diff options
Diffstat (limited to 'java/src')
| -rw-r--r-- | java/src/com/android/inputmethod/keyboard/LatinKeyboardView.java | 43 | ||||
| -rw-r--r-- | java/src/com/android/inputmethod/latin/LatinIME.java | 6 | ||||
| -rw-r--r-- | java/src/com/android/inputmethod/latin/Utils.java | 86 | ||||
| -rw-r--r-- | java/src/com/android/inputmethod/latin/makedict/BinaryDictInputOutput.java | 1208 | ||||
| -rw-r--r-- | java/src/com/android/inputmethod/latin/makedict/CharGroupInfo.java | 48 | ||||
| -rw-r--r-- | java/src/com/android/inputmethod/latin/makedict/FusionDictionary.java | 671 | ||||
| -rw-r--r-- | java/src/com/android/inputmethod/latin/makedict/MakedictLog.java | 40 | ||||
| -rw-r--r-- | java/src/com/android/inputmethod/latin/makedict/PendingAttribute.java | 32 | ||||
| -rw-r--r-- | java/src/com/android/inputmethod/latin/makedict/UnsupportedFormatException.java (renamed from java/src/com/android/inputmethod/latin/makedict/Dummy.java) | 11 | ||||
| -rw-r--r-- | java/src/com/android/inputmethod/latin/makedict/Word.java | 72 |
10 files changed, 2160 insertions, 57 deletions
diff --git a/java/src/com/android/inputmethod/keyboard/LatinKeyboardView.java b/java/src/com/android/inputmethod/keyboard/LatinKeyboardView.java index 17d15b061..b66d1661d 100644 --- a/java/src/com/android/inputmethod/keyboard/LatinKeyboardView.java +++ b/java/src/com/android/inputmethod/keyboard/LatinKeyboardView.java @@ -66,9 +66,6 @@ public class LatinKeyboardView extends KeyboardView implements PointerTracker.Ke SuddenJumpingTouchEventHandler.ProcessMotionEvent { private static final String TAG = LatinKeyboardView.class.getSimpleName(); - // TODO: Kill process when the usability study mode was changed. - private static final boolean ENABLE_USABILITY_STUDY_LOG = LatinImeLogger.sUsabilityStudy; - /** Listener for {@link KeyboardActionListener}. */ private KeyboardActionListener mKeyboardActionListener; @@ -656,6 +653,8 @@ public class LatinKeyboardView extends KeyboardView implements PointerTracker.Ke final int index = me.getActionIndex(); final int id = me.getPointerId(index); final int x, y; + final float size = me.getSize(index); + final float pressure = me.getPressure(index); if (mMoreKeysPanel != null && id == mMoreKeysPanelPointerTrackerId) { x = mMoreKeysPanel.translateX((int)me.getX(index)); y = mMoreKeysPanel.translateY((int)me.getY(index)); @@ -663,32 +662,11 @@ public class LatinKeyboardView extends KeyboardView implements PointerTracker.Ke x = (int)me.getX(index); y = (int)me.getY(index); } - if (ENABLE_USABILITY_STUDY_LOG) { - final String eventTag; - switch (action) { - case MotionEvent.ACTION_UP: - eventTag = "[Up]"; - break; - case MotionEvent.ACTION_DOWN: - eventTag = "[Down]"; - break; - case MotionEvent.ACTION_POINTER_UP: - eventTag = "[PointerUp]"; - break; - case MotionEvent.ACTION_POINTER_DOWN: - eventTag = "[PointerDown]"; - break; - case MotionEvent.ACTION_MOVE: // Skip this as being logged below - eventTag = ""; - break; - default: - eventTag = "[Action" + action + "]"; - break; - } - if (!TextUtils.isEmpty(eventTag)) { - UsabilityStudyLogUtils.getInstance().write( - eventTag + eventTime + "," + id + "," + x + "," + y + "," - + me.getSize(index) + "," + me.getPressure(index)); + if (LatinImeLogger.sUsabilityStudy) { + if (action != MotionEvent.ACTION_MOVE) { + // Skip ACTION_MOVE events as they are logged below + UsabilityStudyLogUtils.getInstance().writeMotionEvent(action, eventTime, id, x, + y, size, pressure); } } @@ -748,10 +726,9 @@ public class LatinKeyboardView extends KeyboardView implements PointerTracker.Ke py = (int)me.getY(i); } tracker.onMoveEvent(px, py, eventTime); - if (ENABLE_USABILITY_STUDY_LOG) { - UsabilityStudyLogUtils.getInstance().write("[Move]" + eventTime + "," - + me.getPointerId(i) + "," + px + "," + py + "," - + me.getSize(i) + "," + me.getPressure(i)); + if (LatinImeLogger.sUsabilityStudy) { + UsabilityStudyLogUtils.getInstance().writeMotionEvent(action, eventTime, id, + px, py, size, pressure); } } } else { diff --git a/java/src/com/android/inputmethod/latin/LatinIME.java b/java/src/com/android/inputmethod/latin/LatinIME.java index e67f0ea05..48fb79809 100644 --- a/java/src/com/android/inputmethod/latin/LatinIME.java +++ b/java/src/com/android/inputmethod/latin/LatinIME.java @@ -68,6 +68,7 @@ import com.android.inputmethod.keyboard.KeyboardId; import com.android.inputmethod.keyboard.KeyboardSwitcher; import com.android.inputmethod.keyboard.KeyboardView; import com.android.inputmethod.keyboard.LatinKeyboardView; +import com.android.inputmethod.latin.Utils.UsabilityStudyLogUtils; import com.android.inputmethod.latin.suggestions.SuggestionsView; import java.io.FileDescriptor; @@ -1266,6 +1267,11 @@ public class LatinIME extends InputMethodServiceCompatWrapper implements Keyboar mDeleteCount = 0; } mLastKeyTime = when; + + if (LatinImeLogger.sUsabilityStudy) { + UsabilityStudyLogUtils.getInstance().writeKeyEvent(primaryCode, x, y); + } + final KeyboardSwitcher switcher = mKeyboardSwitcher; // The space state depends only on the last character pressed and its own previous // state. Here, we revert the space state to neutral if the key is actually modifying diff --git a/java/src/com/android/inputmethod/latin/Utils.java b/java/src/com/android/inputmethod/latin/Utils.java index 708634529..a3589da0a 100644 --- a/java/src/com/android/inputmethod/latin/Utils.java +++ b/java/src/com/android/inputmethod/latin/Utils.java @@ -31,7 +31,9 @@ import android.os.Process; import android.text.TextUtils; import android.text.format.DateUtils; import android.util.Log; +import android.view.MotionEvent; +import com.android.inputmethod.keyboard.Keyboard; import com.android.inputmethod.latin.SuggestedWords.SuggestedWordInfo; import java.io.BufferedReader; @@ -138,9 +140,6 @@ public class Utils { // TODO: accept code points public void push(char c, int x, int y) { if (!mEnabled) return; - if (mUsabilityStudy) { - UsabilityStudyLogUtils.getInstance().writeChar(c, x, y); - } mCharBuf[mEnd] = c; mXBuf[mEnd] = x; mYBuf[mEnd] = y; @@ -263,28 +262,73 @@ public class Utils { } } - public static void writeBackSpace(int x, int y) { - UsabilityStudyLogUtils.getInstance().write("<backspace>\t" + x + "\t" + y); + /** + * Represents a category of logging events that share the same subfield structure. + */ + public static enum LogGroup { + MOTION_EVENT("m"), + KEY("k"), + CORRECTION("c"), + STATE_CHANGE("s"); + + private final String mLogString; + + private LogGroup(String logString) { + mLogString = logString; + } } - public void writeChar(char c, int x, int y) { - String inputChar = String.valueOf(c); - switch (c) { - case '\n': - inputChar = "<enter>"; - break; - case '\t': - inputChar = "<tab>"; - break; - case ' ': - inputChar = "<space>"; - break; + public void writeMotionEvent(final int action, final long eventTime, final int id, + final int x, final int y, final float size, final float pressure) { + final String eventTag; + switch (action) { + case MotionEvent.ACTION_CANCEL: eventTag = "[Cancel]"; break; + case MotionEvent.ACTION_UP: eventTag = "[Up]"; break; + case MotionEvent.ACTION_DOWN: eventTag = "[Down]"; break; + case MotionEvent.ACTION_POINTER_UP: eventTag = "[PointerUp]"; break; + case MotionEvent.ACTION_POINTER_DOWN: eventTag = "[PointerDown]"; break; + case MotionEvent.ACTION_MOVE: eventTag = "[Move]"; break; + case MotionEvent.ACTION_OUTSIDE: eventTag = "[Outside]"; break; + default: eventTag = "[Action" + action + "]"; break; + } + if (!TextUtils.isEmpty(eventTag)) { + StringBuilder sb = new StringBuilder(); + sb.append(eventTag); + sb.append('\t'); sb.append(eventTime); + sb.append('\t'); sb.append(id); + sb.append('\t'); sb.append(x); + sb.append('\t'); sb.append(y); + sb.append('\t'); sb.append(size); + sb.append('\t'); sb.append(pressure); + write(LogGroup.MOTION_EVENT, sb.toString()); } - UsabilityStudyLogUtils.getInstance().write(inputChar + "\t" + x + "\t" + y); + } + + public void writeKeyEvent(int code, int x, int y) { + final StringBuilder sb = new StringBuilder(); + sb.append(Keyboard.printableCode(code)); + sb.append('\t'); sb.append(x); + sb.append('\t'); sb.append(y); + write(LogGroup.KEY, sb.toString()); + + // TODO: replace with a cleaner flush+retrieve mechanism LatinImeLogger.onPrintAllUsabilityStudyLogs(); } - public void write(final String log) { + public void writeCorrection(String subgroup, String before, String after, int position) { + final StringBuilder sb = new StringBuilder(); + sb.append(subgroup); + sb.append('\t'); sb.append(before); + sb.append('\t'); sb.append(after); + sb.append('\t'); sb.append(position); + write(LogGroup.CORRECTION, sb.toString()); + } + + public void writeStateChange(String subgroup, String details) { + write(LogGroup.STATE_CHANGE, subgroup + "\t" + details); + } + + private void write(final LogGroup logGroup, final String log) { mLoggingHandler.post(new Runnable() { @Override public void run() { @@ -292,8 +336,8 @@ public class Utils { final long currentTime = System.currentTimeMillis(); mDate.setTime(currentTime); - final String printString = String.format("%s\t%d\t%s\n", - mDateFormat.format(mDate), currentTime, log); + final String printString = String.format("%s\t%d\t%s\t%s\n", + mDateFormat.format(mDate), currentTime, logGroup.mLogString, log); if (LatinImeLogger.sDBG) { Log.d(USABILITY_TAG, "Write: " + log); } diff --git a/java/src/com/android/inputmethod/latin/makedict/BinaryDictInputOutput.java b/java/src/com/android/inputmethod/latin/makedict/BinaryDictInputOutput.java new file mode 100644 index 000000000..42dd4df34 --- /dev/null +++ b/java/src/com/android/inputmethod/latin/makedict/BinaryDictInputOutput.java @@ -0,0 +1,1208 @@ +/* + * 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. + */ + +package com.android.inputmethod.latin.makedict; + +import com.android.inputmethod.latin.makedict.FusionDictionary.CharGroup; +import com.android.inputmethod.latin.makedict.FusionDictionary.Node; +import com.android.inputmethod.latin.makedict.FusionDictionary.WeightedString; + +import java.io.FileNotFoundException; +import java.io.IOException; +import java.io.OutputStream; +import java.io.RandomAccessFile; +import java.util.ArrayList; +import java.util.Arrays; +import java.util.Iterator; +import java.util.Map; +import java.util.TreeMap; + +/** + * Reads and writes XML files for a FusionDictionary. + * + * All the methods in this class are static. + */ +public class BinaryDictInputOutput { + + /* Node layout is as follows: + * | addressType xx : mask with MASK_GROUP_ADDRESS_TYPE + * 2 bits, 00 = no children : FLAG_GROUP_ADDRESS_TYPE_NOADDRESS + * f | 01 = 1 byte : FLAG_GROUP_ADDRESS_TYPE_ONEBYTE + * l | 10 = 2 bytes : FLAG_GROUP_ADDRESS_TYPE_TWOBYTES + * a | 11 = 3 bytes : FLAG_GROUP_ADDRESS_TYPE_THREEBYTES + * g | has several chars ? 