aboutsummaryrefslogtreecommitdiffstats
path: root/native
diff options
context:
space:
mode:
Diffstat (limited to 'native')
-rw-r--r--native/jni/Android.mk2
-rw-r--r--native/jni/com_android_inputmethod_latin_makedict_BinaryDictDecoder.cpp (renamed from native/jni/com_android_inputmethod_latin_makedict_BinaryDictInputOutput.cpp)12
-rw-r--r--native/jni/com_android_inputmethod_latin_makedict_BinaryDictDecoder.h (renamed from native/jni/com_android_inputmethod_latin_makedict_BinaryDictInputOutput.h)8
-rw-r--r--native/jni/jni_common.cpp6
-rw-r--r--native/jni/src/suggest/core/dictionary/probability_utils.h2
-rw-r--r--native/jni/src/suggest/policyimpl/dictionary/binary_format.h470
-rw-r--r--native/jni/src/suggest/policyimpl/dictionary/patricia_trie_policy.cpp265
-rw-r--r--native/jni/src/suggest/policyimpl/dictionary/utils/mmaped_buffer.h4
8 files changed, 277 insertions, 492 deletions
diff --git a/native/jni/Android.mk b/native/jni/Android.mk
index 8d51a2f26..9e7407abd 100644
--- a/native/jni/Android.mk
+++ b/native/jni/Android.mk
@@ -43,7 +43,7 @@ LATIN_IME_JNI_SRC_FILES := \
com_android_inputmethod_keyboard_ProximityInfo.cpp \
com_android_inputmethod_latin_BinaryDictionary.cpp \
com_android_inputmethod_latin_DicTraverseSession.cpp \
- com_android_inputmethod_latin_makedict_BinaryDictInputOutput.cpp \
+ com_android_inputmethod_latin_makedict_BinaryDictDecoder.cpp \
jni_common.cpp
LATIN_IME_CORE_SRC_FILES := \
diff --git a/native/jni/com_android_inputmethod_latin_makedict_BinaryDictInputOutput.cpp b/native/jni/com_android_inputmethod_latin_makedict_BinaryDictDecoder.cpp
index f78883c2d..457b226b6 100644
--- a/native/jni/com_android_inputmethod_latin_makedict_BinaryDictInputOutput.cpp
+++ b/native/jni/com_android_inputmethod_latin_makedict_BinaryDictDecoder.cpp
@@ -14,16 +14,16 @@
* limitations under the License.
*/
-#define LOG_TAG "LatinIME: jni: BinaryDictInputOutput"
+#define LOG_TAG "LatinIME: jni: BinaryDictDecoder"
-#include "com_android_inputmethod_latin_makedict_BinaryDictInputOutput.h"
+#include "com_android_inputmethod_latin_makedict_BinaryDictDecoder.h"
#include "defines.h"
#include "jni.h"
#include "jni_common.h"
namespace latinime {
-static int latinime_BinaryDictInputOutput_doNothing(JNIEnv *env, jclass clazz) {
+static int latinime_BinaryDictDecoder_doNothing(JNIEnv *env, jclass clazz) {
// This is a phony method for test - it does nothing. It just returns some value
// unlikely to be in memory by chance for testing purposes.
// TODO: remove this method.
@@ -35,13 +35,13 @@ static const JNINativeMethod sMethods[] = {
// TODO: remove this entry when we have one useful method in here
const_cast<char *>("doNothing"),
const_cast<char *>("()I"),
- reinterpret_cast<void *>(latinime_BinaryDictInputOutput_doNothing)
+ reinterpret_cast<void *>(latinime_BinaryDictDecoder_doNothing)
},
};
-int register_BinaryDictInputOutput(JNIEnv *env) {
+int register_BinaryDictDecoder(JNIEnv *env) {
const char *const kClassPathName =
- "com/android/inputmethod/latin/makedict/BinaryDictInputOutput";
+ "com/android/inputmethod/latin/makedict/BinaryDictDecoder";
return registerNativeMethods(env, kClassPathName, sMethods, NELEMS(sMethods));
}
} // namespace latinime
diff --git a/native/jni/com_android_inputmethod_latin_makedict_BinaryDictInputOutput.h b/native/jni/com_android_inputmethod_latin_makedict_BinaryDictDecoder.h
index e622ed4ba..7f3cb67e6 100644
--- a/native/jni/com_android_inputmethod_latin_makedict_BinaryDictInputOutput.h
+++ b/native/jni/com_android_inputmethod_latin_makedict_BinaryDictDecoder.h
@@ -14,12 +14,12 @@
* limitations under the License.
