Home | History | Annotate | Download | only in latin
      1 /*
      2  * Copyright (C) 2011 The Android Open Source Project
      3  *
      4  * Licensed under the Apache License, Version 2.0 (the "License"); you may not
      5  * use this file except in compliance with the License. You may obtain a copy of
      6  * the License at
      7  *
      8  * http://www.apache.org/licenses/LICENSE-2.0
      9  *
     10  * Unless required by applicable law or agreed to in writing, software
     11  * distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
     12  * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
     13  * License for the specific language governing permissions and limitations under
     14  * the License.
     15  */
     16 
     17 package com.android.inputmethod.latin;
     18 
     19 import com.android.inputmethod.latin.FusionDictionary.CharGroup;
     20 import com.android.inputmethod.latin.FusionDictionary.Node;
     21 import com.android.inputmethod.latin.FusionDictionary.WeightedString;
     22 
     23 import java.io.FileNotFoundException;
     24 import java.io.IOException;
     25 import java.io.OutputStream;
     26 import java.io.RandomAccessFile;
     27 import java.util.ArrayList;
     28 import java.util.Arrays;
     29 import java.util.Map;
     30 import java.util.TreeMap;
     31 
     32 /**
     33  * Reads and writes XML files for a FusionDictionary.
     34  *
     35  * All the methods in this class are static.
     36  */
     37 public class BinaryDictInputOutput {
     38 
     39     /* Node layout is as follows:
     40      *   | addressType                         xx     : mask with MASK_GROUP_ADDRESS_TYPE
     41      *                                 2 bits, 00 = no children : FLAG_GROUP_ADDRESS_TYPE_NOADDRESS
     42      * f |                                     01 = 1 byte      : FLAG_GROUP_ADDRESS_TYPE_ONEBYTE
     43      * l |                                     10 = 2 bytes     : FLAG_GROUP_ADDRESS_TYPE_TWOBYTES
     44      * a |                                     11 = 3 bytes     : FLAG_GROUP_ADDRESS_TYPE_THREEBYTES
     45      * g | has several chars ?         1 bit, 1 = yes, 0 = no   : FLAG_HAS_MULTIPLE_CHARS
     46      * s | has a terminal ?            1 bit, 1 = yes, 0 = no   : FLAG_IS_TERMINAL
     47      *   | reserved                    1 bit, 1 = yes, 0 = no
     48      *   | has bigrams ?               1 bit, 1 = yes, 0 = no   : FLAG_HAS_BIGRAMS
     49      *
     50      * c | IF FLAG_HAS_MULTIPLE_CHARS
     51      * h |   char, char, char, char    n * (1 or 3 bytes) : use CharGroupInfo for i/o helpers
     52      * a |   end                       1 byte, = 0
     53      * r | ELSE
     54      * s |   char                      1 or 3 bytes
     55      *   | END
     56      *
     57      * f |
     58      * r | IF FLAG_IS_TERMINAL
     59      * e |   frequency                 1 byte
     60      * q |
     61      *
     62      * c | IF 00 = FLAG_GROUP_ADDRESS_TYPE_NOADDRESS = addressType
     63      * h |   // nothing
     64      * i | ELSIF 01 = FLAG_GROUP_ADDRESS_TYPE_ONEBYTE == addressType
     65      * l |   children address, 1 byte
     66      * d | ELSIF 10 = FLAG_GROUP_ADDRESS_TYPE_TWOBYTES == addressType
     67      * r |   children address, 2 bytes
     68      * e | ELSE // 11 = FLAG_GROUP_ADDRESS_TYPE_THREEBYTES = addressType
     69      * n |   children address, 3 bytes
     70      * A | END
     71      * d
     72      * dress
     73      *
     74      *   | IF FLAG_IS_TERMINAL && FLAG_HAS_BIGRAMS
     75      *   | bigrams address list
     76      *
     77      * Char format is:
     78      * 1 byte = bbbbbbbb match
     79      * case 000xxxxx: xxxxx << 16 + next byte << 8 + next byte
     80      * else: if 00011111 (= 0x1F) : this is the terminator. This is a relevant choice because
     81      *       unicode code points range from 0 to 0x10FFFF, so any 3-byte value starting with
     82      *       00011111 would be outside unicode.
     83      * else: iso-latin-1 code
     84      * This allows for the whole unicode range to be encoded, including chars outside of
     85      * the BMP. Also everything in the iso-latin-1 charset is only 1 byte, except control
     86      * characters which should never happen anyway (and still work, but take 3 bytes).
     87      *
     88      * bigram and shortcut address list is:
     89      * <flags> = | hasNext = 1 bit, 1 = yes, 0 = no     : FLAG_ATTRIBUTE_HAS_NEXT
     90      *           | addressSign = 1 bit,                 : FLAG_ATTRIBUTE_OFFSET_NEGATIVE
     91      *           |                      1 = must take -address, 0 = must take +address
     92      *           |                         xx : mask with MASK_ATTRIBUTE_ADDRESS_TYPE
     93      *           | addressFormat = 2 bits, 00 = unused  : FLAG_ATTRIBUTE_ADDRESS_TYPE_ONEBYTE
     94      *           |                         01 = 1 byte  : FLAG_ATTRIBUTE_ADDRESS_TYPE_ONEBYTE
     95      *           |                         10 = 2 bytes : FLAG_ATTRIBUTE_ADDRESS_TYPE_TWOBYTES
     96      *           |                         11 = 3 bytes : FLAG_ATTRIBUTE_ADDRESS_TYPE_THREEBYTES
     97      *           | 4 bits : frequency         : mask with FLAG_ATTRIBUTE_FREQUENCY
     98      * <address> | IF (01 == FLAG_ATTRIBUTE_ADDRESS_TYPE_ONEBYTE == addressFormat)
     99      *           |   read 1 byte, add top 4 bits
    100      *           | ELSIF (10 == FLAG_ATTRIBUTE_ADDRESS_TYPE_TWOBYTES == addressFormat)
    101      *           |   read 2 bytes, add top 4 bits
    102      *           | ELSE // 11 == FLAG_ATTRIBUTE_ADDRESS_TYPE_THREEBYTES == addressFormat
    103      *           |   read 3 bytes, add top 4 bits
    104      *           | END
    105      *           | if (FLAG_ATTRIBUTE_OFFSET_NEGATIVE) then address = -address
    106      * if (FLAG_ATTRIBUTE_HAS_NET) goto bigram_and_shortcut_address_list_is
    107      *
    108      */
    109 
    110     private static final int MAGIC_NUMBER = 0x78B1;
    111     private static final int VERSION = 1;
    112     private static final int MAXIMUM_SUPPORTED_VERSION = VERSION;
    113     // No options yet, reserved for future use.
