aboutsummaryrefslogtreecommitdiffstats
path: root/native/src/correction.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'native/src/correction.cpp')
-rw-r--r--native/src/correction.cpp147
1 files changed, 4 insertions, 143 deletions
diff --git a/native/src/correction.cpp b/native/src/correction.cpp
index 63dd283c8..2458bca86 100644
--- a/native/src/correction.cpp
+++ b/native/src/correction.cpp
@@ -269,7 +269,7 @@ bool Correction::needsToPrune() const {
// TODO: use edit distance here
return mOutputIndex - 1 >= mMaxDepth || mProximityCount > mMaxEditDistance
// Allow one char longer word for missing character
- || (!mDoAutoCompletion && (mOutputIndex + 1 >= mInputLength));
+ || (!mDoAutoCompletion && (mOutputIndex > mInputLength));
}
void Correction::addCharToCurrentWord(const int32_t c) {
@@ -555,55 +555,6 @@ Correction::CorrectionType Correction::processCharAndCalcState(
Correction::~Correction() {
}
-/////////////////////////
-// static inline utils //
-/////////////////////////
-
-static const int TWO_31ST_DIV_255 = S_INT_MAX / 255;
-static inline int capped255MultForFullMatchAccentsOrCapitalizationDifference(const int num) {
- return (num < TWO_31ST_DIV_255 ? 255 * num : S_INT_MAX);
-}
-
-static const int TWO_31ST_DIV_2 = S_INT_MAX / 2;
-inline static void multiplyIntCapped(const int multiplier, int *base) {
- const int temp = *base;
- if (temp != S_INT_MAX) {
- // Branch if multiplier == 2 for the optimization
- if (multiplier == 2) {
- *base = TWO_31ST_DIV_2 >= temp ? temp << 1 : S_INT_MAX;
- } else {
- // TODO: This overflow check gives a wrong answer when, for example,
- // temp = 2^16 + 1 and multiplier = 2^17 + 1.
- // Fix this behavior.
- const int tempRetval = temp * multiplier;
- *base = tempRetval >= temp ? tempRetval : S_INT_MAX;
- }
- }
-}
-
-inline static int powerIntCapped(const int base, const int n) {
- if (n <= 0) return 1;
- if (base == 2) {
- return n < 31 ? 1 << n : S_INT_MAX;
- } else {
- int ret = base;
- for (int i = 1; i < n; ++i) multiplyIntCapped(base, &ret);
- return ret;
- }
-}
-
-inline static void multiplyRate(const int rate, int *freq) {
- if (*freq != S_INT_MAX) {
- if (*freq > 1000000) {
- *freq /= 100;
- multiplyIntCapped(rate, freq);
- } else {
- multiplyIntCapped(rate, freq);
- *freq /= 100;
- }
- }
-}
-
inline static int getQuoteCount(const unsigned short* word, const int length) {
int quoteCount = 0;
for (int i = 0; i < length; ++i) {
@@ -939,102 +890,12 @@ int Correction::RankingAlgorithm::calcFreqForSplitTwoWords(
multiplyRate(WORDS_WITH_PROXIMITY_CHARACTER_DEMOTION_RATE, &totalFreq);
}
- multiplyRate(WORDS_WITH_MISSING_SPACE_CHARACTER_DEMOTION_RATE, &totalFreq);
-
- if (capitalizedWordDemotion) {
- multiplyRate(TWO_WORDS_CAPITALIZED_DEMOTION_RATE, &totalFreq);
- }
-
- return totalFreq;
-}
-
-/* static */
-int Correction::RankingAlgorithm::calcFreqForSplitTwoWordsOld(
- const int firstFreq, const int secondFreq, const Correction* correction,
- const unsigned short *word) {
- const int spaceProximityPos = correction->mSpaceProximityPos;
- const int missingSpacePos = correction->mMissingSpacePos;
- if (DEBUG_DICT) {
- int inputCount = 0;
- if (spaceProximityPos >= 0) ++inputCount;
- if (missingSpacePos >= 0) ++inputCount;
- assert(inputCount <= 1);
- }
- const bool isSpaceProximity = spaceProximityPos >= 0;
- const int inputLength = correction->mInputLength;
- const int firstWordLength = isSpaceProximity ? spaceProximityPos : missingSpacePos;