*/
-#ifndef _COM_ANDROID_INPUTMETHOD_LATIN_MAKEDICT_BINARYDICTINPUTOUTPUT_H
-#define _COM_ANDROID_INPUTMETHOD_LATIN_MAKEDICT_BINARYDICTINPUTOUTPUT_H
+#ifndef _COM_ANDROID_INPUTMETHOD_LATIN_MAKEDICT_BINARYDICTDECODER_H
+#define _COM_ANDROID_INPUTMETHOD_LATIN_MAKEDICT_BINARYDICTDECODER_H
#include "jni.h"
namespace latinime {
-int register_BinaryDictInputOutput(JNIEnv *env);
+int register_BinaryDictDecoder(JNIEnv *env);
} // namespace latinime
-#endif // _COM_ANDROID_INPUTMETHOD_LATIN_MAKEDICT_BINARYDICTINPUTOUTPUT_H
+#endif // _COM_ANDROID_INPUTMETHOD_LATIN_MAKEDICT_BINARYDICTDECODER_H
diff --git a/native/jni/jni_common.cpp b/native/jni/jni_common.cpp
index 733e15f73..d44be6705 100644
--- a/native/jni/jni_common.cpp
+++ b/native/jni/jni_common.cpp
@@ -23,7 +23,7 @@
#include "com_android_inputmethod_latin_BinaryDictionary.h"
#include "com_android_inputmethod_latin_DicTraverseSession.h"
#endif
-#include "com_android_inputmethod_latin_makedict_BinaryDictInputOutput.h"
+#include "com_android_inputmethod_latin_makedict_BinaryDictDecoder.h"
#include "defines.h"
/*
@@ -55,8 +55,8 @@ jint JNI_OnLoad(JavaVM *vm, void *reserved) {
return -1;
}
#endif
- if (!latinime::register_BinaryDictInputOutput(env)) {
- AKLOGE("ERROR: BinaryDictInputOutput native registration failed");
+ if (!latinime::register_BinaryDictDecoder(env)) {
+ AKLOGE("ERROR: BinaryDictDecoder native registration failed");
return -1;
}
/* success -- return valid version number */
diff --git a/native/jni/src/suggest/core/dictionary/probability_utils.h b/native/jni/src/suggest/core/dictionary/probability_utils.h
index f450087d8..219213574 100644
--- a/native/jni/src/suggest/core/dictionary/probability_utils.h
+++ b/native/jni/src/suggest/core/dictionary/probability_utils.h
@@ -41,7 +41,7 @@ class ProbabilityUtils {
// the unigram probability to be the median value of the 17th step from the top. A value of
// 0 for the bigram probability represents the middle of the 16th step from the top,
// while a value of 15 represents the middle of the top step.
- // See makedict.BinaryDictInputOutput for details.
+ // See makedict.BinaryDictDecoder for details.
const float stepSize = static_cast<float>(MAX_PROBABILITY - unigramProbability)
/ (1.5f + MAX_BIGRAM_ENCODED_PROBABILITY);
return unigramProbability
diff --git a/native/jni/src/suggest/policyimpl/dictionary/binary_format.h b/native/jni/src/suggest/policyimpl/dictionary/binary_format.h
deleted file mode 100644
index 23f4c7fec..000000000
--- a/native/jni/src/suggest/policyimpl/dictionary/binary_format.h
+++ /dev/null
@@ -1,470 +0,0 @@
-/*
- * Copyright (C) 2011 The Android Open Source Project
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#ifndef LATINIME_BINARY_FORMAT_H
-#define LATINIME_BINARY_FORMAT_H
-
-#include <stdint.h>
-
-#include "suggest/core/dictionary/probability_utils.h"
-#include "utils/char_utils.h"
-
-namespace latinime {
-
-class BinaryFormat {
- public:
- // Mask and flags for children address type selection.
- static const int MASK_GROUP_ADDRESS_TYPE = 0xC0;
-
- // Flag for single/multiple char group
- static const int FLAG_HAS_MULTIPLE_CHARS = 0x20;
-
- // Flag for terminal groups
- static const int FLAG_IS_TERMINAL = 0x10;
-
- // Flag for shortcut targets presence
- static const int FLAG_HAS_SHORTCUT_TARGETS = 0x08;
- // Flag for bigram presence
- static const int FLAG_HAS_BIGRAMS = 0x04;
- // Flag for non-words (typically, shortcut only entries)
- static const int FLAG_IS_NOT_A_WORD = 0x02;
- // Flag for blacklist
- static const int FLAG_IS_BLACKLISTED = 0x01;
-
- // Attribute (bigram/shortcut) related flags:
- // Flag for presence of more attributes
- static const int FLAG_ATTRIBUTE_HAS_NEXT = 0x80;
- // Flag for sign of offset. If this flag is set, the offset value must be negated.
- static const int FLAG_ATTRIBUTE_OFFSET_NEGATIVE = 0x40;
-
- // Mask for attribute probability, stored on 4 bits inside the flags byte.
- static const int MASK_ATTRIBUTE_PROBABILITY = 0x0F;
-
- // Mask and flags for attribute address type selection.
- static const int MASK_ATTRIBUTE_ADDRESS_TYPE = 0x30;
-
- static int getGroupCountAndForwardPointer(const uint8_t *const dict, int *pos);
- static uint8_t getFlagsAndForwardPointer(const uint8_t *const dict, int *pos);
- static int getCodePointAndForwardPointer(const uint8_t *const dict, int *pos);
- static int readProbabilityWithoutMovingPointer(const uint8_t *const dict, const int pos);
- static int skipOtherCharacters(const uint8_t *const dict, const int pos);
- static int skipChildrenPosition(const uint8_t flags, const int pos);
- static int skipProbability(const uint8_t flags, const int pos);
- static int skipShortcuts(const uint8_t *const dict, const uint8_t flags, const int pos);
- static int skipChildrenPosAndAttributes(const uint8_t *const dict, const uint8_t flags,
- const int pos);
- static int readChildrenPosition(const uint8_t *const dict, const uint8_t flags, const int pos);
- static bool hasChildrenInFlags(const uint8_t flags);
- static int getTerminalPosition(const uint8_t *const root, const int *const inWord,
