X-Git-Url: https://git.mdrn.pl/wolnelektury.git/blobdiff_plain/075fb07f19c2f20cd3b6818b973e7aebebdacc9f..2e7619a7bc917eb1a5514af07ce4258e461f74ec:/apps/search/index.py diff --git a/apps/search/index.py b/apps/search/index.py index 10069e4e4..a0bf71588 100644 --- a/apps/search/index.py +++ b/apps/search/index.py @@ -1,7 +1,8 @@ # -*- coding: utf-8 -*- from django.conf import settings -from lucene import SimpleFSDirectory, IndexWriter, CheckIndex, \ +from django.dispatch import Signal +from lucene import SimpleFSDirectory, NIOFSDirectory, IndexWriter, IndexReader, IndexWriterConfig, CheckIndex, \ File, Field, Integer, \ NumericField, Version, Document, JavaError, IndexSearcher, \ QueryParser, PerFieldAnalyzerWrapper, \ @@ -17,7 +18,7 @@ from lucene import SimpleFSDirectory, IndexWriter, CheckIndex, \ # KeywordAnalyzer # Initialize jvm -JVM = initVM(CLASSPATH) +JVM = initVM(CLASSPATH, maxheap=settings.JVM_MAXHEAP) import sys import os @@ -25,12 +26,16 @@ import re import errno from librarian import dcparser from librarian.parser import WLDocument +from lxml import etree import catalogue.models +from pdcounter.models import Author as PDCounterAuthor, BookStub as PDCounterBook from multiprocessing.pool import ThreadPool from threading import current_thread +from itertools import chain import atexit import traceback - +import logging +log = logging.getLogger('search') class WLAnalyzer(PerFieldAnalyzerWrapper): def __init__(self): @@ -54,6 +59,8 @@ class WLAnalyzer(PerFieldAnalyzerWrapper): self.addAnalyzer("source_name", simple) self.addAnalyzer("publisher", simple) self.addAnalyzer("authors", simple) + self.addAnalyzer("title", simple) + self.addAnalyzer("is_book", keyword) # shouldn't the title have two forms? _pl and simple? @@ -71,9 +78,14 @@ class WLAnalyzer(PerFieldAnalyzerWrapper): class IndexStore(object): + """ + Provides access to search index. + + self.store - lucene index directory + """ def __init__(self): self.make_index_dir() - self.store = SimpleFSDirectory(File(settings.SEARCH_INDEX)) + self.store = NIOFSDirectory(File(settings.SEARCH_INDEX)) def make_index_dir(self): try: @@ -83,6 +95,9 @@ class IndexStore(object): pass else: raise + def close(self): + self.store.close() + class IndexChecker(IndexStore): def __init__(self): @@ -95,9 +110,14 @@ class IndexChecker(IndexStore): class Snippets(object): + """ + This class manages snippet files for indexed object (book) + the snippets are concatenated together, and their positions and + lengths are kept in lucene index fields. + """ SNIPPET_DIR = "snippets" - def __init__(self, book_id): + def __init__(self, book_id, revision=None): try: os.makedirs(os.path.join(settings.SEARCH_INDEX, self.SNIPPET_DIR)) except OSError as exc: @@ -105,16 +125,40 @@ class Snippets(object): pass else: raise self.book_id = book_id + self.revision = revision self.file = None + @property + def path(self): + if self.revision: fn = "%d.%d" % (self.book_id, self.revision) + else: fn = "%d" % self.book_id + + return os.path.join(settings.SEARCH_INDEX, self.SNIPPET_DIR, fn) + def open(self, mode='r'): + """ + Open the snippet file. Call .close() afterwards. + """ if not 'b' in mode: mode += 'b' - self.file = open(os.path.join(settings.SEARCH_INDEX, self.SNIPPET_DIR, str(self.book_id)), mode) + + if 'w' in mode: + if os.path.exists(self.path): + self.revision = 1 + while True: + if not os.path.exists(self.path): + break + self.revision += 1 + + self.file = open(self.path, mode) self.position = 0 return self def add(self, snippet): + """ + Append a snippet (unicode) to the snippet file. + Return a (position, length) tuple + """ txt = snippet.encode('utf-8') l = len(txt) self.file.write(txt) @@ -123,27 +167,49 @@ class Snippets(object): return pos def get(self, pos): + """ + Given a tuple of (position, length) return an unicode + of the snippet stored there. + """ self.file.seek(pos[0], 0) txt = self.file.read(pos[1]).decode('utf-8') return txt def close(self): + """Close snippet file""" self.file.close() + def remove(self): + self.revision = None + try: + os.unlink(self.path) + self.revision = 0 + while True: + self.revision += 1 + os.unlink(self.path) + except OSError: + pass -class Index(IndexStore): + +class BaseIndex(IndexStore): + """ + Base index class. + Provides basic operations on index: opening, closing, optimizing. + """ def __init__(self, analyzer=None): - IndexStore.__init__(self) + super(BaseIndex, self).__init__() self.index = None if not analyzer: analyzer = WLAnalyzer() self.analyzer = analyzer - def open(self, analyzer=None): + def open(self, timeout=None): if self.index: raise Exception("Index is already opened") - self.index = IndexWriter(self.store, self.analyzer,\ - IndexWriter.MaxFieldLength.LIMITED) + conf = IndexWriterConfig(Version.LUCENE_34, self.analyzer) + if timeout: + conf.setWriteLockTimeout(long(timeout)) + self.index = IndexWriter(self.store, conf) return self.index def optimize(self): @@ -153,44 +219,146 @@ class Index(IndexStore): try: self.index.optimize() except JavaError, je: - print "Error during optimize phase, check index: %s" % je + log.error("Error during optimize phase, check index: %s" % je) self.index.close() self.index = None - def index_tags(self): - q = NumericRangeQuery.newIntRange("tag_id", 0, Integer.MAX_VALUE, True, True) - self.index.deleteDocuments(q) + index_changed.send_robust(self) + + super(BaseIndex, self).close() + + def __enter__(self): + self.open() + return self + + def __exit__(self, type, value, tb): + self.close() + + +index_changed = Signal() - for tag in catalogue.models.Tag.objects.all(): - doc = Document() - doc.add(NumericField("tag_id", Field.Store.YES, True).setIntValue(tag.id)) - doc.add(Field("tag_name", tag.name, Field.Store.NO, Field.Index.ANALYZED)) - doc.add(Field("tag_name_pl", tag.name, Field.Store.NO, Field.Index.ANALYZED)) - doc.add(Field("tag_category", tag.category, Field.Store.NO, Field.Index.NOT_ANALYZED)) - self.index.addDocument(doc) - def remove_book(self, book): - q = NumericRangeQuery.newIntRange("book_id", book.id, book.id, True, True) +class Index(BaseIndex): + """ + Class indexing books. + """ + def __init__(self, analyzer=None): + super(Index, self).