import org.tukaani.xz.* /* ------------------------------------------------------------------------- */ def sortRegexList(path) { def set = new TreeSet(String.CASE_INSENSITIVE_ORDER) new File(path).eachLine('UTF-8'){ // check if regex compiles set += java.util.regex.Pattern.compile(it.trim()).pattern() } def out = set.join('\n').saveAs(path) println "${out}\n${out.text}\n" } // sort and check shared regex collections sortRegexList("website/data/release-groups.txt") sortRegexList("website/data/query-blacklist.txt") sortRegexList("website/data/exclude-blacklist.txt") sortRegexList("website/data/series-mappings.txt") /* ------------------------------------------------------------------------- */ def reviews = [] new File('reviews.csv').eachLine('UTF-8'){ def s = it.split(';', 3) reviews << [user: s[0], date: s[1], text: s[2].replaceAll(/^\"|\"$/, '').replaceAll(/["]{2}/, '"') ] } reviews = reviews.sort{ it.date } def json = new groovy.json.JsonBuilder() json.call(reviews as List) json.toPrettyString().saveAs('website/reviews.json') println "Reviews: " + reviews.size() /* ------------------------------------------------------------------------- */ def moviedb_out = new File("website/data/moviedb.txt") def thetvdb_out = new File("website/data/thetvdb.txt") def anidb_out = new File("website/data/anidb.txt") def pack(file, lines) { new File(file.parentFile, file.name + '.xz').withOutputStream{ out -> new XZOutputStream(out, new LZMA2Options(LZMA2Options.PRESET_DEFAULT)).withWriter('UTF-8'){ writer -> lines.each{ writer.append(it).append('\n') } } } def rows = lines.size() def columns = lines.collect{ it.split(/\t/).length }.max() println "$file ($rows rows, $columns columns)" } /* ------------------------------------------------------------------------- */ def isValidMovieName(s) { return (s.normalizePunctuation().length() >= 4) || (s=~ /^[A-Z0-9]/ && s =~ /[\p{Alnum}]{3}/) } def getNamePermutations(names) { def normalize = { s -> s.toLowerCase().normalizePunctuation() }.memoize() def fn1 = { s -> s.replaceAll(/(?i)(^(The|A)\s)|([,]\s(The|A)$)/, '') } def fn2 = { s -> s.replaceAll(/\s&\s/, ' and ') } def fn3 = { s -> s.replaceAll(/\([^\)]*\)$/, '') } def out = names*.trim().unique().collectMany{ original -> def simplified = original [fn1, fn2, fn3].each{ fn -> simplified = fn(simplified).trim() } return [original, simplified] }.unique{ normalize(it) }.findAll{ it.length() > 0 } out = out.findAll{ it.length() >= 2 && !(it ==~ /[1][0-9][1-9]/) && !(it =~ /^[a-z]/) && it =~ /^[@.\p{L}\p{Digit}]/ } // MUST START WITH UNICODE LETTER out = out.findAll{ !MediaDetection.releaseInfo.structureRootPattern.matcher(it).matches() } // IGNORE NAMES THAT OVERLAP WITH MEDIA FOLDER NAMES // out = out.findAll{ a -> names.take(1).contains(a) || out.findAll{ b -> normalize(a).startsWith(normalize(b) + ' ') }.size() == 0 } // TRY TO EXCLUDE REDUNDANT SUBSTRING DUPLICATES return out } def treeSort(list, keyFunction) { def sorter = new TreeMap(String.CASE_INSENSITIVE_ORDER) list.each{ sorter.put(keyFunction(it), it) } return sorter.values() } def csv(f, delim, keyIndex, valueIndex) { def values = [:] if (f.isFile()) { f.splitEachLine(delim, 'UTF-8') { line -> values.put(line[keyIndex], tryQuietly{ line[valueIndex] }) } } return values } /* ------------------------------------------------------------------------- */ // BUILD moviedb index def omdb = [] new File('omdb.txt').eachLine('Windows-1252'){ def line = it.split(/\t/) if (line.length > 11 && line[0] ==~ /\d+/ && line[3] ==~ /\d{4}/) { def imdbid = line[1].substring(2).toInteger() def name = line[2].replaceAll(/\s+/, ' ').trim() def year = line[3].toInteger() def runtime = line[5] def genres = line[6] def rating = tryQuietly{ line[12].toFloat() } ?: 0 def votes = tryQuietly{ line[13].replaceAll(/\D/, '').toInteger() } ?: 0 if (!