1 bit, 1 = yes, 0 = no : FLAG_HAS_MULTIPLE_CHARS + * s | has a terminal ? 1 bit, 1 = yes, 0 = no : FLAG_IS_TERMINAL + * | has shortcut targets ? 1 bit, 1 = yes, 0 = no : FLAG_HAS_SHORTCUT_TARGETS + * | has bigrams ? 1 bit, 1 = yes, 0 = no : FLAG_HAS_BIGRAMS + * | is shortcut only ? 1 bit, 1 = yes, 0 = no : FLAG_IS_SHORTCUT_ONLY + * + * c | IF FLAG_HAS_MULTIPLE_CHARS + * h | char, char, char, char n * (1 or 3 bytes) : use CharGroupInfo for i/o helpers + * a | end 1 byte, = 0 + * r | ELSE + * s | char 1 or 3 bytes + * | END + * + * f | + * r | IF FLAG_IS_TERMINAL + * e | frequency 1 byte + * q | + * + * c | IF 00 = FLAG_GROUP_ADDRESS_TYPE_NOADDRESS = addressType + * h | // nothing + * i | ELSIF 01 = FLAG_GROUP_ADDRESS_TYPE_ONEBYTE == addressType + * l | children address, 1 byte + * d | ELSIF 10 = FLAG_GROUP_ADDRESS_TYPE_TWOBYTES == addressType + * r | children address, 2 bytes + * e | ELSE // 11 = FLAG_GROUP_ADDRESS_TYPE_THREEBYTES = addressType + * n | children address, 3 bytes + * A | END + * d + * dress + * + * | IF FLAG_IS_TERMINAL && FLAG_HAS_SHORTCUT_TARGETS + * | shortcut targets address list + * | IF FLAG_IS_TERMINAL && FLAG_HAS_BIGRAMS + * | bigrams address list + * + * Char format is: + * 1 byte = bbbbbbbb match + * case 000xxxxx: xxxxx << 16 + next byte << 8 + next byte + * else: if 00011111 (= 0x1F) : this is the terminator. This is a relevant choice because + * unicode code points range from 0 to 0x10FFFF, so any 3-byte value starting with + * 00011111 would be outside unicode. + * else: iso-latin-1 code + * This allows for the whole unicode range to be encoded, including chars outside of + * the BMP. Also everything in the iso-latin-1 charset is only 1 byte, except control + * characters which should never happen anyway (and still work, but take 3 bytes). + * + * bigram and shortcut address list is: + * <flags> = | hasNext = 1 bit, 1 = yes, 0 = no : FLAG_ATTRIBUTE_HAS_NEXT + * | addressSign = 1 bit, : FLAG_ATTRIBUTE_OFFSET_NEGATIVE + * | 1 = must take -address, 0 = must take +address + * | xx : mask with MASK_ATTRIBUTE_ADDRESS_TYPE + * | addressFormat = 2 bits, 00 = unused : FLAG_ATTRIBUTE_ADDRESS_TYPE_ONEBYTE + * | 01 = 1 byte : FLAG_ATTRIBUTE_ADDRESS_TYPE_ONEBYTE + * | 10 = 2 bytes : FLAG_ATTRIBUTE_ADDRESS_TYPE_TWOBYTES + * | 11 = 3 bytes : FLAG_ATTRIBUTE_ADDRESS_TYPE_THREEBYTES + * | 4 bits : frequency : mask with FLAG_ATTRIBUTE_FREQUENCY + * <address> | IF (01 == FLAG_ATTRIBUTE_ADDRESS_TYPE_ONEBYTE == addressFormat) + * | read 1 byte, add top 4 bits + * | ELSIF (10 == FLAG_ATTRIBUTE_ADDRESS_TYPE_TWOBYTES == addressFormat) + * | read 2 bytes, add top 4 bits + * | ELSE // 11 == FLAG_ATTRIBUTE_ADDRESS_TYPE_THREEBYTES == addressFormat + * | read 3 bytes, add top 4 bits + * | END + * | if (FLAG_ATTRIBUTE_OFFSET_NEGATIVE) then address = -address + * if (FLAG_ATTRIBUTE_HAS_NET) goto bigram_and_shortcut_address_list_is + * + */ + + private static final int VERSION_1_MAGIC_NUMBER = 0x78B1; + private static final int VERSION_2_MAGIC_NUMBER = 0x9BC13AFE; + private static final int MINIMUM_SUPPORTED_VERSION = 1; + private static final int MAXIMUM_SUPPORTED_VERSION = 2; + private static final int NOT_A_VERSION_NUMBER = -1; + private static final int FIRST_VERSION_WITH_HEADER_SIZE = 2; + + // No options yet, reserved for future use. + private static final int OPTIONS = 0; + + // TODO: Make this value adaptative to content data, store it in the header, and + // use it in the reading code. + private static final int MAX_WORD_LENGTH = 48; + + private static final int MASK_GROUP_ADDRESS_TYPE = 0xC0; + private static final int FLAG_GROUP_ADDRESS_TYPE_NOADDRESS = 0x00; + private static final int FLAG_GROUP_ADDRESS_TYPE_ONEBYTE = 0x40; + private static final int FLAG_GROUP_ADDRESS_TYPE_TWOBYTES = 0x80; + private static final int FLAG_GROUP_ADDRESS_TYPE_THREEBYTES = 0xC0; + + private static final int FLAG_HAS_MULTIPLE_CHARS = 0x20; + + private static final int FLAG_IS_TERMINAL = 0x10; + private static final int FLAG_HAS_SHORTCUT_TARGETS = 0x08; + private static final int FLAG_HAS_BIGRAMS = 0x04; + private static final int FLAG_IS_SHORTCUT_ONLY = 0x02; + + private static final int FLAG_ATTRIBUTE_HAS_NEXT = 0x80; + private static final int FLAG_ATTRIBUTE_OFFSET_NEGATIVE = 0x40; + private static final int MASK_ATTRIBUTE_ADDRESS_TYPE = 0x30; + private static final int FLAG_ATTRIBUTE_ADDRESS_TYPE_ONEBYTE = 0x10; + private static final int FLAG_ATTRIBUTE_ADDRESS_TYPE_TWOBYTES = 0x20; + private static final int FLAG_ATTRIBUTE_ADDRESS_TYPE_THREEBYTES = 0x30; + private static final int FLAG_ATTRIBUTE_FREQUENCY = 0x0F; + + private static final int GROUP_CHARACTERS_TERMINATOR = 0x1F; + + private static final int GROUP_TERMINATOR_SIZE = 1; + private static final int GROUP_FLAGS_SIZE = 1; + private static final int GROUP_FREQUENCY_SIZE = 1; + private static final int GROUP_MAX_ADDRESS_SIZE = 3; + private static final int GROUP_ATTRIBUTE_FLAGS_SIZE = 1; + private static final int GROUP_ATTRIBUTE_MAX_ADDRESS_SIZE = 3; + + private static final int NO_CHILDREN_ADDRESS = Integer.MIN_VALUE; + private static final int INVALID_CHARACTER = -1; + + private static final int MAX_CHARGROUPS_FOR_ONE_BYTE_CHARGROUP_COUNT = 0x7F; // 127 + private static final int MAX_CHARGROUPS_IN_A_NODE = 0x7FFF; // 32767 + + private static final int MAX_TERMINAL_FREQUENCY = 255; + + /** + * A class grouping utility function for our specific character encoding. + */ + private static class CharEncoding { + + private static final int MINIMAL_ONE_BYTE_CHARACTER_VALUE = 0x20; + private static final int MAXIMAL_ONE_BYTE_CHARACTER_VALUE = 0xFF; + + /** + * Helper method to find out whether this code fits on one byte + */ + private static boolean fitsOnOneByte(int character) { + return character >= MINIMAL_ONE_BYTE_CHARACTER_VALUE + && character <= MAXIMAL_ONE_BYTE_CHARACTER_VALUE; + } + + /** + * Compute the size of a character given its character code. + * + * Char format is: + * 1 byte = bbbbbbbb match + * case 000xxxxx: xxxxx << 16 + next byte << 8 + next byte + * else: if 00011111 (= 0x1F) : this is the terminator. This is a relevant choice because + * unicode code points range from 0 to 0x10FFFF, so any 3-byte value starting with + * 00011111 would be outside unicode. + * else: iso-latin-1 code + * This allows for the whole unicode range to be encoded, including chars outside of + * the BMP. Also everything in the iso-latin-1 charset is only 1 byte, except control + * characters which should never happen anyway (and still work, but take 3 bytes). + * + * @param character the character code. + * @return the size in binary encoded-form, either 1 or 3 bytes. + */ + private static int getCharSize(int character) { + // See char encoding in FusionDictionary.java + if (fitsOnOneByte(character)) return 1; + if (INVALID_CHARACTER == character) return 1; + return 3; + } + + /** + * Compute the byte size of a character array. + */ + private static int getCharArraySize(final int[] chars) { + int size = 0; + for (int character : chars) size += getCharSize(character); + return size; + } + + /** + * Writes a char array to a byte buffer. + * + * @param characters the character array to write. + * @param buffer the byte buffer to write to. + * @param index the index in buffer to write the character array to. + * @return the index after the last character. + */ + private static int writeCharArray(int[] characters, byte[] buffer, int index) { + for (int character : characters) { + if (1 == getCharSize(character)) { + buffer[index++] = (byte)character; + } else { + buffer[index++] = (byte)(0xFF & (character >> 16)); + buffer[index++] = (byte)(0xFF & (character >> 8)); + buffer[index++] = (byte)(0xFF & character); + } + } + return index; + } + + /** + * Reads a character from the file. + * + * This follows the character format documented earlier in this source file. + * + * @param source the file, positioned over an encoded character. + * @return the character code. + */ + private static int readChar(RandomAccessFile source) throws IOException { + int character = source.readUnsignedByte(); + if (!fitsOnOneByte(character)) { + if (GROUP_CHARACTERS_TERMINATOR == character) + return INVALID_CHARACTER; + character <<= 16; + character += source.readUnsignedShort(); + } + return character; + } + } + + /** + * Compute the binary size of the character array in a group + * + * If only one character, this is the size of this character. If many, it's the sum of their + * sizes + 1 byte for the terminator. + * + * @param group the group + * @return the size of the char array, including the terminator if any + */ + private static int getGroupCharactersSize(CharGroup group) { + int size = CharEncoding.getCharArraySize(group.mChars); + if (group.hasSeveralChars()) size += GROUP_TERMINATOR_SIZE; + return size; + } + + /** + * Compute the binary size of the group count + * @param count the group count + * @return the size of the group count, either 1 or 2 bytes. + */ + private static int getGroupCountSize(final int count) { + if (MAX_CHARGROUPS_FOR_ONE_BYTE_CHARGROUP_COUNT >= count) { + return 1; + } else if (MAX_CHARGROUPS_IN_A_NODE >= count) { + return 2; + } else { + throw new RuntimeException("Can't have more than " + MAX_CHARGROUPS_IN_A_NODE + + " groups in a node (found " + count +")"); + } + } + + /** + * Compute the binary size of the group count for a node + * @param node the node + * @return the size of the group count, either 1 or 2 bytes. + */ + private static int getGroupCountSize(final Node node) { + return getGroupCountSize(node.mData.size()); + } + + /** + * Compute the maximum size of a CharGroup, assuming 3-byte addresses for everything. + * + * @param group the CharGroup to compute the size of. + * @return the maximum size of the group. + */ + private static int getCharGroupMaximumSize(CharGroup group) { + int size = getGroupCharactersSize(group) + GROUP_FLAGS_SIZE; + // If