diff options
16 files changed, 2553 insertions, 78 deletions
diff --git a/java/src/com/android/inputmethod/latin/BinaryDictionaryFileDumper.java b/java/src/com/android/inputmethod/latin/BinaryDictionaryFileDumper.java index 2d50a6f46..3da670e2e 100644 --- a/java/src/com/android/inputmethod/latin/BinaryDictionaryFileDumper.java +++ b/java/src/com/android/inputmethod/latin/BinaryDictionaryFileDumper.java @@ -25,7 +25,6 @@ import android.net.Uri; import android.text.TextUtils; import android.util.Log; -import java.io.File; import java.io.FileInputStream; import java.io.FileNotFoundException; import java.io.FileOutputStream; @@ -55,66 +54,6 @@ public class BinaryDictionaryFileDumper { } /** - * Escapes a string for any characters that may be suspicious for a file or directory name. - * - * Concretely this does a sort of URL-encoding except it will encode everything that's not - * alphanumeric or underscore. (true URL-encoding leaves alone characters like '*', which - * we cannot allow here) - */ - // TODO: create a unit test for this method - private static String replaceFileNameDangerousCharacters(String name) { - // This assumes '%' is fully available as a non-separator, normal - // character in a file name. This is probably true for all file systems. - final StringBuilder sb = new StringBuilder(); - for (int i = 0; i < name.length(); ++i) { - final int codePoint = name.codePointAt(i); - if (Character.isLetterOrDigit(codePoint) || '_' == codePoint) { - sb.appendCodePoint(codePoint); - } else { - sb.append('%'); - sb.append(Integer.toHexString(codePoint)); - } - } - return sb.toString(); - } - - /** - * Find out the cache directory associated with a specific locale. - */ - private static String getCacheDirectoryForLocale(Locale locale, Context context) { - final String relativeDirectoryName = replaceFileNameDangerousCharacters(locale.toString()); - final String absoluteDirectoryName = context.getFilesDir() + File.separator - + relativeDirectoryName; - final File directory = new File(absoluteDirectoryName); - if (!directory.exists()) { - if (!directory.mkdirs()) { - Log.e(TAG, "Could not create the directory for locale" + locale); - } - } - return absoluteDirectoryName; - } - - /** - * Generates a file name for the id and locale passed as an argument. - * - * In the current implementation the file name returned will always be unique for - * any id/locale pair, but please do not expect that the id can be the same for - * different dictionaries with different locales. An id should be unique for any - * dictionary. - * The file name is pretty much an URL-encoded version of the id inside a directory - * named like the locale, except it will also escape characters that look dangerous - * to some file systems. - * @param id the id of the dictionary for which to get a file name - * @param locale the locale for which to get the file name - * @param context the context to use for getting the directory - * @return the name of the file to be created - */ - private static String getCacheFileName(String id, Locale locale, Context context) { - final String fileName = replaceFileNameDangerousCharacters(id); - return getCacheDirectoryForLocale(locale, context) + File.separator + fileName; - } - - /** * Return for a given locale or dictionary id the provider URI to get the dictionary. */ private static Uri getProviderUri(String path) { @@ -149,32 +88,32 @@ public class BinaryDictionaryFileDumper { } /** - * Queries a content provider for dictionary data for some locale and returns the file addresses + * Queries a content provider for dictionary data for some locale and cache the returned files * - * This will query a content provider for dictionary data for a given locale, and return - * the addresses of a file set the members of which are suitable to be mmap'ed. It will copy - * them to local storage if needed. - * It should also check the dictionary versions to avoid unnecessary copies but this is - * still in TODO state. - * This will make the data from the content provider the cached dictionary for this locale, - * overwriting any previous cached data. + * This will query a content provider for dictionary data for a given locale, and copy the + * files locally so that they can be mmap'ed. This may overwrite previously cached dictionaries + * with newer versions if a newer version is made available by the content provider. * @returns the addresses of the files, or null if no data could be obtained. * @throw FileNotFoundException if the provider returns non-existent data. * @throw IOException if the provider-returned data could not be read. */ - public static List<AssetFileAddress> getDictSetFromContentProvider(final Locale locale, + public static List<AssetFileAddress> cacheDictionariesFromContentProvider(final Locale locale, final Context context) throws FileNotFoundException, IOException { final ContentResolver resolver = context.getContentResolver(); final List<String> idList = getDictIdList(locale, context); final List<AssetFileAddress> fileAddressList = new ArrayList<AssetFileAddress>(); for (String id : idList) { - final Uri dictionaryPackUri = getProviderUri(id); + final Uri wordListUri = getProviderUri(id); final AssetFileDescriptor afd = - resolver.openAssetFileDescriptor(dictionaryPackUri, "r"); + resolver.openAssetFileDescriptor(wordListUri, "r"); if (null == afd) continue; final String fileName = copyFileTo(afd.createInputStream(), - getCacheFileName(id, locale, context)); + BinaryDictionaryGetter.getCacheFileName(id, locale, context)); afd.close(); + if (0 >= resolver.delete(wordListUri, null, null)) { + // I'd rather not print the word list ID to the log here out of security concerns + Log.e(TAG, "Could not have the dictionary pack delete a word list"); + } fileAddressList.add(AssetFileAddress.makeFromFileName(fileName)); } return fileAddressList; @@ -192,7 +131,9 @@ public class BinaryDictionaryFileDumper { final Locale savedLocale = Utils.setSystemLocale(res, locale); final InputStream stream = res.openRawResource(resource); Utils.setSystemLocale(res, savedLocale); - return copyFileTo(stream, getCacheFileName(Integer.toString(resource), locale, context)); + return copyFileTo(stream, + BinaryDictionaryGetter.getCacheFileName(Integer.toString(resource), + locale, context)); } /** diff --git a/java/src/com/android/inputmethod/latin/BinaryDictionaryGetter.java b/java/src/com/android/inputmethod/latin/BinaryDictionaryGetter.java index 4b1c05adf..b26731ac5 100644 --- a/java/src/com/android/inputmethod/latin/BinaryDictionaryGetter.java +++ b/java/src/com/android/inputmethod/latin/BinaryDictionaryGetter.java @@ -21,6 +21,7 @@ import android.content.res.AssetFileDescriptor; import android.content.res.Resources; import android.util.Log; +import java.io.File; import java.io.FileNotFoundException; import java.io.IOException; import java.util.Arrays; @@ -41,6 +42,66 @@ class BinaryDictionaryGetter { private BinaryDictionaryGetter() {} /** + * Escapes a string for any characters that may be suspicious for a file or directory name. + * + * Concretely this does a sort of URL-encoding except it will encode everything that's not + * alphanumeric or underscore. (true URL-encoding leaves alone characters like '*', which + * we cannot allow here) + */ + // TODO: create a unit test for this method + private static String replaceFileNameDangerousCharacters(String name) { + // This assumes '%' is fully available as a non-separator, normal + // character in a file name. This is probably true for all file systems. + final StringBuilder sb = new StringBuilder(); + for (int i = 0; i < name.length(); ++i) { + final int codePoint = name.codePointAt(i); + if (Character.isLetterOrDigit(codePoint) || '_' == codePoint) { + sb.appendCodePoint(codePoint); + } else { + sb.append('%'); + sb.append(Integer.toHexString(codePoint)); + } + } + return sb.toString(); + } + + /** + * Find out the cache directory associated with a specific locale. + */ + private static String getCacheDirectoryForLocale(Locale locale, Context context) { + final String relativeDirectoryName = replaceFileNameDangerousCharacters(locale.toString()); + final String absoluteDirectoryName = context.getFilesDir() + File.separator + + relativeDirectoryName; + final File directory = new File(absoluteDirectoryName); + if (!directory.exists()) { + if (!directory.mkdirs()) { + Log.e(TAG, "Could not create the directory for locale" + locale); + } + } + return absoluteDirectoryName; + } + + /** + * Generates a file name for the id and locale passed as an argument. + * + * In the current implementation the file name returned will always be unique for + * any id/locale pair, but please do not expect that the id can be the same for + * different dictionaries with different locales. An id should be unique for any + * dictionary. + * The file name is pretty much an URL-encoded version of the id inside a directory + * named like the locale, except it will also escape characters that look dangerous + * to some file systems. + * @param id the id of the dictionary for which to get a file name + * @param locale the locale for which to get the file name + * @param context the context to use for getting the directory + * @return the name of the file to be created + */ + public static String getCacheFileName(String id, Locale locale, Context context) { + final String fileName = replaceFileNameDangerousCharacters(id); + return getCacheDirectoryForLocale(locale, context) + File.separator + fileName; + } + + /** * Returns a file address from a resource, or null if it cannot be opened. */ private static AssetFileAddress loadFallbackResource(final Context context, @@ -74,10 +135,11 @@ class BinaryDictionaryGetter { public static List<AssetFileAddress> getDictionaryFiles(Locale locale, Context context, int fallbackResId) { try { - List<AssetFileAddress> listFromContentProvider = - BinaryDictionaryFileDumper.getDictSetFromContentProvider(locale, context); - if (null != listFromContentProvider) { - return listFromContentProvider; + List<AssetFileAddress> cachedDictionaryList = + BinaryDictionaryFileDumper.cacheDictionariesFromContentProvider(locale, + context); + if (null != cachedDictionaryList) { + return cachedDictionaryList; } // If the list is null, fall through and return the fallback } catch (FileNotFoundException e) { diff --git a/tools/makedict2/Android.mk b/tools/makedict2/Android.mk new file mode 100644 index 000000000..e056168e9 --- /dev/null +++ b/tools/makedict2/Android.mk @@ -0,0 +1,27 @@ +# +# 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. +# +LOCAL_PATH := $(call my-dir) +include $(CLEAR_VARS) + +LOCAL_SRC_FILES := $(call all-java-files-under,src) +LOCAL_SRC_FILES += $(call all-java-files-under,tests) +LOCAL_JAR_MANIFEST := etc/manifest.txt +LOCAL_MODULE_TAGS := eng +LOCAL_MODULE := makedict2 +LOCAL_JAVA_LIBRARIES := junit + +include $(BUILD_HOST_JAVA_LIBRARY) +include $(LOCAL_PATH)/etc/Android.mk diff --git a/tools/makedict2/etc/Android.mk b/tools/makedict2/etc/Android.mk new file mode 100644 index 000000000..c71377cdf --- /dev/null +++ b/tools/makedict2/etc/Android.mk @@ -0,0 +1,21 @@ +# 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. + +LOCAL_PATH := $(call my-dir) +include $(CLEAR_VARS) + +LOCAL_MODULE_TAGS := eng + +LOCAL_PREBUILT_EXECUTABLES := makedict2 +include $(BUILD_HOST_PREBUILT) diff --git a/tools/makedict2/etc/makedict2 b/tools/makedict2/etc/makedict2 new file mode 100755 index 000000000..1bad825e1 --- /dev/null +++ b/tools/makedict2/etc/makedict2 @@ -0,0 +1,63 @@ +#!