src/com/hughes/android/dictionary/engine/WiktionarySplitter.java

   1 // Copyright 2011 Google Inc. All Rights Reserved.
   2 //
   3 // Licensed under the Apache License, Version 2.0 (the "License");
   4 // you may not use this file except in compliance with the License.
   5 // You may obtain a copy of the License at
   6 //
   7 //     http://www.apache.org/licenses/LICENSE-2.0
   8 //
   9 // Unless required by applicable law or agreed to in writing, software
  10 // distributed under the License is distributed on an "AS IS" BASIS,
  11 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  12 // See the License for the specific language governing permissions and
  13 // limitations under the License.
  14
  15 package com.hughes.android.dictionary.engine;
  16
  17 import java.io.BufferedInputStream;
  18 import java.io.BufferedOutputStream;
  19 import java.io.DataOutputStream;
  20 import java.io.File;
  21 import java.io.FileInputStream;
  22 import java.io.FileOutputStream;
  23 import java.io.InputStream;
  24 import java.io.IOException;
  25 import java.io.OutputStream;
  26 import java.util.ArrayList;
  27 import java.util.LinkedHashMap;
  28 import java.util.List;
  29 import java.util.Map;
  30 import java.util.regex.Matcher;
  31 import java.util.regex.Pattern;
  32
  33 import javax.xml.parsers.ParserConfigurationException;
  34 import javax.xml.parsers.SAXParser;
  35
  36 import org.apache.xerces.jaxp.SAXParserFactoryImpl;
  37 import org.apache.commons.compress.compressors.CompressorStreamFactory;
  38 import org.xml.sax.Attributes;
  39 import org.xml.sax.SAXException;
  40
  41 import com.hughes.android.dictionary.parser.wiktionary.WiktionaryLangs;
  42
  43 public class WiktionarySplitter extends org.xml.sax.helpers.DefaultHandler {
  44
  45     // The matches the whole line, otherwise regexes don't work well on French:
  46     // {{=uk=}}
  47     // Spanish has no initial headings, tried to also detect {{ES as such
  48     // with "^(\\{\\{ES|(=+)[^=]).*$" but that broke English.
  49     static final Pattern headingStart = Pattern.compile("^(=+)[^=].*$", Pattern.MULTILINE);
  50
  51     final Map<String,List<Selector>> pathToSelectors = new LinkedHashMap<String, List<Selector>>();
  52     List<Selector> currentSelectors = null;
  53
  54     StringBuilder titleBuilder;
  55     StringBuilder textBuilder;
  56     StringBuilder currentBuilder = null;
  57
  58     public static void main(final String[] args) throws Exception {
  59         final WiktionarySplitter wiktionarySplitter = new WiktionarySplitter();
  60         wiktionarySplitter.go();
  61     }
  62
  63     private WiktionarySplitter() {
  64         List<Selector> selectors;
  65         for (final String code : WiktionaryLangs.wikiCodeToIsoCodeToWikiName.keySet()) {
  66             //if (!code.equals("fr")) {continue;}
  67             selectors = new ArrayList<WiktionarySplitter.Selector>();
  68             pathToSelectors.put(String.format("data/inputs/%swiktionary-pages-articles.xml", code), selectors);
  69             for (final Map.Entry<String, String> entry : WiktionaryLangs.wikiCodeToIsoCodeToWikiName.get(code).entrySet()) {
  70                 final String dir = String.format("data/inputs/wikiSplit/%s", code);
  71                 new File(dir).mkdirs();
  72                 selectors.add(new Selector(String.format("%s/%s.data", dir, entry.getKey()), entry.getValue()));
  73             }
  74         }
  75     }
  76
  77     private void go() throws Exception {
  78         final SAXParser parser = SAXParserFactoryImpl.newInstance().newSAXParser();
  79
  80         // Configure things.
  81         for (final Map.Entry<String, List<Selector>> pathToSelectorsEntry : pathToSelectors.entrySet()) {
  82
  83             currentSelectors = pathToSelectorsEntry.getValue();
  84
  85             for (final Selector selector : currentSelectors) {
  86                 OutputStream tmp = new FileOutputStream(selector.outFilename);
  87                 tmp = new WriteBuffer(tmp, 20 * 1024 * 1024);
  88                 selector.out = new DataOutputStream(tmp);
  89             }
  90
  91             // Do it.
