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