polishing map fixes
[wolnelektury.git] / src / search / index.py
index f943a4d..fc9e9d5 100644 (file)
-# -*- coding: utf-8 -*-
 # This file is part of Wolnelektury, licensed under GNU Affero GPLv3 or later.
 # Copyright © Fundacja Nowoczesna Polska. See NOTICE for more information.
 #
 # This file is part of Wolnelektury, licensed under GNU Affero GPLv3 or later.
 # Copyright © Fundacja Nowoczesna Polska. See NOTICE for more information.
 #
-from django.conf import settings
-
-import os
 import re
 import re
-from librarian import dcparser
 from librarian.parser import WLDocument
 from lxml import etree
 from librarian.parser import WLDocument
 from lxml import etree
-import catalogue.models
-import picture.models
-from pdcounter.models import Author as PDCounterAuthor, BookStub as PDCounterBook
-from itertools import chain
-import scorched
-from . import custom
-import operator
-import logging
-from wolnelektury.utils import makedirs
-
-log = logging.getLogger('search')
-
-if os.path.isfile(settings.SOLR_STOPWORDS):
-    stopwords = set(
-        line.decode('utf-8').strip()
-        for line in open(settings.SOLR_STOPWORDS) if not line.startswith('#'))
-else:
-    stopwords = set()
-
-
-class SolrIndex(object):
-    def __init__(self, mode=None):
-        self.index = custom.CustomSolrInterface(settings.SOLR, mode=mode)
-
-
-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, revision=None):
-        makedirs(os.path.join(settings.SEARCH_INDEX, self.SNIPPET_DIR))
-        self.book_id = book_id
-        self.revision = revision
-        self.file = None
-        self.position = 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 'b' not in mode:
-            mode += 'b'
-
-        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)
-        pos = (self.position, l)
-        self.position += l
-        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"""
-        if self.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(SolrIndex):
+class Index:
     """
     Class indexing books.
     """
     """
     Class indexing books.
     """
-    def __init__(self):
-        super(Index, self).__init__(mode='rw')
-
-    def delete_query(self, *queries):
-        """
-        index.delete(queries=...) doesn't work, so let's reimplement it
-        using deletion of list of uids.
-        """
-        uids = set()
-        for q in queries:
-            if isinstance(q, scorched.search.LuceneQuery):
-                q = self.index.query(q)
-            q.field_limiter.update(['uid'])
-            st = 0
-            rows = 100
-            while True:
-                ids = q.paginate(start=st, rows=rows).execute()
-                if not len(ids):
-                    break
-                for res in ids:
-                    uids.add(res['uid'])
-                st += rows
-        if uids:
-            self.index.delete(uids)
-            return True
-        else:
-            return False
-
-    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
-        """
-        log.debug("Indexing tags")
-        remove_only = kw.get('remove_only', False)
-        # first, remove tags from index.
-        if tags:
-            tag_qs = []
-            for tag in tags:
-                q_id = self.index.Q(tag_id=tag.id)
-
-                if isinstance(tag, PDCounterAuthor):
-                    q_cat = self.index.Q(tag_category='pd_author')
-                elif isinstance(tag, PDCounterBook):
-                    q_cat = self.index.Q(tag_category='pd_book')
-                else:
-                    q_cat = self.index.Q(tag_category=tag.category)
-
-                q_id_cat = self.index.Q(q_id & q_cat)
-                tag_qs.append(q_id_cat)
-            self.delete_query(*tag_qs)
-        else:  # all
-            q = self.index.Q(tag_id__any=True)
-            self.delete_query(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 = {
-                        "tag_id": int(tag.id),
-                        "tag_name": tag.name,
-                        "tag_name_pl": tag.name,
-                        "tag_category": 'pd_author',
-                        "is_pdcounter": True,
-                        "uid": "tag%d_pd_a" % tag.id
-                        }
-                elif isinstance(tag, PDCounterBook):
-                    doc = {
-                        "tag_id": int(tag.id),
-                        "tag_name": tag.title,
-                        "tag_name_pl": tag.title,
-                        "tag_category": 'pd_book',
-                        "is_pdcounter": True,
-                        "uid": "tag%d_pd_b" % tag.id
-                        }
-                else:
-                    doc = {
-                        "tag_id": int(tag.id),
-                        "tag_name": tag.name,
-                        "tag_name_pl": tag.name,
-                        "tag_category": tag.category,
-                        "is_pdcounter": False,
-                        "uid": "tag%d" % tag.id
-                        }
-                self.index.add(doc)
-
-    def create_book_doc(self, book):
-        """
-        Create a lucene document referring book id.
-        """
-        doc = {'book_id': int(book.id)}
-        if book.parent is not None:
-            doc['parent_id'] = 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
-
-        self.delete_query(self.index.Q(book_id=book_id))
-
-        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:
-            # 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, dc_only=[
-            'source_name', 'authors', 'translators', 'title', 'epochs', 'kinds', 'genres'])
-        # 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 n, f in meta_fields.items():
-            book_doc[n] = f
-
-        book_doc['uid'] = "book%s" % book_doc['book_id']
-        self.index.add(book_doc)
-        del book_doc
-        book_fields = {
-            'title': meta_fields['title'],
-            'authors': meta_fields['authors'],
-            'published_date': meta_fields['published_date']
-            }
-
-        for tag_name in ('translators', 'epochs', 'kinds', 'genres'):
-            if tag_name in meta_fields:
-                book_fields[tag_name] = meta_fields[tag_name]
-
-        self.index_content(book, book_fields=book_fields)
-
     master_tags = [
         'opowiadanie',
         'powiesc',
     master_tags = [
         'opowiadanie',
         'powiesc',
@@ -285,7 +23,7 @@ class Index(SolrIndex):
         'uwaga', 'extra', 'nota_red', 'abstrakt',
         'zastepnik_tekstu', 'sekcja_asterysk', 'separator_linia', 'zastepnik_wersu',
         'didaskalia',
         'uwaga', 'extra', 'nota_red', 'abstrakt',
         'zastepnik_tekstu', 'sekcja_asterysk', 'separator_linia', 'zastepnik_wersu',
         'didaskalia',
-        'naglowek_aktu', 'naglowek_sceny', 'naglowek_czesc',
+        'naglowek_aktu', 'naglowek_sceny', 'naglowek_czesc', 'motyw'
     ]
 