  92             try {
  93                 File input = new File(pathToSelectorsEntry.getKey() + ".bz2");
  94                 if (!input.exists()) input = new File(pathToSelectorsEntry.getKey() + ".gz");
  95                 if (!input.exists()) input = new File(pathToSelectorsEntry.getKey() + ".xz");
  96                 if (!input.exists()) {
  97                     // Fallback to uncompressed file
  98                     parser.parse(new File(pathToSelectorsEntry.getKey()), this);
  99                 } else {
 100                     InputStream compressedIn = new BufferedInputStream(new FileInputStream(input));
 101                     InputStream in = new CompressorStreamFactory().createCompressorInputStream(compressedIn);
 102                     in = new ReadAheadBuffer(in, 20 * 1024 * 1024);
 103                     parser.parse(new BufferedInputStream(in), this);
 104                 }
 105             } catch (Exception e) {
 106                 System.err.println("Exception during parse, lastPageTitle=" + lastPageTitle + ", titleBuilder=" + titleBuilder.toString());
 107                 throw e;
 108             }
 109
 110             // Shutdown.
 111             for (final Selector selector : currentSelectors) {
 112                 selector.out.close();
 113             }
 114
 115         }
 116     }
 117
 118     String lastPageTitle = null;
 119     int pageCount = 0;
 120     Pattern endPatterns[] = new Pattern[100];
 121
 122     private Pattern getEndPattern(int depth) {
 123         if (endPatterns[depth] == null)
 124             endPatterns[depth] = Pattern.compile(String.format("^={1,%d}[^=].*$", depth), Pattern.MULTILINE);
 125         return endPatterns[depth];
 126     }
 127
 128     private void endPage() {
 129         final String title = titleBuilder.toString();
 130         lastPageTitle = title;
 131         if (++pageCount % 1000 == 0) {
 132             System.out.println("endPage: " + title + ", count=" + pageCount);
 133         }
 134         if (title.startsWith("Wiktionary:") ||
 135                 title.startsWith("Appendix:") ||
 136                 title.startsWith("Help:") ||
 137                 title.startsWith("Index:") ||
 138                 title.startsWith("MediaWiki:") ||
 139                 title.startsWith("Citations:") ||
 140                 title.startsWith("Concordance:") ||
 141                 title.startsWith("Glossary:") ||
 142                 title.startsWith("Rhymes:") ||
 143                 title.startsWith("Category:") ||
 144                 title.startsWith("Wikisaurus:") ||
 145                 title.startsWith("Unsupported titles/") ||
 146                 title.startsWith("Transwiki:") ||
 147                 title.startsWith("File:") ||
 148                 title.startsWith("Thread:") ||
 149                 title.startsWith("Template:") ||
 150                 title.startsWith("Summary:") ||
 151                 title.startsWith("Module:") ||
 152                 // DE
 153                 title.startsWith("Datei:") ||
 154                 title.startsWith("Verzeichnis:") ||
 155                 title.startsWith("Vorlage:") ||
 156                 title.startsWith("Thesaurus:") ||
 157                 title.startsWith("Kategorie:") ||
 158                 title.startsWith("Hilfe:") ||
 159                 title.startsWith("Reim:") ||
 160                 // FR:
 161                 title.startsWith("Annexe:") ||
 162                 title.startsWith("Catégori:") ||
 163                 title.startsWith("Modèle:") ||
 164                 title.startsWith("Thésaurus:") ||
 165                 title.startsWith("Projet:") ||
 166                 title.startsWith("Aide:") ||
 167                 title.startsWith("Fichier:") ||
 168                 title.startsWith("Wiktionnaire:") ||
 169                 title.startsWith("Catégorie:") ||
 170                 title.startsWith("Portail:") ||
 171                 title.startsWith("utiliusateur:") ||
 172                 title.startsWith("Kategorio:") ||
 173                 // IT
 174                 title.startsWith("Wikizionario:") ||
 175                 title.startsWith("Appendice:") ||
 176                 title.startsWith("Categoria:") ||
 177                 title.startsWith("Aiuto:") ||
 178                 title.startsWith("Portail:") ||
 179                 // ES
 180                 title.startsWith("Apéndice:") ||
 181                 title.startsWith("Archivo:") ||
 182                 title.startsWith("Ayuda:") ||
 183                 title.startsWith("Categoría:") ||
 184                 title.startsWith("Plantilla:") ||
 185                 title.startsWith("Wikcionario:") ||
 186
 187                 // sentinel
 188                 false
 189            ) {
 190             return;
 191         }
 192         if (title.contains(":")) {
 193             if (!title.startsWith("Sign gloss:")) {
 194                 System.err.println("title with colon: " + title);
 195             }
 196         }
 197
 198         String text = textBuilder.toString();
 199         String translingual = "";
 200
 201         while (text.length() > 0) {
 202             // Find start.