terminal, one byte for the frequency + if (group.isTerminal()) size += GROUP_FREQUENCY_SIZE; + size += GROUP_MAX_ADDRESS_SIZE; // For children address + if (null != group.mShortcutTargets) { + size += (GROUP_ATTRIBUTE_FLAGS_SIZE + GROUP_ATTRIBUTE_MAX_ADDRESS_SIZE) + * group.mShortcutTargets.size(); + } + if (null != group.mBigrams) { + size += (GROUP_ATTRIBUTE_FLAGS_SIZE + GROUP_ATTRIBUTE_MAX_ADDRESS_SIZE) + * group.mBigrams.size(); + } + return size; + } + + /** + * Compute the maximum size of a node, assuming 3-byte addresses for everything, and caches + * it in the 'actualSize' member of the node. + * + * @param node the node to compute the maximum size of. + */ + private static void setNodeMaximumSize(Node node) { + int size = getGroupCountSize(node); + for (CharGroup g : node.mData) { + final int groupSize = getCharGroupMaximumSize(g); + g.mCachedSize = groupSize; + size += groupSize; + } + node.mCachedSize = size; + } + + /** + * Helper method to hide the actual value of the no children address. + */ + private static boolean hasChildrenAddress(int address) { + return NO_CHILDREN_ADDRESS != address; + } + + /** + * Helper method to find out if a character info is a shortcut only. + */ + private static boolean isShortcutOnly(final CharGroupInfo info) { + return 0 != (info.mFlags & FLAG_IS_SHORTCUT_ONLY); + } + + /** + * Compute the size, in bytes, that an address will occupy. + * + * This can be used either for children addresses (which are always positive) or for + * attribute, which may be positive or negative but + * store their sign bit separately. + * + * @param address the address + * @return the byte size. + */ + private static int getByteSize(int address) { + assert(address < 0x1000000); + if (!hasChildrenAddress(address)) { + return 0; + } else if (Math.abs(address) < 0x100) { + return 1; + } else if (Math.abs(address) < 0x10000) { + return 2; + } else { + return 3; + } + } + // End utility methods. + + // This method is responsible for finding a nice ordering of the nodes that favors run-time + // cache performance and dictionary size. + /* package for tests */ static ArrayList<Node> flattenTree(Node root) { + final int treeSize = FusionDictionary.countCharGroups(root); + MakedictLog.i("Counted nodes : " + treeSize); + final ArrayList<Node> flatTree = new ArrayList<Node>(treeSize); + return flattenTreeInner(flatTree, root); + } + + private static ArrayList<Node> flattenTreeInner(ArrayList<Node> list, Node node) { + // Removing the node is necessary if the tails are merged, because we would then + // add the same node several times when we only want it once. A number of places in + // the code also depends on any node being only once in the list. + // Merging tails can only be done if there are no attributes. Searching for attributes + // in LatinIME code depends on a total breadth-first ordering, which merging tails + // breaks. If there are no attributes, it should be fine (and reduce the file size) + // to merge tails, and the following step would be necessary. + // If eventually the code runs on Android, searching through the whole array each time + // may be a performance concern. + list.remove(node); + list.add(node); + final ArrayList<CharGroup> branches = node.mData; + final int nodeSize = branches.size(); + for (CharGroup group : branches) { + if (null != group.mChildren) flattenTreeInner(list, group.mChildren); + } + return list; + } + + /** + * Finds the absolute address of a word in the dictionary. + * + * @param dict the dictionary in which to search. + * @param word the word we are searching for. + * @return the word address. If it is not found, an exception is thrown. + */ + private static int findAddressOfWord(final FusionDictionary dict, final String word) { + return FusionDictionary.findWordInTree(dict.mRoot, word).mCachedAddress; + } + + /** + * Computes the actual node size, based on the cached addresses of the children nodes. + * + * Each node stores its tentative address. During dictionary address computing, these + * are not final, but they can be used to compute the node size (the node size depends + * on the address of the children because the number of bytes necessary to store an + * address depends on its numeric value. + * + * @param node the node to compute the size of. + * @param dict the dictionary in which the word/attributes are to be found. + */ + private static void computeActualNodeSize(Node node, FusionDictionary dict) { + int size = getGroupCountSize(node); + for (CharGroup group : node.mData) { + int groupSize = GROUP_FLAGS_SIZE + getGroupCharactersSize(group); + if (group.isTerminal()) groupSize += GROUP_FREQUENCY_SIZE; + if (null != group.mChildren) { + final int offsetBasePoint= groupSize + node.mCachedAddress + size; + final int offset = group.mChildren.mCachedAddress - offsetBasePoint; + groupSize += getByteSize(offset); + } + if (null != group.mShortcutTargets) { + for (WeightedString target : group.mShortcutTargets) { + final int offsetBasePoint = groupSize + node.mCachedAddress + size + + GROUP_FLAGS_SIZE; + final int addressOfTarget = findAddressOfWord(dict, target.mWord); + final int offset = addressOfTarget - offsetBasePoint; + groupSize += getByteSize(offset) + GROUP_FLAGS_SIZE; + } + } + if (null != group.mBigrams) { + for (WeightedString bigram : group.mBigrams) { + final int offsetBasePoint = groupSize + node.mCachedAddress + size + + GROUP_FLAGS_SIZE; + final int addressOfBigram = findAddressOfWord(dict, bigram.mWord); + final int offset = addressOfBigram - offsetBasePoint; + groupSize += getByteSize(offset) + GROUP_FLAGS_SIZE; + } + } + group.mCachedSize = groupSize; + size += groupSize; + } + node.mCachedSize = size; + } + + /** + * Computes the byte size of a list of nodes and updates each node cached position. + * + * @param flatNodes the array of nodes. + * @return the byte size of the entire stack. + */ + private static int stackNodes(ArrayList<Node> flatNodes) { + int nodeOffset = 0; + for (Node n : flatNodes) { + n.mCachedAddress = nodeOffset; + int groupCountSize = getGroupCountSize(n); + int groupOffset = 0; + for (CharGroup g : n.mData) { + g.mCachedAddress = groupCountSize + nodeOffset + groupOffset; + groupOffset += g.mCachedSize; + } + if (groupOffset + groupCountSize != n.mCachedSize) { + throw new RuntimeException("Bug : Stored and computed node size differ"); + } + nodeOffset += n.mCachedSize; + } + return nodeOffset; + } + + /** + * Compute the addresses and sizes of an ordered node array. + * + * This method takes a node array and will update its cached address and size values + * so that they can be written into a file. It determines the smallest size each of the + * nodes can be given the addresses of its children and attributes, and store that into + * each node. + * The order of the node is given by the order of the array. This method makes no effort + * to find a good order; it only mechanically computes the size this order results in. + * + * @param dict the dictionary + * @param flatNodes the ordered array of nodes + * @return the same array it was passed. The nodes have been updated for address and size. + */ + private static ArrayList<Node> computeAddresses(FusionDictionary dict, + ArrayList<Node> flatNodes) { + // First get the worst sizes and offsets + for (Node n : flatNodes) setNodeMaximumSize(n); + final int offset = stackNodes(flatNodes); + + MakedictLog.i("Compressing the array addresses. Original size : " + offset); + MakedictLog.i("(Recursively seen size : " + offset + ")"); + + int passes = 0; + boolean changesDone = false; + do { + changesDone = false; + for (Node n : flatNodes) { + final int oldNodeSize = n.mCachedSize; + computeActualNodeSize(n, dict); + final int newNodeSize = n.mCachedSize; + if (oldNodeSize < newNodeSize) throw new RuntimeException("Increased size ?!"); + if (oldNodeSize != newNodeSize) changesDone = true; + } + stackNodes(flatNodes); + ++passes; + } while (changesDone); + + final Node lastNode = flatNodes.get(flatNodes.size() - 1); + MakedictLog.i("Compression complete in " + passes + " passes."); + MakedictLog.i("After address compression : " + + (lastNode.mCachedAddress + lastNode.mCachedSize)); + + return flatNodes; + } + + /** + * Sanity-checking method. + * + * This method checks an array of node for juxtaposition, that is, it will do + * nothing if each node's cached address is actually the previous node's address + * plus the previous node's size. + * If this is not the case, it will throw an exception. + * + * @param array the array node to check + */ + private static void checkFlatNodeArray(ArrayList<Node> array) { + int offset = 0; + int index = 0; + for (Node n : array) { + if (n.mCachedAddress != offset) { + throw new RuntimeException("Wrong address for node " + index + + " : expected " + offset + ", got " + n.mCachedAddress); + } + ++index; + offset += n.mCachedSize; + } + } + + /** + * Helper method to write a variable-size address to a file. + * + * @param buffer the buffer to write to. + * @param index the index in the buffer to write the address to. + * @param address the address to write. + * @return the size in bytes the address actually took. + */ + private static int writeVariableAddress(byte[] buffer, int index, int address) { + switch (getByteSize(address)) { + case 1: + buffer[index++] = (byte)address; + return 1; + case 2: + buffer[index++] = (byte)(0xFF & (address >> 8)); + buffer[index++] = (byte)(0xFF & address); + return 2; + case 3: + buffer[index++] = (byte)(0xFF & (address >> 16)); + buffer[index++] = (byte)(0xFF & (address >> 8)); + buffer[index++] = (byte)(0xFF & address); + return 3; + case 0: + return 0; + default: + throw new RuntimeException("Address " + address + " has a strange size"); + } + } + + private static byte makeCharGroupFlags(final CharGroup group, final int groupAddress, + final int childrenOffset) { + byte flags = 0; + if (group.mChars.length > 1) flags |= FLAG_HAS_MULTIPLE_CHARS; + if (group.mFrequency >= 0) { + flags |= FLAG_IS_TERMINAL; + } + if (null != group.mChildren) { + switch (getByteSize(childrenOffset)) { + case 1: + flags |= FLAG_GROUP_ADDRESS_TYPE_ONEBYTE; + break; + case 2: + flags |= FLAG_GROUP_ADDRESS_TYPE_TWOBYTES; + break; + case 3: + flags |= FLAG_GROUP_ADDRESS_TYPE_THREEBYTES; + break; + default: + throw new RuntimeException("Node