diff options
Diffstat (limited to 'native')
22 files changed, 295 insertions, 71 deletions
diff --git a/native/jni/com_android_inputmethod_latin_BinaryDictionary.cpp b/native/jni/com_android_inputmethod_latin_BinaryDictionary.cpp index 6a366121d..7f47493b2 100644 --- a/native/jni/com_android_inputmethod_latin_BinaryDictionary.cpp +++ b/native/jni/com_android_inputmethod_latin_BinaryDictionary.cpp @@ -46,8 +46,7 @@ static jlong latinime_BinaryDictionary_open(JNIEnv *env, jclass clazz, jstring s sourceDirChars[sourceDirUtf8Length] = '\0'; DictionaryStructureWithBufferPolicy *const dictionaryStructureWithBufferPolicy = DictionaryStructureWithBufferPolicyFactory::newDictionaryStructureWithBufferPolicy( - sourceDirChars, static_cast<int>(sourceDirUtf8Length), - static_cast<int>(dictOffset), static_cast<int>(dictSize), + sourceDirChars, static_cast<int>(dictOffset), static_cast<int>(dictSize), isUpdatable == JNI_TRUE); if (!dictionaryStructureWithBufferPolicy) { return 0; @@ -59,6 +58,35 @@ static jlong latinime_BinaryDictionary_open(JNIEnv *env, jclass clazz, jstring s return reinterpret_cast<jlong>(dictionary); } +static void latinime_BinaryDictionary_flush(JNIEnv *env, jclass clazz, jlong dict, + jstring filePath) { + Dictionary *dictionary = reinterpret_cast<Dictionary *>(dict); + if (!dictionary) return; + const jsize filePathUtf8Length = env->GetStringUTFLength(filePath); + char filePathChars[filePathUtf8Length + 1]; + env->GetStringUTFRegion(filePath, 0, env->GetStringLength(filePath), filePathChars); + filePathChars[filePathUtf8Length] = '\0'; + dictionary->flush(filePathChars); +} + +static bool latinime_BinaryDictionary_needsToRunGC(JNIEnv *env, jclass clazz, + jlong dict) { + Dictionary *dictionary = reinterpret_cast<Dictionary *>(dict); + if (!dictionary) return false; + return dictionary->needsToRunGC(); +} + +static void latinime_BinaryDictionary_flushWithGC(JNIEnv *env, jclass clazz, jlong dict, + jstring filePath) { + Dictionary *dictionary = reinterpret_cast<Dictionary *>(dict); + if (!dictionary) return; + const jsize filePathUtf8Length = env->GetStringUTFLength(filePath); + char filePathChars[filePathUtf8Length + 1]; + env->GetStringUTFRegion(filePath, 0, env->GetStringLength(filePath), filePathChars); + filePathChars[filePathUtf8Length] = '\0'; + dictionary->flushWithGC(filePathChars); +} + static void latinime_BinaryDictionary_close(JNIEnv *env, jclass clazz, jlong dict) { Dictionary *dictionary = reinterpret_cast<Dictionary *>(dict); if (!dictionary) return; @@ -160,8 +188,8 @@ static jint latinime_BinaryDictionary_getProbability(JNIEnv *env, jclass clazz, return dictionary->getProbability(codePoints, wordLength); } -static jboolean latinime_BinaryDictionary_isValidBigram(JNIEnv *env, jclass clazz, jlong dict, - jintArray word0, jintArray word1) { +static jint latinime_BinaryDictionary_getBigramProbability(JNIEnv *env, jclass clazz, + jlong dict, jintArray word0, jintArray word1) { Dictionary *dictionary = reinterpret_cast<Dictionary *>(dict); if (!dictionary) return JNI_FALSE; const jsize word0Length = env->GetArrayLength(word0); @@ -170,7 +198,8 @@ static jboolean latinime_BinaryDictionary_isValidBigram(JNIEnv *env, jclass claz int word1CodePoints[word1Length]; env->GetIntArrayRegion(word0, 0, word0Length, word0CodePoints); env->GetIntArrayRegion(word1, 0, word1Length, word1CodePoints); - return dictionary->isValidBigram(word0CodePoints, word0Length, word1CodePoints, word1Length); + return dictionary->getBigramProbability(word0CodePoints, word0Length, word1CodePoints, + word1Length); } static jfloat latinime_BinaryDictionary_calcNormalizedScore(JNIEnv *env, jclass clazz, @@ -241,6 +270,16 @@ static void latinime_BinaryDictionary_removeBigramWords(JNIEnv *env, jclass claz word1Length); } +static int latinime_BinaryDictionary_calculateProbabilityNative(JNIEnv *env, jclass clazz, + jlong dict, jint unigramProbability, jint bigramProbability) { + Dictionary *dictionary = reinterpret_cast<Dictionary *>(dict); + if (!dictionary) { + return NOT_A_PROBABILITY; + } + return dictionary->getDictionaryStructurePolicy()->getProbability(unigramProbability, + bigramProbability); +} + static const JNINativeMethod sMethods[] = { { const_cast<char *>("openNative"), @@ -253,6 +292,21 @@ static const JNINativeMethod sMethods[] = { reinterpret_cast<void *>(latinime_BinaryDictionary_close) }, { + const_cast<char *>("flushNative"), + const_cast<char *>("(JLjava/lang/String;)V"), + reinterpret_cast<void *>(latinime_BinaryDictionary_flush) + }, + { + const_cast<char *>("needsToRunGCNative"), + const_cast<char *>("(J)Z"), + reinterpret_cast<void *>(latinime_BinaryDictionary_needsToRunGC) + }, + { + const_cast<char *>("flushWithGCNative"), + const_cast<char *>("(JLjava/lang/String;)V"), + reinterpret_cast<void *>(latinime_BinaryDictionary_flushWithGC) + }, + { const_cast<char *>("getSuggestionsNative"), const_cast<char *>("(JJJ[I[I[I[I[III[I[I[I[I[I[I[I)I"), reinterpret_cast<void *>(latinime_BinaryDictionary_getSuggestions) @@ -263,9 +317,9 @@ static const JNINativeMethod sMethods[] = { reinterpret_cast<void *>(latinime_BinaryDictionary_getProbability) }, { - const_cast<char *>("isValidBigramNative"), - const_cast<char *>("(J[I[I)Z"), - reinterpret_cast<void *>(latinime_BinaryDictionary_isValidBigram) + const_cast<char *>("getBigramProbabilityNative"), + const_cast<char *>("(J[I[I)I"), + reinterpret_cast<void *>(latinime_BinaryDictionary_getBigramProbability) }, { const_cast<char *>("calcNormalizedScoreNative"), @@ -291,6 +345,11 @@ static const JNINativeMethod sMethods[] = { const_cast<char *>("removeBigramWordsNative"), const_cast<char *>("(J[I[I)V"), reinterpret_cast<void *>(latinime_BinaryDictionary_removeBigramWords) + }, + { + const_cast<char *>("calculateProbabilityNative"), + const_cast<char *>("(JII)I"), + reinterpret_cast<void *>(latinime_BinaryDictionary_calculateProbabilityNative) } }; diff --git a/native/jni/src/suggest/core/dicnode/dic_node.h b/native/jni/src/suggest/core/dicnode/dic_node.h index 377015371..41ef9d2b2 100644 --- a/native/jni/src/suggest/core/dicnode/dic_node.h +++ b/native/jni/src/suggest/core/dicnode/dic_node.h @@ -143,7 +143,7 @@ class DicNode { dicNode->mDicNodeState.mDicNodeStatePrevWord.getPrevWordLength(), dicNode->getOutputWordBuf(), dicNode->mDicNodeProperties.getDepth(), - dicNode->mDicNodeState.mDicNodeStatePrevWord.mPrevSpacePositions, + dicNode->mDicNodeState.mDicNodeStatePrevWord.getSecondWordFirstInputIndex(), mDicNodeState.mDicNodeStateInput.getInputIndex(0) /* lastInputIndex */); PROF_NODE_COPY(&dicNode->mProfiler, mProfiler); } @@ -321,8 +321,13 @@ class DicNode { DUMP_WORD_AND_SCORE("OUTPUT"); } - void outputSpacePositionsResult(int *spaceIndices) const { - mDicNodeState.mDicNodeStatePrevWord.outputSpacePositions(spaceIndices); + int getSecondWordFirstInputIndex(const ProximityInfoState *const pInfoState) const { + const int inputIndex = mDicNodeState.mDicNodeStatePrevWord.getSecondWordFirstInputIndex(); + if (inputIndex == NOT_AN_INDEX) { + return NOT_AN_INDEX; + } else { + return pInfoState->getInputIndexOfSampledPoint(inputIndex); + } } bool hasMultipleWords() const { @@ -573,7 +578,11 @@ class DicNode { } } - AK_FORCE_INLINE void updateInputIndexG(DicNode_InputStateG *inputStateG) { + AK_FORCE_INLINE void updateInputIndexG(const DicNode_InputStateG *const inputStateG) { + if (mDicNodeState.mDicNodeStatePrevWord.getPrevWordCount() == 1 && isFirstLetter()) { + mDicNodeState.mDicNodeStatePrevWord.setSecondWordFirstInputIndex( + inputStateG->mInputIndex); + } mDicNodeState.mDicNodeStateInput.updateInputIndexG(inputStateG->mPointerId, inputStateG->mInputIndex, inputStateG->mPrevCodePoint, inputStateG->mTerminalDiffCost, inputStateG->mRawLength); diff --git a/native/jni/src/suggest/core/dicnode/internal/dic_node_state_prevword.h b/native/jni/src/suggest/core/dicnode/internal/dic_node_state_prevword.h index b7af97018..b8986203d 100644 --- a/native/jni/src/suggest/core/dicnode/internal/dic_node_state_prevword.h +++ b/native/jni/src/suggest/core/dicnode/internal/dic_node_state_prevword.h @@ -22,6 +22,7 @@ #include "defines.h" #include "suggest/core/dicnode/dic_node_utils.h" +#include "suggest/core/layout/proximity_info_state.h" namespace latinime { @@ -29,9 +30,8 @@ class DicNodeStatePrevWord { public: AK_FORCE_INLINE DicNodeStatePrevWord() : mPrevWordCount(0), mPrevWordLength(0), mPrevWordStart(0), mPrevWordProbability(0), - mPrevWordNodePos(NOT_A_DICT_POS) { + mPrevWordNodePos(NOT_A_DICT_POS), mSecondWordFirstInputIndex(NOT_AN_INDEX) { memset(mPrevWord, 0, sizeof(mPrevWord)); - memset(mPrevSpacePositions, 0, sizeof(mPrevSpacePositions)); } virtual ~DicNodeStatePrevWord() {} @@ -42,7 +42,7 @@ class DicNodeStatePrevWord { mPrevWordStart = 0; mPrevWordProbability = -1; mPrevWordNodePos = NOT_A_DICT_POS; - memset(mPrevSpacePositions, 0, sizeof(mPrevSpacePositions)); + mSecondWordFirstInputIndex = NOT_AN_INDEX; } void init(const int prevWordNodePos) { @@ -51,7 +51,7 @@ class DicNodeStatePrevWord { mPrevWordStart = 0; mPrevWordProbability = -1; mPrevWordNodePos = prevWordNodePos; - memset(mPrevSpacePositions, 0, sizeof(mPrevSpacePositions)); + mSecondWordFirstInputIndex = NOT_AN_INDEX; } // Init by copy @@ -61,14 +61,14 @@ class DicNodeStatePrevWord { mPrevWordStart = prevWord->mPrevWordStart; mPrevWordProbability = prevWord->mPrevWordProbability; mPrevWordNodePos = prevWord->mPrevWordNodePos; + mSecondWordFirstInputIndex = prevWord->mSecondWordFirstInputIndex; memcpy(mPrevWord, prevWord->mPrevWord, prevWord->mPrevWordLength * sizeof(mPrevWord[0])); - memcpy(mPrevSpacePositions, prevWord->mPrevSpacePositions, sizeof(mPrevSpacePositions)); } void init(const int16_t prevWordCount, const int16_t prevWordProbability, const int prevWordNodePos, const int *const src0, const int16_t length0, - const int *const src1, const int16_t length1, const int *const prevSpacePositions, - const int lastInputIndex) { + const int *const src1, const int16_t length1, + const int prevWordSecondWordFirstInputIndex, const int lastInputIndex) { mPrevWordCount = min(prevWordCount, static_cast<int16_t>(MAX_RESULTS)); mPrevWordProbability = prevWordProbability; mPrevWordNodePos = prevWordNodePos; @@ -80,8 +80,7 @@ class DicNodeStatePrevWord { mPrevWord[twoWordsLen] = KEYCODE_SPACE; mPrevWordStart = length0; mPrevWordLength = static_cast<int16_t>(twoWordsLen + 1); - memcpy(mPrevSpacePositions, prevSpacePositions, sizeof(mPrevSpacePositions)); - mPrevSpacePositions[mPrevWordCount - 1] = lastInputIndex; + mSecondWordFirstInputIndex = prevWordSecondWordFirstInputIndex; } void truncate(const int offset) { @@ -96,11 +95,12 @@ class DicNodeStatePrevWord { mPrevWordLength = newPrevWordLength; } - void outputSpacePositions(int *spaceIndices) const { - // Convert uint16_t to int - for (int i = 0; i < MAX_RESULTS; i++) { - spaceIndices[i] = mPrevSpacePositions[i]; - } + void setSecondWordFirstInputIndex(const int inputIndex) { + mSecondWordFirstInputIndex = inputIndex; + } + + int getSecondWordFirstInputIndex() const { + return mSecondWordFirstInputIndex; } // TODO: remove @@ -138,8 +138,6 @@ class DicNodeStatePrevWord { // TODO: Move to private int mPrevWord[MAX_WORD_LENGTH]; - // TODO: Move to private - int mPrevSpacePositions[MAX_RESULTS]; private: // Caution!!! @@ -150,6 +148,7 @@ class DicNodeStatePrevWord { int16_t mPrevWordStart; int16_t mPrevWordProbability; int mPrevWordNodePos; + int mSecondWordFirstInputIndex; }; } // namespace latinime #endif // LATINIME_DIC_NODE_STATE_PREVWORD_H diff --git a/native/jni/src/suggest/core/dictionary/bigram_dictionary.cpp b/native/jni/src/suggest/core/dictionary/bigram_dictionary.cpp index 425b07624..5ba71c168 100644 --- a/native/jni/src/suggest/core/dictionary/bigram_dictionary.cpp +++ b/native/jni/src/suggest/core/dictionary/bigram_dictionary.cpp @@ -150,24 +150,26 @@ int BigramDictionary::getBigramListPositionForWord(const int *prevWord, const in return mDictionaryStructurePolicy->getBigramsPositionOfNode(pos); } -bool BigramDictionary::isValidBigram(const int *word0, int length0, const int *word1, +int BigramDictionary::getBigramProbability(const int *word0, int length0, const int *word1, int length1) const { int pos = getBigramListPositionForWord(word0, length0, false /* forceLowerCaseSearch */); // getBigramListPositionForWord returns 0 if this word isn't in the dictionary or has no bigrams - if (NOT_A_DICT_POS == pos) return false; + if (NOT_A_DICT_POS == pos) return NOT_A_PROBABILITY; int nextWordPos = mDictionaryStructurePolicy->getTerminalNodePositionOfWord(word1, length1, false /* forceLowerCaseSearch */); - if (NOT_A_DICT_POS == nextWordPos) return false; + if (NOT_A_DICT_POS == nextWordPos) return NOT_A_PROBABILITY; BinaryDictionaryBigramsIterator bigramsIt( mDictionaryStructurePolicy->getBigramsStructurePolicy(), pos); while (bigramsIt.hasNext()) { bigramsIt.next(); if (bigramsIt.getBigramPos() == nextWordPos) { - return true; + return mDictionaryStructurePolicy->getProbability( + mDictionaryStructurePolicy->getUnigramProbabilityOfPtNode(nextWordPos), + bigramsIt.getProbability()); } } - return false; + return NOT_A_PROBABILITY; } // TODO: Move functions related to bigram to here diff --git a/native/jni/src/suggest/core/dictionary/bigram_dictionary.h b/native/jni/src/suggest/core/dictionary/bigram_dictionary.h index 99b964c49..8af7ee75d 100644 --- a/native/jni/src/suggest/core/dictionary/bigram_dictionary.h +++ b/native/jni/src/suggest/core/dictionary/bigram_dictionary.h @@ -29,7 +29,7 @@ class BigramDictionary { int getPredictions(const int *word, int length, int *outBigramCodePoints, int *outBigramProbability, int *outputTypes) const; - bool isValidBigram(const int *word1, int length1, const int *word2, int length2) const; + int getBigramProbability(const int *word1, int length1, const int *word2, int length2) const; ~BigramDictionary(); private: diff --git a/native/jni/src/suggest/core/dictionary/dictionary.cpp b/native/jni/src/suggest/core/dictionary/dictionary.cpp index 29fe7ab94..ec1b63a12 100644 --- a/native/jni/src/suggest/core/dictionary/dictionary.cpp +++ b/native/jni/src/suggest/core/dictionary/dictionary.cpp @@ -93,8 +93,9 @@ int Dictionary::getProbability(const int *word, int length) const { return getDictionaryStructurePolicy()->getUnigramProbabilityOfPtNode(pos); } -bool