    114     private static final int OPTIONS = 0;
    115 
    116     // TODO: Make this value adaptative to content data, store it in the header, and
    117     // use it in the reading code.
    118     private static final int MAX_WORD_LENGTH = 48;
    119 
    120     private static final int MASK_GROUP_ADDRESS_TYPE = 0xC0;
    121     private static final int FLAG_GROUP_ADDRESS_TYPE_NOADDRESS = 0x00;
    122     private static final int FLAG_GROUP_ADDRESS_TYPE_ONEBYTE = 0x40;
    123     private static final int FLAG_GROUP_ADDRESS_TYPE_TWOBYTES = 0x80;
    124     private static final int FLAG_GROUP_ADDRESS_TYPE_THREEBYTES = 0xC0;
    125 
    126     private static final int FLAG_HAS_MULTIPLE_CHARS = 0x20;
    127 
    128     private static final int FLAG_IS_TERMINAL = 0x10;
    129     private static final int FLAG_HAS_BIGRAMS = 0x04;
    130 
    131     private static final int FLAG_ATTRIBUTE_HAS_NEXT = 0x80;
    132     private static final int FLAG_ATTRIBUTE_OFFSET_NEGATIVE = 0x40;
    133     private static final int MASK_ATTRIBUTE_ADDRESS_TYPE = 0x30;
    134     private static final int FLAG_ATTRIBUTE_ADDRESS_TYPE_ONEBYTE = 0x10;
    135     private static final int FLAG_ATTRIBUTE_ADDRESS_TYPE_TWOBYTES = 0x20;
    136     private static final int FLAG_ATTRIBUTE_ADDRESS_TYPE_THREEBYTES = 0x30;
    137     private static final int FLAG_ATTRIBUTE_FREQUENCY = 0x0F;
    138 
    139     private static final int GROUP_CHARACTERS_TERMINATOR = 0x1F;
    140 
    141     private static final int GROUP_COUNT_SIZE = 1;
    142     private static final int GROUP_TERMINATOR_SIZE = 1;
    143     private static final int GROUP_FLAGS_SIZE = 1;
    144     private static final int GROUP_FREQUENCY_SIZE = 1;
    145     private static final int GROUP_MAX_ADDRESS_SIZE = 3;
    146     private static final int GROUP_ATTRIBUTE_FLAGS_SIZE = 1;
    147     private static final int GROUP_ATTRIBUTE_MAX_ADDRESS_SIZE = 3;
    148 
    149     private static final int NO_CHILDREN_ADDRESS = Integer.MIN_VALUE;
    150     private static final int INVALID_CHARACTER = -1;
    151 
    152     // Limiting to 127 for upward compatibility
    153     // TODO: implement a scheme to be able to shoot 256 chargroups in a node
    154     private static final int MAX_CHARGROUPS_IN_A_NODE = 127;
    155 
    156     private static final int MAX_TERMINAL_FREQUENCY = 255;
    157 
    158     /**
    159      * A class grouping utility function for our specific character encoding.
    160      */
    161     private static class CharEncoding {
    162 
    163         private static final int MINIMAL_ONE_BYTE_CHARACTER_VALUE = 0x20;
    164         private static final int MAXIMAL_ONE_BYTE_CHARACTER_VALUE = 0xFF;
    165 
    166         /**
    167          * Helper method to find out whether this code fits on one byte
    168          */
    169         private static boolean fitsOnOneByte(int character) {
    170             return character >= MINIMAL_ONE_BYTE_CHARACTER_VALUE
    171                     && character <= MAXIMAL_ONE_BYTE_CHARACTER_VALUE;
    172         }
    173 
    174         /**
    175          * Compute the size of a character given its character code.
    176          *
    177          * Char format is:
    178          * 1 byte = bbbbbbbb match
    179          * case 000xxxxx: xxxxx << 16 + next byte << 8 + next byte
    180          * else: if 00011111 (= 0x1F) : this is the terminator. This is a relevant choice because
    181          *       unicode code points range from 0 to 0x10FFFF, so any 3-byte value starting with
    182          *       00011111 would be outside unicode.
    183          * else: iso-latin-1 code
    184          * This allows for the whole unicode range to be encoded, including chars outside of
    185          * the BMP. Also everything in the iso-latin-1 charset is only 1 byte, except control
    186          * characters which should never happen anyway (and still work, but take 3 bytes).
    187          *
    188          * @param character the character code.
    189          * @return the size in binary encoded-form, either 1 or 3 bytes.
    190          */
    191         private static int getCharSize(int character) {
    192             // See char encoding in FusionDictionary.java
    193             if (fitsOnOneByte(character)) return 1;
    194             if (INVALID_CHARACTER == character) return 1;
    195             return 3;
    196         }
    197 
    198         /**
    199          * Compute the byte size of a character array.
    200          */
    201         private static int getCharArraySize(final int[] chars) {
    202             int size = 0;
    203             for (int character : chars) size += getCharSize(character);
    204             return size;
    205         }
    206 
    207         /**
    208          * Writes a char array to a byte buffer.
    209          *
    210          * @param characters the character array to write.
    211          * @param buffer the byte buffer to write to.
    212          * @param index the index in buffer to write the character array to.