- const int length, const bool forceLowerCaseSearch);
- static int getCodePointsAndProbabilityAndReturnCodePointCount(
- const uint8_t *const root, const int nodePos, const int maxCodePointCount,
- int *const outCodePoints, int *const outUnigramProbability);
-
- private:
- DISALLOW_IMPLICIT_CONSTRUCTORS(BinaryFormat);
-
- static const int FLAG_GROUP_ADDRESS_TYPE_NOADDRESS = 0x00;
- static const int FLAG_GROUP_ADDRESS_TYPE_ONEBYTE = 0x40;
- static const int FLAG_GROUP_ADDRESS_TYPE_TWOBYTES = 0x80;
- static const int FLAG_GROUP_ADDRESS_TYPE_THREEBYTES = 0xC0;
- static const int FLAG_ATTRIBUTE_ADDRESS_TYPE_ONEBYTE = 0x10;
- static const int FLAG_ATTRIBUTE_ADDRESS_TYPE_TWOBYTES = 0x20;
- static const int FLAG_ATTRIBUTE_ADDRESS_TYPE_THREEBYTES = 0x30;
-
- static const int CHARACTER_ARRAY_TERMINATOR_SIZE = 1;
- static const int MINIMAL_ONE_BYTE_CHARACTER_VALUE = 0x20;
- static const int CHARACTER_ARRAY_TERMINATOR = 0x1F;
- static const int MULTIPLE_BYTE_CHARACTER_ADDITIONAL_SIZE = 2;
- static const int NO_FLAGS = 0;
- static int skipAllAttributes(const uint8_t *const dict, const uint8_t flags, const int pos);
- static int skipBigrams(const uint8_t *const dict, const uint8_t flags, const int pos);
-};
-
-AK_FORCE_INLINE int BinaryFormat::getGroupCountAndForwardPointer(const uint8_t *const dict,
- int *pos) {
- const int msb = dict[(*pos)++];
- if (msb < 0x80) return msb;
- return ((msb & 0x7F) << 8) | dict[(*pos)++];
-}
-
-inline uint8_t BinaryFormat::getFlagsAndForwardPointer(const uint8_t *const dict, int *pos) {
- return dict[(*pos)++];
-}
-
-AK_FORCE_INLINE int BinaryFormat::getCodePointAndForwardPointer(const uint8_t *const dict,
- int *pos) {
- const int origin = *pos;
- const int codePoint = dict[origin];
- if (codePoint < MINIMAL_ONE_BYTE_CHARACTER_VALUE) {
- if (codePoint == CHARACTER_ARRAY_TERMINATOR) {
- *pos = origin + 1;
- return NOT_A_CODE_POINT;
- } else {
- *pos = origin + 3;
- const int char_1 = codePoint << 16;
- const int char_2 = char_1 + (dict[origin + 1] << 8);
- return char_2 + dict[origin + 2];
- }
- } else {
- *pos = origin + 1;
- return codePoint;
- }
-}
-
-inline int BinaryFormat::readProbabilityWithoutMovingPointer(const uint8_t *const dict,
- const int pos) {
- return dict[pos];
-}
-
-AK_FORCE_INLINE int BinaryFormat::skipOtherCharacters(const uint8_t *const dict, const int pos) {
- int currentPos = pos;
- int character = dict[currentPos++];
- while (CHARACTER_ARRAY_TERMINATOR != character) {
- if (character < MINIMAL_ONE_BYTE_CHARACTER_VALUE) {
- currentPos += MULTIPLE_BYTE_CHARACTER_ADDITIONAL_SIZE;
- }
- character = dict[currentPos++];
- }
- return currentPos;
-}
-
-static inline int attributeAddressSize(const uint8_t flags) {
- static const int ATTRIBUTE_ADDRESS_SHIFT = 4;
- return (flags & BinaryFormat::MASK_ATTRIBUTE_ADDRESS_TYPE) >> ATTRIBUTE_ADDRESS_SHIFT;
- /* Note: this is a value-dependant optimization of what may probably be
- more readably written this way:
- switch (flags * BinaryFormat::MASK_ATTRIBUTE_ADDRESS_TYPE) {
- case FLAG_ATTRIBUTE_ADDRESS_TYPE_ONEBYTE: return 1;
- case FLAG_ATTRIBUTE_ADDRESS_TYPE_TWOBYTES: return 2;
- case FLAG_ATTRIBUTE_ADDRESS_TYPE_THREEBYTE: return 3;
- default: return 0;
- }
- */
-}
-
-static AK_FORCE_INLINE int skipExistingBigrams(const uint8_t *const dict, const int pos) {
- int currentPos = pos;
- uint8_t flags = BinaryFormat::getFlagsAndForwardPointer(dict, &currentPos);
- while (flags & BinaryFormat::FLAG_ATTRIBUTE_HAS_NEXT) {
- currentPos += attributeAddressSize(flags);
- flags = BinaryFormat::getFlagsAndForwardPointer(dict, &currentPos);
- }
- currentPos += attributeAddressSize(flags);
- return currentPos;
-}
-
-static inline int childrenAddressSize(const uint8_t flags) {
- static const int CHILDREN_ADDRESS_SHIFT = 6;
- return (BinaryFormat::MASK_GROUP_ADDRESS_TYPE & flags) >> CHILDREN_ADDRESS_SHIFT;
- /* See the note in attributeAddressSize. The same applies here */
-}
-
-static AK_FORCE_INLINE int shortcutByteSize(const uint8_t *const dict, const int pos) {
- return (static_cast<int>(dict[pos] << 8)) + (dict[pos + 1]);
-}
-
-inline int BinaryFormat::skipChildrenPosition(const uint8_t flags, const int pos) {
- return pos + childrenAddressSize(flags);
-}
-
-inline int BinaryFormat::skipProbability(const uint8_t flags, const int pos) {
- return FLAG_IS_TERMINAL & flags ? pos + 1 : pos;
-}
-
-AK_FORCE_INLINE int BinaryFormat::skipShortcuts(const uint8_t *const dict, const uint8_t flags,
- const int pos) {
- if (FLAG_HAS_SHORTCUT_TARGETS & flags) {
- return pos + shortcutByteSize(dict, pos);
- } else {
- return pos;
- }
-}
-
-AK_FORCE_INLINE int BinaryFormat::skipBigrams(const uint8_t *const dict, const uint8_t flags,
- const int pos) {
- if (FLAG_HAS_BIGRAMS & flags) {
- return skipExistingBigrams(dict, pos);
- } else {
- return pos;
- }
-}
-
-AK_FORCE_INLINE int BinaryFormat::skipAllAttributes(const uint8_t *const dict, const uint8_t flags,
- const int pos) {
- // This function skips all attributes: shortcuts and bigrams.
