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.IOException;
  24 import java.io.InputStream;
  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 import javax.xml.parsers.SAXParserFactory;
  36
  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|(=+)[^=]).*$" but that broke English.
  49     static final Matcher headingStart = Pattern.compile("^(=+)[^=].*$", Pattern.MULTILINE).matcher("");
  50     static final Matcher startSpanish = Pattern.compile("\\{\\{ES(\\|[^{}=]*)?}}").matcher("");
  51
  52     final Map<String,List<Selector>> pathToSelectors = new LinkedHashMap<String, List<Selector>>();
  53     List<Selector> currentSelectors = null;
  54
  55     StringBuilder titleBuilder;
  56     StringBuilder textBuilder;
  57     StringBuilder currentBuilder = null;
  58
  59     public static void main(final String[] args) throws Exception {
  60         final WiktionarySplitter wiktionarySplitter = new WiktionarySplitter();
  61         wiktionarySplitter.go();
  62     }
  63
  64     private WiktionarySplitter() {
  65         List<Selector> selectors;
  66         for (final String code : WiktionaryLangs.wikiCodeToIsoCodeToWikiName.keySet()) {
  67             //if (!code.equals("fr")) {continue;}
  68             selectors = new ArrayList<WiktionarySplitter.Selector>();
  69             pathToSelectors.put(String.format("data/inputs/%swiktionary-pages-articles.xml", code), selectors);
  70             for (final Map.Entry<String, String> entry : WiktionaryLangs.wikiCodeToIsoCodeToWikiName.get(code).entrySet()) {
  71                 final String dir = String.format("data/inputs/wikiSplit/%s", code);
  72                 new File(dir).mkdirs();
  73                 selectors.add(new Selector(String.format("%s/%s.data", dir, entry.getKey()), entry.getValue()));
  74             }
  75         }
  76     }
  77
  78     private void go() throws Exception {
  79         final SAXParser parser = SAXParserFactory.newInstance().newSAXParser();
  80
  81         // Configure things.
  82         for (final Map.Entry<String, List<Selector>> pathToSelectorsEntry : pathToSelectors.entrySet()) {
  83
  84             currentSelectors = pathToSelectorsEntry.getValue();
  85
  86             for (final Selector selector : currentSelectors) {
  87                 OutputStream tmp = new FileOutputStream(selector.outFilename + ".gz");
  88                 tmp = new BufferedOutputStream(tmp);
  89                 tmp = new CompressorStreamFactory().createCompressorOutputStream(CompressorStreamFactory.GZIP, tmp);
  90                 tmp = new WriteBuffer(tmp, 20 * 1024 * 1024);
  91                 selector.out = new DataOutputStream(tmp);
  92             }
  93
  94             // Do it.
  95             try {
  96                 File input = new File(pathToSelectorsEntry.getKey() + ".bz2");
  97                 if (!input.exists()) input = new File(pathToSelectorsEntry.getKey() + ".gz");
  98                 if (!input.exists()) input = new File(pathToSelectorsEntry.getKey() + ".xz");
  99                 if (!input.exists()) {
 100                     // Fallback to uncompressed file
 101                     parser.parse(new File(pathToSelectorsEntry.getKey()), this);
 102                 } else {
 103                     InputStream compressedIn = new BufferedInputStream(new FileInputStream(input));
 104                     InputStream in = new CompressorStreamFactory().createCompressorInputStream(compressedIn);
 105                     in = new ReadAheadBuffer(in, 20 * 1024 * 1024);
 106                     parser.parse(new BufferedInputStream(in), this);
 107                 }
 108             } catch (Exception e) {
 109                 System.err.println("Exception during parse, lastPageTitle=" + lastPageTitle + ", titleBuilder=" + titleBuilder.toString() + " of file " + pathToSelectorsEntry.getKey());
 110                 throw e;
 111             }
 112
 113             // Shutdown.
