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