- int newPos = pos;
- newPos = skipShortcuts(dict, flags, newPos);
- newPos = skipBigrams(dict, flags, newPos);
- return newPos;
-}
-
-AK_FORCE_INLINE int BinaryFormat::skipChildrenPosAndAttributes(const uint8_t *const dict,
- const uint8_t flags, const int pos) {
- int currentPos = pos;
- currentPos = skipChildrenPosition(flags, currentPos);
- currentPos = skipAllAttributes(dict, flags, currentPos);
- return currentPos;
-}
-
-AK_FORCE_INLINE int BinaryFormat::readChildrenPosition(const uint8_t *const dict,
- const uint8_t flags, const int pos) {
- int offset = 0;
- switch (MASK_GROUP_ADDRESS_TYPE & flags) {
- case FLAG_GROUP_ADDRESS_TYPE_ONEBYTE:
- offset = dict[pos];
- break;
- case FLAG_GROUP_ADDRESS_TYPE_TWOBYTES:
- offset = dict[pos] << 8;
- offset += dict[pos + 1];
- break;
- case FLAG_GROUP_ADDRESS_TYPE_THREEBYTES:
- offset = dict[pos] << 16;
- offset += dict[pos + 1] << 8;
- offset += dict[pos + 2];
- break;
- default:
- // If we come here, it means we asked for the children of a word with
- // no children.
- return -1;
- }
- return pos + offset;
-}
-
-inline bool BinaryFormat::hasChildrenInFlags(const uint8_t flags) {
- return (FLAG_GROUP_ADDRESS_TYPE_NOADDRESS != (MASK_GROUP_ADDRESS_TYPE & flags));
-}
-
-// This function gets the byte position of the last chargroup of the exact matching word in the
-// dictionary. If no match is found, it returns NOT_A_VALID_WORD_POS.
-AK_FORCE_INLINE int BinaryFormat::getTerminalPosition(const uint8_t *const root,
- const int *const inWord, const int length, const bool forceLowerCaseSearch) {
- int pos = 0;
- int wordPos = 0;
-
- while (true) {
- // If we already traversed the tree further than the word is long, there means
- // there was no match (or we would have found it).
- if (wordPos >= length) return NOT_A_VALID_WORD_POS;
- int charGroupCount = BinaryFormat::getGroupCountAndForwardPointer(root, &pos);
- const int wChar = forceLowerCaseSearch
- ? CharUtils::toLowerCase(inWord[wordPos]) : inWord[wordPos];
- while (true) {
- // If there are no more character groups in this node, it means we could not
- // find a matching character for this depth, therefore there is no match.
- if (0 >= charGroupCount) return NOT_A_VALID_WORD_POS;
- const int charGroupPos = pos;
- const uint8_t flags = BinaryFormat::getFlagsAndForwardPointer(root, &pos);
- int character = BinaryFormat::getCodePointAndForwardPointer(root, &pos);
- if (character == wChar) {
- // This is the correct node. Only one character group may start with the same
- // char within a node, so either we found our match in this node, or there is
- // no match and we can return NOT_A_VALID_WORD_POS. So we will check all the
- // characters in this character group indeed does match.
- if (FLAG_HAS_MULTIPLE_CHARS & flags) {
- character = BinaryFormat::getCodePointAndForwardPointer(root, &pos);
- while (NOT_A_CODE_POINT != character) {
- ++wordPos;
- // If we shoot the length of the word we search for, or if we find a single
- // character that does not match, as explained above, it means the word is
- // not in the dictionary (by virtue of this chargroup being the only one to
- // match the word on the first character, but not matching the whole word).
- if (wordPos >= length) return NOT_A_VALID_WORD_POS;
- if (inWord[wordPos] != character) return NOT_A_VALID_WORD_POS;
- character = BinaryFormat::getCodePointAndForwardPointer(root, &pos);
- }
- }
- // If we come here we know that so far, we do match. Either we are on a terminal
- // and we match the length, in which case we found it, or we traverse children.
- // If we don't match the length AND don't have children, then a word in the
- // dictionary fully matches a prefix of the searched word but not the full word.
- ++wordPos;
- if (FLAG_IS_TERMINAL & flags) {
- if (wordPos == length) {
- return charGroupPos;
- }
- pos = BinaryFormat::skipProbability(FLAG_IS_TERMINAL, pos);
- }
- if (FLAG_GROUP_ADDRESS_TYPE_NOADDRESS == (MASK_GROUP_ADDRESS_TYPE & flags)) {
- return NOT_A_VALID_WORD_POS;
- }
- // We have children and we are still shorter than the word we are searching for, so
- // we need to traverse children. Put the pointer on the children position, and
- // break
- pos = BinaryFormat::readChildrenPosition(root, flags, pos);
- break;
- } else {
- // This chargroup does not match, so skip the remaining part and go to the next.
- if (FLAG_HAS_MULTIPLE_CHARS & flags) {
- pos = BinaryFormat::skipOtherCharacters(root, pos);
- }
- pos = BinaryFormat::skipProbability(flags, pos);
- pos = BinaryFormat::skipChildrenPosAndAttributes(root, flags, pos);
- }
- --charGroupCount;
- }
- }
-}
-
-// This function searches for a terminal in the dictionary by its address.
-// Due to the fact that words are ordered in the dictionary in a strict breadth-first order,
-// it is possible to check for this with advantageous complexity. For each node, we search
-// for groups with children and compare the children address with the address we look for.