    213          * @return the index after the last character.
    214          */
    215         private static int writeCharArray(int[] characters, byte[] buffer, int index) {
    216             for (int character : characters) {
    217                 if (1 == getCharSize(character)) {
    218                     buffer[index++] = (byte)character;
    219                 } else {
    220                     buffer[index++] = (byte)(0xFF & (character >> 16));
    221                     buffer[index++] = (byte)(0xFF & (character >> 8));
    222                     buffer[index++] = (byte)(0xFF & character);
    223                 }
    224             }
    225             return index;
    226         }
    227 
    228         /**
    229          * Reads a character from the file.
    230          *
    231          * This follows the character format documented earlier in this source file.
    232          *
    233          * @param source the file, positioned over an encoded character.
    234          * @return the character code.
    235          */
    236         private static int readChar(RandomAccessFile source) throws IOException {
    237             int character = source.readUnsignedByte();
    238             if (!fitsOnOneByte(character)) {
    239                 if (GROUP_CHARACTERS_TERMINATOR == character)
    240                     return INVALID_CHARACTER;
    241                 character <<= 16;
    242                 character += source.readUnsignedShort();
    243             }
    244             return character;
    245         }
    246     }
    247 
    248     /**
    249      * Compute the binary size of the character array in a group
    250      *
    251      * If only one character, this is the size of this character. If many, it's the sum of their
    252      * sizes + 1 byte for the terminator.
    253      *
    254      * @param group the group
    255      * @return the size of the char array, including the terminator if any
    256      */
    257     private static int getGroupCharactersSize(CharGroup group) {
    258         int size = CharEncoding.getCharArraySize(group.mChars);
    259         if (group.hasSeveralChars()) size += GROUP_TERMINATOR_SIZE;
    260         return size;
    261     }
    262 
    263     /**
    264      * Compute the maximum size of a CharGroup, assuming 3-byte addresses for everything.
    265      *
    266      * @param group the CharGroup to compute the size of.
    267      * @return the maximum size of the group.
    268      */
    269     private static int getCharGroupMaximumSize(CharGroup group) {
    270         int size = getGroupCharactersSize(group) + GROUP_FLAGS_SIZE;
    271         // If terminal, one byte for the frequency
    272         if (group.isTerminal()) size += GROUP_FREQUENCY_SIZE;
    273         size += GROUP_MAX_ADDRESS_SIZE; // For children address
    274         if (null != group.mBigrams) {
    275             for (WeightedString bigram : group.mBigrams) {
    276                 size += GROUP_ATTRIBUTE_FLAGS_SIZE + GROUP_ATTRIBUTE_MAX_ADDRESS_SIZE;
    277             }
    278         }
    279         return size;
    280     }
    281 
    282     /**
    283      * Compute the maximum size of a node, assuming 3-byte addresses for everything, and caches
    284      * it in the 'actualSize' member of the node.
    285      *
    286      * @param node the node to compute the maximum size of.
    287      */
    288     private static void setNodeMaximumSize(Node node) {
    289         int size = GROUP_COUNT_SIZE;
    290         for (CharGroup g : node.mData) {
    291             final int groupSize = getCharGroupMaximumSize(g);
    292             g.mCachedSize = groupSize;
    293             size += groupSize;
    294         }
    295         node.mCachedSize = size;
    296     }
    297 
    298     /**
    299      * Helper method to hide the actual value of the no children address.
    300      */
    301     private static boolean hasChildrenAddress(int address) {
    302         return NO_CHILDREN_ADDRESS != address;
    303     }
    304 
    305     /**
    306      * Compute the size, in bytes, that an address will occupy.
    307      *
    308      * This can be used either for children addresses (which are always positive) or for
    309      * attribute, which may be positive or negative but
    310      * store their sign bit separately.
    311      *
    312      * @param address the address
    313      * @return the byte size.
    314      */
    315     private static int getByteSize(int address) {
    316         assert(address < 0x1000000);
    317         if (!hasChildrenAddress(address)) {
    318             return 0;
    319         } else if (Math.abs(address) < 0x100) {
    320             return 1;
    321         } else if (Math.abs(address) < 0x10000) {
    322             return 2;
    323         } else {
    324             return 3;
    325         }
    326     }
    327     // End utility methods.
    328 
    329     // This method is responsible for finding a nice ordering of the nodes that favors run-time
    330     // cache performance and dictionary size.
    331     /* package for tests */ static ArrayList<Node> flattenTree(Node root) {
    332         final int treeSize = FusionDictionary.countCharGroups(root);
    333         MakedictLog.i("Counted nodes : " + treeSize);
    334         final ArrayList<Node> flatTree = new ArrayList<Node>(treeSize);
    335         return flattenTreeInner(flatTree, root);
    336     }
    337 
    338     private static ArrayList<Node> flattenTreeInner(ArrayList<Node> list, Node node) {
    339         // Removing the node is necessary if the tails are merged, because we would then
    340         // add the same node several times when we only want it once. A number of places in
    341         // the code also depends on any node being only once in the list.
    342         // Merging tails can only be done if there are no attributes. Searching for attributes
    343         // in LatinIME code depends on a total breadth-first ordering, which merging tails
    344         // breaks. If there are no attributes, it should be fine (and reduce the file size)
    345         // to merge tails, and the following step would be necessary.
    346         // If eventually the code runs on Android, searching through the whole array each time
    347         // may be a performance concern.
    348         list.remove(node);
    349         list.add(node);
    350         final ArrayList<CharGroup> branches = node.mData;
    351         final int nodeSize = branches.size();
    352         for (CharGroup group : branches) {
    353             if (null != group.mChildren) flattenTreeInner(list, group.mChildren);
    354         }
    355         return list;
    356     }
    357 
    358     /**
    359      * Finds the absolute address of a word in the dictionary.
    360      *
    361      * @param dict the dictionary in which to search.
    362      * @param word the word we are searching for.
    363      * @return the word address. If it is not found, an exception is thrown.