Dictionary::isValidBigram(const int *word0, int length0, const int *word1, int length1) const { - return mBigramDictionary->isValidBigram(word0, length0, word1, length1); +int Dictionary::getBigramProbability(const int *word0, int length0, const int *word1, + int length1) const { + return mBigramDictionary->getBigramProbability(word0, length0, word1, length1); } void Dictionary::addUnigramWord(const int *const word, const int length, const int probability) { @@ -112,6 +113,18 @@ void Dictionary::removeBigramWords(const int *const word0, const int length0, mDictionaryStructureWithBufferPolicy->removeBigramWords(word0, length0, word1, length1); } +void Dictionary::flush(const char *const filePath) { + mDictionaryStructureWithBufferPolicy->flush(filePath); +} + +void Dictionary::flushWithGC(const char *const filePath) { + mDictionaryStructureWithBufferPolicy->flushWithGC(filePath); +} + +bool Dictionary::needsToRunGC() { + return mDictionaryStructureWithBufferPolicy->needsToRunGC(); +} + void Dictionary::logDictionaryInfo(JNIEnv *const env) const { const int BUFFER_SIZE = 16; int dictionaryIdCodePointBuffer[BUFFER_SIZE]; diff --git a/native/jni/src/suggest/core/dictionary/dictionary.h b/native/jni/src/suggest/core/dictionary/dictionary.h index 0afe5a54b..974447468 100644 --- a/native/jni/src/suggest/core/dictionary/dictionary.h +++ b/native/jni/src/suggest/core/dictionary/dictionary.h @@ -67,7 +67,7 @@ class Dictionary { int getProbability(const int *word, int length) const; - bool isValidBigram(const int *word0, int length0, const int *word1, int length1) const; + int getBigramProbability(const int *word0, int length0, const int *word1, int length1) const; void addUnigramWord(const int *const word, const int length, const int probability); @@ -77,6 +77,12 @@ class Dictionary { void removeBigramWords(const int *const word0, const int length0, const int *const word1, const int length1); + void flush(const char *const filePath); + + void flushWithGC(const char *const filePath); + + bool needsToRunGC(); + const DictionaryStructureWithBufferPolicy *getDictionaryStructurePolicy() const { return mDictionaryStructureWithBufferPolicy; } diff --git a/native/jni/src/suggest/core/layout/proximity_info_state.h b/native/jni/src/suggest/core/layout/proximity_info_state.h index 01bf81864..c94060fa9 100644 --- a/native/jni/src/suggest/core/layout/proximity_info_state.h +++ b/native/jni/src/suggest/core/layout/proximity_info_state.h @@ -130,6 +130,10 @@ class ProximityInfoState { return mSampledInputYs[index]; } + int getInputIndexOfSampledPoint(const int sampledIndex) const { + return mSampledInputIndice[sampledIndex]; + } + bool hasSpaceProximity(const int index) const; int getLengthCache(const int index) const { diff --git a/native/jni/src/suggest/core/policy/dictionary_structure_with_buffer_policy.h b/native/jni/src/suggest/core/policy/dictionary_structure_with_buffer_policy.h index c8cbbcfdf..24d1b8ba1 100644 --- a/native/jni/src/suggest/core/policy/dictionary_structure_with_buffer_policy.h +++ b/native/jni/src/suggest/core/policy/dictionary_structure_with_buffer_policy.h @@ -74,6 +74,12 @@ class DictionaryStructureWithBufferPolicy { virtual bool removeBigramWords(const int *const word0, const int length0, const int *const word1, const int length1) = 0; + virtual void flush(const char *const filePath) = 0; + + virtual void flushWithGC(const char *const filePath) = 0; + + virtual bool needsToRunGC() const = 0; + protected: DictionaryStructureWithBufferPolicy() {} diff --git a/native/jni/src/suggest/core/session/dic_traverse_session.h b/native/jni/src/suggest/core/session/dic_traverse_session.h index e2ef5fc76..e0b1c67d9 100644 --- a/native/jni/src/suggest/core/session/dic_traverse_session.h +++ b/native/jni/src/suggest/core/session/dic_traverse_session.h @@ -113,7 +113,9 @@ class DicTraverseSession { if (usedPointerCount != 1) { return false; } - *pointerId = usedPointerId; + if (pointerId) { + *pointerId = usedPointerId; + } return true; } diff --git a/native/jni/src/suggest/core/suggest.cpp b/native/jni/src/suggest/core/suggest.cpp index e788e914a..0c925be25 100644 --- a/native/jni/src/suggest/core/suggest.cpp +++ b/native/jni/src/suggest/core/suggest.cpp @@ -117,7 +117,7 @@ void Suggest::initializeSearch(DicTraverseSession *traverseSession, int commitPo * Outputs the final list of suggestions (i.e., terminal nodes). */ int Suggest::outputSuggestions(DicTraverseSession *traverseSession, int *frequencies, - int *outputCodePoints, int *spaceIndices, int *outputTypes) const { + int *outputCodePoints, int *outputIndicesToPartialCommit, int *outputTypes) const { #if DEBUG_EVALUATE_MOST_PROBABLE_STRING const int terminalSize = 0; #else @@ -139,6 +139,7 @@ int