__init__(analyzer) + + def index_tags(self, *tags, **kw): + """ + Re-index global tag list. + Removes all tags from index, then index them again. + Indexed fields include: id, name (with and without polish stems), category + """ + remove_only = kw.get('remove_only', False) + # first, remove tags from index. + if tags: + q = BooleanQuery() + for tag in tags: + b_id_cat = BooleanQuery() + + q_id = NumericRangeQuery.newIntRange("tag_id", tag.id, tag.id, True, True) + b_id_cat.add(q_id, BooleanClause.Occur.MUST) + + if isinstance(tag, PDCounterAuthor): + q_cat = TermQuery(Term('tag_category', 'pd_author')) + elif isinstance(tag, PDCounterBook): + q_cat = TermQuery(Term('tag_category', 'pd_book')) + else: + q_cat = TermQuery(Term('tag_category', tag.category)) + b_id_cat.add(q_cat, BooleanClause.Occur.MUST) + + q.add(b_id_cat, BooleanClause.Occur.SHOULD) + else: # all + q = NumericRangeQuery.newIntRange("tag_id", 0, Integer.MAX_VALUE, True, True) + self.index.deleteDocuments(q) + + if not remove_only: + # then add them [all or just one passed] + if not tags: + tags = chain(catalogue.models.Tag.objects.exclude(category='set'), \ + PDCounterAuthor.objects.all(), \ + PDCounterBook.objects.all()) + + for tag in tags: + if isinstance(tag, PDCounterAuthor): + doc = Document() + doc.add(NumericField("tag_id", Field.Store.YES, True).setIntValue(int(tag.id))) + doc.add(Field("tag_name", tag.name, Field.Store.NO, Field.Index.ANALYZED)) + doc.add(Field("tag_name_pl", tag.name, Field.Store.NO, Field.Index.ANALYZED)) + doc.add(Field("tag_category", 'pd_author', Field.Store.YES, Field.Index.NOT_ANALYZED)) + doc.add(Field("is_pdcounter", 'true', Field.Store.YES, Field.Index.NOT_ANALYZED)) + self.index.addDocument(doc) + elif isinstance(tag, PDCounterBook): + doc = Document() + doc.add(NumericField("tag_id", Field.Store.YES, True).setIntValue(int(tag.id))) + doc.add(Field("tag_name", tag.title, Field.Store.NO, Field.Index.ANALYZED)) + doc.add(Field("tag_name_pl", tag.title, Field.Store.NO, Field.Index.ANALYZED)) + doc.add(Field("tag_category", 'pd_book', Field.Store.YES, Field.Index.NOT_ANALYZED)) + doc.add(Field("is_pdcounter", 'true', Field.Store.YES, Field.Index.NOT_ANALYZED)) + self.index.addDocument(doc) + else: + doc = Document() + doc.add(NumericField("tag_id", Field.Store.YES, True).setIntValue(int(tag.id))) + doc.add(Field("tag_name", tag.name, Field.Store.NO, Field.Index.ANALYZED)) + doc.add(Field("tag_name_pl", tag.name, Field.Store.NO, Field.Index.ANALYZED)) + doc.add(Field("tag_category", tag.category, Field.Store.NO, Field.Index.NOT_ANALYZED)) + self.index.addDocument(doc) + + def create_book_doc(self, book): + """ + Create a lucene document referring book id. + """ + doc = Document() + doc.add(NumericField("book_id", Field.Store.YES, True).setIntValue(int(book.id))) + if book.parent is not None: + doc.add(NumericField("parent_id", Field.Store.YES, True).setIntValue(int(book.parent.id))) + return doc + + def remove_book(self, book_or_id, remove_snippets=True): + """Removes a book from search index. + book - Book instance.""" + if isinstance(book_or_id, catalogue.models.Book): + book_id = book_or_id.id + else: + book_id = book_or_id + + q = NumericRangeQuery.newIntRange("book_id", book_id, book_id, True, True) self.index.deleteDocuments(q) + if remove_snippets: + snippets = Snippets(book_id) + snippets.remove() + def index_book(self, book, book_info=None, overwrite=True): + """ + Indexes the book. + Creates a lucene document for extracted metadata + and calls self.index_content() to index the contents of the book. + """ if overwrite: - self.remove_book(book) + # we don't remove snippets, since they might be still needed by + # threads using not reopened index + self.remove_book(book, remove_snippets=False) book_doc = self.create_book_doc(book) - meta_fields = self.extract_metadata(book, book_info) + meta_fields = self.extract_metadata(book, book_info, dc_only=['source_name', 'authors', 'title']) + # let's not index it - it's only used for extracting publish date + if 'source_name' in meta_fields: + del meta_fields['source_name'] + for f in meta_fields.values(): if isinstance(f, list) or isinstance(f, tuple): for elem in f: book_doc.add(elem) else: book_doc.add(f) - self.index.addDocument(book_doc) del book_doc - self.index_content(book, book_fields=[meta_fields['title'], meta_fields['authors']]) + self.index_content(book, book_fields=[meta_fields['title'], meta_fields['authors'], meta_fields['published_date']]) master_tags = [ 'opowiadanie', @@ -198,22 +366,26 @@ class Index(IndexStore): 'dramat_wierszowany_l', 'dramat_wierszowany_lp', 'dramat_wspolczesny', 'liryka_l', 'liryka_lp', - 'wywiad' + 'wywiad', ] - skip_header_tags = ['autor_utworu', 'nazwa_utworu', 'dzielo_nadrzedne'] + ignore_content_tags = [ + 'uwaga', 'extra', + 'zastepnik_tekstu', 'sekcja_asterysk', 'separator_linia', 'zastepnik_wersu', + 'didaskalia', + 'naglowek_aktu', 'naglowek_sceny', 'naglowek_czesc', + ] - def create_book_doc(self, book): + footnote_tags = ['pa', 'pt', 'pr', 'pe'] + + skip_header_tags = ['autor_utworu', 'nazwa_utworu', 'dzielo_nadrzedne', '{http://www.w3.org/1999/02/22-rdf-syntax-ns#}RDF'] + + published_date_re = re.compile("([0-9]+)[\]. ]*$") + + def extract_metadata(self, book, book_info=None, dc_only=None): """ - Create a lucene document connected to the book + Extract metadata from book and returns a map of fields keyed by fieldname """ - doc = Document() - doc.add(NumericField("book_id", Field.Store.YES, True).setIntValue(book.id)) - if