    364      */
    365     private static int findAddressOfWord(final FusionDictionary dict, final String word) {
    366         return FusionDictionary.findWordInTree(dict.mRoot, word).mCachedAddress;
    367     }
    368 
    369     /**
    370      * Computes the actual node size, based on the cached addresses of the children nodes.
    371      *
    372      * Each node stores its tentative address. During dictionary address computing, these
    373      * are not final, but they can be used to compute the node size (the node size depends
    374      * on the address of the children because the number of bytes necessary to store an
    375      * address depends on its numeric value.
    376      *
    377      * @param node the node to compute the size of.
    378      * @param dict the dictionary in which the word/attributes are to be found.
    379      */
    380     private static void computeActualNodeSize(Node node, FusionDictionary dict) {
    381         int size = GROUP_COUNT_SIZE;
    382         for (CharGroup group : node.mData) {
    383             int groupSize = GROUP_FLAGS_SIZE + getGroupCharactersSize(group);
    384             if (group.isTerminal()) groupSize += GROUP_FREQUENCY_SIZE;
    385             if (null != group.mChildren) {
    386                 final int offsetBasePoint= groupSize + node.mCachedAddress + size;
    387                 final int offset = group.mChildren.mCachedAddress - offsetBasePoint;
    388                 groupSize += getByteSize(offset);
    389             }
    390             if (null != group.mBigrams) {
    391                 for (WeightedString bigram : group.mBigrams) {
    392                     final int offsetBasePoint = groupSize + node.mCachedAddress + size
    393                             + GROUP_FLAGS_SIZE;
    394                     final int addressOfBigram = findAddressOfWord(dict, bigram.mWord);
    395                     final int offset = addressOfBigram - offsetBasePoint;
    396                     groupSize += getByteSize(offset) + GROUP_FLAGS_SIZE;
    397                 }
    398             }
    399             group.mCachedSize = groupSize;
    400             size += groupSize;
    401         }
    402         node.mCachedSize = size;
    403     }
    404 
    405     /**
    406      * Computes the byte size of a list of nodes and updates each node cached position.
    407      *
    408      * @param flatNodes the array of nodes.
    409      * @return the byte size of the entire stack.
    410      */
    411     private static int stackNodes(ArrayList<Node> flatNodes) {
    412         int nodeOffset = 0;
    413         for (Node n : flatNodes) {
    414             n.mCachedAddress = nodeOffset;
    415             int groupOffset = 0;
    416             for (CharGroup g : n.mData) {
    417                 g.mCachedAddress = GROUP_COUNT_SIZE + nodeOffset + groupOffset;
    418                 groupOffset += g.mCachedSize;
    419             }
    420             if (groupOffset + GROUP_COUNT_SIZE != n.mCachedSize) {
    421                 throw new RuntimeException("Bug : Stored and computed node size differ");
    422             }
    423             nodeOffset += n.mCachedSize;
    424         }
    425         return nodeOffset;
    426     }
    427 
    428     /**
    429      * Compute the addresses and sizes of an ordered node array.
    430      *
    431      * This method takes a node array and will update its cached address and size values
    432      * so that they can be written into a file. It determines the smallest size each of the
    433      * nodes can be given the addresses of its children and attributes, and store that into
    434      * each node.
    435      * The order of the node is given by the order of the array. This method makes no effort
    436      * to find a good order; it only mechanically computes the size this order results in.
    437      *
    438      * @param dict the dictionary
    439      * @param flatNodes the ordered array of nodes
    440      * @return the same array it was passed. The nodes have been updated for address and size.
    441      */
    442     private static ArrayList<Node> computeAddresses(FusionDictionary dict,
    443             ArrayList<Node> flatNodes) {
    444         // First get the worst sizes and offsets
    445         for (Node n : flatNodes) setNodeMaximumSize(n);
    446         final int offset = stackNodes(flatNodes);
    447 
    448         MakedictLog.i("Compressing the array addresses. Original size : " + offset);
    449         MakedictLog.i("(Recursively seen size : " + offset + ")");
    450 
    451         int passes = 0;
    452         boolean changesDone = false;
    453         do {
    454             changesDone = false;
    455             for (Node n : flatNodes) {
    456                 final int oldNodeSize = n.mCachedSize;
    457                 computeActualNodeSize(n, dict);
    458                 final int newNodeSize = n.mCachedSize;
    459                 if (oldNodeSize < newNodeSize) throw new RuntimeException("Increased size ?!");
    460                 if (oldNodeSize != newNodeSize) changesDone = true;
    461             }
    462             stackNodes(flatNodes);
    463             ++passes;
    464         } while (changesDone);
    465 
    466         final Node lastNode = flatNodes.get(flatNodes.size() - 1);
    467         MakedictLog.i("Compression complete in " + passes + " passes.");
    468         MakedictLog.i("After address compression : "
    469                 + (lastNode.mCachedAddress + lastNode.mCachedSize));
    470 
    471         return flatNodes;
    472     }
    473 
    474     /**
    475      * Sanity-checking method.
    476      *
    477      * This method checks an array of node for juxtaposition, that is, it will do
    478      * nothing if each node's cached address is actually the previous node's address
    479      * plus the previous node's size.
    480      * If this is not the case, it will throw an exception.
    481      *
    482      * @param array the array node to check
    483      */
    484     private static void checkFlatNodeArray(ArrayList<Node> array) {
    485         int offset = 0;
    486         int index = 0;
    487         for (Node n : array) {
    488             if (n.mCachedAddress != offset) {
    489                 throw new RuntimeException("Wrong address for node " + index
    490                         + " : expected " + offset + ", got " + n.mCachedAddress);
    491             }
    492             ++index;
    493             offset += n.mCachedSize;
    494         }
    495     }
    496 
    497     /**
    498      * Helper method to write a variable-size address to a file.
    499      *
    500      * @param buffer the buffer to write to.
    501      * @param index the index in the buffer to write the address to.
    502      * @param address the address to write.