     footnote_tags = ['pa', 'pt', 'pr', 'pe']
     ]
 
     footnote_tags = ['pa', 'pt', 'pr', 'pe']
@@ -293,89 +31,41 @@ class Index(SolrIndex):
     skip_header_tags = ['autor_utworu', 'nazwa_utworu', 'dzielo_nadrzedne',
                         '{http://www.w3.org/1999/02/22-rdf-syntax-ns#}RDF']
 
     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):
-        """
-        Extract metadata from book and returns a map of fields keyed by fieldname
-        """
-        fields = {}
-
-        if book_info is None:
-            book_info = dcparser.parse(open(book.xml_file.path))
-
-        fields['slug'] = book.slug
-        fields['is_book'] = True
-
-        # 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
-                # since no type information is available, we use validator
-                type_indicator = field.validator
-                if type_indicator == dcparser.as_unicode:
-                    s = getattr(book_info, field.name)
-                    if field.multiple:
-                        s = ', '.join(s)
-                    fields[field.name] = s
-                elif type_indicator == dcparser.as_person:
-                    p = getattr(book_info, field.name)
-                    if isinstance(p, dcparser.Person):
-                        persons = str(p)
-                    else:
-                        persons = ', '.join(map(str, p))
-                    fields[field.name] = persons
-                elif type_indicator == dcparser.as_date:
-                    dt = getattr(book_info, field.name)
-                    fields[field.name] = dt
-
-        # 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"] = pd
-
-        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):
+    @classmethod
+    def get_master(cls, root):
         """
         Returns the first master tag from an etree.
         """
         for master in root.iter():
         """
         Returns the first master tag from an etree.
         """
         for master in root.iter():
-            if master.tag in self.master_tags:
+            if master.tag in cls.master_tags:
                 return master
 