-// When we shoot the address we look for, it means the word we look for is in the children
-// of the previous group. The only tricky part is the fact that if we arrive at the end of a
-// node with the last group's children address still less than what we are searching for, we
-// must descend the last group's children (for example, if the word we are searching for starts
-// with a z, it's the last group of the root node, so all children addresses will be smaller
-// than the address we look for, and we have to descend the z node).
-/* Parameters :
- * root: the dictionary buffer
- * address: the byte position of the last chargroup of the word we are searching for (this is
- * what is stored as the "bigram address" in each bigram)
- * outword: an array to write the found word, with MAX_WORD_LENGTH size.
- * outUnigramProbability: a pointer to an int to write the probability into.
- * Return value : the length of the word, of 0 if the word was not found.
- */
-AK_FORCE_INLINE int BinaryFormat::getCodePointsAndProbabilityAndReturnCodePointCount(
- const uint8_t *const root, const int nodePos, const int maxCodePointCount,
- int *const outCodePoints, int *const outUnigramProbability) {
- int pos = 0;
- int wordPos = 0;
-
- // One iteration of the outer loop iterates through nodes. As stated above, we will only
- // traverse nodes that are actually a part of the terminal we are searching, so each time
- // we enter this loop we are one depth level further than last time.
- // The only reason we count nodes is because we want to reduce the probability of infinite
- // looping in case there is a bug. Since we know there is an upper bound to the depth we are
- // supposed to traverse, it does not hurt to count iterations.
- for (int loopCount = maxCodePointCount; loopCount > 0; --loopCount) {
- int lastCandidateGroupPos = 0;
- // Let's loop through char groups in this node searching for either the terminal
- // or one of its ascendants.
- for (int charGroupCount = getGroupCountAndForwardPointer(root, &pos); charGroupCount > 0;
- --charGroupCount) {
- const int startPos = pos;
- const uint8_t flags = getFlagsAndForwardPointer(root, &pos);
- const int character = getCodePointAndForwardPointer(root, &pos);
- if (nodePos == startPos) {
- // We found the address. Copy the rest of the word in the buffer and return
- // the length.
- outCodePoints[wordPos] = character;
- if (FLAG_HAS_MULTIPLE_CHARS & flags) {
- int nextChar = getCodePointAndForwardPointer(root, &pos);
- // We count chars in order to avoid infinite loops if the file is broken or
- // if there is some other bug
- int charCount = maxCodePointCount;
- while (NOT_A_CODE_POINT != nextChar && --charCount > 0) {
- outCodePoints[++wordPos] = nextChar;
- nextChar = getCodePointAndForwardPointer(root, &pos);
- }
- }
- *outUnigramProbability = readProbabilityWithoutMovingPointer(root, pos);
- return ++wordPos;
- }
- // We need to skip past this char group, so skip any remaining chars after the
- // first and possibly the probability.
- if (FLAG_HAS_MULTIPLE_CHARS & flags) {
- pos = skipOtherCharacters(root, pos);
- }
- pos = skipProbability(flags, pos);
-
- // The fact that this group has children is very important. Since we already know
- // that this group does not match, if it has no children we know it is irrelevant
- // to what we are searching for.
- const bool hasChildren = (FLAG_GROUP_ADDRESS_TYPE_NOADDRESS !=
- (MASK_GROUP_ADDRESS_TYPE & flags));
- // We will write in `found' whether we have passed the children address we are
- // searching for. For example if we search for "beer", the children of b are less
- // than the address we are searching for and the children of c are greater. When we
- // come here for c, we realize this is too big, and that we should descend b.
- bool found;
- if (hasChildren) {
- // Here comes the tricky part. First, read the children position.
- const int childrenPos = readChildrenPosition(root, flags, pos);
- if (childrenPos > nodePos) {
- // If the children pos is greater than address, it means the previous chargroup,
- // which address is stored in lastCandidateGroupPos, was the right one.
- found = true;
- } else if (1 >= charGroupCount) {
- // However if we are on the LAST group of this node, and we have NOT shot the
- // address we should descend THIS node. So we trick the lastCandidateGroupPos
- // so that we will descend this node, not the previous one.
- lastCandidateGroupPos = startPos;
- found = true;
- } else {
- // Else, we should continue looking.
- found = false;
- }
- } else {
- // Even if we don't have children here, we could still be on the last group of this
- // node. If this is the case, we should descend the last group that had children,
- // and their address is already in lastCandidateGroup.
- found = (1 >= charGroupCount);
- }
-
- if (found) {
- // Okay, we found the group we should descend. Its address is in
- // the lastCandidateGroupPos variable, so we just re-read it.
- if (0 != lastCandidateGroupPos) {
- const uint8_t lastFlags =
- getFlagsAndForwardPointer(root, &lastCandidateGroupPos);
- const int lastChar =
- getCodePointAndForwardPointer(root, &lastCandidateGroupPos);
- // We copy all the characters in this group to the buffer
- outCodePoints[wordPos] = lastChar;
- if (FLAG_HAS_MULTIPLE_CHARS & lastFlags) {
- int nextChar = getCodePointAndForwardPointer(root, &lastCandidateGroupPos);
- int charCount = maxCodePointCount;
- while (-1 != nextChar && --charCount > 0) {
- outCodePoints[++wordPos] = nextChar;
- nextChar = getCodePointAndForwardPointer(root, &lastCandidateGroupPos);
- }
- }
- ++wordPos;
- // Now we only need to branch to the children address. Skip the probability if
- // it's there, read pos, and break to resume the search at pos.
- lastCandidateGroupPos = skipProbability(lastFlags, lastCandidateGroupPos);
- pos = readChildrenPosition(root, lastFlags, lastCandidateGroupPos);
- break;
- } else {
- // Here is a little tricky part: we come here if we found out that all children
- // addresses in this group are bigger than the address we are searching for.