with a strange address"); + } + } + if (null != group.mShortcutTargets) { + if (0 == group.mShortcutTargets.size()) { + throw new RuntimeException("0-sized shortcut list must be null"); + } + flags |= FLAG_HAS_SHORTCUT_TARGETS; + } + if (null != group.mBigrams) { + if (0 == group.mBigrams.size()) { + throw new RuntimeException("0-sized bigram list must be null"); + } + flags |= FLAG_HAS_BIGRAMS; + } + if (group.mIsShortcutOnly) { + flags |= FLAG_IS_SHORTCUT_ONLY; + } + return flags; + } + + /** + * Makes the flag value for an attribute. + * + * @param more whether there are more attributes after this one. + * @param offset the offset of the attribute. + * @param frequency the frequency of the attribute, 0..15 + * @return the flags + */ + private static final int makeAttributeFlags(final boolean more, final int offset, + final int frequency) { + int bigramFlags = (more ? FLAG_ATTRIBUTE_HAS_NEXT : 0) + + (offset < 0 ? FLAG_ATTRIBUTE_OFFSET_NEGATIVE : 0); + switch (getByteSize(offset)) { + case 1: + bigramFlags |= FLAG_ATTRIBUTE_ADDRESS_TYPE_ONEBYTE; + break; + case 2: + bigramFlags |= FLAG_ATTRIBUTE_ADDRESS_TYPE_TWOBYTES; + break; + case 3: + bigramFlags |= FLAG_ATTRIBUTE_ADDRESS_TYPE_THREEBYTES; + break; + default: + throw new RuntimeException("Strange offset size"); + } + bigramFlags += frequency & FLAG_ATTRIBUTE_FREQUENCY; + return bigramFlags; + } + + /** + * Write a node to memory. The node is expected to have its final position cached. + * + * This can be an empty map, but the more is inside the faster the lookups will be. It can + * be carried on as long as nodes do not move. + * + * @param dict the dictionary the node is a part of (for relative offsets). + * @param buffer the memory buffer to write to. + * @param node the node to write. + * @return the address of the END of the node. + */ + private static int writePlacedNode(FusionDictionary dict, byte[] buffer, Node node) { + int index = node.mCachedAddress; + + final int groupCount = node.mData.size(); + final int countSize = getGroupCountSize(node); + if (1 == countSize) { + buffer[index++] = (byte)groupCount; + } else if (2 == countSize) { + // We need to signal 2-byte size by setting the top bit of the MSB to 1, so + // we | 0x80 to do this. + buffer[index++] = (byte)((groupCount >> 8) | 0x80); + buffer[index++] = (byte)(groupCount & 0xFF); + } else { + throw new RuntimeException("Strange size from getGroupCountSize : " + countSize); + } + int groupAddress = index; + for (int i = 0; i < groupCount; ++i) { + CharGroup group = node.mData.get(i); + if (index != group.mCachedAddress) throw new RuntimeException("Bug: write index is not " + + "the same as the cached address of the group"); + groupAddress += GROUP_FLAGS_SIZE + getGroupCharactersSize(group); + // Sanity checks. + if (group.mFrequency > MAX_TERMINAL_FREQUENCY) { + throw new RuntimeException("A node has a frequency > " + MAX_TERMINAL_FREQUENCY + + " : " + group.mFrequency); + } + if (group.mFrequency >= 0) groupAddress += GROUP_FREQUENCY_SIZE; + final int childrenOffset = null == group.mChildren + ? NO_CHILDREN_ADDRESS : group.mChildren.mCachedAddress - groupAddress; + byte flags = makeCharGroupFlags(group, groupAddress, childrenOffset); + buffer[index++] = flags; + index = CharEncoding.writeCharArray(group.mChars, buffer, index); + if (group.hasSeveralChars()) { + buffer[index++] = GROUP_CHARACTERS_TERMINATOR; + } + if (group.mFrequency >= 0) { + buffer[index++] = (byte) group.mFrequency; + } + final int shift = writeVariableAddress(buffer, index, childrenOffset); + index += shift; + groupAddress += shift; + + // Write shortcuts + if (null != group.mShortcutTargets) { + final Iterator shortcutIterator = group.mShortcutTargets.iterator(); + while (shortcutIterator.hasNext()) { + final WeightedString target = (WeightedString)shortcutIterator.next(); + final int addressOfTarget = findAddressOfWord(dict, target.mWord); + ++groupAddress; + final int offset = addressOfTarget - groupAddress; + int shortcutFlags = makeAttributeFlags(shortcutIterator.hasNext(), offset, + target.mFrequency); + buffer[index++] = (byte)shortcutFlags; + final int shortcutShift = writeVariableAddress(buffer, index, Math.abs(offset)); + index += shortcutShift; + groupAddress += shortcutShift; + } + } + // Write bigrams + if (null != group.mBigrams) { + final Iterator bigramIterator = group.mBigrams.iterator(); + while (bigramIterator.hasNext()) { + final WeightedString bigram = (WeightedString)bigramIterator.next(); + final int addressOfBigram = findAddressOfWord(dict, bigram.mWord); + ++groupAddress; + final int offset = addressOfBigram - groupAddress; + int bigramFlags = makeAttributeFlags(bigramIterator.hasNext(), offset, + bigram.mFrequency); + buffer[index++] = (byte)bigramFlags; + final int bigramShift = writeVariableAddress(buffer, index, Math.abs(offset)); + index += bigramShift; + groupAddress += bigramShift; + } + } + + } + if (index != node.mCachedAddress + node.mCachedSize) throw new RuntimeException( + "Not the same size : written " + + (index - node.mCachedAddress) + " bytes out of a node that should have " + + node.mCachedSize + " bytes"); + return index; + } + + /** + * Dumps a collection of useful statistics about a node array. + * + * This prints purely informative stuff, like the total estimated file size, the + * number of nodes, of character groups, the repartition of each address size, etc + * + * @param nodes the node array. + */ + private static void showStatistics(ArrayList<Node> nodes) { + int firstTerminalAddress = Integer.MAX_VALUE; + int lastTerminalAddress = Integer.MIN_VALUE; + int size = 0; + int charGroups = 0; + int maxGroups = 0; + int maxRuns = 0; + for (Node n : nodes) { + if (maxGroups < n.mData.size()) maxGroups = n.mData.size(); + for (CharGroup cg : n.mData) { + ++charGroups; + if (cg.mChars.length > maxRuns) maxRuns = cg.mChars.length; + if (cg.mFrequency >= 0) { + if (n.mCachedAddress < firstTerminalAddress) + firstTerminalAddress = n.mCachedAddress; + if (n.mCachedAddress > lastTerminalAddress) + lastTerminalAddress = n.mCachedAddress; + } + } + if (n.mCachedAddress + n.mCachedSize > size) size = n.mCachedAddress + n.mCachedSize; + } + final int[] groupCounts = new int[maxGroups + 1]; + final int[] runCounts = new int[maxRuns + 1]; + for (Node n : nodes) { + ++groupCounts[n.mData.size()]; + for (CharGroup cg : n.mData) { + ++runCounts[cg.mChars.length]; + } + } + + MakedictLog.i("Statistics:\n" + + " total file size " + size + "\n" + + " " + nodes.size() + " nodes\n" + + " " + charGroups + " groups (" + ((float)charGroups / nodes.size()) + + " groups per node)\n" + + " first terminal at " + firstTerminalAddress + "\n" + + " last terminal at " + lastTerminalAddress + "\n" + + " Group stats : max = " + maxGroups); + for (int i = 0; i < groupCounts.length; ++i) { + MakedictLog.i(" " + i + " : " + groupCounts[i]); + } + MakedictLog.i(" Character run stats : max = " + maxRuns); + for (int i = 0; i < runCounts.length; ++i) { + MakedictLog.i(" " + i + " : " + runCounts[i]); + } + } + + /** + * Dumps a FusionDictionary to a file. + * + * This is the public entry point to write a dictionary to a file. + * + * @param destination the stream to write the binary data to. + * @param dict the dictionary to write. + * @param version the version of the format to write, currently either 1 or 2. + */ + public static void writeDictionaryBinary(final OutputStream destination, + final FusionDictionary dict, final int version) + throws IOException, UnsupportedFormatException { + + // Addresses are limited to 3 bytes, so we'll just make a 16MB buffer. Since addresses + // can be relative to each node, the structure itself is not limited to 16MB at all, but + // I doubt this will ever be shot. If it is, deciding the order of the nodes becomes + // a quite complicated problem, because though the dictionary itself does not have a + // size limit, each node must still be within 16MB of all its children and parents. + // As long as this is ensured, the dictionary file may grow to any size. + // Anyway, to make a dictionary bigger than 16MB just increase the size of this buffer. + final byte[] buffer = new byte[1 << 24]; + int index = 0; + + if (version < MINIMUM_SUPPORTED_VERSION || version > MAXIMUM_SUPPORTED_VERSION) { + throw new UnsupportedFormatException("Requested file format version " + version + + ", but this implementation only supports versions " + + MINIMUM_SUPPORTED_VERSION + " through " + MAXIMUM_SUPPORTED_VERSION); + } + + // The magic number in big-endian order. + if (version >= FIRST_VERSION_WITH_HEADER_SIZE) { + // Magic number for version 2+. + buffer[index++] = (byte) (0xFF & (VERSION_2_MAGIC_NUMBER >> 24)); + buffer[index++] = (byte) (0xFF & (VERSION_2_MAGIC_NUMBER >> 16)); + buffer[index++] = (byte) (0xFF & (VERSION_2_MAGIC_NUMBER >> 8)); + buffer[index++] = (byte) (0xFF & VERSION_2_MAGIC_NUMBER); + // Dictionary version. + buffer[index++] = (byte) (0xFF & (version >> 8)); + buffer[index++] = (byte) (0xFF & version); + } else { + // Magic number for version 1. + buffer[index++] = (byte) (0xFF & (VERSION_1_MAGIC_NUMBER >> 8)); + buffer[index++] = (byte) (0xFF & VERSION_1_MAGIC_NUMBER); + // Dictionary version. + buffer[index++] = (byte) (0xFF & version); + } + // Options flags + buffer[index++] = (byte) (0xFF & (OPTIONS >> 8)); + buffer[index++] = (byte) (0xFF & OPTIONS); + if (version >= FIRST_VERSION_WITH_HEADER_SIZE) { + final int headerSizeOffset = index; + index += 4; // Size of the header size + // TODO: Write out the header contents here. + // Write out the header size. + buffer[headerSizeOffset] = (byte) (0xFF & (index >> 24)); + buffer[headerSizeOffset + 1] = (byte) (0xFF & (index >> 16)); + buffer[headerSizeOffset + 2] = (byte) (0xFF & (index >> 8)); + buffer[headerSizeOffset + 3] = (byte) (0xFF & (index >> 0)); + } + + destination.write(buffer, 0, index); + index = 0; + + // Leave the choice of the optimal node order to the flattenTree function. + MakedictLog.i("Flattening the tree..."); + ArrayList<Node> flatNodes = flattenTree(dict.mRoot); + + MakedictLog.i("Computing addresses..."); + computeAddresses(dict, flatNodes); + MakedictLog.i("Checking array..."); + checkFlatNodeArray(flatNodes); + + MakedictLog.i("Writing file..."); + int dataEndOffset = 0; + for (Node n : flatNodes) { + dataEndOffset = writePlacedNode(dict, buffer, n); + } + + showStatistics(flatNodes); + + destination.write(buffer, 0, dataEndOffset); + + destination.close(); + MakedictLog.i("Done"); + } + + + // Input methods: Read a binary dictionary to memory. + // readDictionaryBinary is the public entry point for them. + + static final int[] characterBuffer = new int[MAX_WORD_LENGTH]; + private static CharGroupInfo readCharGroup(RandomAccessFile source, + final int originalGroupAddress) throws IOException { + int addressPointer = originalGroupAddress; + final int flags = source.readUnsignedByte(); + ++addressPointer; + final int characters[]; + if (0 != (flags & FLAG_HAS_MULTIPLE_CHARS)) { + int index = 0; + int character = CharEncoding.readChar(source); + addressPointer += CharEncoding.getCharSize(character); + while (-1 != character) { + characterBuffer[index++] = character; + character = CharEncoding.readChar(source); + addressPointer += CharEncoding.getCharSize(character); + } + characters = Arrays.copyOfRange(characterBuffer, 0, index); + } else { + final int character = CharEncoding.readChar(source); + addressPointer += CharEncoding.getCharSize(character); + characters = new int[] { character }; + } + final int frequency; + if (0 != (FLAG_IS_TERMINAL & flags)) { + ++addressPointer; + frequency = source.readUnsignedByte(); + } else { + frequency = CharGroup.NOT_A_TERMINAL; + } + int childrenAddress = addressPointer; + switch (flags & MASK_GROUP_ADDRESS_TYPE) { + case FLAG_GROUP_ADDRESS_TYPE_ONEBYTE: + childrenAddress += source.readUnsignedByte(); + addressPointer += 1; + break; + case FLAG_GROUP_ADDRESS_TYPE_TWOBYTES: + childrenAddress += source.readUnsignedShort(); + addressPointer += 2; + break; + case FLAG_GROUP_ADDRESS_TYPE_THREEBYTES: + childrenAddress += (source.readUnsignedByte() << 16) + source.readUnsignedShort(); + addressPointer += 3; + break; + case FLAG_GROUP_ADDRESS_TYPE_NOADDRESS: + default: + childrenAddress = NO_CHILDREN_ADDRESS; + break; + } + ArrayList<PendingAttribute> shortcutTargets = null; + if (0 != (flags & FLAG_HAS_SHORTCUT_TARGETS)) { + shortcutTargets = new ArrayList<PendingAttribute>(); + while (true) { + final int targetFlags = source.readUnsignedByte(); + ++addressPointer; + final int sign = 0 == (targetFlags & FLAG_ATTRIBUTE_OFFSET_NEGATIVE) ? 1 : -1; + int targetAddress = addressPointer; + switch (targetFlags & MASK_ATTRIBUTE_ADDRESS_TYPE) { + case FLAG_ATTRIBUTE_ADDRESS_TYPE_ONEBYTE: + targetAddress += sign * source.readUnsignedByte(); + addressPointer += 1; + break; + case FLAG_ATTRIBUTE_ADDRESS_TYPE_TWOBYTES: + targetAddress += sign * source.readUnsignedShort(); + addressPointer += 2; + break; + case FLAG_ATTRIBUTE_ADDRESS_TYPE_THREEBYTES: + final int offset = ((source.readUnsignedByte() << 16) + + source.readUnsignedShort()); + targetAddress += sign * offset; + addressPointer += 3; + break; + default: + throw new RuntimeException("Has shortcut targets with no address"); + } + shortcutTargets.add(new PendingAttribute(targetFlags & FLAG_ATTRIBUTE_FREQUENCY, + targetAddress)); + if (0 == (targetFlags & FLAG_ATTRIBUTE_HAS_NEXT)) break; + } + } + ArrayList<PendingAttribute> bigrams = null; + if (0 != (flags & FLAG_HAS_BIGRAMS)) { + bigrams = new ArrayList<PendingAttribute>(); + while (true) { + final int bigramFlags = source.readUnsignedByte(); + ++addressPointer; + final int sign = 0 == (bigramFlags & FLAG_ATTRIBUTE_OFFSET_NEGATIVE) ? 1 : -1; + int bigramAddress = addressPointer; + switch (bigramFlags & MASK_ATTRIBUTE_ADDRESS_TYPE) { + case FLAG_ATTRIBUTE_ADDRESS_TYPE_ONEBYTE: + bigramAddress += sign * source.readUnsignedByte(); + addressPointer += 1; + break; + case FLAG_ATTRIBUTE_ADDRESS_TYPE_TWOBYTES: + bigramAddress += sign * source.readUnsignedShort(); + addressPointer += 2; + break; + case FLAG_ATTRIBUTE_ADDRESS_TYPE_THREEBYTES: + final int offset = ((source.readUnsignedByte() << 16) + + source.readUnsignedShort()); + bigramAddress += sign * offset; + addressPointer += 3; + break; + default: + throw new RuntimeException("Has bigrams with no address"); + } + bigrams.add(new PendingAttribute(bigramFlags & FLAG_ATTRIBUTE_FREQUENCY, + bigramAddress)); + if (0 == (bigramFlags & FLAG_ATTRIBUTE_HAS_NEXT)) break; + } + } + return new CharGroupInfo(originalGroupAddress, addressPointer, flags, characters, frequency, + childrenAddress, shortcutTargets, bigrams); + } + + /** + * Reads and returns the char group count out of a file and forwards the pointer. + */ + private static int readCharGroupCount(RandomAccessFile source) throws IOException { + final int msb = source.readUnsignedByte(); + if (MAX_CHARGROUPS_FOR_ONE_BYTE_CHARGROUP_COUNT >= msb) { + return msb; + } else { + return ((MAX_CHARGROUPS_FOR_ONE_BYTE_CHARGROUP_COUNT & msb) << 8) + + source.readUnsignedByte(); + } + } + + /** + * Finds, as a string, the word at the address passed as an argument. + * + * @param source the file to read from. + * @param headerSize the size of the header. + * @param address the address to seek. + * @return the word, as a string. + * @throws IOException if the file can't be read. + */ + private static String getWordAtAddress(RandomAccessFile source, long headerSize, + int address) throws IOException { + final long originalPointer = source.getFilePointer(); + source.seek(headerSize); + final int count = readCharGroupCount(source); + int groupOffset = getGroupCountSize(count); + final StringBuilder builder = new StringBuilder(); + String result = null; + + CharGroupInfo last = null; + for (int i = count - 1; i >= 0; --i) { + CharGroupInfo info = readCharGroup(source, groupOffset); + groupOffset = info.mEndAddress; + if (info.mOriginalAddress == address) { + builder.append(new String(info.mCharacters, 0, info.mCharacters.length)); + result = builder.toString(); + break; // and return + } + if (hasChildrenAddress(info.mChildrenAddress)) { + if (info.mChildrenAddress > address) { + if (null == last) continue; + builder.append(new String(last.mCharacters, 0, last.mCharacters.length)); + source.seek(last.mChildrenAddress + headerSize); + groupOffset = last.mChildrenAddress + 1; + i = source.readUnsignedByte(); + last = null; + continue; + } + last = info; + } + if (0 == i && hasChildrenAddress(last.mChildrenAddress)) { + builder.append(new String(last.mCharacters, 0, last.mCharacters.length)); + source.seek(last.mChildrenAddress + headerSize); + groupOffset = last.mChildrenAddress + 1; + i = source.readUnsignedByte(); + last = null; + continue; + } + } + source.seek(originalPointer); + return result; + } + + /** + * Reads a single node from a binary file. + * + * This methods reads the file at the current position of its file pointer. A node is + * fully expected to start at the current position. + * This will recursively read other nodes into the structure, populating the reverse + * maps on the fly and using them to keep track of already read nodes. + * + * @param source the data file, correctly positioned at the start of a node. + * @param headerSize the size, in bytes, of the file header. + * @param reverseNodeMap a mapping from addresses to already read nodes. + * @param reverseGroupMap a mapping from addresses to already read character groups. + * @return the read node with all his children already read. + */ + private static Node readNode(RandomAccessFile source, long headerSize, + Map<Integer, Node> reverseNodeMap, Map<Integer, CharGroup> reverseGroupMap) + throws IOException { + final int nodeOrigin = (int)(source.getFilePointer() - headerSize); + final int count = readCharGroupCount(source); + final ArrayList<CharGroup> nodeContents = new ArrayList<CharGroup>(); + int groupOffset = nodeOrigin + getGroupCountSize(count); + for (int i = count; i > 0; --i) { + CharGroupInfo info = readCharGroup(source, groupOffset); + ArrayList<WeightedString> shortcutTargets = null; + if (null != info.mShortcutTargets) { + shortcutTargets = new ArrayList<WeightedString>(); + for (PendingAttribute target : info.mShortcutTargets) { + final String word = getWordAtAddress(source, headerSize, target.mAddress); + shortcutTargets.add(new WeightedString(word, target.mFrequency)); + } + } + ArrayList<WeightedString> bigrams = null; + if (null != info.mBigrams) { + bigrams = new ArrayList<WeightedString>(); + for (PendingAttribute bigram : info.mBigrams) { + final String word = getWordAtAddress(source, headerSize, bigram.mAddress); + bigrams.add(new WeightedString(word, bigram.mFrequency)); + } + } + if (hasChildrenAddress(info.mChildrenAddress)) { + Node children = reverseNodeMap.get(info.mChildrenAddress); + if (null == children) { + final long currentPosition = source.getFilePointer(); + source.seek(info.mChildrenAddress + headerSize); + children = readNode(source, headerSize, reverseNodeMap, reverseGroupMap); + source.seek(currentPosition); + } + nodeContents.add( + new CharGroup(info.mCharacters, shortcutTargets, bigrams, info.mFrequency, + children, isShortcutOnly(info))); + } else { + nodeContents.add( + new CharGroup(info.mCharacters, shortcutTargets, bigrams, info.mFrequency, + isShortcutOnly(info))); + } + groupOffset = info.mEndAddress; + } + final Node node = new Node(nodeContents); + node.mCachedAddress = nodeOrigin; + reverseNodeMap.put(node.mCachedAddress, node); + return node; + } + + /** + * Helper function to get the binary format version from the header. + */ + private static int getFormatVersion(final RandomAccessFile source) throws IOException { + final int magic_v1 = source.readUnsignedShort(); + if (VERSION_1_MAGIC_NUMBER == magic_v1) return source.readUnsignedByte(); + final int magic_v2 = (magic_v1 << 16) + source.readUnsignedShort(); + if (VERSION_2_MAGIC_NUMBER == magic_v2) return source.readUnsignedShort(); + return NOT_A_VERSION_NUMBER; + } + + /** + * Reads a random access file and returns the memory representation of the dictionary. + * + * This high-level method takes a binary file and reads its contents, populating a + * FusionDictionary structure. The optional dict argument is an existing dictionary to + * which words from the file should be added. If it is null, a new dictionary is created. + * + * @param source the file to read. + * @param dict an optional dictionary to add words to, or null. + * @return the created (or merged) dictionary. + */ + public static FusionDictionary readDictionaryBinary(final RandomAccessFile source, + final