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 + ".gz");
  87                 tmp = new BufferedOutputStream(tmp);
  88                 tmp = new CompressorStreamFactory().createCompressorOutputStream(CompressorStreamFactory.GZIP, tmp);
  89                 tmp = new WriteBuffer(tmp, 20 * 1024 * 1024);
  90                 selector.out = new DataOutputStream(tmp);
  91             }
  92
  93             // Do it.
  94             try {
  95                 File input = new File(pathToSelectorsEntry.getKey() + ".bz2");
  96                 if (!input.exists()) input = new File(pathToSelectorsEntry.getKey() + ".gz");
  97                 if (!input.exists()) input = new File(pathToSelectorsEntry.getKey() + ".xz");
  98                 if (!input.exists()) {
  99                     // Fallback to uncompressed file
 100                     parser.parse(new File(pathToSelectorsEntry.getKey()), this);
 101                 } else {
 102                     InputStream compressedIn = new BufferedInputStream(new FileInputStream(input));
 103                     InputStream in = new CompressorStreamFactory().createCompressorInputStream(compressedIn);
 104                     in = new ReadAheadBuffer(in, 20 * 1024 * 1024);
 105                     parser.parse(new BufferedInputStream(in), this);
 106                 }
 107             } catch (Exception e) {
 108                 System.err.println("Exception during parse, lastPageTitle=" + lastPageTitle + ", titleBuilder=" + titleBuilder.toString() + " of file " + pathToSelectorsEntry.getKey());
 109                 throw e;
 110             }
 111
 112             // Shutdown.
 113             for (final Selector selector : currentSelectors) {
 114                 selector.out.close();
 115             }
 116
 117         }
 118     }
 119
 120     String lastPageTitle = null;
 121     int pageCount = 0;
 122     Pattern endPatterns[] = new Pattern[100];
 123
 124     private Pattern getEndPattern(int depth) {
 125         if (endPatterns[depth] == null)
 126             endPatterns[depth] = Pattern.compile(String.format("^={1,%d}[^=].*$", depth), Pattern.MULTILINE);
 127         return endPatterns[depth];
 128     }
 129
 130     private void endPage() {
 131         final String title = titleBuilder.toString();
 132         lastPageTitle = title;
 133         if (++pageCount % 100000 == 0) {
 134             System.out.println("endPage: " + title + ", count=" + pageCount);
 135         }
 136         if (title.startsWith("Unsupported titles/")) return;
 137         if (title.contains(":")) {
 138             if (title.startsWith("Wiktionary:") ||
 139                 title.startsWith("Appendix:") ||
 140                 title.startsWith("Help:") ||
 141                 title.startsWith("Index:") ||
 142                 title.startsWith("MediaWiki:") ||
 143                 title.startsWith("Citations:") ||
 144                 title.startsWith("Concordance:") ||
 145                 title.startsWith("Glossary:") ||
 146                 title.startsWith("Rhymes:") ||
 147                 title.startsWith("Category:") ||
 148                 title.startsWith("Wikisaurus:") ||
 149                 title.startsWith("Transwiki:") ||
 150                 title.startsWith("File:") ||
 151                 title.startsWith("Thread:") ||
 152                 title.startsWith("Template:") ||
 153                 title.startsWith("Summary:") ||
 154                 title.startsWith("Module:") ||
 155                 // DE
 156                 title.startsWith("Datei:") ||
 157                 title.startsWith("Verzeichnis:") ||
 158                 title.startsWith("Vorlage:") ||
 159                 title.startsWith("Thesaurus:") ||
 160                 title.startsWith("Kategorie:") ||
 161                 title.startsWith("Hilfe:") ||
 162                 title.startsWith("Reim:") ||
 163                 // FR:
 164                 title.startsWith("Annexe:") ||
 165                 title.startsWith("Catégori:") ||
 166                 title.startsWith("Modèle:") ||
 167                 title.startsWith("Thésaurus:") ||
 168                 title.startsWith("Projet:") ||
 169                 title.startsWith("Aide:") ||
 170                 title.startsWith("Fichier:") ||
 171                 title.startsWith("Wiktionnaire:") ||
 172                 title.startsWith("Catégorie:") ||
 173                 title.startsWith("Portail:") ||
 174                 title.startsWith("utiliusateur:") ||
 175                 title.startsWith("Kategorio:") ||
 176                 // IT
 177                 title.startsWith("Wikizionario:") ||
 178                 title.startsWith("Appendice:") ||
 179                 title.startsWith("Categoria:") ||
 180                 title.startsWith("Aiuto:") ||
 181                 title.startsWith("Portail:") ||
 182                 // ES
 183                 title.startsWith("Apéndice:") ||
 184                 title.startsWith("Archivo:") ||
 185                 title.startsWith("Ayuda:") ||
 186                 title.startsWith("Categoría:") ||
 187                 title.startsWith("Plantilla:") ||
 188                 title.startsWith("Wikcionario:") ||
 189
 190                 // sentinel
 191                 false
 192                ) return;
 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         int start = 0;
 201         final Matcher startMatcher = headingStart.matcher(text);
 202
 203         while (start < text.length()) {
 204             // Find start.