book.parent is not None: - doc.add(NumericField("parent_id", Field.Store.YES, True).setIntValue(book.parent.id)) - return doc - - def extract_metadata(self, book, book_info=None): fields = {} if book_info is None: @@ -225,6 +397,8 @@ class Index(IndexStore): # validator, name for field in dcparser.BookInfo.FIELDS: + if dc_only and field.name not in dc_only: + continue if hasattr(book_info, field.name): if not getattr(book_info, field.name): continue @@ -250,20 +424,40 @@ class Index(IndexStore): fields[field.name] = Field(field.name, "%04d%02d%02d" %\ (dt.year, dt.month, dt.day), Field.Store.NO, Field.Index.NOT_ANALYZED) - return fields + # get published date + pd = None + if hasattr(book_info, 'source_name') and book_info.source_name: + match = self.published_date_re.search(book_info.source_name) + if match is not None: + pd = str(match.groups()[0]) + if not pd: pd = "" + fields["published_date"] = Field("published_date", pd, Field.Store.YES, Field.Index.NOT_ANALYZED) - def get_master(self, root): - for master in root.iter(): - if master.tag in self.master_tags: - return master + return fields def add_gaps(self, fields, fieldname): + """ + Interposes a list of fields with gap-fields, which are indexed spaces and returns it. + This allows for doing phrase queries which do not overlap the gaps (when slop is 0). + """ def gap(): while True: yield Field(fieldname, ' ', Field.Store.NO, Field.Index.NOT_ANALYZED) return reduce(lambda a, b: a + b, zip(fields, gap()))[0:-1] + def get_master(self, root): + """ + Returns the first master tag from an etree. + """ + for master in root.iter(): + if master.tag in self.master_tags: + return master + def index_content(self, book, book_fields=[]): + """ + Walks the book XML and extract content from it. + Adds parts for each header tag and for each fragment. + """ wld = WLDocument.from_file(book.xml_file.path, parse_dublincore=False) root = wld.edoc.getroot() @@ -271,16 +465,34 @@ class Index(IndexStore): if master is None: return [] - def walker(node): - yield node, None - for child in list(node): - for b, e in walker(child): - yield b, e - yield None, node + def walker(node, ignore_tags=[]): + + if node.tag not in ignore_tags: + yield node, None, None + if node.text is not None: + yield None, node.text, None + for child in list(node): + for b, t, e in walker(child): + yield b, t, e + yield None, None, node + + if node.tail is not None: + yield None, node.tail, None return def fix_format(text): - return re.sub("/$", "", text, flags=re.M) + # separator = [u" ", u"\t", u".", u";", u","] + if isinstance(text, list): + # need to join it first + text = filter(lambda s: s is not None, content) + text = u' '.join(text) + # for i in range(len(text)): + # if i > 0: + # if text[i][0] not in separator\ + # and text[i - 1][-1] not in separator: + # text.insert(i, u" ") + + return re.sub("(?m)/$", "", text) def add_part(snippets, **fields): doc = self.create_book_doc(book) @@ -298,6 +510,8 @@ class Index(IndexStore): snip_pos = snippets.add(fields["content"]) doc.add(NumericField("snippets_position", Field.Store.YES, True).setIntValue(snip_pos[0])) doc.add(NumericField("snippets_length", Field.Store.YES, True).setIntValue(snip_pos[1])) + if snippets.revision: + doc.add(NumericField("snippets_revision", Field.Store.YES, True).setIntValue(snippets.revision)) if 'fragment_anchor' in fields: doc.add(Field("fragment_anchor", fields['fragment_anchor'], @@ -325,7 +539,6 @@ class Index(IndexStore): else: return s - fragments = {} snippets = Snippets(book.id).open('w') try: @@ -333,24 +546,51 @@ class Index(IndexStore): if header.tag in self.skip_header_tags: continue + if header.tag is etree.Comment: + continue - content = u' '.join([t for t in header.itertext()]) - content = fix_format(content) - - doc = add_part(snippets, header_index=position, header_type=header.tag, content=content) - - self.index.addDocument(doc) - - for start, end in walker(header): + # section content + content = [] + footnote = [] + + def all_content(text): + for frag in fragments.values(): + frag['content'].append(text) + content.append(text) + handle_text = [all_content] + + + for start, text, end in walker(header, ignore_tags=self.ignore_content_tags): + # handle footnotes + if start is not None and start.tag in self.footnote_tags: + footnote = [] + def collect_footnote(t): + footnote.append(t) + handle_text.append(collect_footnote) + elif end is not None and footnote is not [] and end.tag in self.footnote_tags: + handle_text.pop() + doc = add_part(snippets, header_index=position, header_type=header.tag, + content=u''.join(footnote), + is_footnote=Field("is_footnote", 'true', Field.Store.NO, Field.Index.NOT_ANALYZED)) + + self.index.addDocument(doc) + #print "@ footnote text: %s" % footnote + footnote = [] + + # handle fragments and themes. if start is not None and start.tag == 'begin': fid = start.attrib['id'][1:] fragments[fid] = {'content': [], 'themes': [], 'start_section': position, 'start_header': header.tag} - fragments[fid]['content'].append(start.tail) + + # themes for this fragment elif start is not None and start.tag == 'motyw': fid = start.attrib['id'][1:] + handle_text.append(None) if start.text is not None: fragments[fid]['themes'] += map(str.strip, map(give_me_utf8, start.text.split(','))) - fragments[fid]['content'].append(start.tail) + elif end is not None and end.tag == 'motyw': + handle_text.pop() + elif start is not None and start.tag == 'end': fid = start.attrib['id'][1:] if fid not in fragments: @@ -360,36 +600,32 @@ class Index(IndexStore): continue # empty themes list. del fragments[fid] - def jstr(l): - return u' '.join(map( - lambda x: x == None and u'(none)' or unicode(x), - l)) - doc = add_part(snippets, header_type=frag['start_header'], header_index=frag['start_section'], header_span=position - frag['start_section'] + 1, fragment_anchor=fid, - content=u' '.join(filter(lambda s: s is not None, frag['content'])), + content=fix_format(frag['content']), themes=frag['themes']) - + #print '@ FRAG %s' % frag['content'] self.index.addDocument(doc) - elif start is not None: - for frag in fragments.values(): - frag['content'].append(start.text) - elif end is not None: - for frag in fragments.values(): - frag['content'].append(end.tail) - finally: - snippets.close() + # Collect content. - def __enter__(self): - self.open() - return self + if text is not None and handle_text is not []: + hdl = handle_text[-1] + if hdl is not None: + hdl(text) - def __exit__(self, type, value, tb): - self.close() + # in the end, add a section text. + doc = add_part(snippets, header_index=position, header_type=header.tag, + content=fix_format(content)) + #print '@ CONTENT: %s' % fix_format(content) + + self.index.addDocument(doc) + + finally: + snippets.close() def log_exception_wrapper(f): @@ -397,7 +633,7 @@ def log_exception_wrapper(f): try: f(*a) except Exception, e: - print("Error in indexing thread: %s" % e) + log.error("Error in indexing thread: %s" % e) traceback.print_exc() raise e return _wrap @@ -413,12 +649,11 @@ class ReusableIndex(Index): """ index = None - def open(self, analyzer=None, threads=4): - if ReusableIndex.index is not None: + def open(self, analyzer=None, **kw): + if ReusableIndex.index: self.index = ReusableIndex.index else: - print("opening index") - Index.open(self, analyzer) + Index.open(self, analyzer, **kw) ReusableIndex.index = self.index atexit.register(ReusableIndex.close_reusable) @@ -428,29 +663,25 @@ class ReusableIndex(Index): @staticmethod def close_reusable(): - if ReusableIndex.index is not None: + if ReusableIndex.index: ReusableIndex.index.optimize() ReusableIndex.index.close() ReusableIndex.index = None - def close(self): - pass + index_changed.send_robust(None) + def close(self): + if ReusableIndex.index: + ReusableIndex.index.commit() -class Search(IndexStore): - def __init__(self, default_field="content"): - IndexStore.__init__(self) - self.analyzer = WLAnalyzer() #PolishAnalyzer(Version.LUCENE_34) - ## self.analyzer = WLAnalyzer() - self.searcher = IndexSearcher(self.store, True) - self.parser = QueryParser(Version.LUCENE_34, default_field, - self.analyzer) - - self.parent_filter = TermsFilter() - self.parent_filter.addTerm(Term("is_book", "true")) - def query(self, query): - return self.parser.parse(query) +class JoinSearch(object): + """ + This mixin could be used to handle block join queries. + (currently unused) + """ + def __init__(self, *args, **kw): + super(JoinSearch, self).__init__(*args, **kw) def wrapjoins(self, query, fields=[]): """ @@ -475,29 +706,6 @@ class Search(IndexStore): return BlockJoinQuery(query, self.parent_filter, BlockJoinQuery.ScoreMode.Total) - def simple_search(self, query, max_results=50): - """Returns (books, total_hits) - """ - - tops = self.searcher.search(self.query(query), max_results) - bks = [] - for found in tops.scoreDocs: - doc = self.searcher.doc(found.doc) - bks.append(catalogue.models.Book.objects.get(id=doc.get("book_id"))) - return (bks, tops.totalHits) - - - def search(self, query, max_results=50): - query = self.query(query) - query = self.wrapjoins(query, ["content", "themes"]) - - tops = self.searcher.search(query, max_results) - bks = [] - for found in tops.scoreDocs: - doc = self.searcher.doc(found.doc) - bks.append(catalogue.models.Book.objects.get(id=doc.get("book_id"))) - return (bks, tops.totalHits) - def bsearch(self, query, max_results=50): q = self.query(query) bjq = BlockJoinQuery(q, self.parent_filter, BlockJoinQuery.ScoreMode.Avg) @@ -509,86 +717,158 @@ class Search(IndexStore): bks.append(catalogue.models.Book.objects.get(id=doc.get("book_id"))) return (bks, tops.totalHits) -# TokenStream tokenStream = analyzer.tokenStream(fieldName, reader); -# OffsetAttribute offsetAttribute = tokenStream.getAttribute(OffsetAttribute.class); -# CharTermAttribute charTermAttribute = tokenStream.getAttribute(CharTermAttribute.class); - -# while (tokenStream.incrementToken()) { -# int startOffset = offsetAttribute.startOffset(); -# int endOffset = offsetAttribute.endOffset(); -# String term = charTermAttribute.toString(); -# } - class SearchResult(object): - def __init__(self, searcher, scoreDocs, score=None, how_found=None, snippets=None): - self.snippets = [] + def __init__(self, search, scoreDocs, score=None, how_found=None, snippets=None, searched=None, tokens_cache=None): + if tokens_cache is None: tokens_cache = {} if score: - self.score = score + self._score = score else: - self.score = scoreDocs.score + self._score = scoreDocs.score + + self.boost = 1.0 - self.hits = [] + self._hits = [] + self._processed_hits = None # processed hits - stored = searcher.doc(scoreDocs.doc) + stored = search.searcher.doc(scoreDocs.doc) self.book_id = int(stored.get("book_id")) + pd = stored.get("published_date") + try: + self.published_date = int(pd) + except ValueError: + self.published_date = 0 + header_type = stored.get("header_type") - if not header_type: - return + # we have a content hit in some header of fragment + if header_type is not None: + sec = (header_type, int(stored.get("header_index"))) + header_span = stored.get('header_span') + header_span = header_span is not None and int(header_span) or 1 + + fragment = stored.get("fragment_anchor") - sec = (header_type, int(stored.get("header_index"))) - header_span = stored.get('header_span') - header_span = header_span is not None and int(header_span) or 1 + if snippets: + snippets = snippets.replace("/\n", "\n") + hit = (sec + (header_span,), fragment, scoreDocs.score, {'how_found': how_found, 'snippets': snippets and [snippets] or []}) - fragment = stored.get("fragment_anchor") + self._hits.append(hit) - hit = (sec + (header_span,), fragment, scoreDocs.score, {'how_found': how_found, 'snippets': snippets}) + self.search = search + self.searched = searched + self.tokens_cache = tokens_cache - self.hits.append(hit) + @property + def score(self): + return self._score * self.boost def merge(self, other): if self.book_id != other.book_id: raise ValueError("this search result is or book %d; tried to merge with %d" % (self.book_id, other.book_id)) - self.hits += other.hits + self._hits += other._hits if other.score > self.score: - self.score = other.score + self._score = other._score return self def get_book(self): + if hasattr(self, '_book'): + return self._book return catalogue.models.Book.objects.get(id=self.book_id) book = property(get_book) - def process_hits(self): - frags = filter(lambda r: r[1] is not None, self.hits) - sect = filter(lambda r: r[1] is None, self.hits) + @property + def hits(self): + if self._processed_hits is not None: + return self._processed_hits + + POSITION = 0 + FRAGMENT = 1 + POSITION_INDEX = 1 + POSITION_SPAN = 2 + SCORE = 2 + OTHER = 3 + + # to sections and fragments + frags = filter(lambda r: r[FRAGMENT] is not None, self._hits) + + sect = filter(lambda r: r[FRAGMENT] is None, self._hits) + + # sections not covered by fragments sect = filter(lambda s: 0 == len(filter( - lambda f: s[0][1] >= f[0][1] and s[0][1] < f[0][1] + f[0][2], + lambda f: s[POSITION][POSITION_INDEX] >= f[POSITION][POSITION_INDEX] + and s[POSITION][POSITION_INDEX] < f[POSITION][POSITION_INDEX] + f[POSITION][POSITION_SPAN], frags)), sect) hits = [] + def remove_duplicates(lst, keyfn, compare): + els = {} + for e in lst: + eif = keyfn(e) + if eif in els: + if compare(els[eif], e) >= 1: + continue + els[eif] = e + return els.values() + + # remove fragments with duplicated fid's and duplicated snippets + frags = remove_duplicates(frags, lambda f: f[FRAGMENT], lambda a, b: cmp(a[SCORE], b[SCORE])) + frags = remove_duplicates(frags, lambda f: f[OTHER]['snippets'] and f[OTHER]['snippets'][0] or f[FRAGMENT], + lambda a, b: cmp(a[SCORE], b[SCORE])) + + # remove duplicate sections + sections = {} + for s in sect: - m = {'score': s[2], - 'header_index': s[0][1] + si = s[POSITION][POSITION_INDEX] + # skip existing + if si in sections: + if sections[si]['score'] >= s[SCORE]: + continue + + m = {'score': s[SCORE], + 'section_number': s[POSITION][POSITION_INDEX] + 1, } - m.update(s[3]) - hits.append(m) + m.update(s[OTHER]) + sections[si] = m + + hits = sections.values() for f in frags: - frag = catalogue.models.Fragment.objects.get(anchor=f[1]) - m = {'score': f[2], + try: + frag = catalogue.models.Fragment.objects.get(anchor=f[FRAGMENT], book__id=self.book_id) + except catalogue.models.Fragment.DoesNotExist: + # stale index + continue + + # Figure out if we were searching for a token matching some word in theme name. + themes = frag.tags.filter(category='theme') + themes_hit = [] + if self.searched is not None: + tokens = self.search.get_tokens(self.searched, 'POLISH', cached=self.tokens_cache) + for theme in themes: + name_tokens = self.search.get_tokens(theme.name, 'POLISH') + for t in tokens: + if t in name_tokens: + if not theme in themes_hit: + themes_hit.append(theme) + break + + m = {'score': f[SCORE], 'fragment': frag, - 'themes': frag.tags.filter(category='theme') + 'section_number': f[POSITION][POSITION_INDEX] + 1, + 'themes': themes, + 'themes_hit': themes_hit } - m.update(f[3]) + m.update(f[OTHER]) hits.append(m) hits.sort(lambda a, b: cmp(a['score'], b['score']), reverse=True) - print("--- %s" % hits) + self._processed_hits = hits return hits @@ -602,35 +882,58 @@ class SearchResult(object): for r in rl: if r.book_id in books: books[r.book_id].merge(r) - #print(u"already have one with score %f, and this one has score %f" % (books[book.id][0], found.score)) else: books[r.book_id] = r return books.values() def __cmp__(self, other): - return cmp(self.score, other.score) + c = cmp(self.score, other.score) + if c == 0: + # this is inverted, because earlier date is better + return cmp(other.published_date, self.published_date) + else: + return c class Hint(object): + """ + Given some hint information (information we already know about) + our search target - like author, title (specific book), epoch, genre, kind + we can narrow down search using filters. + """ def __init__(self, search): + """ + Accepts a Searcher instance. + """ self.search = search self.book_tags = {} self.part_tags = [] self._books = [] def books(self, *books): + """ + Give a hint that we search these books. + """ self._books = books def tags(self, tags): + """ + Give a hint that these Tag objects (a list of) + is necessary. + """ for t in tags: if t.category in ['author', 'title', 'epoch', 'genre', 'kind']: lst = self.book_tags.get(t.category, []) lst.append(t) self.book_tags[t.category] = lst - if t.category in ['theme']: + if t.category in ['theme', 'theme_pl']: self.part_tags.append(t) def tag_filter(self, tags, field='tags'): + """ + Given a lsit of tags and an optional field (but they are normally in tags field) + returns a filter accepting only books with specific tags. + """ q = BooleanQuery() for tag in tags: @@ -643,6 +946,9 @@ class Hint(object): return QueryWrapperFilter(q) def book_filter(self): + """ + Filters using book tags (all tag kinds except a theme) + """ tags = reduce(lambda a, b: a + b, self.book_tags.values(), []) if tags: return self.tag_filter(tags) @@ -650,6 +956,10 @@ class Hint(object): return None def part_filter(self): + """ + This filter can be used to look for book parts. + It filters on book id and/or themes. + """ fs = [] if self.part_tags: fs.append(self.tag_filter(self.part_tags, field='themes')) @@ -661,7 +971,7 @@ class Hint(object): bf.add(FilterClause(id_filter, BooleanClause.Occur.SHOULD)) fs.append(bf) - return MultiSearch.chain_filters(fs) + return Search.chain_filters(fs) def should_search_for_book(self): return self._books == [] @@ -680,13 +990,65 @@ class Hint(object): return some -class MultiSearch(Search): - """Class capable of IMDb-like searching""" - def get_tokens(self, searched, field='content'): +class Search(IndexStore): + """ + Search facilities. + """ + def __init__(self, default_field="content"): + IndexStore.