is a bloom filter of said frequencies for even faster rejection.
    933         const int probability = BinaryFormat::getProbability(initialPos, bigramMap, bigramFilter,
    934                 unigramFreq);
    935         onTerminal(probability, terminalAttributes, correction, queuePool, needsToInvokeOnTerminal,
    936                 currentWordIndex);
    937 
    938         // If there are more chars in this node, then this virtual node has children.
    939         // If we are on the last char, this virtual node has children if this node has.
    940         const bool hasChildren = BinaryFormat::hasChildrenInFlags(flags);
    941 
    942         // This character matched the typed character (enough to traverse the node at least)
    943         // so we just evaluated it. Now we should evaluate this virtual node's children - that
    944         // is, if it has any. If it has no children, we're done here - so we skip the end of
    945         // the node, output the siblings position, and return false "don't traverse children".
    946         // Note that !hasChildren implies isLastChar, so we know we don't have to skip any
    947         // remaining char in this group for there can't be any.
    948         if (!hasChildren) {
    949             pos = BinaryFormat::skipFrequency(flags, pos);
    950             *nextSiblingPosition =
    951                     BinaryFormat::skipChildrenPosAndAttributes(DICT_ROOT, flags, pos);
    952             return false;
    953         }
    954 
    955         // Optimization: Prune out words that are too long compared to how much was typed.
    956         if (correction->needsToPrune()) {
    957             pos = BinaryFormat::skipFrequency(flags, pos);
    958             *nextSiblingPosition =
    959                     BinaryFormat::skipChildrenPosAndAttributes(DICT_ROOT, flags, pos);
    960             if (DEBUG_DICT_FULL) {
    961                 AKLOGI("Traversing was pruned.");
    962             }
    963             return false;
    964         }
    965     }
    966 
    967     // Now we finished processing this node, and we want to traverse children. If there are no
    968     // children, we can't come here.
    969     assert(BinaryFormat::hasChildrenInFlags(flags));
    970 
    971     // If this node was a terminal it still has the frequency under the pointer (it may have been
    972     // read, but not skipped - see readFrequencyWithoutMovingPointer).
    973     // Next come the children position, then possibly attributes (attributes are bigrams only for
    974     // now, maybe something related to shortcuts in the future).
    975     // Once this is read, we still need to output the number of nodes in the immediate children of
    976     // this node, so we read and output it before returning true, as in "please traverse children".
    977     pos = BinaryFormat::skipFrequency(flags, pos);
    978     int childrenPos = BinaryFormat::readChildrenPosition(DICT_ROOT, flags, pos);
    979     *nextSiblingPosition = BinaryFormat::skipChildrenPosAndAttributes(DICT_ROOT, flags, pos);
    980     *newCount = BinaryFormat::getGroupCountAndForwardPointer(DICT_ROOT, &childrenPos);
    981     *newChildrenPosition = childrenPos;
    982     return true;
    983 }
    984 
    985 } // namespace latinime
    986