- // Should we conclude the word is not in the dictionary? No! It could still be
- // one of the remaining chargroups in this node, so we have to keep looking in
- // this node until we find it (or we realize it's not there either, in which
- // case it's actually not in the dictionary). Pass the end of this group, ready
- // to start the next one.
- pos = skipChildrenPosAndAttributes(root, flags, pos);
- }
- } else {
- // If we did not find it, we should record the last children address for the next
- // iteration.
- if (hasChildren) lastCandidateGroupPos = startPos;
- // Now skip the end of this group (children pos and the attributes if any) so that
- // our pos is after the end of this char group, at the start of the next one.
- pos = skipChildrenPosAndAttributes(root, flags, pos);
- }
-
- }
- }
- // If we have looked through all the chargroups and found no match, the address is
- // not the address of a terminal in this dictionary.
- return 0;
-}
-
-} // namespace latinime
-#endif // LATINIME_BINARY_FORMAT_H
diff --git a/native/jni/src/suggest/policyimpl/dictionary/patricia_trie_policy.cpp b/native/jni/src/suggest/policyimpl/dictionary/patricia_trie_policy.cpp
index 3e664a29b..15eb0674d 100644
--- a/native/jni/src/suggest/policyimpl/dictionary/patricia_trie_policy.cpp
+++ b/native/jni/src/suggest/policyimpl/dictionary/patricia_trie_policy.cpp
@@ -20,7 +20,6 @@
#include "defines.h"
#include "suggest/core/dicnode/dic_node.h"
#include "suggest/core/dicnode/dic_node_vector.h"
-#include "suggest/policyimpl/dictionary/binary_format.h"
#include "suggest/policyimpl/dictionary/patricia_trie_reading_utils.h"
namespace latinime {
@@ -38,17 +37,273 @@ void PatriciaTriePolicy::createAndGetAllChildNodes(const DicNode *const dicNode,
}
}
+// This retrieves code points and the probability of the word by its terminal position.
+// Due to the fact that words are ordered in the dictionary in a strict breadth-first order,
+// it is possible to check for this with advantageous complexity. For each node, we search
+// for groups with children and compare the children position with the position we look for.
+// When we shoot the position we look for, it means the word we look for is in the children
+// of the previous group. The only tricky part is the fact that if we arrive at the end of a
+// node with the last group's children position still less than what we are searching for, we
+// must descend the last group's children (for example, if the word we are searching for starts
+// with a z, it's the last group of the root node, so all children addresses will be smaller
+// than the position we look for, and we have to descend the z node).
+/* Parameters :
+ * nodePos: the byte position of the terminal chargroup of the word we are searching for (this is
+ * what is stored as the "bigram position" in each bigram)
+ * outCodePoints: an array to write the found word, with MAX_WORD_LENGTH size.
+ * outUnigramProbability: a pointer to an int to write the probability into.
+ * Return value : the code point count, of 0 if the word was not found.
+ */
+// TODO: Split this function to be more readable
int PatriciaTriePolicy::getCodePointsAndProbabilityAndReturnCodePointCount(
const int nodePos, const int maxCodePointCount, int *const outCodePoints,
int *const outUnigramProbability) const {
- return BinaryFormat::getCodePointsAndProbabilityAndReturnCodePointCount(mDictRoot, nodePos,
- maxCodePointCount, outCodePoints, outUnigramProbability);
+ int pos = getRootPosition();
+ int wordPos = 0;
+ // One iteration of the outer loop iterates through nodes. As stated above, we will only
+ // traverse nodes that are actually a part of the terminal we are searching, so each time
+ // we enter this loop we are one depth level further than last time.
+ // The only reason we count nodes is because we want to reduce the probability of infinite
+ // looping in case there is a bug. Since we know there is an upper bound to the depth we are
+ // supposed to traverse, it does not hurt to count iterations.
+ for (int loopCount = maxCodePointCount; loopCount > 0; --loopCount) {
+ int lastCandidateGroupPos = 0;
+ // Let's loop through char groups in this node searching for either the terminal
+ // or one of its ascendants.
+ for (int charGroupCount = PatriciaTrieReadingUtils::getGroupCountAndAdvancePosition(
+ mDictRoot, &pos); charGroupCount > 0; --charGroupCount) {
+ const int startPos = pos;
+ const PatriciaTrieReadingUtils::NodeFlags flags =
+ PatriciaTrieReadingUtils::getFlagsAndAdvancePosition(mDictRoot, &pos);
+ const int character = PatriciaTrieReadingUtils::getCodePointAndAdvancePosition(
+ mDictRoot, &pos);
+ if (nodePos == startPos) {
+ // We found the position. Copy the rest of the code points in the buffer and return
+ // the length.
+ outCodePoints[wordPos] = character;
+ if (PatriciaTrieReadingUtils::hasMultipleChars(flags)) {
+ int nextChar = PatriciaTrieReadingUtils::getCodePointAndAdvancePosition(
+ mDictRoot, &pos);
+ // We count code points in order to avoid infinite loops if the file is broken
+ // or if there is some other bug
+ int charCount = maxCodePointCount;
+ while (NOT_A_CODE_POINT != nextChar && --charCount > 0) {
+ outCodePoints[++wordPos] = nextChar;
+ nextChar = PatriciaTrieReadingUtils::getCodePointAndAdvancePosition(
+ mDictRoot, &pos);
+ }
+ }
+ *outUnigramProbability =
+ PatriciaTrieReadingUtils::readProbabilityAndAdvancePosition(mDictRoot,
+ &pos);
+ return ++wordPos;
+ }
+ // We need to skip past this char group, so skip any remaining code points after the
+ // first and possibly the probability.