                 return master
 
-    def index_content(self, book, book_fields):
+    @staticmethod
+    def add_snippet(book, text, position):
+        book.snippet_set.create(
+            sec=position + 1,
+            text=text
+        )
+
+    @classmethod
+    def index_book(cls, book):
         """
         Walks the book XML and extract content from it.
         Adds parts for each header tag and for each fragment.
         """
         """
         Walks the book XML and extract content from it.
         Adds parts for each header tag and for each fragment.
         """
+        if not book.xml_file: return
+
+        book.snippet_set.all().delete()
+
         wld = WLDocument.from_file(book.xml_file.path, parse_dublincore=False)
         root = wld.edoc.getroot()
 
         wld = WLDocument.from_file(book.xml_file.path, parse_dublincore=False)
         root = wld.edoc.getroot()
 
-        master = self.get_master(root)
+        master = cls.get_master(root)
         if master is None:
             return []
 
         def walker(node):
         if master is None:
             return []
 
         def walker(node):
-            if node.tag not in self.ignore_content_tags:
+            if node.tag not in cls.ignore_content_tags:
                 yield node, None, None
                 if node.text is not None:
                     yield None, node.text, None
                 yield node, None, None
                 if node.text is not None:
                     yield None, node.text, None
@@ -389,611 +79,43 @@ class Index(SolrIndex):
             return
 
         def fix_format(text):
             return
 
         def fix_format(text):
-            # separator = [u" ", u"\t", u".", u";", u","]
             if isinstance(text, list):
             if isinstance(text, list):
-                # need to join it first
                 text = filter(lambda s: s is not None, content)
                 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" ")
+                text = ' '.join(text)
 
             return re.sub("(?m)/$", "", text)
 
 
             return re.sub("(?m)/$", "", text)
 
