some speedups for batch indexing
[wolnelektury.git] / apps / search / index.py
index cc478ee..12554d2 100644 (file)
@@ -25,7 +25,9 @@ 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
 from multiprocessing.pool import ThreadPool
 from threading import current_thread
 import atexit
@@ -212,20 +214,29 @@ class Index(BaseIndex):
 
         for tag in catalogue.models.Tag.objects.all():
             doc = Document()
-            doc.add(NumericField("tag_id", Field.Store.YES, True).setIntValue(tag.id))
+            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)
 
+        for pdtag in PDCounterAuthor.objects.all():
+            doc = Document()
+            doc.add(NumericField("tag_id", Field.Store.YES, True).setIntValue(int(pdtag.id)))
+            doc.add(Field("tag_name", pdtag.name, Field.Store.NO, Field.Index.ANALYZED))
+            doc.add(Field("tag_name_pl", pdtag.name, Field.Store.NO, Field.Index.ANALYZED))
+            doc.add(Field("tag_category", 'pdcounter', Field.Store.NO, Field.Index.NOT_ANALYZED))
+            doc.add(Field("is_pdcounter", 'true', Field.Store.YES, 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(book.id))
+        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(book.parent.id))
+            doc.add(NumericField("parent_id", Field.Store.YES, True).setIntValue(int(book.parent.id)))
         return doc
 
     def remove_book(self, book):
@@ -255,7 +266,7 @@ class Index(BaseIndex):
         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',
@@ -263,10 +274,21 @@ class Index(BaseIndex):
         'dramat_wierszowany_l',
         'dramat_wierszowany_lp',
         'dramat_wspolczesny', 'liryka_l', 'liryka_lp',
-        'wywiad'
+        'wywiad',
+        ]
+
+    ignore_content_tags = [
+        'uwaga', 'extra',
+        'zastepnik_tekstu', 'sekcja_asterysk', 'separator_linia', 'zastepnik_wersu',
+        'didaskalia',
+        'naglowek_aktu', 'naglowek_sceny', 'naglowek_czesc',
         ]
 
-    skip_header_tags = ['autor_utworu', 'nazwa_utworu', 'dzielo_nadrzedne']
+    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):
         """
@@ -308,6 +330,15 @@ class Index(BaseIndex):
                     fields[field.name] = Field(field.name, "%04d%02d%02d" %\
                                                (dt.year, dt.month, dt.day), Field.Store.NO, Field.Index.NOT_ANALYZED)
 
+        # 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)
+
         return fields
 
     def add_gaps(self, fields, fieldname):
@@ -340,15 +371,33 @@ class Index(BaseIndex):
         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):
+            #            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):
@@ -401,24 +450,51 @@ class Index(BaseIndex):
 
                 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:
@@ -428,26 +504,30 @@ class Index(BaseIndex):
                             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)
+
+                        # Collect content.
+
+                    if text is not None and handle_text is not []:
+                        hdl = handle_text[-1]
+                        if hdl is not None:
+                            hdl(text)
+
+                        # 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()
 
@@ -541,38 +621,56 @@ class JoinSearch(object):
 
 
 class SearchResult(object):
-    def __init__(self, searcher, scoreDocs, score=None, how_found=None, snippets=None):
+    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 = None  # processed 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")
+        if pd is None:
+            pd = 0
+        self.published_date = int(pd)
+
         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 and [snippets] or []})
+        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
         if other.score > self.score:
-            self.score = other.score
+            self._score = other._score
         return self
 
     def get_book(self):
@@ -580,7 +678,11 @@ class SearchResult(object):
 
     book = property(get_book)
 
-    def process_hits(self):
+    @property
+    def hits(self):
+        if self._processed_hits is not None:
+            return self._processed_hits
+
         POSITION = 0
         FRAGMENT = 1
         POSITION_INDEX = 1
@@ -627,19 +729,39 @@ class SearchResult(object):
         hits = sections.values()
 
         for f in frags:
-            frag = catalogue.models.Fragment.objects.get(anchor=f[FRAGMENT])
+            try:
+                frag = catalogue.models.Fragment.objects.get(anchor=f[FRAGMENT])
+            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[OTHER])
             hits.append(m)
 
         hits.sort(lambda a, b: cmp(a['score'], b['score']), reverse=True)
 