__init__(self) + self.analyzer = WLAnalyzer() # PolishAnalyzer(Version.LUCENE_34) + # self.analyzer = WLAnalyzer() + reader = IndexReader.open(self.store, True) + self.searcher = IndexSearcher(reader) + self.parser = QueryParser(Version.LUCENE_34, default_field, + self.analyzer) + + self.parent_filter = TermsFilter() + self.parent_filter.addTerm(Term("is_book", "true")) + index_changed.connect(self.reopen) + + def close(self): + reader = self.searcher.getIndexReader() + self.searcher.close() + reader.close() + super(Search, self).close() + index_changed.disconnect(self.reopen) + + def reopen(self, **unused): + reader = self.searcher.getIndexReader() + rdr = reader.reopen() + if not rdr.equals(reader): + log.debug('Reopening index') + oldsearch = self.searcher + self.searcher = IndexSearcher(rdr) + oldsearch.close() + reader.close() + + def query(self, query): + """Parse query in default Lucene Syntax. (for humans) + """ + return self.parser.parse(query) + + def simple_search(self, query, max_results=50): + """Runs a query for books using lucene syntax. (for humans) + Returns (books, total_hits) + """ + + tops = self.searcher.search(self.query(query), max_results) + bks = [] + for found in tops.scoreDocs: + doc = self.searcher.doc(found.doc) + bks.append(catalogue.models.Book.objects.get(id=doc.get("book_id"))) + return (bks, tops.totalHits) + + def get_tokens(self, searched, field='content', cached=None): """returns tokens analyzed by a proper (for a field) analyzer argument can be: StringReader, string/unicode, or tokens. In the last case they will just be returned (so we can reuse tokens, if we don't change the analyzer) """ + if cached is not None and field in cached: + return cached[field] + if isinstance(searched, str) or isinstance(searched, unicode): searched = StringReader(searched) elif isinstance(searched, list): @@ -698,9 +1060,15 @@ class MultiSearch(Search): while tokens.incrementToken(): cta = tokens.getAttribute(CharTermAttribute.class_) toks.append(cta.toString()) + + if cached is not None: + cached[field] = toks + return toks - def fuzziness(self, fuzzy): + @staticmethod + def fuzziness(fuzzy): + """Helper method to sanitize fuzziness""" if not fuzzy: return None if isinstance(fuzzy, float) and fuzzy > 0.0 and fuzzy <= 1.0: @@ -709,6 +1077,9 @@ class MultiSearch(Search): return 0.5 def make_phrase(self, tokens, field='content', slop=2, fuzzy=False): + """ + Return a PhraseQuery with a series of tokens. + """ if fuzzy: phrase = MultiPhraseQuery() for t in tokens: @@ -717,7 +1088,6 @@ class MultiSearch(Search): fuzzterms = [] while True: - # print("fuzz %s" % unicode(fuzzterm.term()).encode('utf-8')) ft = fuzzterm.term() if ft: fuzzterms.append(ft) @@ -734,7 +1104,13 @@ class MultiSearch(Search): phrase.add(term) return phrase - def make_term_query(self, tokens, field='content', modal=BooleanClause.Occur.SHOULD, fuzzy=False): + @staticmethod + def make_term_query(tokens, field='content', modal=BooleanClause.Occur.SHOULD, fuzzy=False): + """ + Returns term queries joined by boolean query. + modal - applies to boolean query + fuzzy - should the query by fuzzy. + """ q = BooleanQuery() for t in tokens: term = Term(field, t) @@ -745,11 +1121,46 @@ class MultiSearch(Search): q.add(BooleanClause(term, modal)) return q - # def content_query(self, query): - # return BlockJoinQuery(query, self.parent_filter, - # BlockJoinQuery.ScoreMode.Total) + def search_phrase(self, searched, field, book=True, max_results=20, fuzzy=False, + filters=None, tokens_cache=None, boost=None, snippets=False, slop=2): + if filters is None: filters = [] + if tokens_cache is None: tokens_cache = {} + + tokens = self.get_tokens(searched, field, cached=tokens_cache) + + query = self.make_phrase(tokens, field=field, fuzzy=fuzzy, slop=slop) + if book: + filters.append(self.term_filter(Term('is_book', 'true'))) + top = self.searcher.search(query, self.chain_filters(filters), max_results) + + return [SearchResult(self, found, snippets=(snippets and self.get_snippets(found, query) or None), searched=searched) for found in top.scoreDocs] + + def search_some(self, searched, fields, book=True, max_results=20, fuzzy=False, + filters=None, tokens_cache=None, boost=None, snippets=True): + if filters is None: filters = [] + if tokens_cache is None: tokens_cache = {} + + if book: + filters.append(self.term_filter(Term('is_book', 'true'))) + + query = BooleanQuery() + + for fld in fields: + tokens = self.get_tokens(searched, fld, cached=tokens_cache) + + query.add(BooleanClause(self.make_term_query(tokens, field=fld, + fuzzy=fuzzy), BooleanClause.Occur.SHOULD)) + + top = self.searcher.search(query, self.chain_filters(filters), max_results) + + return [SearchResult(self, found, searched=searched, tokens_cache=tokens_cache, + snippets=(snippets and self.get_snippets(found, query) or None)) for found in top.scoreDocs] def search_perfect_book(self, searched, max_results=20, fuzzy=False, hint=None): + """ + Search for perfect book matches. Just see if the query matches with some author or title, + taking hints into account. + """ fields_to_search = ['authors', 'title'] only_in = None if hint: @@ -763,13 +1174,44 @@ class MultiSearch(Search): books = [] for q in qrys: top = self.searcher.search(q, - self.chain_filters([only_in, self.term_filter(Term('is_book', 'true'))]), + self.chain_filters([only_in, self.term_filter(Term('is_book', 