-        def add_part(snippets, **fields):
-            doc = self.create_book_doc(book)
-            for n, v in book_fields.items():
-                doc[n] = v
-
-            doc['header_index'] = fields["header_index"]
-            doc['header_span'] = 'header_span' in fields and fields['header_span'] or 1
-            doc['header_type'] = fields['header_type']
-
-            doc['text'] = fields['text']
-
-            # snippets
-            snip_pos = snippets.add(fields["text"])
-
-            doc['snippets_position'] = snip_pos[0]
-            doc['snippets_length'] = snip_pos[1]
-            if snippets.revision:
-                doc["snippets_revision"] = snippets.revision
-
-            if 'fragment_anchor' in fields:
-                doc["fragment_anchor"] = fields['fragment_anchor']
-
-            if 'themes' in fields:
-                doc['themes'] = fields['themes']
-            doc['uid'] = "part%s-%s-%s-%s" % (
-                book.id, doc['header_index'], doc['header_span'], doc.get('fragment_anchor', ''))
-            return doc
-
-        fragments = {}
-        snippets = Snippets(book.id).open('w')
-        try:
-            for header, position in zip(list(master), range(len(master))):
-
-                if header.tag in self.skip_header_tags:
-                    continue
-                if header.tag is etree.Comment:
-                    continue
-
-                # section content
-                content = []
-                footnote = []
-
-                def all_content(text):
-                    for frag in fragments.values():
-                        frag['text'].append(text)
-                    content.append(text)
-                handle_text = [all_content]
-
-                for start, text, end in walker(header):
-                    # 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,
-                                       text=u''.join(footnote),
-                                       is_footnote=True)
-                        self.index.add(doc)
-                        footnote = []
-
-                    # handle fragments and themes.
-                    if start is not None and start.tag == 'begin':
-                        fid = start.attrib['id'][1:]
-                        fragments[fid] = {
-                            'text': [], 'themes': [], 'start_section': position, 'start_header': header.tag}
-
-                    # themes for this fragment
-                    elif start is not None and start.tag == 'motyw':
-                        fid = start.attrib['id'][1:]
-                        handle_text.append(lambda text: None)
-                        if start.text is not None:
-                            fragments[fid]['themes'] += map(str.strip, map(str, (start.text.split(','))))
-                    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:
-                            continue  # a broken <end> node, skip it
-                        frag = fragments[fid]
-                        if not frag['themes']:
-                            continue  # empty themes list.
-                        del fragments[fid]
-
-                        doc = add_part(snippets,
-                                       header_type=frag['start_header'],
-                                       header_index=frag['start_section'],
-                                       header_span=position - frag['start_section'] + 1,
-                                       fragment_anchor=fid,
-                                       text=fix_format(frag['text']),
-                                       themes=frag['themes'])
-                        self.index.add(doc)
-
-                        # Collect content.
-
-                    if text is not None and handle_text is not []:
-                        hdl = handle_text[-1]
-                        hdl(text)
-
-                        # in the end, add a section text.
-                doc = add_part(snippets, header_index=position,
-                               header_type=header.tag, text=fix_format(content))
-
-                self.index.add(doc)
-
-        finally:
-            snippets.close()
-
-    def remove_picture(self, picture_or_id):
-        """Removes a picture from search index."""
-        if isinstance(picture_or_id, picture.models.Picture):
-            picture_id = picture_or_id.id
-        else:
-            picture_id = picture_or_id
-        self.delete_query(self.index.Q(picture_id=picture_id))
-
-    def index_picture(self, picture, picture_info=None, overwrite=True):
-        """
-        Indexes the picture.
-        Creates a lucene document for extracted metadata
-        and calls self.index_area() to index the contents of the picture.
-        """
-        if overwrite:
-            # we don't remove snippets, since they might be still needed by
-            # threads using not reopened index
-            self.remove_picture(picture)
-
-        picture_doc = {'picture_id': int(picture.id)}
-        meta_fields = self.extract_metadata(picture, picture_info, dc_only=[
-            'authors', 'title', 'epochs', 'kinds', 'genres'])
-
-        picture_doc.update(meta_fields)
-
-        picture_doc['uid'] = "picture%s" % picture_doc['picture_id']
-        self.index.add(picture_doc)
-        del picture_doc['is_book']
-        for area in picture.areas.all():
-            self.index_area(area, picture_fields=picture_doc)
-
-    def index_area(self, area, picture_fields):
-        """
-        Indexes themes and objects on the area.
-        """
-        doc = dict(picture_fields)
-        doc['area_id'] = area.id
-        doc['themes'] = list(area.tags.filter(category__in=('theme', 'thing')).values_list('name', flat=True))
-        doc['uid'] = 'area%s' % area.id
-        self.index.add(doc)
-
-
-class SearchResult(object):
-    def __init__(self, doc, how_found=None, query_terms=None):
-        self.boost = 1.0
-        self._hits = []
-        self._processed_hits = None  # processed hits
-        self.snippets = []
-        self.query_terms = query_terms
-        self._book = None
-
-        if 'score' in doc:
-            self._score = doc['score']
-        else:
-            self._score = 0
-
-        self.book_id = int(doc["book_id"])
-
-        try:
-            self.published_date = int(doc.get("published_date"))
-        except ValueError:
-            self.published_date = 0
-
-        # content hits
-        header_type = doc.get("header_type", None)
-        # we have a content hit in some header of fragment
-        if header_type is not None:
-            sec = (header_type, int(doc["header_index"]))
-            header_span = doc['header_span']
-            header_span = header_span is not None and int(header_span) or 1
-            fragment = doc.get("fragment_anchor", None)
-            snippets_pos = (doc['snippets_position'], doc['snippets_length'])
-            snippets_rev = doc.get('snippets_revision', None)
-
-            hit = (sec + (header_span,), fragment, self._score, {
-                'how_found': how_found,
-                'snippets_pos': snippets_pos,
-                'snippets_revision': snippets_rev,
-                'themes': doc.get('themes', []),
-                'themes_pl': doc.get('themes_pl', [])
-                })
-
-            self._hits.append(hit)
-
-    @classmethod
-    def from_book(cls, book, how_found=None, query_terms=None):
-        doc = {
-            'score': book.popularity.count,
-            'book_id': book.id,
-            'published_date': 0,
-        }
-        result = cls(doc, how_found=how_found, query_terms=query_terms)
-        result._book = book
-        return result
-
-    def __str__(self):
-        return u"<SR id=%d %d(%d) hits score=%f %d snippets>" % \