 114             for (final Selector selector : currentSelectors) {
 115                 selector.out.close();
 116             }
 117
 118         }
 119     }
 120
 121     String lastPageTitle = null;
 122     int pageCount = 0;
 123     Matcher endPatterns[] = new Matcher[100];
 124
 125     private Matcher getEndPattern(int depth) {
 126         if (endPatterns[depth] == null)
 127             endPatterns[depth] = Pattern.compile(String.format("^={1,%d}[^=].*$", depth), Pattern.MULTILINE).matcher("");
 128         return endPatterns[depth];
 129     }
 130
 131     private void endPage() {
 132         final String title = titleBuilder.toString();
 133         lastPageTitle = title;
 134         if (++pageCount % 100000 == 0) {
 135             System.out.println("endPage: " + title + ", count=" + pageCount);
 136         }
 137         if (title.startsWith("Unsupported titles/")) return;
 138         if (title.contains(":")) {
 139             if (title.startsWith("Wiktionary:") ||
 140                 title.startsWith("Appendix:") ||
 141                 title.startsWith("Help:") ||
 142                 title.startsWith("Index:") ||
 143                 title.startsWith("MediaWiki:") ||
 144                 title.startsWith("Citations:") ||
 145                 title.startsWith("Concordance:") ||
 146                 title.startsWith("Glossary:") ||
 147                 title.startsWith("Rhymes:") ||
 148                 title.startsWith("Category:") ||
 149                 title.startsWith("Wikisaurus:") ||
 150                 title.startsWith("Transwiki:") ||
 151                 title.startsWith("File:") ||
 152                 title.startsWith("Thread:") ||
 153                 title.startsWith("Template:") ||
 154                 title.startsWith("Summary:") ||
 155                 title.startsWith("Module:") ||
 156                 title.startsWith("Reconstruction:") ||
 157                 // DE
 158                 title.startsWith("Datei:") ||
 159                 title.startsWith("Verzeichnis:") ||
 160                 title.startsWith("Vorlage:") ||
 161                 title.startsWith("Thesaurus:") ||
 162                 title.startsWith("Kategorie:") ||
 163                 title.startsWith("Hilfe:") ||
 164                 title.startsWith("Reim:") ||
 165                 title.startsWith("Modul:") ||
 166                 // FR:
 167                 title.startsWith("Annexe:") ||
 168                 title.startsWith("Catégori:") ||
 169                 title.startsWith("Modèle:") ||
 170                 title.startsWith("Thésaurus:") ||
 171                 title.startsWith("Projet:") ||
 172                 title.startsWith("Aide:") ||
 173                 title.startsWith("Fichier:") ||
 174                 title.startsWith("Wiktionnaire:") ||
 175                 title.startsWith("Translations:Wiktionnaire:") ||
 176                 title.startsWith("Translations:Projet:") ||
 177                 title.startsWith("Catégorie:") ||
 178                 title.startsWith("Portail:") ||
 179                 title.startsWith("utiliusateur:") ||
 180                 title.startsWith("Kategorio:") ||
 181                 title.startsWith("Tutoriel:") ||
 182                 // IT
 183                 title.startsWith("Wikizionario:") ||
 184                 title.startsWith("Appendice:") ||
 185                 title.startsWith("Categoria:") ||
 186                 title.startsWith("Aiuto:") ||
 187                 title.startsWith("Portail:") ||
 188                 title.startsWith("Modulo:") ||
 189                 // ES
 190                 title.startsWith("Apéndice:") ||
 191                 title.startsWith("Archivo:") ||
 192                 title.startsWith("Ayuda:") ||
 193                 title.startsWith("Categoría:") ||
 194                 title.startsWith("Plantilla:") ||
 195                 title.startsWith("Wikcionario:") ||
 196
 197                 // PT
 198                 title.startsWith("Ajuda:") ||
 199                 title.startsWith("Apêndice:") ||
 200                 title.startsWith("Citações:") ||
 201                 title.startsWith("Portal:") ||
 202                 title.startsWith("Predefinição:") ||
 203                 title.startsWith("Vocabulário:") ||
 204                 title.startsWith("Wikcionário:") ||
 205                 title.startsWith("Módulo:") ||
 206
 207                 // sentinel
 208                 false
 209                ) return;
 210             // leave the Flexion: pages in for now and do not warn about them
 211             if (!title.startsWith("Sign gloss:") && !title.startsWith("Flexion:")) {
 212                 System.err.println("title with colon: " + title);
 213             }
 214         }
 215
 216         String text = textBuilder.toString();
 217         // Workaround for Spanish wiktionary {{ES}} and {{ES|word}} patterns
 218         text = startSpanish.reset(text).replaceAll("== {{lengua|es}} ==");
 219         String translingual = "";
 220         int start = 0;
 221         headingStart.reset(text);
 222
 223         while (start < text.length()) {
 224             // Find start.