- const int secondWordLength = isSpaceProximity ? (inputLength - spaceProximityPos - 1)
- : (inputLength - missingSpacePos);
- const int typedLetterMultiplier = correction->TYPED_LETTER_MULTIPLIER;
-
- bool firstCapitalizedWordDemotion = false;
- if (firstWordLength >= 2) {
- firstCapitalizedWordDemotion = isUpperCase(word[0]);
- }
-
- bool secondCapitalizedWordDemotion = false;
- if (secondWordLength >= 2) {
- secondCapitalizedWordDemotion = isUpperCase(word[firstWordLength + 1]);
- }
-
- const bool capitalizedWordDemotion =
- firstCapitalizedWordDemotion ^ secondCapitalizedWordDemotion;
-
- if (DEBUG_DICT_FULL) {
- AKLOGI("Two words: %c, %c, %d",
- word[0], word[firstWordLength + 1], capitalizedWordDemotion);
- }
-
- if (firstWordLength == 0 || secondWordLength == 0) {
- return 0;
- }
- const int firstDemotionRate = 100 - 100 / (firstWordLength + 1);
- int tempFirstFreq = firstFreq;
- multiplyRate(firstDemotionRate, &tempFirstFreq);
-
- const int secondDemotionRate = 100 - 100 / (secondWordLength + 1);
- int tempSecondFreq = secondFreq;
- multiplyRate(secondDemotionRate, &tempSecondFreq);
-
- const int totalLength = firstWordLength + secondWordLength;
-
- // Promote pairFreq with multiplying by 2, because the word length is the same as the typed
- // length.
- int totalFreq = tempFirstFreq + tempSecondFreq;
-
- // This is a workaround to try offsetting the not-enough-demotion which will be done in
- // calcNormalizedScore in Utils.java.
- // In calcNormalizedScore the score will be demoted by (1 - 1 / length)
- // but we demoted only (1 - 1 / (length + 1)) so we will additionally adjust freq by
- // (1 - 1 / length) / (1 - 1 / (length + 1)) = (1 - 1 / (length * length))
- const int normalizedScoreNotEnoughDemotionAdjustment = 100 - 100 / (totalLength * totalLength);
- multiplyRate(normalizedScoreNotEnoughDemotionAdjustment, &totalFreq);
-
- // At this moment, totalFreq is calculated by the following formula:
- // (firstFreq * (1 - 1 / (firstWordLength + 1)) + secondFreq * (1 - 1 / (secondWordLength + 1)))
- // * (1 - 1 / totalLength) / (1 - 1 / (totalLength + 1))
-
- multiplyIntCapped(powerIntCapped(typedLetterMultiplier, totalLength), &totalFreq);
-
- // This is another workaround to offset the demotion which will be done in
- // calcNormalizedScore in Utils.java.
- // In calcNormalizedScore the score will be demoted by (1 - 1 / length) so we have to promote
- // the same amount because we already have adjusted the synthetic freq of this "missing or
- // mistyped space" suggestion candidate above in this method.
- const int normalizedScoreDemotionRateOffset = (100 + 100 / totalLength);
- multiplyRate(normalizedScoreDemotionRateOffset, &totalFreq);
-
if (isSpaceProximity) {
- // A word pair with one space proximity correction
- if (DEBUG_DICT) {
- AKLOGI("Found a word pair with space proximity correction.");
- }
- multiplyIntCapped(typedLetterMultiplier, &totalFreq);
- multiplyRate(WORDS_WITH_PROXIMITY_CHARACTER_DEMOTION_RATE, &totalFreq);
+ multiplyRate(WORDS_WITH_MISTYPED_SPACE_DEMOTION_RATE, &totalFreq);
+ } else {
+ multiplyRate(WORDS_WITH_MISSING_SPACE_CHARACTER_DEMOTION_RATE, &totalFreq);
}
- multiplyRate(WORDS_WITH_MISSING_SPACE_CHARACTER_DEMOTION_RATE, &totalFreq);
-
if (capitalizedWordDemotion) {
multiplyRate(TWO_WORDS_CAPITALIZED_DEMOTION_RATE, &totalFreq);
}
generated by cgit v1.2.3-83-g751a (git 2.47.1) at 2025-05-07 18:13:45 +0000