 205             if (!startMatcher.find(start)) {
 206                 return;
 207             }
 208             start = startMatcher.end();
 209
 210             final String heading = startMatcher.group();
 211
 212             // For Translingual entries just store the text for later
 213             // use in the per-language sections
 214             if (heading.indexOf("Translingual") != -1) {
 215                 // Find end.
 216                 final int depth = startMatcher.group(1).length();
 217                 final Pattern endPattern = getEndPattern(depth);
 218
 219                 final Matcher endMatcher = endPattern.matcher(text);
 220                 if (endMatcher.find(start)) {
 221                     int end = endMatcher.start();
 222                     translingual = text.substring(start, end);
 223                     start = end;
 224                     continue;
 225                 }
 226             }
 227
 228             for (final Selector selector : currentSelectors) {
 229                 if (selector.pattern.matcher(heading).find()) {
 230                     // Find end.
 231                     final int depth = startMatcher.group(1).length();
 232                     final Pattern endPattern = getEndPattern(depth);
 233
 234                     final Matcher endMatcher = endPattern.matcher(text);
 235                     final int end;
 236                     if (endMatcher.find(start)) {
 237                         end = endMatcher.start();
 238                     } else {
 239                         end = text.length();
 240                     }
 241
 242                     String sectionText = text.substring(start, end);
 243                     // Hack to remove empty dummy section from French
 244                     if (sectionText.startsWith("\n=== {{S|étymologie}} ===\n: {{ébauche-étym")) {
 245                         int dummy_end = sectionText.indexOf("}}", 41) + 2;
 246                         while (dummy_end + 1 < sectionText.length() &&
 247                                 sectionText.charAt(dummy_end) == '\n' &&
 248                                 sectionText.charAt(dummy_end + 1) == '\n') ++dummy_end;
 249                         sectionText = sectionText.substring(dummy_end);
 250                     }
 251                     if (heading.indexOf("Japanese") == -1) sectionText += translingual;
 252                     final Section section = new Section(title, heading, sectionText);
 253
 254                     try {
 255                         selector.out.writeUTF(section.title);
 256                         selector.out.writeUTF(section.heading);
 257                         final byte[] bytes = section.text.getBytes("UTF8");
 258                         selector.out.writeInt(bytes.length);
 259                         selector.out.write(bytes);
 260                     } catch (IOException e) {
 261                         throw new RuntimeException(e);
 262                     }
 263
 264                     start = end;
 265                     break;
 266                 }
 267             }
 268         }
 269
 270     }
 271
 272     // -----------------------------------------------------------------------
 273
 274     static class Section implements java.io.Serializable {
 275         private static final long serialVersionUID = -7676549898325856822L;
 276
 277         final String title;
 278         final String heading;
 279         final String text;
 280
 281         public Section(final String title, final String heading, final String text) {
 282             this.title = title;
 283             this.heading = heading;
 284             this.text = text;
 285
 286             //System.out.printf("TITLE:%s\nHEADING:%s\nTEXT:%s\n\n\n\n\n\n", title, heading, text);
 287         }
 288     }
 289
 290     static class Selector {
 291         final String outFilename;
 292         final Pattern pattern;
 293
 294         DataOutputStream out;
 295
 296         public Selector(final String filename, final String pattern) {
 297             this.outFilename = filename;
 298             this.pattern = Pattern.compile(pattern, Pattern.CASE_INSENSITIVE);
 299         }
 300     }
 301
 302     // -----------------------------------------------------------------------
 303
 304     @Override
 305     public void startElement(String uri, String localName, String qName,
 306                              Attributes attributes) {
 307         currentBuilder = null;
 308         if ("page".equals(qName)) {
 309             titleBuilder = new StringBuilder();
 310
 311             // Start with "\n" to better match certain strings.
 312             textBuilder = new StringBuilder("\n");
 313         } else if ("title".equals(qName)) {
 314             currentBuilder = titleBuilder;
 315         } else if ("text".equals(qName)) {
 316             currentBuilder = textBuilder;
 317         }
 318     }
 319
 320     @Override
 321     public void characters(char[] ch, int start, int length) throws SAXException {
 322         if (currentBuilder != null) {
 323             currentBuilder.append(ch, start, length);
 324         }
 325     }
 326
 327     @Override
 328     public void endElement(String uri, String localName, String qName)
 329     throws SAXException {
 330         currentBuilder = null;
 331         if ("page".equals(qName)) {
 332             endPage();
 333         }
 334     }
 335
 336     public void parse(final File file) throws ParserConfigurationException,
 337         SAXException, IOException {
 338         final SAXParser parser = SAXParserFactoryImpl.newInstance().newSAXParser();
 339         parser.parse(file, this);
 340     }
 341
 342 }