/bin/sh +# Copyright 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. + +# Set up prog to be the path of this script, including following symlinks, +# and set up progdir to be the fully-qualified pathname of its directory. +prog="$0" +while [ -h "${prog}" ]; do + newProg=`/bin/ls -ld "${prog}"` + newProg=`expr "${newProg}" : ".* -> \(.*\)$"` + if expr "x${newProg}" : 'x/' >/dev/null; then + prog="${newProg}" + else + progdir=`dirname "${prog}"` + prog="${progdir}/${newProg}" + fi +done +oldwd=`pwd` +progdir=`dirname "${prog}"` +cd "${progdir}" +progdir=`pwd` +prog="${progdir}"/`basename "${prog}"` +cd "${oldwd}" + +jarfile=makedict2.jar +frameworkdir="$progdir" +if [ ! -r "$frameworkdir/$jarfile" ] +then + frameworkdir=`dirname "$progdir"`/tools/lib + libdir=`dirname "$progdir"`/tools/lib +fi +if [ ! -r "$frameworkdir/$jarfile" ] +then + frameworkdir=`dirname "$progdir"`/framework + libdir=`dirname "$progdir"`/lib +fi +if [ ! -r "$frameworkdir/$jarfile" ] +then + echo `basename "$prog"`": can't find $jarfile" + exit 1 +fi + +if [ "$OSTYPE" = "cygwin" ] ; then + jarpath=`cygpath -w "$frameworkdir/$jarfile"` + progdir=`cygpath -w "$progdir"` +else + jarpath="$frameworkdir/$jarfile" +fi + +# need to use "java.ext.dirs" because "-jar" causes classpath to be ignored +# might need more memory, e.g. -Xmx128M +exec java -ea -Djava.ext.dirs="$frameworkdir" -jar "$jarpath" "$@" diff --git a/tools/makedict2/etc/manifest.txt b/tools/makedict2/etc/manifest.txt new file mode 100644 index 000000000..948609da6 --- /dev/null +++ b/tools/makedict2/etc/manifest.txt @@ -0,0 +1 @@ +Main-Class: com.android.inputmethod.latin.DictionaryMaker diff --git a/tools/makedict2/src/com/android/inputmethod/latin/BinaryDictInputOutput.java b/tools/makedict2/src/com/android/inputmethod/latin/BinaryDictInputOutput.java new file mode 100644 index 000000000..92f402d3e --- /dev/null +++ b/tools/makedict2/src/com/android/inputmethod/latin/BinaryDictInputOutput.java @@ -0,0 +1,1024 @@ +/* + * 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; + +import com.android.inputmethod.latin.FusionDictionary.CharGroup; +import com.android.inputmethod.latin.FusionDictionary.Node; +import com.android.inputmethod.latin.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.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 + * | reserved 1 bit, 1 = yes, 0 = no + * | has bigrams ? 1 bit, 1 = yes, 0 = no : FLAG_HAS_BIGRAMS + * + * 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_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 MAGIC_NUMBER = 0x78B1; + private static final int VERSION = 1; + private static final int MAXIMUM_SUPPORTED_VERSION = VERSION; + // 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_BIGRAMS = 0x04; + + 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_COUNT_SIZE = 1; + 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; + + // Limiting to 127 for upward compatibility + // TODO: implement a scheme to be able to shoot 256 chargroups in a node + private static final int MAX_CHARGROUPS_IN_A_NODE = 127; + + 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 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.mBigrams) { + for (WeightedString bigram : group.mBigrams) { + size += GROUP_ATTRIBUTE_FLAGS_SIZE + GROUP_ATTRIBUTE_MAX_ADDRESS_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 = GROUP_COUNT_SIZE; + 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; + } + + /** + * 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 = GROUP_COUNT_SIZE; + 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.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 groupOffset = 0; + for (CharGroup g : n.mData) { + g.mCachedAddress = GROUP_COUNT_SIZE + nodeOffset + groupOffset; + groupOffset += g.mCachedSize; + } + if (groupOffset + GROUP_COUNT_SIZE != 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.mBigrams) flags |= FLAG_HAS_BIGRAMS; + 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 size = node.mData.size(); + if (size > MAX_CHARGROUPS_IN_A_NODE) + throw new RuntimeException("A node has a group count over 127 (" + size + ")."); + + buffer[index++] = (byte)size; + int groupAddress = index; + for (int i = 0; i < size; ++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 bigrams + if (null != group.mBigrams) { + int remainingBigrams = group.mBigrams.size(); + for (WeightedString bigram : group.mBigrams) { + boolean more = remainingBigrams > 1; + final int addressOfBigram = findAddressOfWord(dict, bigram.mWord); + ++groupAddress; + final int offset = addressOfBigram - groupAddress; + int bigramFlags = makeAttributeFlags(more, offset, bigram.mFrequency); + buffer[index++] = (byte)bigramFlags; + final int bigramShift = writeVariableAddress(buffer, index, Math.abs(offset)); + index += bigramShift; + groupAddress += bigramShift; + --remainingBigrams; + } + } + + } + 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. + */ + public static void writeDictionaryBinary(OutputStream destination, FusionDictionary dict) + throws IOException { + + // 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; + + // Magic number in big-endian order. + buffer[index++] = (byte) (0xFF & (MAGIC_NUMBER >> 8)); + buffer[index++] = (byte) (0xFF & MAGIC_NUMBER); + // Dictionary version. + buffer[index++] = (byte) (0xFF & VERSION); + // Options flags + buffer[index++] = (byte) (0xFF & (OPTIONS >> 8)); + buffer[index++] = (byte) (0xFF & OPTIONS); + + // Should we include the locale and title of the dictionary ? + + 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> bigrams = null; + if (0 != (flags & FLAG_HAS_BIGRAMS)) { + bigrams = new ArrayList<PendingAttribute>(); + boolean more = true; + while (more) { + int bigramFlags = source.readUnsignedByte(); + ++addressPointer; + more = (0 != (bigramFlags & FLAG_ATTRIBUTE_HAS_NEXT)); + 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 attribute with no address"); + } + bigrams.add(new PendingAttribute(bigramFlags & FLAG_ATTRIBUTE_FREQUENCY, + bigramAddress)); + } + } + return new CharGroupInfo(originalGroupAddress, addressPointer, flags, characters, frequency, + childrenAddress, bigrams); + } + + /** + * 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 = source.readUnsignedByte(); + int groupOffset = 1; // 1 for the group 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 = source.readUnsignedByte(); + final ArrayList<CharGroup> nodeContents = new ArrayList<CharGroup>(); + int groupOffset = nodeOrigin + 1; // 1 byte for the group count + for (int i = count; i > 0; --i) { + CharGroupInfo info = readCharGroup(source, groupOffset); + 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, bigrams, info.mFrequency, + children)); + } else { + nodeContents.add( + new CharGroup(info.mCharacters, bigrams, info.mFrequency)); + } + groupOffset = info.mEndAddress; + } + final Node node = new Node(nodeContents); + node.mCachedAddress = nodeOrigin; + reverseNodeMap.put(node.mCachedAddress, node); + return node; + } + + /** + * 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(RandomAccessFile source, + FusionDictionary dict) throws IOException, UnsupportedFormatException { + // Check magic number + final int magic = source.readUnsignedShort(); + if (MAGIC_NUMBER != magic) { + throw new UnsupportedFormatException("The magic number in this file does not match " + + "the expected value"); + } + + // Check file version + final int version = source.readUnsignedByte(); + if (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(); + + long headerSize = source.getFilePointer(); + 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.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(String filename) { + try { + RandomAccessFile f = new RandomAccessFile(filename, "r"); + return MAGIC_NUMBER == f.readUnsignedShort(); + } catch (FileNotFoundException e) { + return false; + } catch (IOException e) { + return false; + } + } +} diff --git a/tools/makedict2/src/com/android/inputmethod/latin/CharGroupInfo.java b/tools/makedict2/src/com/android/inputmethod/latin/CharGroupInfo.java new file mode 100644 index 000000000..6badfd13a --- /dev/null +++ b/tools/makedict2/src/com/android/inputmethod/latin/CharGroupInfo.java @@ -0,0 +1,45 @@ +/* + * 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; + +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> mBigrams; + + public CharGroupInfo(final int originalAddress, final int endAddress, final int flags, + final int[] characters, final int frequency, final int childrenAddress, + final ArrayList<PendingAttribute> bigrams) { + mOriginalAddress = originalAddress; + mEndAddress = endAddress; + mFlags = flags; + mCharacters = characters; + mFrequency = frequency; + mChildrenAddress = childrenAddress; + mBigrams = bigrams; + } +} diff --git a/tools/makedict2/src/com/android/inputmethod/latin/DictionaryMaker.java b/tools/makedict2/src/com/android/inputmethod/latin/DictionaryMaker.java new file mode 100644 index 000000000..61944545a --- /dev/null +++ b/tools/makedict2/src/com/android/inputmethod/latin/DictionaryMaker.java @@ -0,0 +1,256 @@ +/* + * 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; + +import java.io.File; +import java.io.FileInputStream; +import java.io.FileNotFoundException; +import java.io.FileOutputStream; +import java.io.FileWriter; +import java.io.IOException; +import java.io.RandomAccessFile; +import java.util.Arrays; +import java.util.LinkedList; + +import javax.xml.parsers.ParserConfigurationException; + +import org.xml.sax.SAXException; + +/** + * Main class/method for DictionaryMaker. + */ +public class DictionaryMaker { + + static class Arguments { + private final static String OPTION_VERSION_2 = "-2"; + private final static String OPTION_INPUT_SOURCE = "-s"; + private final static String OPTION_INPUT_BIGRAM_XML = "-b"; + private final static String OPTION_OUTPUT_BINARY = "-d"; + private final static String OPTION_OUTPUT_XML = "-x"; + private final static String OPTION_HELP = "-h"; + public final String mInputBinary; + public final String mInputUnigramXml; + public final String mInputBigramXml; + public final String mOutputBinary; + public final String mOutputXml; + + private void checkIntegrity() { + checkHasExactlyOneInput(); + checkHasAtLeastOneOutput(); + } + + private