Suggest::outputSuggestions(DicTraverseSession *traverseSession, int *frequen SCORING->getMostProbableString(traverseSession, terminalSize, languageWeight, &outputCodePoints[0], &outputTypes[0], &frequencies[0]); if (hasMostProbableString) { + outputIndicesToPartialCommit[outputWordIndex] = NOT_AN_INDEX; ++outputWordIndex; } @@ -160,6 +161,9 @@ int Suggest::outputSuggestions(DicTraverseSession *traverseSession, int *frequen || (traverseSession->getInputSize() >= MIN_LEN_FOR_MULTI_WORD_AUTOCORRECT && terminals[0].hasMultipleWords())) : false; + // TODO: have partial commit work even with multiple pointers. + const bool outputSecondWordFirstLetterInputIndex = + traverseSession->isOnlyOnePointerUsed(0 /* pointerId */); // Output suggestion results here for (int terminalIndex = 0; terminalIndex < terminalSize && outputWordIndex < MAX_RESULTS; ++terminalIndex) { @@ -194,18 +198,21 @@ int Suggest::outputSuggestions(DicTraverseSession *traverseSession, int *frequen terminalDicNode->isExactMatch() || (forceCommitMultiWords && terminalDicNode->hasMultipleWords()) || (isValidWord && SCORING->doesAutoCorrectValidWord())); - maxScore = max(maxScore, finalScore); - - // TODO: Implement a smarter auto-commit method for handling multi-word suggestions. - // Index for top typing suggestion should be 0. - if (isValidWord && outputWordIndex == 0) { - terminalDicNode->outputSpacePositionsResult(spaceIndices); + if (maxScore < finalScore && isValidWord) { + maxScore = finalScore; } // Don't output invalid words. However, we still need to submit their shortcuts if any. if (isValidWord) { outputTypes[outputWordIndex] = Dictionary::KIND_CORRECTION | outputTypeFlags; frequencies[outputWordIndex] = finalScore; + if (outputSecondWordFirstLetterInputIndex) { + outputIndicesToPartialCommit[outputWordIndex] = + terminalDicNode->getSecondWordFirstInputIndex( + traverseSession->getProximityInfoState(0)); + } else { + outputIndicesToPartialCommit[outputWordIndex] = NOT_AN_INDEX; + } // Populate the outputChars array with the suggested word. const int startIndex = outputWordIndex * MAX_WORD_LENGTH; terminalDicNode->outputResult(&outputCodePoints[startIndex]); @@ -220,8 +227,19 @@ int Suggest::outputSuggestions(DicTraverseSession *traverseSession, int *frequen // Shortcut is not supported for multiple words suggestions. // TODO: Check shortcuts during traversal for multiple words suggestions. const bool sameAsTyped = TRAVERSAL->sameAsTyped(traverseSession, terminalDicNode); - outputWordIndex = ShortcutUtils::outputShortcuts(&shortcutIt, outputWordIndex, - finalScore, outputCodePoints, frequencies, outputTypes, sameAsTyped); + const int updatedOutputWordIndex = ShortcutUtils::outputShortcuts(&shortcutIt, + outputWordIndex, finalScore, outputCodePoints, frequencies, outputTypes, + sameAsTyped); + const int secondWordFirstInputIndex = terminalDicNode->getSecondWordFirstInputIndex( + traverseSession->getProximityInfoState(0)); + for (int i = outputWordIndex; i < updatedOutputWordIndex; ++i) { + if (outputSecondWordFirstLetterInputIndex) { + outputIndicesToPartialCommit[i] = secondWordFirstInputIndex; + } else { + outputIndicesToPartialCommit[i] = NOT_AN_INDEX; + } + } + outputWordIndex = updatedOutputWordIndex; } DicNode::managedDelete(terminalDicNode); } diff --git a/native/jni/src/suggest/core/suggest.h b/native/jni/src/suggest/core/suggest.h index 875cbe4e0..b24019632 100644 --- a/native/jni/src/suggest/core/suggest.h +++ b/native/jni/src/suggest/core/suggest.h @@ -55,7 +55,7 @@ class Suggest : public SuggestInterface { void createNextWordDicNode(DicTraverseSession *traverseSession, DicNode *dicNode, const bool spaceSubstitution) const; int outputSuggestions(DicTraverseSession *traverseSession, int *frequencies, - int *outputCodePoints, int *outputIndices, int *outputTypes) const; + int *outputCodePoints, int *outputIndicesToPartialCommit, int *outputTypes) const; void initializeSearch(DicTraverseSession *traverseSession, int commitPoint) const; void expandCurrentDicNodes(DicTraverseSession *traverseSession) const; void processTerminalDicNode(DicTraverseSession *traverseSession, DicNode *dicNode) const; diff --git a/native/jni/src/suggest/policyimpl/dictionary/bigram/dynamic_bigram_list_policy.cpp b/native/jni/src/suggest/policyimpl/dictionary/bigram/dynamic_bigram_list_policy.cpp index b8a5f27e9..4c44d22fd 100644 --- a/native/jni/src/suggest/policyimpl/dictionary/bigram/dynamic_bigram_list_policy.cpp +++ b/native/jni/src/suggest/policyimpl/dictionary/bigram/dynamic_bigram_list_policy.cpp @@ -54,11 +54,13 @@ void DynamicBigramListPolicy::skipAllBigrams(int *const pos) const { } } -bool DynamicBigramListPolicy::copyAllBigrams(int *const fromPos, int *const toPos) { +bool DynamicBigramListPolicy::copyAllBigrams(int *const