'true'))]), max_results) for found in top.scoreDocs: - books.append(SearchResult(self.searcher, found)) + books.append(SearchResult(self, found, how_found="search_perfect_book")) + return books + + def search_book(self, searched, max_results=20, fuzzy=False, hint=None): + fields_to_search = ['tags', 'authors', 'title'] + + only_in = None + if hint: + if not hint.should_search_for_book(): + return [] + fields_to_search = hint.just_search_in(fields_to_search) + only_in = hint.book_filter() + + tokens = self.get_tokens(searched, field='SIMPLE') + + q = BooleanQuery() + + for fld in fields_to_search: + q.add(BooleanClause(self.make_term_query(tokens, field=fld, + fuzzy=fuzzy), BooleanClause.Occur.SHOULD)) + + books = [] + top = self.searcher.search(q, + self.chain_filters([only_in, self.term_filter(Term('is_book', 'true'))]), + max_results) + for found in top.scoreDocs: + books.append(SearchResult(self, found, how_found="search_book")) + return books def search_perfect_parts(self, searched, max_results=20, fuzzy=False, hint=None): + """ + Search for book parts which contains a phrase perfectly matching (with a slop of 2, default for make_phrase()) + some part/fragment of the book. + """ qrys = [self.make_phrase(self.get_tokens(searched), field=fld, fuzzy=fuzzy) for fld in ['content']] flt = None @@ -780,15 +1222,20 @@ class MultiSearch(Search): for q in qrys: top = self.searcher.search(q, self.chain_filters([self.term_filter(Term('is_book', 'true'), inverse=True), - flt - ]), + flt]), max_results) for found in top.scoreDocs: - books.append(SearchResult(self.searcher, found, snippets=self.get_snippets(found, q))) + books.append(SearchResult(self, found, snippets=self.get_snippets(found, q), how_found='search_perfect_parts')) return books - def search_everywhere(self, searched, max_results=20, fuzzy=False, hint=None): + def search_everywhere(self, searched, max_results=20, fuzzy=False, hint=None, tokens_cache=None): + """ + Tries to use search terms to match different fields of book (or its parts). + E.g. one word can be an author survey, another be a part of the title, and the rest + are some words from third chapter. + """ + if tokens_cache is None: tokens_cache = {} books = [] only_in = None @@ -798,43 +1245,52 @@ class MultiSearch(Search): # content only query : themes x content q = BooleanQuery() - tokens = self.get_tokens(searched) - if hint is None or hint.just_search_in(['themes_pl']) != []: - q.add(BooleanClause(self.make_term_query(tokens, field='themes_pl', + tokens_pl = self.get_tokens(searched, field='content', cached=tokens_cache) + tokens = self.get_tokens(searched, field='SIMPLE', cached=tokens_cache) + + # only search in themes when we do not already filter by themes + if hint is None or hint.just_search_in(['themes']) != []: + q.add(BooleanClause(self.make_term_query(tokens_pl, field='themes_pl', fuzzy=fuzzy), BooleanClause.Occur.MUST)) - q.add(BooleanClause(self.make_term_query(tokens, field='content', + q.add(BooleanClause(self.make_term_query(tokens_pl, field='content', fuzzy=fuzzy), BooleanClause.Occur.SHOULD)) topDocs = self.searcher.search(q, only_in, max_results) for found in topDocs.scoreDocs: - books.append(SearchResult(self.searcher, found)) + books.append(SearchResult(self, found, how_found='search_everywhere_themesXcontent', searched=searched)) # query themes/content x author/title/tags q = BooleanQuery() - # in_meta = BooleanQuery() in_content = BooleanQuery() + in_meta = BooleanQuery() + + for fld in ['themes_pl', 'content']: + in_content.add(BooleanClause(self.make_term_query(tokens_pl, field=fld, fuzzy=False), BooleanClause.Occur.SHOULD)) - for fld in ['themes', 'content', 'tags', 'authors', 'title']: - in_content.add(BooleanClause(self.make_term_query(tokens, field=fld, fuzzy=False), BooleanClause.Occur.SHOULD)) + for fld in ['tags', 'authors', 'title']: + in_meta.add(BooleanClause(self.make_term_query(tokens, field=fld, fuzzy=False), BooleanClause.Occur.SHOULD)) + + q.add(BooleanClause(in_content, BooleanClause.Occur.MUST)) + q.add(BooleanClause(in_meta, BooleanClause.Occur.SHOULD)) topDocs = self.searcher.search(q, only_in, max_results) for found in topDocs.scoreDocs: - books.append(SearchResult(self.searcher, found)) + books.append(SearchResult(self, found, how_found='search_everywhere', searched=searched)) return books - def multisearch(self, query, max_results=50): - """ - Search strategy: - - (phrase) OR -> content - -> title - -> authors - - (keywords) -> authors - -> motyw - -> tags - -> content - """ + # def multisearch(self, query, max_results=50): + # """ + # Search strategy: + # - (phrase) OR -> content + # -> title + # -> authors + # - (keywords) -> authors + # -> motyw + # -> tags + # -> content + # """ # queryreader = StringReader(query) # tokens = self.get_tokens(queryreader) @@ -864,40 +1320,52 @@ class MultiSearch(Search): # top_level.add(BooleanClause(phrase_level, Should)) # top_level.add(BooleanClause(kw_level, Should)) - return None - - def book_search(self, query, filter=None, max_results=50, collector=None): - tops = self.searcher.search(query, filter, max_results) - #tops = self.searcher.search(p_content, max_results) - - bks = [] - for found in tops.scoreDocs: - doc = self.searcher.doc(found.doc) - b = catalogue.models.Book.objects.get(id=doc.get("book_id")) - bks.append(b) - print "%s (%d) -> %f" % (b, b.id, found.score) - return bks + # return None def get_snippets(self, scoreDoc, query, field='content'): + """ + Returns a snippet for found scoreDoc. + """ htmlFormatter = SimpleHTMLFormatter() highlighter = Highlighter(htmlFormatter, QueryScorer(query)) stored = self.searcher.doc(scoreDoc.doc) + position = stored.get('snippets_position') + length = stored.get('snippets_length') + if position is None or length is None: + return None + revision = stored.get('snippets_revision') + if