 203             final Matcher startMatcher = headingStart.matcher(text);
 204             if (!startMatcher.find()) {
 205                 return;
 206             }
 207             text = text.substring(startMatcher.end());
 208
 209             final String heading = startMatcher.group();
 210             for (final Selector selector : currentSelectors) {
 211                 if (heading.indexOf("Translingual") != -1) {
 212                     // Find end.
 213                     final int depth = startMatcher.group(1).length();
 214                     final Pattern endPattern = getEndPattern(depth);
 215
 216                     final Matcher endMatcher = endPattern.matcher(text);
 217                     if (endMatcher.find()) {
 218                         int end = endMatcher.start();
 219                         translingual = text.substring(0, endMatcher.start());
 220                         text = text.substring(end);
 221                         break;
 222                     }
 223                 }
 224                 if (selector.pattern.matcher(heading).find()) {
 225
 226                     // Find end.
 227                     final int depth = startMatcher.group(1).length();
 228                     final Pattern endPattern = getEndPattern(depth);
 229
 230                     final Matcher endMatcher = endPattern.matcher(text);
 231                     final int end;
 232                     if (endMatcher.find()) {
 233                         end = endMatcher.start();
 234                     } else {
 235                         end = text.length();
 236                     }
 237
 238                     String sectionText = text.substring(0, end);
 239                     // Hack to remove empty dummy section from French
 240                     if (sectionText.startsWith("\n=== {{S|étymologie}} ===\n: {{ébauche-étym")) {
 241                         int dummy_end = sectionText.indexOf("}}", 41) + 2;
 242                         while (dummy_end + 1 < sectionText.length() &&
 243                                 sectionText.charAt(dummy_end) == '\n' &&
 244                                 sectionText.charAt(dummy_end + 1) == '\n') ++dummy_end;
 245                         sectionText = sectionText.substring(dummy_end);
 246                     }
 247                     if (heading.indexOf("Japanese") == -1) sectionText += translingual;
 248                     final Section section = new Section(title, heading, sectionText);
 249
 250                     try {
 251                         selector.out.writeUTF(section.title);
 252                         selector.out.writeUTF(section.heading);
 253                         final byte[] bytes = section.text.getBytes("UTF8");
 254                         selector.out.writeInt(bytes.length);
 255                         selector.out.write(bytes);
 256                     } catch (IOException e) {
 257                         throw new RuntimeException(e);
 258                     }
 259
 260                     text = text.substring(end);
 261                     break;
 262                 }
 263             }
 264         }
 265
 266     }
 267
 268     // -----------------------------------------------------------------------
 269
 270     static class Section implements java.io.Serializable {
 271         private static final long serialVersionUID = -7676549898325856822L;
 272
 273         final String title;
 274         final String heading;
 275         final String text;
 276
 277         public Section(final String title, final String heading, final String text) {
 278             this.title = title;
 279             this.heading = heading;
 280             this.text = text;
 281
 282             //System.out.printf("TITLE:%s\nHEADING:%s\nTEXT:%s\n\n\n\n\n\n", title, heading, text);
 283         }
 284     }
 285
 286     static class Selector {
 287         final String outFilename;
 288         final Pattern pattern;
 289
 290         DataOutputStream out;
 291
 292         public Selector(final String filename, final String pattern) {
 293             this.outFilename = filename;
 294             this.pattern = Pattern.compile(pattern, Pattern.CASE_INSENSITIVE);
 295         }
 296     }
 297
 298     // -----------------------------------------------------------------------
 299
 300     @Override
 301     public void startElement(String uri, String localName, String qName,
 302                              Attributes attributes) {
 303         currentBuilder = null;
 304         if ("page".equals(qName)) {
 305             titleBuilder = new StringBuilder();
 306
 307             // Start with "\n" to better match certain strings.
 308             textBuilder = new StringBuilder("\n");
 309         } else if ("title".equals(qName)) {
 310             currentBuilder = titleBuilder;
 311         } else if ("text".equals(qName)) {
 312             currentBuilder = textBuilder;
 313         }
 314     }
 315
 316     @Override
 317     public void characters(char[] ch, int start, int length) throws SAXException {
 318         if (currentBuilder != null) {
 319             currentBuilder.append(ch, start, length);
 320         }
 321     }
 322
 323     @Override
 324     public void endElement(String uri, String localName, String qName)
 325     throws SAXException {
 326         currentBuilder = null;
 327         if ("page".equals(qName)) {
 328             endPage();
 329         }
 330     }
 331
 332     public void parse(final File file) throws ParserConfigurationException,
 333         SAXException, IOException {
 334         final SAXParser parser = SAXParserFactoryImpl.newInstance().newSAXParser();
 335         parser.parse(file, this);
 336     }
 337
 338 }