-            (self.book_id, len(self._hits),
-             len(self._processed_hits) if self._processed_hits else -1,
-             self._score, len(self.snippets))
-
-    def __bytes__(self):
-        return str(self).encode('utf-8')
-
-    @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 for book %d; tried to merge with %d" % (self.book_id, other.book_id))
-        self._hits += other._hits
-        self._score += max(other._score, 0)
-        return self
-
-    def get_book(self):
-        if self._book is not None:
-            return self._book
-        self._book = catalogue.models.Book.objects.get(id=self.book_id)
-        return self._book
-
-    book = property(get_book)
-
-    POSITION = 0
-    FRAGMENT = 1
-    POSITION_INDEX = 1
-    POSITION_SPAN = 2
-    SCORE = 2
-    OTHER = 3
-
-    @property
-    def hits(self):
-        if self._processed_hits is not None:
-            return self._processed_hits
-
-        # to sections and fragments
-        frags = filter(lambda r: r[self.FRAGMENT] is not None, self._hits)
-
-        sect = filter(lambda r: r[self.FRAGMENT] is None, self._hits)
-
-        # sections not covered by fragments
-        sect = filter(lambda s: 0 == len(filter(
-            lambda f: f[self.POSITION][self.POSITION_INDEX] <= s[self.POSITION][self.POSITION_INDEX] <
-                      f[self.POSITION][self.POSITION_INDEX] + f[self.POSITION][self.POSITION_SPAN], frags)), sect)
-
-        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[self.FRAGMENT], lambda a, b: cmp(a[self.SCORE], b[self.SCORE]))
-        # frags = remove_duplicates(frags, lambda f: f[OTHER]['snippet_pos'] and f[OTHER]['snippet_pos'] or f[FRAGMENT],
-        #                           lambda a, b: cmp(a[SCORE], b[SCORE]))
-
-        # remove duplicate sections
-        sections = {}
-
-        for s in sect:
-            si = s[self.POSITION][self.POSITION_INDEX]
-            # skip existing
-            if si in sections:
-                if sections[si]['score'] >= s[self.SCORE]:
-                    continue
-
-            m = {'score': s[self.SCORE],
-                 'section_number': s[self.POSITION][self.POSITION_INDEX] + 1,
-                 }
-            m.update(s[self.OTHER])
-            sections[si] = m
-
-        hits = sections.values()
-
-        for f in frags:
-            try:
-                frag = catalogue.models.Fragment.objects.get(anchor=f[self.FRAGMENT], book__id=self.book_id)
-            except catalogue.models.Fragment.DoesNotExist:
-                # stale index
+        for position, header in enumerate(master):
+            if header.tag in cls.skip_header_tags:
                 continue
                 continue
-            # Figure out if we were searching for a token matching some word in theme name.
-            themes = frag.tags.filter(category='theme')
-            themes_hit = set()
-            if self.query_terms is not None:
-                for i in range(0, len(f[self.OTHER]['themes'])):
-                    tms = f[self.OTHER]['themes'][i].split(r' +') + f[self.OTHER]['themes_pl'][i].split(' ')
-                    tms = map(str.lower, tms)
-                    for qt in self.query_terms:
-                        if qt in tms:
-                            themes_hit.add(f[self.OTHER]['themes'][i])
-                            break
-
-            def theme_by_name(n):
-                th = filter(lambda t: t.name == n, themes)
-                if th:
-                    return th[0]
-                else:
-                    return None
-            themes_hit = filter(lambda a: a is not None, map(theme_by_name, themes_hit))
-
-            m = {'score': f[self.SCORE],
-                 'fragment': frag,
-                 'section_number': f[self.POSITION][self.POSITION_INDEX] + 1,
-                 'themes': themes,
-                 'themes_hit': themes_hit
-                 }
-            m.update(f[self.OTHER])
-            hits.append(m)
-
-        hits.sort(key=lambda h: h['score'], reverse=True)
-
-        self._processed_hits = hits
-
-        return hits
-
-    @staticmethod
-    def aggregate(*result_lists):
-        books = {}
-        for rl in result_lists:
-            for r in rl:
-                if r.book_id in books:
-                    books[r.book_id].merge(r)
-                else:
-                    books[r.book_id] = r
-        return books.values()
-
-    def __cmp__(self, other):
-        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
-
-    def __len__(self):
-        return len(self.hits)
-
-    def snippet_pos(self, idx=0):
-        return self.hits[idx]['snippets_pos']
-
-    def snippet_revision(self, idx=0):
-        try:
-            return self.hits[idx]['snippets_revision']
-        except (IndexError, KeyError):
-            return None
-
-
-class PictureResult(object):
-    def __init__(self, doc, how_found=None, query_terms=None):
-        self.boost = 1.0
-        self.query_terms = query_terms
-        self._picture = None
-        self._hits = []
-        self._processed_hits = None
-
-        if 'score' in doc:
-            self._score = doc['score']
-        else:
-            self._score = 0
-
-        self.picture_id = int(doc["picture_id"])
-
-        if doc.get('area_id'):
-            hit = (self._score, {
-                'how_found': how_found,
-                'area_id': doc['area_id'],
-                'themes': doc.get('themes', []),
-                'themes_pl': doc.get('themes_pl', []),
-            })
-
-            self._hits.append(hit)
-
-    def __str__(self):
-        return u"<PR id=%d score=%f >" % (self.picture_id, self._score)
-
-    def __repr__(self):
-        return str(self)
-
-    @property
-    def score(self):
-        return self._score * self.boost
-
-    def merge(self, other):
-        if self.picture_id != other.picture_id:
-            raise ValueError(
-                "this search result is for picture %d; tried to merge with %d" % (self.picture_id, other.picture_id))
-        self._hits += other._hits
-        self._score += max(other._score, 0)
-        return self
-
-    SCORE = 0
-    OTHER = 1
-
-    @property
-    def hits(self):
-        if self._processed_hits is not None:
-            return self._processed_hits
-
-        hits = []
-        for hit in self._hits:
-            try:
-                area = picture.models.PictureArea.objects.get(id=hit[self.OTHER]['area_id'])
-            except picture.models.PictureArea.DoesNotExist:
-                # stale index
+            if header.tag is etree.Comment:
                 continue
                 continue
-            # Figure out if we were searching for a token matching some word in theme name.
-            themes_hit = set()
-            if self.query_terms is not None:
-                for i in range(0, len(hit[self.OTHER]['themes'])):
-                    tms = hit[self.OTHER]['themes'][i].split(r' +') + hit[self.OTHER]['themes_pl'][i].split(' ')
-                    tms = map(str.lower, tms)
-                    for qt in self.query_terms:
-                        if qt in tms:
-                            themes_hit.add(hit[self.OTHER]['themes'][i])
-                            break
-
-            m = {
-                'score': hit[self.SCORE],
-                'area': area,
-                'themes_hit': themes_hit,
-            }
-            m.update(hit[self.OTHER])
-            hits.append(m)
-
-        hits.sort(key=lambda h: h['score'], reverse=True)
-        hits = hits[:1]
-        self._processed_hits = hits
-        return hits
-
-    def get_picture(self):
-        if self._picture is None:
-            self._picture = picture.models.Picture.objects.get(id=self.picture_id)
-        return self._picture
 