(genres =~ /Short/ || votes <= 50 || rating <= 2) && ((year >= 1970 && (runtime =~ /(\d.h)|(\d{3}.min)/ || votes >= 500)) || (year >= 1950 && votes >= 20000))) { omdb << [imdbid.pad(7), name, year] } } } omdb = omdb.findAll{ (it[0] as int) <= 9999999 && isValidMovieName(it[1]) } def tmdb_txt = new File('tmdb.txt') def tmdb_index = csv(tmdb_txt, '\t', 1, [0..-1]) def tmdb = [] omdb.each{ m -> def sync = System.currentTimeMillis() if (tmdb_index.containsKey(m[0]) && (sync - tmdb_index[m[0]][0].toLong()) < ((m[2].toInteger() < 2000 ? 360 : 120) * 24 * 60 * 60 * 1000L) ) { tmdb << tmdb_index[m[0]] return } try { def info = WebServices.TheMovieDB.getMovieInfo("tt${m[0]}", Locale.ENGLISH, true) if (info.votes <= 1 || info.rating <= 2) throw new FileNotFoundException('Insufficient movie data') def names = [info.name, info.originalName] + info.alternativeTitles [info?.released?.year, m[2]].findResults{ it?.toInteger() }.unique().each{ y -> def row = [sync, m[0].pad(7), info.id.pad(7), y.pad(4)] + names println row tmdb << row } } catch(FileNotFoundException e) { def row = [sync, m[0].pad(7), 0, m[2], m[1]] println row tmdb << row } } tmdb*.join('\t').join('\n').saveAs(tmdb_txt) movies = tmdb.findResults{ def ity = it[1..3] // imdb id, tmdb id, year def names = getNamePermutations(it[4..-1]).findAll{ isValidMovieName(it) } if (ity[0].toInteger() > 0 && ity[1].toInteger() > 0 && names.size() > 0) return ity + names else return null } movies = treeSort(movies, { it[3, 2].join(' ') }) // sanity check if (movies.size() < 20000) { die('Movie index sanity failed:' + movies.size()) } pack(moviedb_out, movies*.join('\t')) /* ------------------------------------------------------------------------- */ // BUILD tvdb index def tvdb_txt = new File('tvdb.txt') def tvdb = [:] if (tvdb_txt.exists()) { tvdb_txt.eachLine('UTF-8'){ def line = it.split('\t').toList() def names = line.subList(5, line.size()) tvdb.put(line[1] as Integer, [line[0] as Long, line[1] as Integer, line[2], line[3] as Float, line[4] as Float] + names) } } def tvdb_updates = [:] as TreeMap new File('updates_all.xml').eachLine('UTF-8'){ def m = (it =~ '(\\d+)') while(m.find()) { def id = m.group(1) as Integer def time = m.group(2) as Integer tvdb_updates[id] = [id: id, time: time] } } // blacklist crap entries tvdb_updates.remove(219901) tvdb_updates.remove(256135) tvdb_updates.values().each{ update -> if (tvdb[update.id] == null || update.time > tvdb[update.id][0]) { try { retry(2, 500) { def seriesNames = [] def xml = new XmlSlurper().parse("http://thetvdb.com/api/BA864DEE427E384A/series/${update.id}/en.xml") def imdbid = xml.Series.IMDB_ID.text() seriesNames += xml.Series.SeriesName.text() def rating = tryQuietly{ xml.Series.Rating.text().toFloat() } def votes = tryQuietly{ xml.Series.RatingCount.text().toFloat() } // only retrieve additional data for reasonably popular shows if (votes >= 5 && rating >= 4) { tryLogCatch{ if (imdbid =~ /tt(\d+)/) { seriesNames += OMDb.getMovieDescriptor(new Movie(null, 0, imdbid.match(/tt(\d+)/) as int, -1), Locale.ENGLISH).getName() } } // scrape extra alias titles from webpage (not supported yet by API) def jsoup = org.jsoup.Jsoup.connect("http://thetvdb.com/?tab=series&id=${update.id}").get() def akaseries = jsoup.select('#akaseries table tr table tr') .findAll{ it.select('td').any{ it.text() ==~ /en/ } } .findResults{ it.select('td').first().text() } .findAll{ it?.length() > 0 } def intlseries = jsoup.select('#seriesform input') .findAll{ it.attr('name') =~ /SeriesName/ } .sort{ it.attr('name').match(/\d+/) as