fromPos, int *const toPos, + int *outBigramsCount) { const bool usesAdditionalBuffer = mBuffer->isInAdditionalBuffer(*fromPos); if (usesAdditionalBuffer) { *fromPos -= mBuffer->getOriginalBufferSize(); } + *outBigramsCount = 0; BigramListReadWriteUtils::BigramFlags flags; do { // The buffer address can be changed after calling buffer writing methods. @@ -91,6 +93,7 @@ bool DynamicBigramListPolicy::copyAllBigrams(int *const fromPos, int *const toPo toPos)) { return false; } + (*outBigramsCount)++; } while(BigramListReadWriteUtils::hasNext(flags)); if (usesAdditionalBuffer) { *fromPos += mBuffer->getOriginalBufferSize(); diff --git a/native/jni/src/suggest/policyimpl/dictionary/bigram/dynamic_bigram_list_policy.h b/native/jni/src/suggest/policyimpl/dictionary/bigram/dynamic_bigram_list_policy.h index c45e26acc..dafb62d80 100644 --- a/native/jni/src/suggest/policyimpl/dictionary/bigram/dynamic_bigram_list_policy.h +++ b/native/jni/src/suggest/policyimpl/dictionary/bigram/dynamic_bigram_list_policy.h @@ -45,8 +45,9 @@ class DynamicBigramListPolicy : public DictionaryBigramsStructurePolicy { void skipAllBigrams(int *const pos) const; // Copy bigrams from the bigram list that starts at fromPos to toPos and advance these - // positions after bigram lists. This method skips invalid bigram entries. - bool copyAllBigrams(int *const fromPos, int *const toPos); + // positions after bigram lists. This method skips invalid bigram entries and write the valid + // bigram entry count to outBigramsCount. + bool copyAllBigrams(int *const fromPos, int *const toPos, int *outBigramsCount); bool addNewBigramEntryToBigramList(const int bigramPos, const int probability, int *const pos); diff --git a/native/jni/src/suggest/policyimpl/dictionary/dictionary_structure_with_buffer_policy_factory.cpp b/native/jni/src/suggest/policyimpl/dictionary/dictionary_structure_with_buffer_policy_factory.cpp index 434858d67..ff80dd2f6 100644 --- a/native/jni/src/suggest/policyimpl/dictionary/dictionary_structure_with_buffer_policy_factory.cpp +++ b/native/jni/src/suggest/policyimpl/dictionary/dictionary_structure_with_buffer_policy_factory.cpp @@ -27,12 +27,12 @@ namespace latinime { /* static */ DictionaryStructureWithBufferPolicy *DictionaryStructureWithBufferPolicyFactory - ::newDictionaryStructureWithBufferPolicy(const char *const path, const int pathLength, - const int bufOffset, const int size, const bool isUpdatable) { + ::newDictionaryStructureWithBufferPolicy(const char *const path, const int bufOffset, + const int size, const bool isUpdatable) { // Allocated buffer in MmapedBuffer::openBuffer() will be freed in the destructor of // impl classes of DictionaryStructureWithBufferPolicy. - const MmappedBuffer *const mmapedBuffer = MmappedBuffer::openBuffer(path, pathLength, bufOffset, - size, isUpdatable); + const MmappedBuffer *const mmapedBuffer = MmappedBuffer::openBuffer(path, bufOffset, size, + isUpdatable); if (!mmapedBuffer) { return 0; } diff --git a/native/jni/src/suggest/policyimpl/dictionary/dictionary_structure_with_buffer_policy_factory.h b/native/jni/src/suggest/policyimpl/dictionary/dictionary_structure_with_buffer_policy_factory.h index 1cb7a89c4..8cebc3b16 100644 --- a/native/jni/src/suggest/policyimpl/dictionary/dictionary_structure_with_buffer_policy_factory.h +++ b/native/jni/src/suggest/policyimpl/dictionary/dictionary_structure_with_buffer_policy_factory.h @@ -27,8 +27,7 @@ namespace latinime { class DictionaryStructureWithBufferPolicyFactory { public: static DictionaryStructureWithBufferPolicy *newDictionaryStructureWithBufferPolicy( - const char *const path, const int pathLength, const int bufOffset, const int size, - const bool isUpdatable); + const char *const path, const int bufOffset, const int size, const bool isUpdatable); private: DISALLOW_IMPLICIT_CONSTRUCTORS(DictionaryStructureWithBufferPolicyFactory); diff --git a/native/jni/src/suggest/policyimpl/dictionary/dynamic_patricia_trie_policy.cpp b/native/jni/src/suggest/policyimpl/dictionary/dynamic_patricia_trie_policy.cpp index cccc09041..2198a13c9 100644 --- a/native/jni/src/suggest/policyimpl/dictionary/dynamic_patricia_trie_policy.cpp +++ b/native/jni/src/suggest/policyimpl/dictionary/dynamic_patricia_trie_policy.cpp @@ -243,4 +243,31 @@ bool DynamicPatriciaTriePolicy::removeBigramWords(const int *const word0, const return writingHelper.removeBigramWords(word0Pos, word1Pos); } +void DynamicPatriciaTriePolicy::flush(const char *const filePath) { + if (!mBuffer->isUpdatable()) { + AKLOGI("Warning: flush() is called for non-updatable dictionary."); + return; + } + DynamicPatriciaTrieWritingHelper writingHelper(&mBufferWithExtendableBuffer, + &mBigramListPolicy, &mShortcutListPolicy); + writingHelper.writeToDictFile(filePath, mBuffer->getBuffer(), mHeaderPolicy.getSize()); +} + +void