 
-    picture = property(get_picture)
+            # section content
+            content = []
+            footnote = []
 
 
-    @staticmethod
-    def aggregate(*result_lists):
-        books = {}
-        for rl in result_lists:
-            for r in rl:
-                if r.picture_id in books:
-                    books[r.picture_id].merge(r)
-                else:
-                    books[r.picture_id] = r
-        return books.values()
-
-    def __cmp__(self, other):
-        return cmp(self.score, other.score)
-
-
-class Search(SolrIndex):
-    """
-    Search facilities.
-    """
-    def __init__(self, default_field="text"):
-        super(Search, self).__init__(mode='r')
-
-    def make_term_query(self, query, field='text', modal=operator.or_):
-        """
-        Returns term queries joined by boolean query.
-        modal - applies to boolean query
-        fuzzy - should the query by fuzzy.
-        """
-        if query is None:
-            query = ''
-        q = self.index.Q()
-        q = reduce(modal, map(lambda s: self.index.Q(**{field: s}), query.split(r" ")), q)
-
-        return q
-
-    def search_by_author(self, words):
-        from catalogue.models import Book
-        books = Book.objects.filter(parent=None).order_by('-popularity__count')
-        for word in words:
-            books = books.filter(cached_author__iregex='\m%s\M' % word).select_related('popularity__count')
-        return [SearchResult.from_book(book, how_found='search_by_author', query_terms=words) for book in books[:30]]
+            def all_content(text):
+                content.append(text)
+            handle_text = [all_content]
 