void checkHasExactlyOneInput() { + if (null == mInputUnigramXml && null == mInputBinary) { + throw new RuntimeException("No input file specified"); + } else if (null != mInputUnigramXml && null != mInputBinary) { + throw new RuntimeException("Both input XML and binary specified"); + } else if (null != mInputBinary && null != mInputBigramXml) { + throw new RuntimeException("Cannot specify a binary input and a separate bigram " + + "file"); + } + } + + private void checkHasAtLeastOneOutput() { + if (null == mOutputBinary && null == mOutputXml) { + throw new RuntimeException("No output specified"); + } + } + + private void displayHelp() { + MakedictLog.i("Usage: makedict2 " + + "[-s <unigrams.xml> [-b <bigrams.xml>] | -s <binary input>] " + + " [-d <binary output>] [-x <xml output>] [-2]\n" + + "\n" + + " Converts a source dictionary file to one or several outputs.\n" + + " Source can be an XML file, with an optional XML bigrams file, or a\n" + + " binary dictionary file.\n" + + " Both binary and XML outputs are supported. Both can be output at\n" + + " the same time but outputting several files of the same type is not\n" + + " supported."); + } + + public Arguments(String[] argsArray) { + final LinkedList<String> args = new LinkedList<String>(Arrays.asList(argsArray)); + if (args.isEmpty()) { + displayHelp(); + } + String inputBinary = null; + String inputUnigramXml = null; + String inputBigramXml = null; + String outputBinary = null; + String outputXml = null; + + while (!args.isEmpty()) { + final String arg = args.get(0); + args.remove(0); + if (arg.charAt(0) == '-') { + if (OPTION_VERSION_2.equals(arg)) { + // Do nothing, this is the default + } else if (OPTION_HELP.equals(arg)) { + displayHelp(); + } else { + // All these options need an argument + if (args.isEmpty()) { + throw new RuntimeException("Option " + arg + " requires an argument"); + } + String filename = args.get(0); + args.remove(0); + if (OPTION_INPUT_SOURCE.equals(arg)) { + if (BinaryDictInputOutput.isBinaryDictionary(filename)) { + inputBinary = filename; + } else { + inputUnigramXml = filename; + } + } else if (OPTION_INPUT_BIGRAM_XML.equals(arg)) { + inputBigramXml = filename; + } else if (OPTION_OUTPUT_BINARY.equals(arg)) { + outputBinary = filename; + } else if (OPTION_OUTPUT_XML.equals(arg)) { + outputXml = filename; + } + } + } else { + if (null == inputBinary && null == inputUnigramXml) { + if (BinaryDictInputOutput.isBinaryDictionary(arg)) { + inputBinary = arg; + } else { + inputUnigramXml = arg; + } + } else if (null == outputBinary) { + outputBinary = arg; + } else { + throw new RuntimeException("Several output binary files specified"); + } + } + } + + mInputBinary = inputBinary; + mInputUnigramXml = inputUnigramXml; + mInputBigramXml = inputBigramXml; + mOutputBinary = outputBinary; + mOutputXml = outputXml; + checkIntegrity(); + } + } + + public static void main(String[] args) + throws FileNotFoundException, ParserConfigurationException, SAXException, IOException, + UnsupportedFormatException { + final Arguments parsedArgs = new Arguments(args); + FusionDictionary dictionary = readInputFromParsedArgs(parsedArgs); + writeOutputToParsedArgs(parsedArgs, dictionary); + } + + /** + * Invoke the right input method according to args. + * + * @param args the parsed command line arguments. + * @return the read dictionary. + */ + private static FusionDictionary readInputFromParsedArgs(final Arguments args) + throws IOException, UnsupportedFormatException, ParserConfigurationException, + SAXException, FileNotFoundException { + if (null != args.mInputBinary) { + return readBinaryFile(args.mInputBinary); + } else if (null != args.mInputUnigramXml) { + return readXmlFile(args.mInputUnigramXml, args.mInputBigramXml); + } else { + throw new RuntimeException("No input file specified"); + } + } + + /** + * Read a dictionary from the name of a binary file. + * + * @param binaryFilename the name of the file in the binary dictionary format. + * @return the read dictionary. + * @throws FileNotFoundException if the file can't be found + * @throws IOException if the input file can't be read + * @throws UnsupportedFormatException if the binary file is not in the expected format + */ + private static FusionDictionary readBinaryFile(final String binaryFilename) + throws FileNotFoundException, IOException, UnsupportedFormatException { + final RandomAccessFile inputFile = new RandomAccessFile(binaryFilename, "r"); + return BinaryDictInputOutput.readDictionaryBinary(inputFile, null); + } + + /** + * Read a dictionary from a unigram XML file, and optionally a bigram XML file. + * + * @param unigramXmlFilename the name of the unigram XML file. May not be null. + * @param bigramXmlFilename the name of the bigram XML file. Pass null if there are no bigrams. + * @return the read dictionary. + * @throws FileNotFoundException if one of the files can't be found + * @throws SAXException if one or more of the XML files is not well-formed + * @throws IOException if one the input files can't be read + * @throws ParserConfigurationException if the system can't create a SAX parser + */ + private static FusionDictionary readXmlFile(final String unigramXmlFilename, + final String bigramXmlFilename) throws FileNotFoundException, SAXException, + IOException, ParserConfigurationException { + final FileInputStream unigrams = new FileInputStream(new File(unigramXmlFilename)); + final FileInputStream bigrams = null == bigramXmlFilename ? null : + new FileInputStream(new File(bigramXmlFilename)); + return XmlDictInputOutput.readDictionaryXml(unigrams, bigrams); + } + + /** + * Invoke the right output method according to args. + * + * This will write the passed dictionary to the file(s) passed in the command line arguments. + * @param args the parsed arguments. + * @param dict the file to output. + * @throws FileNotFoundException if one of the output files can't be created. + * @throws IOException if one of the output files can't be written to. + */ + private static void writeOutputToParsedArgs(final Arguments args, final FusionDictionary dict) + throws FileNotFoundException, IOException { + if (null != args.mOutputBinary) { + writeBinaryDictionary(args.mOutputBinary, dict); + } + if (null != args.mOutputXml) { + writeXmlDictionary(args.mOutputXml, dict); + } + } + + /** + * Write the dictionary in binary format to the specified filename. + * + * @param outputFilename the name of the file to write to. + * @param dict the dictionary to write. + * @throws FileNotFoundException if the output file can't be created. + * @throws IOException if the output file can't be written to. + */ + private static void writeBinaryDictionary(final String outputFilename, + final FusionDictionary dict) throws FileNotFoundException, IOException { + final File outputFile = new File(outputFilename); + BinaryDictInputOutput.writeDictionaryBinary(new FileOutputStream(outputFilename), dict); + } + + /** + * Write the dictionary in XML format to the specified filename. + * + * @param outputFilename the name of the file to write to. + * @param dict the dictionary to write. + * @throws FileNotFoundException if the output file can't be created. + * @throws IOException if the output file can't be written to. + */ + private static void writeXmlDictionary(final String outputFilename, + final FusionDictionary dict) throws FileNotFoundException, IOException { + XmlDictInputOutput.writeDictionaryXml(new FileWriter(outputFilename), dict); + } +} diff --git a/tools/makedict2/src/com/android/inputmethod/latin/FusionDictionary.java b/tools/makedict2/src/com/android/inputmethod/latin/FusionDictionary.java new file mode 100644 index 000000000..031f35dfe --- /dev/null +++ b/tools/makedict2/src/com/android/inputmethod/latin/FusionDictionary.java @@ -0,0 +1,602 @@ +/* + * 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; + +import java.util.ArrayList; +import java.util.Arrays; +import java.util.Collections; +import java.util.Iterator; +import java.util.LinkedList; +import java.util.List; + +/** + * 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, shortcuts, 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> mBigrams; + final int mFrequency; // NOT_A_TERMINAL == mFrequency indicates this is not 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> bigrams, final int frequency) { + mChars = chars; + mFrequency = frequency; + mBigrams = bigrams; + mChildren = null; + } + + public CharGroup(final int[] chars, + final ArrayList<WeightedString> bigrams, final int frequency, final Node children) { + mChars = chars; + mFrequency = frequency; + 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) { + array[i] = word.codePointAt(i); + } + return array; + } + + /** + * 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 bigrams a list of bigrams, or null. + */ + public void add(String word, int frequency, ArrayList<WeightedString> bigrams) { + if (null != bigrams) { + for (WeightedString bigram : bigrams) { + final CharGroup t = findWordInTree(mRoot, bigram.mWord); + if (null == t) { + add(getCodePoints(bigram.mWord), 0, null); + } + } + } + add(getCodePoints(word), frequency, bigrams); + } + + /** + * 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; + } + } + + /** + * Add a word to this dictionary. + * + * The 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 bigrams an optional list of bigrams for this word (null if none). + */ + private void add(int[] word, int frequency, ArrayList<WeightedString> bigrams) { + 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), bigrams, frequency); + 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, + bigrams, frequency, currentGroup.mChildren); + 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), + bigrams, frequency); + currentGroup.mChildren = new Node(); + currentGroup.mChildren.mData.add(newNode); + } + } else { + if (0 == differentCharIndex) { + // Exact same word. Check the frequency is 0 or -1, 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.mBigrams, frequency); + 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.mBigrams, currentGroup.mFrequency, currentGroup.mChildren); + newChildren.mData.add(newOldWord); + + final CharGroup newParent; + if (charIndex + differentCharIndex >= word.length) { + newParent = new CharGroup( + Arrays.copyOfRange(currentGroup.mChars, 0, differentCharIndex), + bigrams, frequency, newChildren); + } else { + newParent = new CharGroup( + Arrays.copyOfRange(currentGroup.mChars, 0, differentCharIndex), + null, -1, newChildren); + final CharGroup newWord = new CharGroup( + Arrays.copyOfRange(word, charIndex + differentCharIndex, + word.length), bigrams, frequency); + 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 { + public int compare(Object o1, Object o2) { + final CharGroup c1 = (CharGroup)o1; + final CharGroup c2 = (CharGroup)o2; + if (c1.mChars[0] == c2.mChars[0]) return 