+ if (PatriciaTrieReadingUtils::hasMultipleChars(flags)) {
+ PatriciaTrieReadingUtils::skipCharacters(mDictRoot, flags, MAX_WORD_LENGTH, &pos);
+ }
+ if (PatriciaTrieReadingUtils::isTerminal(flags)) {
+ PatriciaTrieReadingUtils::readProbabilityAndAdvancePosition(mDictRoot, &pos);
+ }
+ // The fact that this group has children is very important. Since we already know
+ // that this group does not match, if it has no children we know it is irrelevant
+ // to what we are searching for.
+ const bool hasChildren = PatriciaTrieReadingUtils::hasChildrenInFlags(flags);
+ // We will write in `found' whether we have passed the children position we are
+ // searching for. For example if we search for "beer", the children of b are less
+ // than the address we are searching for and the children of c are greater. When we
+ // come here for c, we realize this is too big, and that we should descend b.
+ bool found;
+ if (hasChildren) {
+ int currentPos = pos;
+ // Here comes the tricky part. First, read the children position.
+ const int childrenPos = PatriciaTrieReadingUtils
+ ::readChildrenPositionAndAdvancePosition(mDictRoot, flags, &currentPos);
+ if (childrenPos > nodePos) {
+ // If the children pos is greater than the position, it means the previous
+ // chargroup, which position is stored in lastCandidateGroupPos, was the right
+ // one.
+ found = true;
+ } else if (1 >= charGroupCount) {
+ // However if we are on the LAST group of this node, and we have NOT shot the
+ // position we should descend THIS node. So we trick the lastCandidateGroupPos
+ // so that we will descend this node, not the previous one.
+ lastCandidateGroupPos = startPos;
+ found = true;
+ } else {
+ // Else, we should continue looking.
+ found = false;
+ }
+ } else {
+ // Even if we don't have children here, we could still be on the last group of this
+ // node. If this is the case, we should descend the last group that had children,
+ // and their position is already in lastCandidateGroup.
+ found = (1 >= charGroupCount);
+ }
+
+ if (found) {
+ // Okay, we found the group we should descend. Its position is in
+ // the lastCandidateGroupPos variable, so we just re-read it.
+ if (0 != lastCandidateGroupPos) {
+ const PatriciaTrieReadingUtils::NodeFlags lastFlags =
+ PatriciaTrieReadingUtils::getFlagsAndAdvancePosition(
+ mDictRoot, &lastCandidateGroupPos);
+ const int lastChar = PatriciaTrieReadingUtils::getCodePointAndAdvancePosition(
+ mDictRoot, &lastCandidateGroupPos);
+ // We copy all the characters in this group to the buffer
+ outCodePoints[wordPos] = lastChar;
+ if (PatriciaTrieReadingUtils::hasMultipleChars(lastFlags)) {
+ int nextChar = PatriciaTrieReadingUtils::getCodePointAndAdvancePosition(
+ mDictRoot, &lastCandidateGroupPos);
+ int charCount = maxCodePointCount;
+ while (-1 != nextChar && --charCount > 0) {
+ outCodePoints[++wordPos] = nextChar;
+ nextChar = PatriciaTrieReadingUtils::getCodePointAndAdvancePosition(
+ mDictRoot, &lastCandidateGroupPos);
+ }
+ }
+ ++wordPos;
+ // Now we only need to branch to the children address. Skip the probability if
+ // it's there, read pos, and break to resume the search at pos.
+ if (PatriciaTrieReadingUtils::isTerminal(lastFlags)) {
+ PatriciaTrieReadingUtils::readProbabilityAndAdvancePosition(mDictRoot,
+ &lastCandidateGroupPos);
+ }
+ pos = PatriciaTrieReadingUtils::readChildrenPositionAndAdvancePosition(
+ mDictRoot, lastFlags, &lastCandidateGroupPos);
+ break;
+ } else {
+ // Here is a little tricky part: we come here if we found out that all children
+ // addresses in this group are bigger than the address we are searching for.
+ // Should we conclude the word is not in the dictionary? No! It could still be
+ // one of the remaining chargroups in this node, so we have to keep looking in
+ // this node until we find it (or we realize it's not there either, in which
+ // case it's actually not in the dictionary). Pass the end of this group, ready
+ // to start the next one.
+ if (PatriciaTrieReadingUtils::hasChildrenInFlags(flags)) {
+ PatriciaTrieReadingUtils::readChildrenPositionAndAdvancePosition(
+ mDictRoot, flags, &pos);
+ }
+ if (PatriciaTrieReadingUtils::hasShortcutTargets(flags)) {
+ mShortcutListPolicy.skipAllShortcuts(&pos);
+ }
+ if (PatriciaTrieReadingUtils::hasBigrams(flags)) {
+ mBigramListPolicy.skipAllBigrams(&pos);
+ }
+ }
+ } else {
+ // If we did not find it, we should record the last children address for the next
+ // iteration.
+ if (hasChildren) lastCandidateGroupPos = startPos;
+ // Now skip the end of this group (children pos and the attributes if any) so that
+ // our pos is after the end of this char group, at the start of the next one.
+ if (PatriciaTrieReadingUtils::hasChildrenInFlags(flags)) {
+ PatriciaTrieReadingUtils::readChildrenPositionAndAdvancePosition(
+ mDictRoot, flags, &pos);
+ }
+ if (PatriciaTrieReadingUtils::hasShortcutTargets(flags)) {
+ mShortcutListPolicy.skipAllShortcuts(&pos);
+ }
+ if (PatriciaTrieReadingUtils::hasBigrams(flags)) {
+ mBigramListPolicy.skipAllBigrams(&pos);
+ }
+ }
+
+ }
+ }
+ // If we have looked through all the chargroups and found no match, the nodePos is
+ // not the position of a terminal in this dictionary.