 
-    def search_words(self, words, fields, required=None, book=True, picture=False):
-        if book and not picture and fields == ['authors']:
-            return self.search_by_author(words)
-        filters = []
-        for word in words:
-            if book or picture or (word not in stopwords):
-                word_filter = None
-                for field in fields:
-                    q = self.index.Q(**{field: word})
-                    if word_filter is None:
-                        word_filter = q
-                    else:
-                        word_filter |= q
-                filters.append(word_filter)
-        if required:
-            required_filter = None
-            for field in required:
-                for word in words:
-                    if book or picture or (word not in stopwords):
-                        q = self.index.Q(**{field: word})
-                        if required_filter is None:
-                            required_filter = q
-                        else:
-                            required_filter |= q
-            filters.append(required_filter)
-        if not filters:
-            return []
-        params = {}
-        if book:
-            params['is_book'] = True
-        if picture:
-            params['picture_id__gt'] = 0
-        else:
-            params['book_id__gt'] = 0
-        query = self.index.query(**params)
-        query = self.apply_filters(query, filters).field_limit(score=True, all_fields=True)
-        result_class = PictureResult if picture else SearchResult
-        return [result_class(found, how_found='search_words', query_terms=words) for found in query.execute()]
-
-    def get_snippets(self, searchresult, query, field='text', num=1):
-        """
-        Returns a snippet for found scoreDoc.
-        """
-        maxnum = len(searchresult)
-        if num is None or num < 0 or num > maxnum:
-            num = maxnum
-        book_id = searchresult.book_id
-        revision = searchresult.snippet_revision()
-        snippets = Snippets(book_id, revision=revision)
-        snips = [None] * maxnum
-        try:
-            snippets.open()
-            idx = 0
-            while idx < maxnum and num > 0:
-                position, length = searchresult.snippet_pos(idx)
-                if position is None or length is None:
-                    continue
-                text = snippets.get((int(position),
-                                     int(length)))
-                snip = self.index.highlight(text=text, field=field, q=query)
-                if not snip and field == 'text':
-                    snip = self.index.highlight(text=text, field='text_nonstem', q=query)
-                if snip not in snips:
-                    snips[idx] = snip
-                    if snip:
-                        num -= 1
-                idx += 1
-
-        except IOError as e:
-            book = catalogue.models.Book.objects.filter(id=book_id)
-            if not book:
-                log.error("Book does not exist for book id = %d" % book_id)
-            elif not book.get().children.exists():
-                log.error("Cannot open snippet file for book id = %d [rev=%s], %s" % (book_id, revision, e))
-            return []
-        finally:
-            snippets.close()
+            for start, text, end in walker(header):
+                # handle footnotes
+                if start is not None and start.tag in cls.footnote_tags:
+                    footnote = []
 
 
-            # remove verse end markers..
-        snips = map(lambda s: s and s.replace("/\n", "\n"), snips)
+                    def collect_footnote(t):
+                        footnote.append(t)
 
 
-        searchresult.snippets = snips
-
-        return snips
-
-    @staticmethod
-    def apply_filters(query, filters):
-        """
-        Apply filters to a query
-        """
-        if filters is None:
-            filters = []
-        filters = filter(lambda x: x is not None, filters)
-        for f in filters:
-            query = query.query(f)
-        return query
+                    handle_text.append(collect_footnote)
+                elif end is not None and footnote is not [] and end.tag in cls.footnote_tags:
+                    handle_text.pop()
+                    cls.add_snippet(book, ''.join(footnote), position)
+                    footnote = []
 
 
+                if text is not None and handle_text is not []:
+                    hdl = handle_text[-1]
+                    hdl(text)
 
 
-if getattr(settings, 'SEARCH_MOCK', False):
-    from .mock_search import Search
+            # in the end, add a section text.
+            cls.add_snippet(book, fix_format(content), position)