revision: revision = int(revision) + # locate content. - snippets = Snippets(stored.get('book_id')).open() + book_id = int(stored.get('book_id')) + snippets = Snippets(book_id, revision=revision) + try: - text = snippets.get((int(stored.get('snippets_position')), - int(stored.get('snippets_length')))) - finally: - snippets.close() + snippets.open() + except IOError, e: + log.error("Cannot open snippet file for book id = %d [rev=%d], %s" % (book_id, revision, e)) + return [] - tokenStream = TokenSources.getAnyTokenStream(self.searcher.getIndexReader(), scoreDoc.doc, field, self.analyzer) - # highlighter.getBestTextFragments(tokenStream, text, False, 10) - # import pdb; pdb.set_trace() - snip = highlighter.getBestFragments(tokenStream, text, 3, "...") + try: + try: + text = snippets.get((int(position), + int(length))) + finally: + snippets.close() - return [snip] + tokenStream = TokenSources.getAnyTokenStream(self.searcher.getIndexReader(), scoreDoc.doc, field, self.analyzer) + # highlighter.getBestTextFragments(tokenStream, text, False, 10) + snip = highlighter.getBestFragments(tokenStream, text, 3, "...") + + except Exception, e: + e2 = e + if hasattr(e, 'getJavaException'): + e2 = unicode(e.getJavaException()) + raise Exception("Problem fetching snippets for book %d, @%d len=%d" % (book_id, int(position), int(length)), + e2) + return snip @staticmethod def enum_to_array(enum): @@ -916,32 +1384,61 @@ class MultiSearch(Search): if terms: return JArray('object')(terms, Term) - def search_tags(self, query, filter=None, max_results=40): - tops = self.searcher.search(query, filter, max_results) + def search_tags(self, query, filt=None, max_results=40, pdcounter=False): + """ + Search for Tag objects using query. + """ + if not pdcounter: + filters = self.chain_filters([filt, self.term_filter(Term('is_pdcounter', 'true'), inverse=True)]) + tops = self.searcher.search(query, filt, max_results) tags = [] for found in tops.scoreDocs: doc = self.searcher.doc(found.doc) - tag = catalogue.models.Tag.objects.get(id=doc.get("tag_id")) - tags.append(tag) - print "%s (%d) -> %f" % (tag, tag.id, found.score) + is_pdcounter = doc.get('is_pdcounter') + category = doc.get('tag_category') + try: + if is_pdcounter == 'true': + if category == 'pd_author': + tag = PDCounterAuthor.objects.get(id=doc.get('tag_id')) + elif category == 'pd_book': + tag = PDCounterBook.objects.get(id=doc.get('tag_id')) + tag.category = 'pd_book' # make it look more lik a tag. + else: + print "Warning. cannot get pdcounter tag_id=%d from db; cat=%s" % (int(doc.get('tag_id')), category) + else: + tag = catalogue.models.Tag.objects.get(id=doc.get("tag_id")) + # don't add the pdcounter tag if same tag already exists + + tags.append(tag) + + except catalogue.models.Tag.DoesNotExist: pass + except PDCounterAuthor.DoesNotExist: pass + except PDCounterBook.DoesNotExist: pass + + log.debug('search_tags: %s' % tags) return tags - def search_books(self, query, filter=None, max_results=10): + def search_books(self, query, filt=None, max_results=10): + """ + Searches for Book objects using query + """ bks = [] - tops = self.searcher.search(query, filter, max_results) + tops = self.searcher.search(query, filt, max_results) for found in tops.scoreDocs: doc = self.searcher.doc(found.doc) - bks.append(catalogue.models.Book.objects.get(id=doc.get("book_id"))) + try: + bks.append(catalogue.models.Book.objects.get(id=doc.get("book_id"))) + except catalogue.models.Book.DoesNotExist: pass return bks - def create_prefix_phrase(self, toks, field): + def make_prefix_phrase(self, toks, field): q = MultiPhraseQuery() for i in range(len(toks)): t = Term(field, toks[i]) if i == len(toks) - 1: - pterms = MultiSearch.enum_to_array(PrefixTermEnum(self.searcher.getIndexReader(), t)) + pterms = Search.enum_to_array(PrefixTermEnum(self.searcher.getIndexReader(), t)) if pterms: q.add(pterms) else: @@ -962,34 +1459,55 @@ class MultiSearch(Search): return only_term - def hint_tags(self, string, max_results=50): + def hint_tags(self, string, max_results=50, pdcounter=True, prefix=True, fuzzy=False): + """ + Return auto-complete hints for tags + using prefix search. + """ toks = self.get_tokens(string, field='SIMPLE') top = BooleanQuery() for field in ['tag_name', 'tag_name_pl']: - q = self.create_prefix_phrase(toks, field) + if prefix: + q = self.make_prefix_phrase(toks, field) + else: + q = self.make_term_query(toks, field, fuzzy=fuzzy) top.add(BooleanClause(q, BooleanClause.Occur.SHOULD)) no_book_cat = self.term_filter(Term("tag_category", "book"), inverse=True) - return self.search_tags(top, no_book_cat, max_results=max_results) + return self.search_tags(top, no_book_cat, max_results=max_results, pdcounter=pdcounter) - def hint_books(self, string, max_results=50): + def hint_books(self, string, max_results=50, prefix=True, fuzzy=False): + """ + Returns auto-complete hints for book titles + Because we do not index 'pseudo' title-tags. + Prefix search. + """ toks = self.get_tokens(string, field='SIMPLE') - q = self.create_prefix_phrase(toks, 'title') + if prefix: + q = self.make_prefix_phrase(toks, 'title') + else: + q = self.make_term_query(toks, 'title', fuzzy=fuzzy) - return self.book_search(q, self.term_filter(Term("is_book", "true")), max_results=max_results) + return self.search_books(q, self.term_filter(Term("is_book", "true")), max_results=max_results) @staticmethod def chain_filters(filters, op=ChainedFilter.AND): + """ + Chains a filter list together + """ filters = filter(lambda x: x is not None, filters) - if not filters: + if not filters or filters is []: return None chf = ChainedFilter(JArray('object')(filters, Filter), op) return chf def filtered_categories(self, tags): + """ + Return a list of tag categories, present in tags list. + """ cats = {} for t in tags: cats[t.category] = True