DynamicPatriciaTriePolicy::flushWithGC(const char *const filePath) { + if (!mBuffer->isUpdatable()) { + AKLOGI("Warning: flushWithGC() is called for non-updatable dictionary."); + return; + } + // TODO: Implement. +} + +bool DynamicPatriciaTriePolicy::needsToRunGC() const { + if (!mBuffer->isUpdatable()) { + AKLOGI("Warning: needsToRunGC() is called for non-updatable dictionary."); + return false; + } + // TODO: Implement. + return false; +} + } // namespace latinime diff --git a/native/jni/src/suggest/policyimpl/dictionary/dynamic_patricia_trie_policy.h b/native/jni/src/suggest/policyimpl/dictionary/dynamic_patricia_trie_policy.h index 50c724012..2cbb0ff3b 100644 --- a/native/jni/src/suggest/policyimpl/dictionary/dynamic_patricia_trie_policy.h +++ b/native/jni/src/suggest/policyimpl/dictionary/dynamic_patricia_trie_policy.h @@ -85,6 +85,12 @@ class DynamicPatriciaTriePolicy : public DictionaryStructureWithBufferPolicy { bool removeBigramWords(const int *const word0, const int length0, const int *const word1, const int length1); + void flush(const char *const filePath); + + void flushWithGC(const char *const filePath); + + bool needsToRunGC() const; + private: DISALLOW_IMPLICIT_CONSTRUCTORS(DynamicPatriciaTriePolicy); diff --git a/native/jni/src/suggest/policyimpl/dictionary/dynamic_patricia_trie_writing_helper.cpp b/native/jni/src/suggest/policyimpl/dictionary/dynamic_patricia_trie_writing_helper.cpp index b80b9b33f..a67c0d94a 100644 --- a/native/jni/src/suggest/policyimpl/dictionary/dynamic_patricia_trie_writing_helper.cpp +++ b/native/jni/src/suggest/policyimpl/dictionary/dynamic_patricia_trie_writing_helper.cpp @@ -16,6 +16,9 @@ #include "suggest/policyimpl/dictionary/dynamic_patricia_trie_writing_helper.h" +#include <cstdio> +#include <cstring> + #include "suggest/policyimpl/dictionary/bigram/dynamic_bigram_list_policy.h" #include "suggest/policyimpl/dictionary/dynamic_patricia_trie_node_reader.h" #include "suggest/policyimpl/dictionary/dynamic_patricia_trie_reading_helper.h" @@ -27,6 +30,8 @@ namespace latinime { const int DynamicPatriciaTrieWritingHelper::CHILDREN_POSITION_FIELD_SIZE = 3; +const char *const DynamicPatriciaTrieWritingHelper::TEMP_FILE_SUFFIX_FOR_WRITING_DICT_FILE = + ".tmp"; bool DynamicPatriciaTrieWritingHelper::addUnigramWord( DynamicPatriciaTrieReadingHelper *const readingHelper, @@ -125,11 +130,45 @@ bool DynamicPatriciaTrieWritingHelper::addBigramWords(const int word0Pos, const bool DynamicPatriciaTrieWritingHelper::removeBigramWords(const int word0Pos, const int word1Pos) { DynamicPatriciaTrieNodeReader nodeReader(mBuffer, mBigramPolicy, mShortcutPolicy); nodeReader.fetchNodeInfoFromBuffer(word0Pos); - if (nodeReader.isDeleted() || nodeReader.getBigramsPos() == NOT_A_DICT_POS) { + if (nodeReader.getBigramsPos() == NOT_A_DICT_POS) { return false; } - // TODO: Implement. - return false; + return mBigramPolicy->removeBigram(nodeReader.getBigramsPos(), word1Pos); +} + +void DynamicPatriciaTrieWritingHelper::writeToDictFile(const char *const fileName, + const uint8_t *const headerBuf, const int headerSize) { + const int tmpFileNameBufSize = strlen(fileName) + + strlen(TEMP_FILE_SUFFIX_FOR_WRITING_DICT_FILE) + 1; + char tmpFileName[tmpFileNameBufSize]; + snprintf(tmpFileName, tmpFileNameBufSize, "%s%s", fileName, + TEMP_FILE_SUFFIX_FOR_WRITING_DICT_FILE); + FILE *const file = fopen(tmpFileName, "wb"); + if (!file) { + return; + } + // Write header. + if (fwrite(headerBuf, headerSize, 1, file) < 1) { + fclose(file); + remove(tmpFileName); + return; + } + // Write data in original buffer. + if (fwrite(mBuffer->getBuffer(false /* usesAdditionalBuffer */), + mBuffer->getOriginalBufferSize(), 1, file) < 1) { + fclose(file); + remove(tmpFileName); + return; + } + // Write data in additional buffer. + if (fwrite(mBuffer->getBuffer(true /* usesAdditionalBuffer */), + mBuffer->getTailPosition() - mBuffer->getOriginalBufferSize(), 1, file) < 1) { + fclose(file); + remove(tmpFileName); + return; + } + fclose(file); + rename(tmpFileName, fileName); } bool DynamicPatriciaTrieWritingHelper::markNodeAsMovedAndSetPosition( @@ -193,13 +232,9 @@ bool DynamicPatriciaTrieWritingHelper::writePtNodeWithFullInfoToBuffer(const boo const int originalBigramListPos, const int originalShortcutListPos, int *const writingPos) { const int nodePos = *writingPos; - // Create node flags and write them. - const PatriciaTrieReadingUtils::NodeFlags nodeFlags = - PatriciaTrieReadingUtils::createAndGetFlags(isBlacklisted, isNotAWord, - probability != NOT_A_PROBABILITY, originalShortcutListPos != NOT_A_DICT_POS, - originalBigramListPos != NOT_A_DICT_POS, codePointCount > 1, - 3 /* childrenPositionFieldSize */); - if (!DynamicPatriciaTrieWritingUtils::writeFlagsAndAdvancePosition(mBuffer, nodeFlags, + // Write dummy flags. The Node flags are updated with appropriate flags at the last step of the + // PtNode writing. + if (!DynamicPatriciaTrieWritingUtils::writeFlagsAndAdvancePosition(mBuffer, 0 /* nodeFlags */, writingPos)) { return false; } @@ -212,7 +247,7 @@ bool DynamicPatriciaTrieWritingHelper::writePtNodeWithFullInfoToBuffer(const boo // Write code points if (!DynamicPatriciaTrieWritingUtils::writeCodePointsAndAdvancePosition(mBuffer, codePoints, codePointCount, writingPos)) { - return false;; + return false; } // Write probability when the probability is a valid probability, which means this node is // terminal. @@ -235,12 +270,25 @@ bool DynamicPatriciaTrieWritingHelper::writePtNodeWithFullInfoToBuffer(const boo } } // Copy bigram list when the originalBigramListPos is valid dictionary position. + int bigramCount = 0; if (originalBigramListPos != NOT_A_DICT_POS) { int fromPos = originalBigramListPos; - if (!mBigramPolicy->copyAllBigrams(&fromPos, writingPos)) { + if (!mBigramPolicy->copyAllBigrams(&fromPos, writingPos, &bigramCount)) { return false; } } + // Create node flags and write them. + PatriciaTrieReadingUtils::NodeFlags nodeFlags = + PatriciaTrieReadingUtils::createAndGetFlags(isBlacklisted, isNotAWord, + probability != NOT_A_PROBABILITY /* isTerminal */, + originalShortcutListPos != NOT_A_DICT_POS /* hasShortcutTargets */, + bigramCount > 0 /* hasBigrams */, codePointCount > 1 /* hasMultipleChars */, + CHILDREN_POSITION_FIELD_SIZE); + int flagsFieldPos = nodePos; + if (!DynamicPatriciaTrieWritingUtils::writeFlagsAndAdvancePosition(mBuffer, nodeFlags, + &flagsFieldPos)) { + return false; + } return true; } diff --git a/native/jni/src/suggest/policyimpl/dictionary/dynamic_patricia_trie_writing_helper.h b/native/jni/src/suggest/policyimpl/dictionary/dynamic_patricia_trie_writing_helper.h index 20e35abcf..faf7a4e1b 100644 --- a/native/jni/src/suggest/policyimpl/dictionary/dynamic_patricia_trie_writing_helper.h +++ b/native/jni/src/suggest/policyimpl/dictionary/dynamic_patricia_trie_writing_helper.h @@ -17,6 +17,8 @@ #ifndef LATINIME_DYNAMIC_PATRICIA_TRIE_WRITING_HELPER_H #define LATINIME_DYNAMIC_PATRICIA_TRIE_WRITING_HELPER_H +#include <stdint.h> + #include "defines.h" namespace latinime { @@ -46,10 +48,14 @@ class DynamicPatriciaTrieWritingHelper { // Remove a bigram relation from word0Pos to word1Pos. bool removeBigramWords(const int word0Pos, const int word1Pos); + void writeToDictFile(const char *const fileName, const uint8_t *const headerBuf, + const int headerSize); + private: DISALLOW_IMPLICIT_CONSTRUCTORS(DynamicPatriciaTrieWritingHelper); static const int CHILDREN_POSITION_FIELD_SIZE; + static const char *const TEMP_FILE_SUFFIX_FOR_WRITING_DICT_FILE; BufferWithExtendableBuffer *const mBuffer; DynamicBigramListPolicy *const mBigramPolicy; diff --git a/native/jni/src/suggest/policyimpl/dictionary/patricia_trie_policy.h b/native/jni/src/suggest/policyimpl/dictionary/patricia_trie_policy.h index 75d976205..cee3e4ab2 100644 --- a/native/jni/src/suggest/policyimpl/dictionary/patricia_trie_policy.h +++ b/native/jni/src/suggest/policyimpl/dictionary/patricia_trie_policy.h @@ -96,6 +96,22 @@ class PatriciaTriePolicy : public DictionaryStructureWithBufferPolicy { return false; } + void flush(const char *const filePath) { + // This method should not be called for non-updatable dictionary. + AKLOGI("Warning: flush() is called for non-updatable dictionary."); + } + + void flushWithGC(const char *const filePath) { + // This method should not be called for non-updatable dictionary. + AKLOGI("Warning: flushWithGC() is called for non-updatable dictionary."); + } + + bool needsToRunGC() const { + // This method should not be called for non-updatable dictionary. + AKLOGI("Warning: needsToRunGC() is called for non-updatable dictionary."); + return false; + } + private: DISALLOW_IMPLICIT_CONSTRUCTORS(PatriciaTriePolicy); diff --git a/native/jni/src/suggest/policyimpl/dictionary/utils/mmapped_buffer.h b/native/jni/src/suggest/policyimpl/dictionary/utils/mmapped_buffer.h index 6febd7832..6b69116eb 100644 --- a/native/jni/src/suggest/policyimpl/dictionary/utils/mmapped_buffer.h +++ b/native/jni/src/suggest/policyimpl/dictionary/utils/mmapped_buffer.h @@ -29,8 +29,8 @@ namespace latinime { class MmappedBuffer { public: - static MmappedBuffer* openBuffer(const char *const path, const int pathLength, - const int bufferOffset, const int bufferSize, const bool isUpdatable) { + static MmappedBuffer* openBuffer(const char *const path, const int bufferOffset, + const int bufferSize, const bool isUpdatable) { const int openMode = isUpdatable ? O_RDWR : O_RDONLY; const int mmapFd = open(path, openMode); if (mmapFd < 0) { |