FusionDictionary dict) throws IOException, UnsupportedFormatException { + // Check file version + final int version = getFormatVersion(source); + if (version < MINIMUM_SUPPORTED_VERSION || version > MAXIMUM_SUPPORTED_VERSION ) { + throw new UnsupportedFormatException("This file has version " + version + + ", but this implementation does not support versions above " + + MAXIMUM_SUPPORTED_VERSION); + } + + // Read options + source.readUnsignedShort(); + + final long headerSize; + if (version < FIRST_VERSION_WITH_HEADER_SIZE) { + headerSize = source.getFilePointer(); + } else { + headerSize = (source.readUnsignedByte() << 24) + (source.readUnsignedByte() << 16) + + (source.readUnsignedByte() << 8) + source.readUnsignedByte(); + // read the header body + source.seek(headerSize); + } + + Map<Integer, Node> reverseNodeMapping = new TreeMap<Integer, Node>(); + Map<Integer, CharGroup> reverseGroupMapping = new TreeMap<Integer, CharGroup>(); + final Node root = readNode(source, headerSize, reverseNodeMapping, reverseGroupMapping); + + FusionDictionary newDict = new FusionDictionary(root, + new FusionDictionary.DictionaryOptions()); + if (null != dict) { + for (Word w : dict) { + newDict.add(w.mWord, w.mFrequency, w.mShortcutTargets, w.mBigrams); + } + } + + return newDict; + } + + /** + * Basic test to find out whether the file is a binary dictionary or not. + * + * Concretely this only tests the magic number. + * + * @param filename The name of the file to test. + * @return true if it's a binary dictionary, false otherwise + */ + public static boolean isBinaryDictionary(final String filename) { + try { + RandomAccessFile f = new RandomAccessFile(filename, "r"); + final int version = getFormatVersion(f); + return (version >= MINIMUM_SUPPORTED_VERSION && version <= MAXIMUM_SUPPORTED_VERSION); + } catch (FileNotFoundException e) { + return false; + } catch (IOException e) { + return false; + } + } +} diff --git a/java/src/com/android/inputmethod/latin/makedict/CharGroupInfo.java b/java/src/com/android/inputmethod/latin/makedict/CharGroupInfo.java new file mode 100644 index 000000000..444b11732 --- /dev/null +++ b/java/src/com/android/inputmethod/latin/makedict/CharGroupInfo.java @@ -0,0 +1,48 @@ +/* + * 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. + */ + +package com.android.inputmethod.latin.makedict; + +import java.util.ArrayList; + +/** + * Raw char group info straight out of a file. This will contain numbers for addresses. + */ +public class CharGroupInfo { + + public final int mOriginalAddress; + public final int mEndAddress; + public final int mFlags; + public final int[] mCharacters; + public final int mFrequency; + public final int mChildrenAddress; + public final ArrayList<PendingAttribute> mShortcutTargets; + public final ArrayList<PendingAttribute> mBigrams; + + public CharGroupInfo(final int originalAddress, final int endAddress, final int flags, + final int[] characters, final int frequency, final int childrenAddress, + final ArrayList<PendingAttribute> shortcutTargets, + final ArrayList<PendingAttribute> bigrams) { + mOriginalAddress = originalAddress; + mEndAddress = endAddress; + mFlags = flags; + mCharacters = characters; + mFrequency = frequency; + mChildrenAddress = childrenAddress; + mShortcutTargets = shortcutTargets; + mBigrams = bigrams; + } +} diff --git a/java/src/com/android/inputmethod/latin/makedict/FusionDictionary.java b/java/src/com/android/inputmethod/latin/makedict/FusionDictionary.java new file mode 100644 index 000000000..e88ab685a --- /dev/null +++ b/java/src/com/android/inputmethod/latin/makedict/FusionDictionary.java @@ -0,0 +1,671 @@ +/* + * 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. + */ + +package com.android.inputmethod.latin.makedict; + +import java.util.ArrayList; +import java.util.Arrays; +import java.util.Collections; +import java.util.Iterator; +import java.util.LinkedList; + +/** + * A dictionary that can fusion heads and tails of words for more compression. + */ +public class FusionDictionary implements Iterable<Word> { + + /** + * A node of the dictionary, containing several CharGroups. + * + * A node is but an ordered array of CharGroups, which essentially contain all the + * real information. + * This class also contains fields to cache size and address, to help with binary + * generation. + */ + public static class Node { + ArrayList<CharGroup> mData; + // To help with binary generation + int mCachedSize; + int mCachedAddress; + public Node() { + mData = new ArrayList<CharGroup>(); + mCachedSize = Integer.MIN_VALUE; + mCachedAddress = Integer.MIN_VALUE; + } + public Node(ArrayList<CharGroup> data) { + mData = data; + mCachedSize = Integer.MIN_VALUE; + mCachedAddress = Integer.MIN_VALUE; + } + } + + /** + * A string with a frequency. + * + * This represents an "attribute", that is either a bigram or a shortcut. + */ + public static class WeightedString { + final String mWord; + final int mFrequency; + public WeightedString(String word, int frequency) { + mWord = word; + mFrequency = frequency; + } + } + + /** + * A group of characters, with a frequency, shortcut targets, bigrams, and children. + * + * This is the central class of the in-memory representation. A CharGroup is what can + * be seen as a traditional "trie node", except it can hold several characters at the + * same time. A CharGroup essentially represents one or several characters in the middle + * of the trie trie; as such, it can be a terminal, and it can have children. + * In this in-memory representation, whether the CharGroup is a terminal or not is represented + * in the frequency, where NOT_A_TERMINAL (= -1) means this is not a terminal and any other + * value is the frequency of this terminal. A terminal may have non-null shortcuts and/or + * bigrams, but a non-terminal may not. Moreover, children, if present, are null. + */ + public static class CharGroup { + public static final int NOT_A_TERMINAL = -1; + final int mChars[]; + final ArrayList<WeightedString> mShortcutTargets; + final ArrayList<WeightedString> mBigrams; + final int mFrequency; // NOT_A_TERMINAL == mFrequency indicates this is not a terminal. + final boolean mIsShortcutOnly; // Only valid if this is a terminal. + Node mChildren; + // The two following members to help with binary generation + int mCachedSize; + int mCachedAddress; + + public CharGroup(final int[] chars, final ArrayList<WeightedString> shortcutTargets, + final ArrayList<WeightedString> bigrams, final int frequency, + final boolean isShortcutOnly) { + mChars = chars; + mFrequency = frequency; + mIsShortcutOnly = isShortcutOnly; + if (mIsShortcutOnly && NOT_A_TERMINAL == mFrequency) { + throw new RuntimeException("A node must be a terminal to be a shortcut only"); + } + mShortcutTargets = shortcutTargets; + mBigrams = bigrams; + mChildren = null; + } + + public CharGroup(final int[] chars, final ArrayList<WeightedString> shortcutTargets, + final ArrayList<WeightedString> bigrams, final int frequency, final Node children, + final boolean isShortcutOnly) { + mChars = chars; + mFrequency = frequency; + mIsShortcutOnly = isShortcutOnly; + if (mIsShortcutOnly && NOT_A_TERMINAL == mFrequency) { + throw new RuntimeException("A node must be a terminal to be a shortcut only"); + } + mShortcutTargets = shortcutTargets; + mBigrams = bigrams; + mChildren = children; + } + + public void addChild(CharGroup n) { + if (null == mChildren) { + mChildren = new Node(); + } + mChildren.mData.add(n); + } + + public boolean isTerminal() { + return NOT_A_TERMINAL != mFrequency; + } + + public boolean hasSeveralChars() { + assert(mChars.length > 0); + return 1 < mChars.length; + } + } + + /** + * Options global to the dictionary. + * + * There are no options at the moment, so this class is empty. + */ + public static class DictionaryOptions { + } + + + public final DictionaryOptions mOptions; + public final Node mRoot; + + public FusionDictionary() { + mOptions = new DictionaryOptions(); + mRoot = new Node(); + } + + public FusionDictionary(final Node root, final DictionaryOptions options) { + mRoot = root; + mOptions = options; + } + + /** + * Helper method to convert a String to an int array. + */ + static private int[] getCodePoints(String word) { + final int wordLength = word.length(); + int[] array = new int[word.codePointCount(0, wordLength)]; + for (int i = 0; i < wordLength; i = word.offsetByCodePoints(i, 1)) { + array[i] = word.codePointAt(i); + } + return array; + } + + /** + * Helper method to add all words in a list as 0-frequency entries + * + * These words are added when shortcuts targets or bigrams are not found in the dictionary + * yet. The same words may be added later with an actual frequency - this is handled by + * the private version of add(). + */ + private void addNeutralWords(final ArrayList<WeightedString> words) { + if (null != words) { + for (WeightedString word : words) { + final CharGroup t = findWordInTree(mRoot, word.mWord); + if (null == t) { + add(getCodePoints(word.mWord), 0, null, null, false /* isShortcutOnly */); + } + } + } + } + + /** + * Helper method to add a word as a string. + * + * This method adds a word to the dictionary with the given frequency. Optional + * lists of bigrams and shortcuts can be passed here. For each word inside, + * they will be added to the dictionary as necessary. + * + * @param word the word to add. + * @param frequency the frequency of the word, in the range [0..255]. + * @param shortcutTargets a list of shortcut targets for this word, or null. + * @param bigrams a list of bigrams, or null. + */ + public void add(final String word, final int frequency, + final ArrayList<WeightedString> shortcutTargets, + final ArrayList<WeightedString> bigrams) { + if (null != shortcutTargets) { + addNeutralWords(shortcutTargets); + } + if (null != bigrams) { + addNeutralWords(bigrams); + } + add(getCodePoints(word), frequency, shortcutTargets, bigrams, false /* isShortcutOnly */); + } + + /** + * Sanity check for a node. + * + * This method checks that all CharGroups in a node are ordered as expected. + * If they are, nothing happens. If they aren't, an exception is thrown. + */ + private void checkStack(Node node) { + ArrayList<CharGroup> stack = node.mData; + int lastValue = -1; + for (int i = 0; i < stack.size(); ++i) { + int currentValue = stack.get(i).mChars[0]; + if (currentValue <= lastValue) + throw new RuntimeException("Invalid stack"); + else + lastValue = currentValue; + } + } + + /** + * Helper method to add a shortcut that should not be a dictionary word. + * + * @param word the word to add. + * @param frequency the frequency of the word, in the range [0..255]. + * @param shortcutTargets a list of shortcut targets. May not be null. + */ + public void addShortcutOnly(final String word, final int frequency, + final ArrayList<WeightedString> shortcutTargets) { + if (null == shortcutTargets) { + throw new RuntimeException("Can't add a shortcut without targets"); + } + addNeutralWords(shortcutTargets); + add(getCodePoints(word), frequency, shortcutTargets, null, true /* isShortcutOnly */); + } + + /** + * Add a word to this dictionary. + * + * The shortcuts and bigrams, if any, have to be in the dictionary already. If they aren't, + * an exception is thrown. + * + * @param word the word, as an int array. + * @param frequency the frequency of the word, in the range [0..255]. + * @param shortcutTargets an optional list of shortcut targets for this word (null if none). + * @param bigrams an optional list of bigrams for this word (null if none). + * @param isShortcutOnly whether this should be a shortcut only. + */ + private void add(final int[] word, final int frequency, + final ArrayList<WeightedString> shortcutTargets, + final ArrayList<WeightedString> bigrams, + final boolean isShortcutOnly) { + assert(frequency >= 0 && frequency <= 255); + Node currentNode = mRoot; + int charIndex = 0; + + CharGroup currentGroup = null; + int differentCharIndex = 0; // Set by the loop to the index of the char that differs + int nodeIndex = findIndexOfChar(mRoot, word[charIndex]); + while (CHARACTER_NOT_FOUND != nodeIndex) { + currentGroup = currentNode.mData.get(nodeIndex); + differentCharIndex = compareArrays(currentGroup.mChars, word, charIndex); + if (ARRAYS_ARE_EQUAL != differentCharIndex + && differentCharIndex < currentGroup.mChars.length) break; + if (null == currentGroup.mChildren) break; + charIndex += currentGroup.mChars.length; + if (charIndex >= word.length) break; + currentNode = currentGroup.mChildren; + nodeIndex = findIndexOfChar(currentNode, word[charIndex]); + } + + if (-1 == nodeIndex) { + // No node at this point to accept the word. Create one. + final int insertionIndex = findInsertionIndex(currentNode, word[charIndex]); + final CharGroup newGroup = new CharGroup( + Arrays.copyOfRange(word, charIndex, word.length), + shortcutTargets, bigrams, frequency, isShortcutOnly); + currentNode.mData.add(insertionIndex, newGroup); + checkStack(currentNode); + } else { + // There is a word with a common prefix. + if (differentCharIndex == currentGroup.mChars.length) { + if (charIndex + differentCharIndex >= word.length) { + // The new word is a prefix of an existing word, but the node on which it + // should end already exists as is. + if (currentGroup.mFrequency > 0) { + throw new RuntimeException("Such a word already exists in the dictionary : " + + new String(word, 0, word.length)); + } else { + final CharGroup newNode = new CharGroup(currentGroup.mChars, + shortcutTargets, bigrams, frequency, currentGroup.mChildren, + isShortcutOnly); + currentNode.mData.set(nodeIndex, newNode); + checkStack(currentNode); + } + } else { + // The new word matches the full old word and extends past it. + // We only have to create a new node and add it to the end of this. + final CharGroup newNode = new CharGroup( + Arrays.copyOfRange(word, charIndex + differentCharIndex, word.length), + shortcutTargets, bigrams, frequency, isShortcutOnly); + currentGroup.mChildren = new Node(); + currentGroup.mChildren.mData.add(newNode); + } + } else { + if (0 == differentCharIndex) { + // Exact same word. Check the frequency is 0 or NOT_A_TERMINAL, and update. + if (0 != frequency) { + if (0 < currentGroup.mFrequency) { + throw new RuntimeException("This word already exists with frequency " + + currentGroup.mFrequency + " : " + + new String(word, 0, word.length)); + } + final CharGroup newGroup = new CharGroup(word, + currentGroup.mShortcutTargets, currentGroup.mBigrams, + frequency, currentGroup.mChildren, + currentGroup.mIsShortcutOnly && isShortcutOnly); + currentNode.mData.set(nodeIndex, newGroup); + } + } else { + // Partial prefix match only. We have to replace the current node with a node + // containing the current prefix and create two new ones for the tails. + Node newChildren = new Node(); + final CharGroup newOldWord = new CharGroup( + Arrays.copyOfRange(currentGroup.mChars, differentCharIndex, + currentGroup.mChars.length), currentGroup.mShortcutTargets, + currentGroup.mBigrams, currentGroup.mFrequency, currentGroup.mChildren, + currentGroup.mIsShortcutOnly); + newChildren.mData.add(newOldWord); + + final CharGroup newParent; + if (charIndex + differentCharIndex >= word.length) { + newParent = new CharGroup( + Arrays.copyOfRange(currentGroup.mChars, 0, differentCharIndex), + shortcutTargets, bigrams, frequency, newChildren, isShortcutOnly); + } else { + // isShortcutOnly makes no sense for non-terminal nodes. The following node + // is non-terminal (frequency 0 in FusionDictionary representation) so we + // pass false for isShortcutOnly + newParent = new CharGroup( + Arrays.copyOfRange(currentGroup.mChars, 0, differentCharIndex), + null, null, -1, newChildren, false /* isShortcutOnly */); + final CharGroup newWord = new CharGroup( + Arrays.copyOfRange(word, charIndex + differentCharIndex, + word.length), shortcutTargets, bigrams, frequency, + isShortcutOnly); + final int addIndex = word[charIndex + differentCharIndex] + > currentGroup.mChars[differentCharIndex] ? 1 : 0; + newChildren.mData.add(addIndex, newWord); + } + currentNode.mData.set(nodeIndex, newParent); + } + checkStack(currentNode); + } + } + } + + /** + * Custom comparison of two int arrays taken to contain character codes. + * + * This method compares the two arrays passed as an argument in a lexicographic way, + * with an offset in the dst string. + * This method does NOT test for the first character. It is taken to be equal. + * I repeat: this method starts the comparison at 1 <> dstOffset + 1. + * The index where the strings differ is returned. ARRAYS_ARE_EQUAL = 0 is returned if the + * strings are equal. This works BECAUSE we don't look at the first character. + * + * @param src the left-hand side string of the comparison. + * @param dst the right-hand side string of the comparison. + * @param dstOffset the offset in the right-hand side string. + * @return the index at which the strings differ, or ARRAYS_ARE_EQUAL = 0 if they don't. + */ + private static int ARRAYS_ARE_EQUAL = 0; + private static int compareArrays(final int[] src, final int[] dst, int dstOffset) { + // We do NOT test the first char, because we come from a method that already + // tested it. + for (int i = 1; i < src.length; ++i) { + if (dstOffset + i >= dst.length) return i; + if (src[i] != dst[dstOffset + i]) return i; + } + if (dst.length > src.length) return src.length; + return ARRAYS_ARE_EQUAL; + } + + /** + * Helper class that compares and sorts two chargroups according to their + * first element only. I repeat: ONLY the first element is considered, the rest + * is ignored. + * This comparator imposes orderings that are inconsistent with equals. + */ + static private class CharGroupComparator implements java.util.Comparator<CharGroup> { + public int compare(CharGroup c1, CharGroup c2) { + if (c1.mChars[0] == c2.mChars[0]) return 0; + return c1.mChars[0] < c2.mChars[0] ? -1 : 1; + } + } + final static private CharGroupComparator CHARGROUP_COMPARATOR = new CharGroupComparator(); + + /** + * Finds the insertion index of a character within a node. + */ + private static int findInsertionIndex(final Node node, int character) { + final ArrayList<CharGroup> data = node.mData; + final CharGroup reference = new CharGroup(new int[] { character }, null, null, 0, + false /* isShortcutOnly */); + int result = Collections.binarySearch(data, reference, CHARGROUP_COMPARATOR); + return result >= 0 ? result : -result - 1; + } + + /** + * Find the index of a char in a node, if it exists. + * + * @param node the node to search in. + * @param character the character to search for. + * @return the position of the character if it's there, or CHARACTER_NOT_FOUND = -1 else. + */ + private static int CHARACTER_NOT_FOUND = -1; + private static int findIndexOfChar(final Node node, int character) { + final int insertionIndex = findInsertionIndex(node, character); + if (node.mData.size() <= insertionIndex) return CHARACTER_NOT_FOUND; + return character == node.mData.get(insertionIndex).mChars[0] ? insertionIndex + : CHARACTER_NOT_FOUND; + } + + /** + * Helper method to find a word in a given branch. + */ + public static CharGroup findWordInTree(Node node, final String s) { + int index = 0; + final StringBuilder checker = new StringBuilder(); + + CharGroup currentGroup; + do { + int indexOfGroup = findIndexOfChar(node, s.codePointAt(index)); + if (CHARACTER_NOT_FOUND == indexOfGroup) return null; + currentGroup = node.mData.get(indexOfGroup); + checker.append(new String(currentGroup.mChars, 0, currentGroup.mChars.length)); + index += currentGroup.mChars.length; + if (index < s.length()) { + node = currentGroup.mChildren; + } + } while (null != node && index < s.length()); + + if (!s.equals(checker.toString())) return null; + return currentGroup; + } + + /** + * Helper method to find out whether a word is in the dict or not. + */ + public boolean hasWord(final String s) { + if (null == s || "".equals(s)) { + throw new RuntimeException("Can't search for a null or empty string"); + } + return null != findWordInTree(mRoot, s); + } + + /** + * Recursively count the number of character groups in a given branch of the trie. + * + * @param node the parent node. + * @return the number of char groups in all the branch under this node. + */ + public static int countCharGroups(final Node node) { + final int nodeSize = node.mData.size(); + int size = nodeSize; + for (int i = nodeSize - 1; i >= 0; --i) { + CharGroup group = node.mData.get(i); + if (null != group.mChildren) + size += countCharGroups(group.mChildren); + } + return size; + } + + /** + * Recursively count the number of nodes in a given branch of