diff --git a/java/src/com/android/inputmethod/latin/BinaryDictionary.java b/java/src/com/android/inputmethod/latin/BinaryDictionary.java index dbc2b9082..03f7d1c10 100644 --- a/java/src/com/android/inputmethod/latin/BinaryDictionary.java +++ b/java/src/com/android/inputmethod/latin/BinaryDictionary.java @@ -147,10 +147,20 @@ public final class BinaryDictionary extends Dictionary { ++len; } if (len > 0) { - final int score = SuggestedWordInfo.KIND_WHITELIST == mOutputTypes[j] + final int flags = mOutputTypes[j] & SuggestedWordInfo.KIND_MASK_FLAGS; + if (0 != (flags & SuggestedWordInfo.KIND_FLAG_POSSIBLY_OFFENSIVE) + && 0 == (flags & SuggestedWordInfo.KIND_FLAG_EXACT_MATCH)) { + // If the word is possibly offensive, we don't output it unless it's also + // an exact match. + continue; + } + final int kind = mOutputTypes[j] & SuggestedWordInfo.KIND_MASK_KIND; + final int score = SuggestedWordInfo.KIND_WHITELIST == kind ? SuggestedWordInfo.MAX_SCORE : mOutputScores[j]; + // TODO: check that all users of the `kind' parameter are ready to accept + // flags too and pass mOutputTypes[j] instead of kind suggestions.add(new SuggestedWordInfo(new String(mOutputCodePoints, start, len), - score, mOutputTypes[j], mDictType)); + score, kind, mDictType)); } } return suggestions; diff --git a/java/src/com/android/inputmethod/latin/SuggestedWords.java b/java/src/com/android/inputmethod/latin/SuggestedWords.java index 616e1911b..dfddb0ffe 100644 --- a/java/src/com/android/inputmethod/latin/SuggestedWords.java +++ b/java/src/com/android/inputmethod/latin/SuggestedWords.java @@ -122,6 +122,7 @@ public final class SuggestedWords { public static final class SuggestedWordInfo { public static final int MAX_SCORE = Integer.MAX_VALUE; + public static final int KIND_MASK_KIND = 0xFF; // Mask to get only the kind public static final int KIND_TYPED = 0; // What user typed public static final int KIND_CORRECTION = 1; // Simple correction/suggestion public static final int KIND_COMPLETION = 2; // Completion (suggestion with appended chars) @@ -132,6 +133,11 @@ public final class SuggestedWords { public static final int KIND_SHORTCUT = 7; // A shortcut public static final int KIND_PREDICTION = 8; // A prediction (== a suggestion with no input) public static final int KIND_RESUMED = 9; // A resumed suggestion (comes from a span) + + public static final int KIND_MASK_FLAGS = 0xFFFFFF00; // Mask to get the flags + public static final int KIND_FLAG_POSSIBLY_OFFENSIVE = 0x80000000; + public static final int KIND_FLAG_EXACT_MATCH = 0x40000000; + public final String mWord; public final int mScore; public final int mKind; // one of the KIND_* constants above diff --git a/native/jni/src/dictionary.h b/native/jni/src/dictionary.h index 0653d3ca9..2ad5b6c0b 100644 --- a/native/jni/src/dictionary.h +++ b/native/jni/src/dictionary.h @@ -31,6 +31,7 @@ class UnigramDictionary; class Dictionary { public: // Taken from SuggestedWords.java + static const int KIND_MASK_KIND = 0xFF; // Mask to get only the kind static const int KIND_TYPED = 0; // What user typed static const int KIND_CORRECTION = 1; // Simple correction/suggestion static const int KIND_COMPLETION = 2; // Completion (suggestion with appended chars) @@ -41,6 +42,10 @@ class Dictionary { static const int KIND_SHORTCUT = 7; // A shortcut static const int KIND_PREDICTION = 8; // A prediction (== a suggestion with no input) + static const int KIND_MASK_FLAGS = 0xFFFFFF00; // Mask to get the flags + static const int KIND_FLAG_POSSIBLY_OFFENSIVE = 0x80000000; + static const int KIND_FLAG_EXACT_MATCH = 0x40000000; + Dictionary(void *dict, int dictSize, int mmapFd, int dictBufAdjust); int getSuggestions(ProximityInfo *proximityInfo, void *traverseSession, int *xcoordinates, diff --git a/native/jni/src/suggest/core/suggest.cpp b/native/jni/src/suggest/core/suggest.cpp index 0cf4e4a68..b1a5ff24e 100644 --- a/native/jni/src/suggest/core/suggest.cpp +++ b/native/jni/src/suggest/core/suggest.cpp @@ -161,12 +161,15 @@ int Suggest::outputSuggestions(DicTraverseSession *traverseSession, int *frequen + doubleLetterCost; const TerminalAttributes terminalAttributes(traverseSession->getOffsetDict(), terminalDicNode->getFlags(), terminalDicNode->getAttributesPos()); - const int originalTerminalProbability = terminalDicNode->getProbability(); + const bool isPossiblyOffensiveWord = terminalDicNode->getProbability() <= 0; + const bool isExactMatch = terminalDicNode->isExactMatch(); + const int outputTypeFlags = + isPossiblyOffensiveWord ? Dictionary::KIND_FLAG_POSSIBLY_OFFENSIVE : 0 + | isExactMatch ? Dictionary::KIND_FLAG_EXACT_MATCH : 0; + + // Entries that are blacklisted or do not represent a word should not be output. + const bool isValidWord = !terminalAttributes.isBlacklistedOrNotAWord(); - // Do not suggest words with a 0 probability, or entries that are blacklisted or do not - // represent a word. However, we should still submit their shortcuts if any. - const bool isValidWord = - originalTerminalProbability > 0 && !terminalAttributes.isBlacklistedOrNotAWord(); // Increase output score of top typing suggestion to ensure autocorrection. // TODO: Better integration with java side autocorrection logic. // Force autocorrection for obvious long multi-word suggestions. @@ -188,10 +191,9 @@ int Suggest::outputSuggestions(DicTraverseSession *traverseSession, int *frequen } } - // Do not suggest words with a 0 probability, or entries that are blacklisted or do not - // represent a word. However, we should still submit their shortcuts if any. + // Don't output invalid words. However, we still need to submit their shortcuts if any. if (isValidWord) { - outputTypes[outputWordIndex] = Dictionary::KIND_CORRECTION; + outputTypes[outputWordIndex] = Dictionary::KIND_CORRECTION | outputTypeFlags; frequencies[outputWordIndex] = finalScore; // Populate the outputChars array with the suggested word. const int startIndex = outputWordIndex * MAX_WORD_LENGTH;