int } .collect{ it.attr('value') } .findAll{ it?.length() > 0 } println "Scraped data $akaseries and $intlseries for series $seriesNames" seriesNames += akaseries seriesNames += intlseries } def data = [update.time, update.id, imdbid, rating ?: 0, votes ?: 0] + seriesNames.findAll{ it != null && it.length() > 0 } tvdb.put(update.id, data) println "Update $update => $data" } } catch(Throwable e) { printException(e, false) def data = [update.time, update.id, '', 0, 0] tvdb.put(update.id, data) println "Update $update => $data" } } } // remove entries that have become invalid tvdb.keySet().toList().each{ id -> if (tvdb_updates[id] == null) { println "Invalid ID found: ${tvdb[id]}" tvdb.remove(id) } } tvdb.values().findResults{ it.collect{ it.toString().replace('\t', '').trim() }.join('\t') }.join('\n').saveAs(tvdb_txt) def thetvdb_index = [] tvdb.values().each{ r -> def tvdb_id = r[1] def rating = r[3] def votes = r[4] def names = r.subList(5, r.size()) if ((votes >= 5 && rating >= 4) || (votes >= 2 && rating >= 7) || (votes >= 1 && rating >= 10)) { getNamePermutations(names).each{ n -> thetvdb_index << [tvdb_id, n] } } } def addSeriesAlias = { from, to -> def se = thetvdb_index.find{ from == it[1] && !it.contains(to) } if (se == null) die("Unabled to find series '${from}': '${to}'") thetvdb_index << [se[0], to] } // additional custom mappings addSeriesAlias('Law & Order: Special Victims Unit', 'Law and Order SVU') addSeriesAlias('Law & Order: Special Victims Unit', 'Law & Order SVU') addSeriesAlias('CSI: Crime Scene Investigation', 'CSI') addSeriesAlias('M*A*S*H', 'MASH') addSeriesAlias('M*A*S*H', 'M.A.S.H.') addSeriesAlias('NCIS: Los Angeles', 'NCIS LA') addSeriesAlias('NCIS: Los Angeles', 'NCIS LosAngeles') addSeriesAlias('How I Met Your Mother', 'HIMYM') addSeriesAlias('Battlestar Galactica (2003)', 'BSG') addSeriesAlias('World Series of Poker', 'WSOP') addSeriesAlias('House of Cards', 'HOC') addSeriesAlias('The Big Bang Theory', 'TBBT') addSeriesAlias('The Walking Dead', 'TWD') addSeriesAlias('@midnight', 'At Midnight') addSeriesAlias('The Late Late Show with Craig Ferguson', 'Craig Ferguson') addSeriesAlias('Naruto Shippuden', 'Naruto Shippuuden') addSeriesAlias('Resurrection', 'Resurrection (US)') addSeriesAlias('Revolution', 'Revolution (2012)') addSeriesAlias('Cosmos: A Spacetime Odyssey', 'Cosmos A Space Time Odyssey') addSeriesAlias('The Bridge (2013)', 'The Bridge (US)') thetvdb_index = thetvdb_index.findResults{ [it[0] as Integer, it[1].replaceAll(/\s+/, ' ').trim()] }.findAll{ !(it[1] =~ /(?i:duplicate)/ || it[1] =~ /\d{6,}/ || it[1].startsWith('*') || it[1].endsWith('*') || it[1].length() < 2) } thetvdb_index = thetvdb_index.sort{ a, b -> a[0] <=> b[0] } // join and sort def thetvdb_txt = thetvdb_index.groupBy{ it[0] }.findResults{ k, v -> ([k.pad(6)] + v*.getAt(1).unique{ it.toLowerCase() }).join('\t') } // sanity check if (thetvdb_txt.size() < 4000) { die('TheTVDB index sanity failed: ' + thetvdb_txt.size()) } pack(thetvdb_out, thetvdb_txt) /* ------------------------------------------------------------------------- */ // BUILD anidb index def anidb = new AnidbClient('filebot', 5).getAnimeTitles() def anidb_index = anidb.findResults{ def names = it.effectiveNames*.replaceAll(/\s+/, ' ')*.trim()*.replaceAll(/['`´‘’ʻ]+/, /'/) names = getNamePermutations(names) names = names.findAll{ stripReleaseInfo(it)?.length() > 0 } return names.empty ? null : [it.getAnimeId().pad(5)] + names.take(4) } // join and sort def anidb_txt = anidb_index.findResults{ row -> row.join('\t') }.sort().unique() // sanity check if (anidb_txt.size() < 8000) { die('AniDB index sanity failed:' + anidb_txt.size()) } pack(anidb_out, anidb_txt)