the trie. + * + * @param node the node to count. + * @result the number of nodes in this branch. + */ + public static int countNodes(final Node node) { + int size = 1; + for (int i = node.mData.size() - 1; i >= 0; --i) { + CharGroup group = node.mData.get(i); + if (null != group.mChildren) + size += countNodes(group.mChildren); + } + return size; + } + + // Historically, the tails of the words were going to be merged to save space. + // However, that would prevent the code to search for a specific address in log(n) + // time so this was abandoned. + // The code is still of interest as it does add some compression to any dictionary + // that has no need for attributes. Implementations that does not read attributes should be + // able to read a dictionary with merged tails. + // Also, the following code does support frequencies, as in, it will only merges + // tails that share the same frequency. Though it would result in the above loss of + // performance while searching by address, it is still technically possible to merge + // tails that contain attributes, but this code does not take that into account - it does + // not compare attributes and will merge terminals with different attributes regardless. + public void mergeTails() { + MakedictLog.i("Do not merge tails"); + return; + +// MakedictLog.i("Merging nodes. Number of nodes : " + countNodes(root)); +// MakedictLog.i("Number of groups : " + countCharGroups(root)); +// +// final HashMap<String, ArrayList<Node>> repository = +// new HashMap<String, ArrayList<Node>>(); +// mergeTailsInner(repository, root); +// +// MakedictLog.i("Number of different pseudohashes : " + repository.size()); +// int size = 0; +// for (ArrayList<Node> a : repository.values()) { +// size += a.size(); +// } +// MakedictLog.i("Number of nodes after merge : " + (1 + size)); +// MakedictLog.i("Recursively seen nodes : " + countNodes(root)); + } + + // The following methods are used by the deactivated mergeTails() +// private static boolean isEqual(Node a, Node b) { +// if (null == a && null == b) return true; +// if (null == a || null == b) return false; +// if (a.data.size() != b.data.size()) return false; +// final int size = a.data.size(); +// for (int i = size - 1; i >= 0; --i) { +// CharGroup aGroup = a.data.get(i); +// CharGroup bGroup = b.data.get(i); +// if (aGroup.frequency != bGroup.frequency) return false; +// if (aGroup.alternates == null && bGroup.alternates != null) return false; +// if (aGroup.alternates != null && !aGroup.equals(bGroup.alternates)) return false; +// if (!Arrays.equals(aGroup.chars, bGroup.chars)) return false; +// if (!isEqual(aGroup.children, bGroup.children)) return false; +// } +// return true; +// } + +// static private HashMap<String, ArrayList<Node>> mergeTailsInner( +// final HashMap<String, ArrayList<Node>> map, final Node node) { +// final ArrayList<CharGroup> branches = node.data; +// final int nodeSize = branches.size(); +// for (int i = 0; i < nodeSize; ++i) { +// CharGroup group = branches.get(i); +// if (null != group.children) { +// String pseudoHash = getPseudoHash(group.children); +// ArrayList<Node> similarList = map.get(pseudoHash); +// if (null == similarList) { +// similarList = new ArrayList<Node>(); +// map.put(pseudoHash, similarList); +// } +// boolean merged = false; +// for (Node similar : similarList) { +// if (isEqual(group.children, similar)) { +// group.children = similar; +// merged = true; +// break; +// } +// } +// if (!merged) { +// similarList.add(group.children); +// } +// mergeTailsInner(map, group.children); +// } +// } +// return map; +// } + +// private static String getPseudoHash(final Node node) { +// StringBuilder s = new StringBuilder(); +// for (CharGroup g : node.data) { +// s.append(g.frequency); +// for (int ch : g.chars){ +// s.append(Character.toChars(ch)); +// } +// } +// return s.toString(); +// } + + /** + * Iterator to walk through a dictionary. + * + * This is purely for convenience. + */ + public static class DictionaryIterator implements Iterator<Word> { + + private static class Position { + public Iterator<CharGroup> pos; + public int length; + public Position(ArrayList<CharGroup> groups) { + pos = groups.iterator(); + length = 0; + } + } + final StringBuilder mCurrentString; + final LinkedList<Position> mPositions; + + public DictionaryIterator(ArrayList<CharGroup> root) { + mCurrentString = new StringBuilder(); + mPositions = new LinkedList<Position>(); + final Position rootPos = new Position(root); + mPositions.add(rootPos); + } + + @Override + public boolean hasNext() { + for (Position p : mPositions) { + if (p.pos.hasNext()) { + return true; + } + } + return false; + } + + @Override + public Word next() { + Position currentPos = mPositions.getLast(); + mCurrentString.setLength(mCurrentString.length() - currentPos.length); + + do { + if (currentPos.pos.hasNext()) { + final CharGroup currentGroup = currentPos.pos.next(); + currentPos.length = currentGroup.mChars.length; + for (int i : currentGroup.mChars) + mCurrentString.append(Character.toChars(i)); + if (null != currentGroup.mChildren) { + currentPos = new Position(currentGroup.mChildren.mData); + mPositions.addLast(currentPos); + } + if (currentGroup.mFrequency >= 0) + return new Word(mCurrentString.toString(), currentGroup.mFrequency, + currentGroup.mShortcutTargets, currentGroup.mBigrams, + currentGroup.mIsShortcutOnly); + } else { + mPositions.removeLast(); + currentPos = mPositions.getLast(); + mCurrentString.setLength(mCurrentString.length() - mPositions.getLast().length); + } + } while(true); + } + + @Override + public void remove() { + throw new UnsupportedOperationException("Unsupported yet"); + } + + } + + /** + * Method to return an iterator. + * + * This method enables Java's enhanced for loop. With this you can have a FusionDictionary x + * and say : for (Word w : x) {} + */ + @Override + public Iterator<Word> iterator() { + return new DictionaryIterator(mRoot.mData); + } +} diff --git a/java/src/com/android/inputmethod/latin/makedict/MakedictLog.java b/java/src/com/android/inputmethod/latin/makedict/MakedictLog.java new file mode 100644 index 000000000..cff8d6fd0 --- /dev/null +++ b/java/src/com/android/inputmethod/latin/makedict/MakedictLog.java @@ -0,0 +1,40 @@ +/* + * 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. + */ + +package com.android.inputmethod.latin.makedict; + +/** + * Wrapper to redirect log events to the right output medium. + */ +public class MakedictLog { + + private static void print(String message) { + System.out.println(message); + } + + public static void d(String message) { + print(message); + } + public static void e(String message) { + print(message); + } + public static void i(String message) { + print(message); + } + public static void w(String message) { + print(message); + } +} diff --git a/java/src/com/android/inputmethod/latin/makedict/PendingAttribute.java b/java/src/com/android/inputmethod/latin/makedict/PendingAttribute.java new file mode 100644 index 000000000..5b41d27f2 --- /dev/null +++ b/java/src/com/android/inputmethod/latin/makedict/PendingAttribute.java @@ -0,0 +1,32 @@ +/* + * 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. + */ + +package com.android.inputmethod.latin.makedict; + +/** + * A not-yet-resolved attribute. + * + * An attribute is either a bigram or a shortcut. + * All instances of this class are always immutable. + */ +public class PendingAttribute { + public final int mFrequency; + public final int mAddress; + public PendingAttribute(final int frequency, final int address) { + mFrequency = frequency; + mAddress = address; + } +} diff --git a/java/src/com/android/inputmethod/latin/makedict/Dummy.java b/java/src/com/android/inputmethod/latin/makedict/UnsupportedFormatException.java index 27ea5ace6..bd42fb8fa 100644 --- a/java/src/com/android/inputmethod/latin/makedict/Dummy.java +++ b/java/src/com/android/inputmethod/latin/makedict/UnsupportedFormatException.java @@ -1,5 +1,5 @@ /* - * Copyright (C) 2012 The Android Open Source Project + * 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 @@ -16,6 +16,11 @@ package com.android.inputmethod.latin.makedict; -public class Dummy { - +/** + * Simple exception thrown when a file format is not recognized. + */ +public class UnsupportedFormatException extends Exception { + public UnsupportedFormatException(String description) { + super(description); + } } diff --git a/java/src/com/android/inputmethod/latin/makedict/Word.java b/java/src/com/android/inputmethod/latin/makedict/Word.java new file mode 100644 index 000000000..c2c01e1f8 --- /dev/null +++ b/java/src/com/android/inputmethod/latin/makedict/Word.java @@ -0,0 +1,72 @@ +/* + * 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. + */ + +package com.android.inputmethod.latin.makedict; + +import com.android.inputmethod.latin.makedict.FusionDictionary.WeightedString; + +import java.util.ArrayList; + +/** + * Utility class for a word with a frequency. + * + * This is chiefly used to iterate a dictionary. + */ +public class Word implements Comparable<Word> { + final String mWord; + final int mFrequency; + final boolean mIsShortcutOnly; + final ArrayList<WeightedString> mShortcutTargets; + final ArrayList<WeightedString> mBigrams; + + public Word(final String word, final int frequency, + final ArrayList<WeightedString> shortcutTargets, + final ArrayList<WeightedString> bigrams, final boolean isShortcutOnly) { + mWord = word; + mFrequency = frequency; + mShortcutTargets = shortcutTargets; + mBigrams = bigrams; + mIsShortcutOnly = isShortcutOnly; + } + + /** + * Three-way comparison. + * + * A Word x is greater than a word y if x has a higher frequency. If they have the same + * frequency, they are sorted in lexicographic order. + */ + @Override + public int compareTo(Word w) { + if (mFrequency < w.mFrequency) return 1; + if (mFrequency > w.mFrequency) return -1; + return mWord.compareTo(w.mWord); + } + + /** + * Equality test. + * + * Words are equal if they have the same frequency, the same spellings, and the same + * attributes. + */ + @Override + public boolean equals(Object o) { + if (!(o instanceof Word)) return false; + Word w = (Word)o; + return mFrequency == w.mFrequency && mWord.equals(w.mWord) + && mShortcutTargets.equals(w.mShortcutTargets) + && mBigrams.equals(w.mBigrams); + } +} |