    503      * @return the size in bytes the address actually took.
    504      */
    505     private static int writeVariableAddress(byte[] buffer, int index, int address) {
    506         switch (getByteSize(address)) {
    507         case 1:
    508             buffer[index++] = (byte)address;
    509             return 1;
    510         case 2:
    511             buffer[index++] = (byte)(0xFF & (address >> 8));
    512             buffer[index++] = (byte)(0xFF & address);
    513             return 2;
    514         case 3:
    515             buffer[index++] = (byte)(0xFF & (address >> 16));
    516             buffer[index++] = (byte)(0xFF & (address >> 8));
    517             buffer[index++] = (byte)(0xFF & address);
    518             return 3;
    519         case 0:
    520             return 0;
    521         default:
    522             throw new RuntimeException("Address " + address + " has a strange size");
    523         }
    524     }
    525 
    526     private static byte makeCharGroupFlags(final CharGroup group, final int groupAddress,
    527             final int childrenOffset) {
    528         byte flags = 0;
    529         if (group.mChars.length > 1) flags |= FLAG_HAS_MULTIPLE_CHARS;
    530         if (group.mFrequency >= 0) {
    531             flags |= FLAG_IS_TERMINAL;
    532         }
    533         if (null != group.mChildren) {
    534             switch (getByteSize(childrenOffset)) {
    535              case 1:
    536                  flags |= FLAG_GROUP_ADDRESS_TYPE_ONEBYTE;
    537                  break;
    538              case 2:
    539                  flags |= FLAG_GROUP_ADDRESS_TYPE_TWOBYTES;
    540                  break;
    541              case 3:
    542                  flags |= FLAG_GROUP_ADDRESS_TYPE_THREEBYTES;
    543                  break;
    544              default:
    545                  throw new RuntimeException("Node with a strange address");
    546              }
    547         }
    548         if (null != group.mBigrams) flags |= FLAG_HAS_BIGRAMS;
    549         return flags;
    550     }
    551 
    552     /**
    553      * Makes the flag value for an attribute.
    554      *
    555      * @param more whether there are more attributes after this one.
    556      * @param offset the offset of the attribute.
    557      * @param frequency the frequency of the attribute, 0..15
    558      * @return the flags
    559      */
    560     private static final int makeAttributeFlags(final boolean more, final int offset,
    561             final int frequency) {
    562         int bigramFlags = (more ? FLAG_ATTRIBUTE_HAS_NEXT : 0)
    563                 + (offset < 0 ? FLAG_ATTRIBUTE_OFFSET_NEGATIVE : 0);
    564         switch (getByteSize(offset)) {
    565         case 1:
    566             bigramFlags |= FLAG_ATTRIBUTE_ADDRESS_TYPE_ONEBYTE;
    567             break;
    568         case 2:
    569             bigramFlags |= FLAG_ATTRIBUTE_ADDRESS_TYPE_TWOBYTES;
    570             break;
    571         case 3:
    572             bigramFlags |= FLAG_ATTRIBUTE_ADDRESS_TYPE_THREEBYTES;
    573             break;
    574         default:
    575             throw new RuntimeException("Strange offset size");
    576         }
    577         bigramFlags += frequency & FLAG_ATTRIBUTE_FREQUENCY;
    578         return bigramFlags;
    579     }
    580 
    581     /**
    582      * Write a node to memory. The node is expected to have its final position cached.
    583      *
    584      * This can be an empty map, but the more is inside the faster the lookups will be. It can
    585      * be carried on as long as nodes do not move.
    586      *
    587      * @param dict the dictionary the node is a part of (for relative offsets).
    588      * @param buffer the memory buffer to write to.
    589      * @param node the node to write.
    590      * @return the address of the END of the node.
    591      */
    592     private static int writePlacedNode(FusionDictionary dict, byte[] buffer, Node node) {
    593         int index = node.mCachedAddress;
    594 
    595         final int size = node.mData.size();
    596         if (size > MAX_CHARGROUPS_IN_A_NODE)
    597             throw new RuntimeException("A node has a group count over 127 (" + size + ").");
    598 
    599         buffer[index++] = (byte)size;
    600         int groupAddress = index;
    601         for (int i = 0; i < size; ++i) {
    602             CharGroup group = node.mData.get(i);
    603             if (index != group.mCachedAddress) throw new RuntimeException("Bug: write index is not "
    604                     + "the same as the cached address of the group");
    605             groupAddress += GROUP_FLAGS_SIZE + getGroupCharactersSize(group);
    606             // Sanity checks.
    607             if (group.mFrequency > MAX_TERMINAL_FREQUENCY) {
    608                 throw new RuntimeException("A node has a frequency > " + MAX_TERMINAL_FREQUENCY
    609                         + " : " + group.mFrequency);
    610             }
    611             if (group.mFrequency >= 0) groupAddress += GROUP_FREQUENCY_SIZE;
    612             final int childrenOffset = null == group.mChildren
    613                     ? NO_CHILDREN_ADDRESS : group.mChildren.mCachedAddress - groupAddress;
    614             byte flags = makeCharGroupFlags(group, groupAddress, childrenOffset);
    615             buffer[index++] = flags;
    616             index = CharEncoding.writeCharArray(group.mChars, buffer, index);
    617             if (group.hasSeveralChars()) {
    618                 buffer[index++] = GROUP_CHARACTERS_TERMINATOR;
    619             }
    620             if (group.mFrequency >= 0) {
    621                 buffer[index++] = (byte) group.mFrequency;
    622             }
    623             final int shift = writeVariableAddress(buffer, index, childrenOffset);
    624             index += shift;
    625             groupAddress += shift;
    626 
    627             // Write bigrams
    628             if (null != group.mBigrams) {
    629                 int remainingBigrams = group.mBigrams.size();
    630                 for (WeightedString bigram : group.mBigrams) {
    631                     boolean more = remainingBigrams > 1;
    632                     final int addressOfBigram = findAddressOfWord(dict, bigram.mWord);
    633                     ++groupAddress;
    634                     final int offset = addressOfBigram - groupAddress;
    635                     int bigramFlags = makeAttributeFlags(more, offset, bigram.mFrequency);
    636                     buffer[index++] = (byte)bigramFlags;
    637                     final int bigramShift = writeVariableAddress(buffer, index, Math.abs(offset));
    638                     index += bigramShift;
    639                     groupAddress += bigramShift;
    640                     --remainingBigrams;
    641                 }
    642             }
    643 
    644         }
    645         if (index != node.mCachedAddress + node.mCachedSize) throw new RuntimeException(
    646                 "Not the same size : written "
    647                 + (index - node.mCachedAddress) + " bytes out of a node that should have "
    648                 + node.mCachedSize + " bytes");
    649         return index;
    650     }
    651 
    652     /**
    653      * Dumps a collection of useful statistics about a node array.