 225             if (!headingStart.find(start)) {
 226                 return;
 227             }
 228             start = headingStart.end();
 229
 230             final String heading = headingStart.group();
 231
 232             // For Translingual entries just store the text for later
 233             // use in the per-language sections
 234             if (heading.indexOf("Translingual") != -1) {
 235                 // Find end.
 236                 final int depth = headingStart.group(1).length();
 237                 final Matcher endMatcher = getEndPattern(depth).reset(text);
 238
 239                 if (endMatcher.find(start)) {
 240                     int end = endMatcher.start();
 241                     translingual = text.substring(start, end);
 242                     start = end;
 243                     continue;
 244                 }
 245             }
 246
 247             for (final Selector selector : currentSelectors) {
 248                 if (selector.pattern.reset(heading).find()) {
 249                     // Find end.
 250                     final int depth = headingStart.group(1).length();
 251                     final Matcher endMatcher = getEndPattern(depth).reset(text);
 252
 253                     final int end;
 254                     if (endMatcher.find(start)) {
 255                         end = endMatcher.start();
 256                     } else {
 257                         end = text.length();
 258                     }
 259
 260                     String sectionText = text.substring(start, end);
 261                     // Hack to remove empty dummy section from French
 262                     if (sectionText.startsWith("\n=== {{S|étymologie}} ===\n: {{ébauche-étym")) {
 263                         int dummy_end = sectionText.indexOf("}}", 41) + 2;
 264                         while (dummy_end + 1 < sectionText.length() &&
 265                                 sectionText.charAt(dummy_end) == '\n' &&
 266                                 sectionText.charAt(dummy_end + 1) == '\n') ++dummy_end;
 267                         sectionText = sectionText.substring(dummy_end);
 268                     }
 269                     if (heading.indexOf("Japanese") == -1) sectionText += translingual;
 270                     final Section section = new Section(title, heading, sectionText);
 271
 272                     try {
 273                         selector.out.writeUTF(section.title);
 274                         selector.out.writeUTF(section.heading);
 275                         final byte[] bytes = section.text.getBytes("UTF8");
 276                         selector.out.writeInt(bytes.length);
 277                         selector.out.write(bytes);
 278                     } catch (IOException e) {
 279                         throw new RuntimeException(e);
 280                     }
 281
 282                     start = end;
 283                     break;
 284                 }
 285             }
 286         }
 287
 288     }
 289
 290     // -----------------------------------------------------------------------
 291
 292     static class Section implements java.io.Serializable {
 293         private static final long serialVersionUID = -7676549898325856822L;
 294
 295         final String title;
 296         final String heading;
 297         final String text;
 298
 299         public Section(final String title, final String heading, final String text) {
 300             this.title = title;
 301             this.heading = heading;
 302             this.text = text;
 303
 304             //System.out.printf("TITLE:%s\nHEADING:%s\nTEXT:%s\n\n\n\n\n\n", title, heading, text);
 305         }
 306     }
 307
 308     static class Selector {
 309         final String outFilename;
 310         final Matcher pattern;
 311
 312         DataOutputStream out;
 313
 314         public Selector(final String filename, final String pattern) {
 315             this.outFilename = filename;
 316             this.pattern = Pattern.compile(pattern, Pattern.CASE_INSENSITIVE).matcher("");
 317         }
 318     }
 319
 320     // -----------------------------------------------------------------------
 321
 322     @Override
 323     public void startElement(String uri, String localName, String qName,
 324                              Attributes attributes) {
 325         currentBuilder = null;
 326         if ("page".equals(qName)) {
 327             titleBuilder = new StringBuilder();
 328
 329             // Start with "\n" to better match certain strings.
 330             textBuilder = new StringBuilder("\n");
 331         } else if ("title".equals(qName)) {
 332             currentBuilder = titleBuilder;
 333         } else if ("text".equals(qName)) {
 334             currentBuilder = textBuilder;
 335         }
 336     }
 337
 338     @Override
 339     public void characters(char[] ch, int start, int length) throws SAXException {
 340         if (currentBuilder != null) {
 341             currentBuilder.append(ch, start, length);
 342         }
 343     }
 344
 345     @Override
 346     public void endElement(String uri, String localName, String qName)
 347     throws SAXException {
 348         currentBuilder = null;
 349         if ("page".equals(qName)) {
 350             endPage();
 351         }
 352     }
 353
 354     public void parse(final File file) throws ParserConfigurationException,
 355         SAXException, IOException {
 356         final SAXParser parser = SAXParserFactory.newInstance().newSAXParser();
 357         parser.parse(file, this);
 358     }
 359
 360 }