+ return 0;
}
+// This function gets the position of the terminal node of the exact matching word in the
+// dictionary. If no match is found, it returns NOT_A_VALID_WORD_POS.
int PatriciaTriePolicy::getTerminalNodePositionOfWord(const int *const inWord,
const int length, const bool forceLowerCaseSearch) const {
- return BinaryFormat::getTerminalPosition(mDictRoot, inWord,
- length, forceLowerCaseSearch);
+ int pos = getRootPosition();
+ int wordPos = 0;
+
+ while (true) {
+ // If we already traversed the tree further than the word is long, there means
+ // there was no match (or we would have found it).
+ if (wordPos >= length) return NOT_A_VALID_WORD_POS;
+ int charGroupCount = PatriciaTrieReadingUtils::getGroupCountAndAdvancePosition(mDictRoot,
+ &pos);
+ const int wChar = forceLowerCaseSearch
+ ? CharUtils::toLowerCase(inWord[wordPos]) : inWord[wordPos];
+ while (true) {
+ // If there are no more character groups in this node, it means we could not
+ // find a matching character for this depth, therefore there is no match.
+ if (0 >= charGroupCount) return NOT_A_VALID_WORD_POS;
+ const int charGroupPos = pos;
+ const PatriciaTrieReadingUtils::NodeFlags flags =
+ PatriciaTrieReadingUtils::getFlagsAndAdvancePosition(mDictRoot, &pos);
+ int character = PatriciaTrieReadingUtils::getCodePointAndAdvancePosition(mDictRoot,
+ &pos);
+ if (character == wChar) {
+ // This is the correct node. Only one character group may start with the same
+ // char within a node, so either we found our match in this node, or there is
+ // no match and we can return NOT_A_VALID_WORD_POS. So we will check all the
+ // characters in this character group indeed does match.
+ if (PatriciaTrieReadingUtils::hasMultipleChars(flags)) {
+ character = PatriciaTrieReadingUtils::getCodePointAndAdvancePosition(mDictRoot,
+ &pos);
+ while (NOT_A_CODE_POINT != character) {
+ ++wordPos;
+ // If we shoot the length of the word we search for, or if we find a single
+ // character that does not match, as explained above, it means the word is
+ // not in the dictionary (by virtue of this chargroup being the only one to
+ // match the word on the first character, but not matching the whole word).
+ if (wordPos >= length) return NOT_A_VALID_WORD_POS;
+ if (inWord[wordPos] != character) return NOT_A_VALID_WORD_POS;
+ character = PatriciaTrieReadingUtils::getCodePointAndAdvancePosition(
+ mDictRoot, &pos);
+ }
+ }
+ // If we come here we know that so far, we do match. Either we are on a terminal
+ // and we match the length, in which case we found it, or we traverse children.
+ // If we don't match the length AND don't have children, then a word in the
+ // dictionary fully matches a prefix of the searched word but not the full word.
+ ++wordPos;
+ if (PatriciaTrieReadingUtils::isTerminal(flags)) {
+ if (wordPos == length) {
+ return charGroupPos;
+ }
+ PatriciaTrieReadingUtils::readProbabilityAndAdvancePosition(mDictRoot, &pos);
+ }
+ if (!PatriciaTrieReadingUtils::hasChildrenInFlags(flags)) {
+ return NOT_A_VALID_WORD_POS;
+ }
+ // We have children and we are still shorter than the word we are searching for, so
+ // we need to traverse children. Put the pointer on the children position, and
+ // break
+ pos = PatriciaTrieReadingUtils::readChildrenPositionAndAdvancePosition(mDictRoot,
+ flags, &pos);
+ break;
+ } else {
+ // This chargroup does not match, so skip the remaining part and go to the next.
+ if (PatriciaTrieReadingUtils::hasMultipleChars(flags)) {
+ PatriciaTrieReadingUtils::skipCharacters(mDictRoot, flags, MAX_WORD_LENGTH,
+ &pos);
+ }
+ if (PatriciaTrieReadingUtils::isTerminal(flags)) {
+ PatriciaTrieReadingUtils::readProbabilityAndAdvancePosition(mDictRoot, &pos);
+ }
+ if (PatriciaTrieReadingUtils::hasChildrenInFlags(flags)) {
+ PatriciaTrieReadingUtils::readChildrenPositionAndAdvancePosition(mDictRoot,
+ flags, &pos);
+ }
+ if (PatriciaTrieReadingUtils::hasShortcutTargets(flags)) {
+ mShortcutListPolicy.skipAllShortcuts(&pos);
+ }
+ if (PatriciaTrieReadingUtils::hasBigrams(flags)) {
+ mBigramListPolicy.skipAllBigrams(&pos);
+ }
+ }
+ --charGroupCount;
+ }
+ }
}
int PatriciaTriePolicy::getUnigramProbability(const int nodePos) const {
diff --git a/native/jni/src/suggest/policyimpl/dictionary/utils/mmaped_buffer.h b/native/jni/src/suggest/policyimpl/dictionary/utils/mmaped_buffer.h
index e3aabb084..08add03c6 100644
--- a/native/jni/src/suggest/policyimpl/dictionary/utils/mmaped_buffer.h
+++ b/native/jni/src/suggest/policyimpl/dictionary/utils/mmaped_buffer.h
@@ -48,13 +48,13 @@ class MmapedBuffer {
close(fd);
return 0;
}
- uint8_t *const buffer = static_cast<uint8_t *>(mmapedBuffer) + bufOffset;
+ uint8_t *const buffer = static_cast<uint8_t *>(mmapedBuffer) + offset;
if (!buffer) {
AKLOGE("DICT: buffer is null");
close(fd);
return 0;
}
- return new MmapedBuffer(buffer, adjSize, fd, adjOffset, isUpdatable);
+ return new MmapedBuffer(buffer, adjSize, fd, offset, isUpdatable);
}
~MmapedBuffer() {
generated by cgit v1.2.3-83-g751a (git 2.47.1) at 2025-05-06 01:26:53 +0000