    654      *
    655      * This prints purely informative stuff, like the total estimated file size, the
    656      * number of nodes, of character groups, the repartition of each address size, etc
    657      *
    658      * @param nodes the node array.
    659      */
    660     private static void showStatistics(ArrayList<Node> nodes) {
    661         int firstTerminalAddress = Integer.MAX_VALUE;
    662         int lastTerminalAddress = Integer.MIN_VALUE;
    663         int size = 0;
    664         int charGroups = 0;
    665         int maxGroups = 0;
    666         int maxRuns = 0;
    667         for (Node n : nodes) {
    668             if (maxGroups < n.mData.size()) maxGroups = n.mData.size();
    669             for (CharGroup cg : n.mData) {
    670                 ++charGroups;
    671                 if (cg.mChars.length > maxRuns) maxRuns = cg.mChars.length;
    672                 if (cg.mFrequency >= 0) {
    673                     if (n.mCachedAddress < firstTerminalAddress)
    674                         firstTerminalAddress = n.mCachedAddress;
    675                     if (n.mCachedAddress > lastTerminalAddress)
    676                         lastTerminalAddress = n.mCachedAddress;
    677                 }
    678             }
    679             if (n.mCachedAddress + n.mCachedSize > size) size = n.mCachedAddress + n.mCachedSize;
    680         }
    681         final int[] groupCounts = new int[maxGroups + 1];
    682         final int[] runCounts = new int[maxRuns + 1];
    683         for (Node n : nodes) {
    684             ++groupCounts[n.mData.size()];
    685             for (CharGroup cg : n.mData) {
    686                 ++runCounts[cg.mChars.length];
    687             }
    688         }
    689 
    690         MakedictLog.i("Statistics:\n"
    691                 + "  total file size " + size + "\n"
    692                 + "  " + nodes.size() + " nodes\n"
    693                 + "  " + charGroups + " groups (" + ((float)charGroups / nodes.size())
    694                         + " groups per node)\n"
    695                 + "  first terminal at " + firstTerminalAddress + "\n"
    696                 + "  last terminal at " + lastTerminalAddress + "\n"
    697                 + "  Group stats : max = " + maxGroups);
    698         for (int i = 0; i < groupCounts.length; ++i) {
    699             MakedictLog.i("    " + i + " : " + groupCounts[i]);
    700         }
    701         MakedictLog.i("  Character run stats : max = " + maxRuns);
    702         for (int i = 0; i < runCounts.length; ++i) {
    703             MakedictLog.i("    " + i + " : " + runCounts[i]);
    704         }
    705     }
    706 
    707     /**
    708      * Dumps a FusionDictionary to a file.
    709      *
    710      * This is the public entry point to write a dictionary to a file.
    711      *
    712      * @param destination the stream to write the binary data to.
    713      * @param dict the dictionary to write.
    714      */
    715     public static void writeDictionaryBinary(OutputStream destination, FusionDictionary dict)
    716             throws IOException {
    717 
    718         // Addresses are limited to 3 bytes, so we'll just make a 16MB buffer. Since addresses
    719         // can be relative to each node, the structure itself is not limited to 16MB at all, but
    720         // I doubt this will ever be shot. If it is, deciding the order of the nodes becomes
    721         // a quite complicated problem, because though the dictionary itself does not have a
    722         // size limit, each node must still be within 16MB of all its children and parents.
    723         // As long as this is ensured, the dictionary file may grow to any size.
    724         // Anyway, to make a dictionary bigger than 16MB just increase the size of this buffer.
    725         final byte[] buffer = new byte[1 << 24];
    726         int index = 0;
    727 
    728         // Magic number in big-endian order.
    729         buffer[index++] = (byte) (0xFF & (MAGIC_NUMBER >> 8));
    730         buffer[index++] = (byte) (0xFF & MAGIC_NUMBER);
    731         // Dictionary version.
    732         buffer[index++] = (byte) (0xFF & VERSION);
    733         // Options flags
    734         buffer[index++] = (byte) (0xFF & (OPTIONS >> 8));
    735         buffer[index++] = (byte) (0xFF & OPTIONS);
    736 
    737         // Should we include the locale and title of the dictionary ?
    738 
    739         destination.write(buffer, 0, index);
    740         index = 0;
    741 
    742         // Leave the choice of the optimal node order to the flattenTree function.
    743         MakedictLog.i("Flattening the tree...");
    744         ArrayList<Node> flatNodes = flattenTree(dict.mRoot);
    745 
    746         MakedictLog.i("Computing addresses...");
    747         computeAddresses(dict, flatNodes);
    748         MakedictLog.i("Checking array...");
    749         checkFlatNodeArray(flatNodes);
    750 
    751         MakedictLog.i("Writing file...");
    752         int dataEndOffset = 0;
    753         for (Node n : flatNodes) {
    754             dataEndOffset = writePlacedNode(dict, buffer, n);
    755         }
    756 
    757         showStatistics(flatNodes);
    758 
    759         destination.write(buffer, 0, dataEndOffset);
    760 
    761         destination.close();
    762         MakedictLog.i("Done");
    763     }
    764 
    765 
    766     // Input methods: Read a binary dictionary to memory.