0; + return c1.mChars[0] < c2.mChars[0] ? -1 : 1; + } + public boolean equals(Object o) { + return o instanceof CharGroupComparator; + } + } + 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 List data = node.mData; + final CharGroup reference = new CharGroup(new int[] { character }, null, 0); + 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; + } + + /** + * 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.mBigrams); + } 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/tools/makedict2/src/com/android/inputmethod/latin/MakedictLog.java b/tools/makedict2/src/com/android/inputmethod/latin/MakedictLog.java new file mode 100644 index 000000000..badb2ff3d --- /dev/null +++ b/tools/makedict2/src/com/android/inputmethod/latin/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; + +/** + * 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/tools/makedict2/src/com/android/inputmethod/latin/PendingAttribute.java b/tools/makedict2/src/com/android/inputmethod/latin/PendingAttribute.java new file mode 100644 index 000000000..e50202184 --- /dev/null +++ b/tools/makedict2/src/com/android/inputmethod/latin/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; + +/** + * A not-yet-resolved attribute. + * + * An attribute is either a bigram or an 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/tools/makedict2/src/com/android/inputmethod/latin/UnsupportedFormatException.java b/tools/makedict2/src/com/android/inputmethod/latin/UnsupportedFormatException.java new file mode 100644 index 000000000..511ad5675 --- /dev/null +++ b/tools/makedict2/src/com/android/inputmethod/latin/UnsupportedFormatException.java @@ -0,0 +1,26 @@ +/* + * 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; + +/** + * Simple exception thrown when a file format is not recognized. + */ +public class UnsupportedFormatException extends Exception { + public UnsupportedFormatException(String description) { + super(description); + } +} diff --git a/tools/makedict2/src/com/android/inputmethod/latin/Word.java b/tools/makedict2/src/com/android/inputmethod/latin/Word.java new file mode 100644 index 000000000..916165a41 --- /dev/null +++ b/tools/makedict2/src/com/android/inputmethod/latin/Word.java @@ -0,0 +1,65 @@ +/* + * 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; + +import com.android.inputmethod.latin.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 ArrayList<WeightedString> mBigrams; + + public Word(String word, int frequency, ArrayList<WeightedString> bigrams) { + mWord = word; + mFrequency = frequency; + mBigrams = bigrams; + } + + /** + * 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) + && mBigrams.equals(w.mBigrams); + } +} diff --git a/tools/makedict2/src/com/android/inputmethod/latin/XmlDictInputOutput.java b/tools/makedict2/src/com/android/inputmethod/latin/XmlDictInputOutput.java new file mode 100644 index 000000000..096bfd182 --- /dev/null +++ b/tools/makedict2/src/com/android/inputmethod/latin/XmlDictInputOutput.java @@ -0,0 +1,215 @@ +/* + * 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; + +import com.android.inputmethod.latin.FusionDictionary.WeightedString; + +import java.io.IOException; +import java.io.InputStream; +import java.io.Writer; +import java.util.ArrayList; +import java.util.HashMap; +import java.util.TreeSet; + +import javax.xml.parsers.ParserConfigurationException; +import javax.xml.parsers.SAXParser; +import javax.xml.parsers.SAXParserFactory; + +import org.xml.sax.Attributes; +import org.xml.sax.SAXException; +import org.xml.sax.helpers.DefaultHandler; + +/** + * Reads and writes XML files for a FusionDictionary. + * + * All functions in this class are static. + */ +public class XmlDictInputOutput { + + private static final String WORD_TAG = "w"; + private static final String BIGRAM_TAG = "bigram"; + private static final String FREQUENCY_ATTR = "f"; + private static final String WORD_ATTR = "word"; + + /** + * SAX handler for a unigram XML file. + */ + static private class UnigramHandler extends DefaultHandler { + // Parser states + private static final int NONE = 0; + private static final int START = 1; + private static final int WORD = 2; + private static final int BIGRAM = 4; + private static final int END = 5; + private static final int UNKNOWN = 6; + + final FusionDictionary mDictionary; + int mState; // the state of the parser + int mFreq; // the currently read freq + final HashMap<String, ArrayList<WeightedString>> mBigramsMap; + + /** + * Create the handler. + * + * @param dict the dictionary to construct. + * @param bigrams the bigrams as a map. This may be empty, but may not be null. + */ + public UnigramHandler(FusionDictionary dict, + HashMap<String, ArrayList<WeightedString>> bigrams) { + mDictionary = dict; + mBigramsMap = bigrams; + mState = START; + mFreq = 0; + } + + @Override + public void startElement(String uri, String localName, String qName, Attributes attrs) { + if (WORD_TAG.equals(localName)) { + mState = WORD; + for (int attrIndex = 0; attrIndex < attrs.getLength(); ++attrIndex) { + final String attrName = attrs.getLocalName(attrIndex); + if (FREQUENCY_ATTR.equals(attrName)) { + mFreq = Integer.parseInt(attrs.getValue(attrIndex)); + } + } + } else { + mState = UNKNOWN; + } + } + + @Override + public void characters(char[] ch, int start, int length) { + if (WORD == mState) { + final String word = String.copyValueOf(ch, start, length); + mDictionary.add(word, mFreq, mBigramsMap.get(word)); + } + } + + @Override + public void endElement(String uri, String localName, String qName) { + if (WORD == mState) mState = START; + } + } + + /** + * SAX handler for a bigram XML file. + */ + static private class BigramHandler extends DefaultHandler { + private final static String BIGRAM_W1_TAG = "bi"; + private final static String BIGRAM_W2_TAG = "w"; + private final static String BIGRAM_W1_ATTRIBUTE = "w1"; + private final static String BIGRAM_W2_ATTRIBUTE = "w2"; + private final static String BIGRAM_FREQ_ATTRIBUTE = "p"; + + String mW1; + final HashMap<String, ArrayList<WeightedString>> mBigramsMap; + + public BigramHandler() { + mW1 = null; + mBigramsMap = new HashMap<String, ArrayList<WeightedString>>(); + } + + @Override + public void startElement(String uri, String localName, String qName, Attributes attrs) { + if (BIGRAM_W1_TAG.equals(localName)) { + mW1 = attrs.getValue(uri, BIGRAM_W1_ATTRIBUTE); + } else if (BIGRAM_W2_TAG.equals(localName)) { + String w2 = attrs.getValue(uri, BIGRAM_W2_ATTRIBUTE); + int freq = Integer.parseInt(attrs.getValue(uri, BIGRAM_FREQ_ATTRIBUTE)); + WeightedString bigram = new WeightedString(w2, freq / 8); + ArrayList<WeightedString> bigramList = mBigramsMap.get(mW1); + if (null == bigramList) bigramList = new ArrayList<WeightedString>(); + bigramList.add(bigram); + mBigramsMap.put(mW1, bigramList); + } + } + + public HashMap<String, ArrayList<WeightedString>> getBigramMap() { + return mBigramsMap; + } + } + + /** + * Reads a dictionary from an XML file. + * + * This is the public method that will parse an XML file and return the corresponding memory + * representation. + * + * @param unigrams the file to read the data from. + * @return the in-memory representation of the dictionary. + */ + public static FusionDictionary readDictionaryXml(InputStream unigrams, InputStream bigrams) + throws SAXException, IOException, ParserConfigurationException { + final SAXParserFactory factory = SAXParserFactory.newInstance(); + factory.setNamespaceAware(true); + final SAXParser parser = factory.newSAXParser(); + final BigramHandler bigramHandler = new BigramHandler(); + if (null != bigrams) parser.parse(bigrams, bigramHandler); + + final FusionDictionary dict = new FusionDictionary(); + final UnigramHandler unigramHandler = + new UnigramHandler(dict, bigramHandler.getBigramMap()); + parser.parse(unigrams, unigramHandler); + return dict; + } + + /** + * Reads a dictionary in the first, legacy XML format + * + * This method reads data from the parser and creates a new FusionDictionary with it. + * The format parsed by this method is the format used before Ice Cream Sandwich, + * which has no support for bigrams or shortcuts. + * It is important to note that this method expects the parser to have already eaten + * the first, all-encompassing tag. + * + * @param xpp the parser to read the data from. + * @return the parsed dictionary. + */ + + /** + * Writes a dictionary to an XML file. + * + * The output format is the "second" format, which supports bigrams and shortcuts. + * + * @param destination a destination stream to write to. + * @param dict the dictionary to write. + */ + public static void writeDictionaryXml(Writer destination, FusionDictionary dict) + throws IOException { + final TreeSet<Word> set = new TreeSet<Word>(); + for (Word word : dict) { + set.add(word); + } + // TODO: use an XMLSerializer if this gets big + destination.write("<wordlist format=\"2\">\n"); + for (Word word : set) { + destination.write(" <" + WORD_TAG + " " + WORD_ATTR + "=\"" + word.mWord + "\" " + + FREQUENCY_ATTR + "=\"" + word.mFrequency + "\">"); + if (null != word.mBigrams) { + destination.write("\n"); + for (WeightedString bigram : word.mBigrams) { + destination.write(" <" + BIGRAM_TAG + " " + FREQUENCY_ATTR + "=\"" + + bigram.mFrequency + "\">" + bigram.mWord + "</" + BIGRAM_TAG + ">\n"); + } + destination.write(" "); + } + destination.write("</" + WORD_TAG + ">\n"); + } + destination.write("</wordlist>\n"); + destination.close(); + } +} diff --git a/tools/makedict2/tests/com/android/inputmethod/latin/BinaryDictInputOutputTest.java b/tools/makedict2/tests/com/android/inputmethod/latin/BinaryDictInputOutputTest.java new file mode 100644 index 000000000..79cf14b2b --- /dev/null +++ b/tools/makedict2/tests/com/android/inputmethod/latin/BinaryDictInputOutputTest.java @@ -0,0 +1,55 @@ +/* + * 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; + +import com.android.inputmethod.latin.FusionDictionary.Node; + +import java.util.ArrayList; + +import junit.framework.TestCase; + +/** + * Unit tests for BinaryDictInputOutput. + */ +public class BinaryDictInputOutputTest extends TestCase { + + public void setUp() throws Exception { + super.setUp(); + } + + public void tearDown() throws Exception { + super.tearDown(); + } + + // Test the flattened array contains the expected number of nodes, and + // that it does not contain any duplicates. + public void testFlattenNodes() { + final FusionDictionary dict = new FusionDictionary(); + dict.add("foo", 1, null); + dict.add("fta", 1, null); + dict.add("ftb", 1, null); + dict.add("bar", 1, null); + dict.add("fool", 1, null); + final ArrayList<Node> result = BinaryDictInputOutput.flattenTree(dict.mRoot); + assertEquals(4, result.size()); + while (!result.isEmpty()) { + final Node n = result.remove(0); + assertFalse("Flattened array contained the same node twice", result.contains(n)); + } + } + +} |