-        self.hits = hits
+        self._processed_hits = hits
 
-        return self
+        return hits
 
     def __unicode__(self):
         return u'SearchResult(book_id=%d, score=%d)' % (self.book_id, self.score)
@@ -657,7 +779,12 @@ class SearchResult(object):
         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):
@@ -787,11 +914,14 @@ class Search(IndexStore):
             bks.append(catalogue.models.Book.objects.get(id=doc.get("book_id")))
         return (bks, tops.totalHits)
 
-    def get_tokens(self, searched, field='content'):
+    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):
@@ -803,6 +933,10 @@ class Search(IndexStore):
         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):
@@ -859,9 +993,40 @@ class Search(IndexStore):
             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):
         """
@@ -884,7 +1049,7 @@ class Search(IndexStore):
                 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):
@@ -910,13 +1075,13 @@ class Search(IndexStore):
                                    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_book"))
 
         return books
 
     def search_perfect_parts(self, searched, max_results=20, fuzzy=False, hint=None):
         """
-        Search for book parts which containt a phrase perfectly matching (with a slop of 2, default for make_phrase())
+        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']]
@@ -932,16 +1097,17 @@ class Search(IndexStore):
                                                            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
 
@@ -951,8 +1117,8 @@ class Search(IndexStore):
         # content only query : themes x content
         q = BooleanQuery()
 
-        tokens_pl = self.get_tokens(searched, field='content')
-        tokens = self.get_tokens(searched, field='SIMPLE')
+        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']) != []:
@@ -964,7 +1130,7 @@ class Search(IndexStore):
 
         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))
             print "* %s theme x content: %s" % (searched, books[-1]._hits)
 
         # query themes/content x author/title/tags
@@ -983,7 +1149,7 @@ class Search(IndexStore):
 
         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))
             print "* %s scatter search: %s" % (searched, books[-1]._hits)
 
         return books
@@ -1039,11 +1205,15 @@ class Search(IndexStore):
 
         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
         # locate content.
         snippets = Snippets(stored.get('book_id')).open()
         try:
-            text = snippets.get((int(stored.get('snippets_position')),
-                                 int(stored.get('snippets_length'))))
+            text = snippets.get((int(position),
+                                 int(length)))
         finally:
             snippets.close()
 
@@ -1070,19 +1240,27 @@ class Search(IndexStore):
         if terms:
             return JArray('object')(terms, Term)
 
-    def search_tags(self, query, filter=None, max_results=40):
+    def search_tags(self, query, filters=None, max_results=40, pdcounter=False):
         """
         Search for Tag objects using query.
         """
-        tops = self.searcher.search(query, filter, max_results)
+        if not pdcounter:
+            filters = self.chain_filters([filter, self.term_filter(Term('is_pdcounter', 'true'), inverse=True)])
+        tops = self.searcher.search(query, filters, 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')
+            if is_pdcounter:
+                tag = PDCounterAuthor.objects.get(id=doc.get('tag_id'))
+            else:
+                tag = catalogue.models.Tag.objects.get(id=doc.get("tag_id"))
+                # don't add the pdcounter tag if same tag already exists
+            if not (is_pdcounter and filter(lambda t: tag.slug == t.slug, tags)):
+                tags.append(tag)
+                #            print "%s (%d) -> %f" % (tag, tag.id, found.score)
+        print 'returning %s' % tags
         return tags
 
     def search_books(self, query, filter=None, max_results=10):
@@ -1096,7 +1274,7 @@ class Search(IndexStore):
             bks.append(catalogue.models.Book.objects.get(id=doc.get("book_id")))
         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])
@@ -1122,7 +1300,7 @@ class Search(IndexStore):
 
         return only_term
 
-    def hint_tags(self, string, max_results=50):
+    def hint_tags(self, string, max_results=50, pdcounter=True, prefix=True):
         """
         Return auto-complete hints for tags
         using prefix search.
@@ -1131,14 +1309,17 @@ class Search(IndexStore):
         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)
             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):
         """
         Returns auto-complete hints for book titles
         Because we do not index 'pseudo' title-tags.
@@ -1146,7 +1327,10 @@ class Search(IndexStore):
         """
         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')
 
         return self.search_books(q, self.term_filter(Term("is_book", "true")), max_results=max_results)
 
@@ -1156,7 +1340,7 @@ class Search(IndexStore):
         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