    767     // readDictionaryBinary is the public entry point for them.
    768 
    769     static final int[] characterBuffer = new int[MAX_WORD_LENGTH];
    770     private static CharGroupInfo readCharGroup(RandomAccessFile source,
    771             final int originalGroupAddress) throws IOException {
    772         int addressPointer = originalGroupAddress;
    773         final int flags = source.readUnsignedByte();
    774         ++addressPointer;
    775         final int characters[];
    776         if (0 != (flags & FLAG_HAS_MULTIPLE_CHARS)) {
    777             int index = 0;
    778             int character = CharEncoding.readChar(source);
    779             addressPointer += CharEncoding.getCharSize(character);
    780             while (-1 != character) {
    781                 characterBuffer[index++] = character;
    782                 character = CharEncoding.readChar(source);
    783                 addressPointer += CharEncoding.getCharSize(character);
    784             }
    785             characters = Arrays.copyOfRange(characterBuffer, 0, index);
    786         } else {
    787             final int character = CharEncoding.readChar(source);
    788             addressPointer += CharEncoding.getCharSize(character);
    789             characters = new int[] { character };
    790         }
    791         final int frequency;
    792         if (0 != (FLAG_IS_TERMINAL & flags)) {
    793             ++addressPointer;
    794             frequency = source.readUnsignedByte();
    795         } else {
    796             frequency = CharGroup.NOT_A_TERMINAL;
    797         }
    798         int childrenAddress = addressPointer;
    799         switch (flags & MASK_GROUP_ADDRESS_TYPE) {
    800         case FLAG_GROUP_ADDRESS_TYPE_ONEBYTE:
    801             childrenAddress += source.readUnsignedByte();
    802             addressPointer += 1;
    803             break;
    804         case FLAG_GROUP_ADDRESS_TYPE_TWOBYTES:
    805             childrenAddress += source.readUnsignedShort();
    806             addressPointer += 2;
    807             break;
    808         case FLAG_GROUP_ADDRESS_TYPE_THREEBYTES:
    809             childrenAddress += (source.readUnsignedByte() << 16) + source.readUnsignedShort();
    810             addressPointer += 3;
    811             break;
    812         case FLAG_GROUP_ADDRESS_TYPE_NOADDRESS:
    813         default:
    814             childrenAddress = NO_CHILDREN_ADDRESS;
    815             break;
    816         }
    817         ArrayList<PendingAttribute> bigrams = null;
    818         if (0 != (flags & FLAG_HAS_BIGRAMS)) {
    819             bigrams = new ArrayList<PendingAttribute>();
    820             boolean more = true;
    821             while (more) {
    822                 int bigramFlags = source.readUnsignedByte();
    823                 ++addressPointer;
    824                 more = (0 != (bigramFlags & FLAG_ATTRIBUTE_HAS_NEXT));
    825                 final int sign = 0 == (bigramFlags & FLAG_ATTRIBUTE_OFFSET_NEGATIVE) ? 1 : -1;
    826                 int bigramAddress = addressPointer;
    827                 switch (bigramFlags & MASK_ATTRIBUTE_ADDRESS_TYPE) {
    828                 case FLAG_ATTRIBUTE_ADDRESS_TYPE_ONEBYTE:
    829                     bigramAddress += sign * source.readUnsignedByte();
    830                     addressPointer += 1;
    831                     break;
    832                 case FLAG_ATTRIBUTE_ADDRESS_TYPE_TWOBYTES:
    833                     bigramAddress += sign * source.readUnsignedShort();
    834                     addressPointer += 2;
    835                     break;
    836                 case FLAG_ATTRIBUTE_ADDRESS_TYPE_THREEBYTES:
    837                     final int offset = ((source.readUnsignedByte() << 16)
    838                             + source.readUnsignedShort());
    839                     bigramAddress += sign * offset;
    840                     addressPointer += 3;
    841                     break;
    842                 default:
    843                     throw new RuntimeException("Has attribute with no address");
    844                 }
    845                 bigrams.add(new PendingAttribute(bigramFlags & FLAG_ATTRIBUTE_FREQUENCY,
    846                         bigramAddress));
    847             }
    848         }
    849         return new CharGroupInfo(originalGroupAddress, addressPointer, flags, characters, frequency,
    850                 childrenAddress, bigrams);
    851     }
    852 
    853     /**
    854      * Finds, as a string, the word at the address passed as an argument.
    855      *
    856      * @param source the file to read from.
    857      * @param headerSize the size of the header.
    858      * @param address the address to seek.
    859      * @return the word, as a string.
    860      * @throws IOException if the file can't be read.
    861      */
    862     private static String getWordAtAddress(RandomAccessFile source, long headerSize,
    863             int address) throws IOException {
    864         final long originalPointer = source.getFilePointer();
    865         source.seek(headerSize);
    866         final int count = source.readUnsignedByte();
    867         int groupOffset = 1; // 1 for the group count
    868         final StringBuilder builder = new StringBuilder();
    869         String result = null;
    870 
    871         CharGroupInfo last = null;
    872         for (int i = count - 1; i >= 0; --i) {
    873             CharGroupInfo info = readCharGroup(source, groupOffset);
    874             groupOffset = info.mEndAddress;
    875             if (info.mOriginalAddress == address) {
    876                 builder.append(new String(info.mCharacters, 0, info.mCharacters.length));
    877                 result = builder.toString();
    878                 break; // and return
    879             }
    880             if (hasChildrenAddress(info.mChildrenAddress)) {
    881                 if (info.mChildrenAddress > address) {
    882                     if (null == last) continue;
    883                     builder.append(new String(last.mCharacters, 0, last.mCharacters.length));
    884                     source.seek(last.mChildrenAddress + headerSize);
    885                     groupOffset = last.mChildrenAddress + 1;
    886                     i = source.readUnsignedByte();
    887                     last = null;
    888                     continue;
    889                 }
    890                 last = info;
    891             }
    892             if (0 == i && hasChildrenAddress(last.mChildrenAddress)) {
    893                 builder.append(new String(last.mCharacters, 0, last.mCharacters.length));
    894                 source.seek(last.mChildrenAddress + headerSize);
    895                 groupOffset = last.mChildrenAddress + 1;
    896                 i = source.readUnsignedByte();
    897                 last = null;
    898                 continue;
    899             }
    900         }
    901         source.seek(originalPointer);
    902         return result;
    903     }
    904 
    905     /**
    906      * Reads a single node from a binary file.
    907      *
    908      * This methods reads the file at the current position of its file pointer. A node is
    909      * fully expected to start at the current position.
    910      * This will recursively read other nodes into the structure, populating the reverse
    911      * maps on the fly and using them to keep track of already read nodes.
    912      *
    913      * @param source the data file, correctly positioned at the start of a node.
    914      * @param headerSize the size, in bytes, of the file header.
    915      * @param reverseNodeMap a mapping from addresses to already read nodes.
    916      * @param reverseGroupMap a mapping from addresses to already read character groups.
    917      * @return the read node with all his children already read.
    918      */
    919     private static Node readNode(RandomAccessFile source, long headerSize,
    920             Map<Integer, Node> reverseNodeMap, Map<Integer, CharGroup> reverseGroupMap)
    921             throws IOException {
    922         final int nodeOrigin = (int)(source.getFilePointer() - headerSize);
    923         final int count = source.readUnsignedByte();
    924         final ArrayList<CharGroup> nodeContents = new ArrayList<CharGroup>();
    925         int groupOffset = nodeOrigin + 1; // 1 byte for the group count
    926         for (int i = count; i > 0; --i) {
    927             CharGroupInfo info = readCharGroup(source, groupOffset);
    928             ArrayList<WeightedString> bigrams = null;
    929             if (null != info.mBigrams) {
    930                 bigrams = new ArrayList<WeightedString>();
    931                 for (PendingAttribute bigram : info.mBigrams) {
    932                     final String word = getWordAtAddress(source, headerSize, bigram.mAddress);
    933                     bigrams.add(new WeightedString(word, bigram.mFrequency));
    934                 }
    935             }
    936             if (hasChildrenAddress(info.mChildrenAddress)) {
    937                 Node children = reverseNodeMap.get(info.mChildrenAddress);
    938                 if (null == children) {
    939                     final long currentPosition = source.getFilePointer();
    940                     source.seek(info.mChildrenAddress + headerSize);
    941                     children = readNode(source, headerSize, reverseNodeMap, reverseGroupMap);
    942                     source.seek(currentPosition);
    943                 }
    944                 nodeContents.add(
    945                         new CharGroup(info.mCharacters, bigrams, info.mFrequency,
    946                         children));
    947             } else {
    948                 nodeContents.add(
    949                         new CharGroup(info.mCharacters, bigrams, info.mFrequency));
    950             }
    951             groupOffset = info.mEndAddress;
    952         }
    953         final Node node = new Node(nodeContents);
    954         node.mCachedAddress = nodeOrigin;
    955         reverseNodeMap.put(node.mCachedAddress, node);
    956         return node;
    957     }
    958 
    959     /**
    960      * Reads a random access file and returns the memory representation of the dictionary.
    961      *
    962      * This high-level method takes a binary file and reads its contents, populating a
    963      * FusionDictionary structure. The optional dict argument is an existing dictionary to
    964      * which words from the file should be added. If it is null, a new dictionary is created.
    965      *
    966      * @param source the file to read.
    967      * @param dict an optional dictionary to add words to, or null.
    968      * @return the created (or merged) dictionary.
    969      */
    970     public static FusionDictionary readDictionaryBinary(RandomAccessFile source,
    971             FusionDictionary dict) throws IOException, UnsupportedFormatException {
    972         // Check magic number
    973         final int magic = source.readUnsignedShort();
    974         if (MAGIC_NUMBER != magic) {
    975             throw new UnsupportedFormatException("The magic number in this file does not match "
    976                     + "the expected value");
    977         }
    978 
    979         // Check file version
    980         final int version = source.readUnsignedByte();
    981         if (version > MAXIMUM_SUPPORTED_VERSION) {
    982             throw new UnsupportedFormatException("This file has version " + version
    983                     + ", but this implementation does not support versions above "
    984                     + MAXIMUM_SUPPORTED_VERSION);
    985         }
    986 
    987         // Read options
    988         source.readUnsignedShort();
    989 
    990         long headerSize = source.getFilePointer();
    991         Map<Integer, Node> reverseNodeMapping = new TreeMap<Integer, Node>();
    992         Map<Integer, CharGroup> reverseGroupMapping = new TreeMap<Integer, CharGroup>();
    993         final Node root = readNode(source, headerSize, reverseNodeMapping, reverseGroupMapping);
    994 
    995         FusionDictionary newDict = new FusionDictionary(root,
    996                 new FusionDictionary.DictionaryOptions());
    997         if (null != dict) {
    998             for (Word w : dict) {
    999                 newDict.add(w.mWord, w.mFrequency, w.mBigrams);
   1000             }
   1001         }
   1002 
   1003         return newDict;
   1004     }
   1005 
   1006     /**
   1007      * Basic test to find out whether the file is a binary dictionary or not.
   1008      *
   1009      * Concretely this only tests the magic number.
   1010      *
   1011      * @param filename The name of the file to test.
   1012      * @return true if it's a binary dictionary, false otherwise
   1013      */
   1014     public static boolean isBinaryDictionary(String filename) {
   1015         try {
   1016             RandomAccessFile f = new RandomAccessFile(filename, "r");
   1017             return MAGIC_NUMBER == f.readUnsignedShort();
   1018         } catch (FileNotFoundException e) {
   1019             return false;
   1020         } catch (IOException e) {
   1021             return false;
   1022         }
   1023     }
   1024 }
   1025