Search engine now should be db change change-proof
authorMarcin Koziej <marcin.koziej@nowoczesnapolska.org.pl>
Wed, 21 Mar 2012 16:16:54 +0000 (17:16 +0100)
committerMarcin Koziej <marcin.koziej@nowoczesnapolska.org.pl>
Thu, 22 Mar 2012 09:07:20 +0000 (10:07 +0100)
apps/catalogue/models.py
apps/pdcounter/models.py
apps/search/index.py
apps/search/views.py
wolnelektury/settings/cache.py
wolnelektury/settings/static.py

index 556a8a1..ea09c04 100644 (file)
@@ -1087,9 +1087,11 @@ post_save.connect(_post_save_handler)
 @django.dispatch.receiver(post_delete, sender=Book)
 def _remove_book_from_index_handler(sender, instance, **kwargs):
     """ remove the book from search index, when it is deleted."""
+    search.JVM.attachCurrentThread()
     idx = search.Index()
     idx.open(timeout=10000)  # 10 seconds timeout.
     try:
         idx.remove_book(instance)
+        idx.index_tags()
     finally:
         idx.close()
index f483293..af88bdb 100644 (file)
@@ -6,7 +6,8 @@ from django.db import models
 from django.db.models import permalink
 from django.utils.translation import ugettext as _
 from datetime import datetime
-
+from django.db.models.signals import post_save, post_delete
+import search
 
 class Author(models.Model):
     name = models.CharField(_('name'), max_length=50, db_index=True)
@@ -85,3 +86,19 @@ class BookStub(models.Model):
 
     def pretty_title(self, html_links=False):
         return ', '.join((self.author, self.title))
+
+
+def update_index(sender, instance, **kwargs):
+    print "update pd index %s [update %s]" % (instance, 'created' in kwargs)
+    search.JVM.attachCurrentThread()
+    idx = search.Index()
+    idx.open()
+    try:
+        idx.index_tags(instance, remove_only=not 'created' in kwargs)
+    finally:
+        idx.close()
+
+post_delete.connect(update_index, Author)
+post_delete.connect(update_index, BookStub)
+post_save.connect(update_index, Author)
+post_save.connect(update_index, BookStub)
index b3e9322..9d6d598 100644 (file)
@@ -1,7 +1,8 @@
 # -*- coding: utf-8 -*-
 
 from django.conf import settings
-from lucene import SimpleFSDirectory, NIOFSDirectory, IndexWriter, IndexWriterConfig, 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, \
@@ -92,6 +93,9 @@ class IndexStore(object):
                 pass
             else: raise
 
+    def close(self):
+        self.store.close()
+
 
 class IndexChecker(IndexStore):
     def __init__(self):
@@ -111,7 +115,7 @@ class Snippets(object):
     """
     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:
@@ -119,15 +123,33 @@ 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
+            print "using %s" % self.path
+
+        self.file = open(self.path, mode)
         self.position = 0
         return self
 
@@ -156,6 +178,17 @@ class Snippets(object):
         """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 BaseIndex(IndexStore):
     """
@@ -190,6 +223,10 @@ class BaseIndex(IndexStore):
         self.index.close()
         self.index = None
 
+        index_changed.send_robust(self)
+
+        super(BaseIndex, self).close()
+
     def __enter__(self):
         self.open()
         return self
@@ -198,6 +235,9 @@ class BaseIndex(IndexStore):
         self.close()
 
 
+index_changed = Signal()
+
+
 class Index(BaseIndex):
     """
     Class indexing books.
@@ -205,40 +245,66 @@ class Index(BaseIndex):
     def __init__(self, analyzer=None):
         super(Index, self).__init__(analyzer)
 
-    def index_tags(self):
+    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
         """
-        q = NumericRangeQuery.newIntRange("tag_id", 0, Integer.MAX_VALUE, True, True)
-        self.index.deleteDocuments(q)
+        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)
 
-        for tag in catalogue.models.Tag.objects.exclude(category='set'):
-            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)
-
-        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", '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)
-
-        for pdtag in PDCounterBook.objects.all():
-            doc = Document()
-            doc.add(NumericField("tag_id", Field.Store.YES, True).setIntValue(int(pdtag.id)))
-            doc.add(Field("tag_name", pdtag.title, Field.Store.NO, Field.Index.ANALYZED))
-            doc.add(Field("tag_name_pl", pdtag.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)
+                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 = 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):
         """
@@ -250,12 +316,16 @@ class Index(BaseIndex):
             doc.add(NumericField("parent_id", Field.Store.YES, True).setIntValue(int(book.parent.id)))
         return doc
 
-    def remove_book(self, book):
+    def remove_book(self, book, remove_snippets=True):
         """Removes a book from search index.
         book - Book instance."""
         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.
@@ -263,7 +333,9 @@ class Index(BaseIndex):
         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)
@@ -420,12 +492,16 @@ class Index(BaseIndex):
                     .setIntValue('header_span' in fields and fields['header_span'] or 1))
             doc.add(Field('header_type', fields["header_type"], Field.Store.YES, Field.Index.NOT_ANALYZED))
 
+            print ">>[%s]>%s<<<" % (fields.get('fragment_anchor', ''), fields['content'])
+
             doc.add(Field('content', fields["content"], Field.Store.NO, Field.Index.ANALYZED, \
                           Field.TermVector.WITH_POSITIONS_OFFSETS))
 
             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'],
@@ -584,6 +660,8 @@ class ReusableIndex(Index):
             ReusableIndex.index.close()
             ReusableIndex.index = None
 
+            index_changed.send_robust(None)
+
     def close(self):
         if ReusableIndex.index:
             ReusableIndex.index.commit()
@@ -687,6 +765,8 @@ class SearchResult(object):
         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)
@@ -705,7 +785,10 @@ class SearchResult(object):
 
         # 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[POSITION][POSITION_INDEX] >= f[POSITION][POSITION_INDEX]
             and s[POSITION][POSITION_INDEX] < f[POSITION][POSITION_INDEX] + f[POSITION][POSITION_SPAN],
@@ -713,15 +796,20 @@ class SearchResult(object):
 
         hits = []
 
-        # remove duplicate fragments
-        fragments = {}
-        for f in frags:
-            fid = f[FRAGMENT]
-            if fid in fragments:
-                if fragments[fid][SCORE] >= f[SCORE]:
-                    continue
-            fragments[fid] = f
-        frags = fragments.values()
+        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 hash(f),
+                                  lambda a, b: cmp(a[SCORE], b[SCORE]))
 
         # remove duplicate sections
         sections = {}
@@ -903,12 +991,32 @@ class Search(IndexStore):
         IndexStore.__init__(self)
         self.analyzer = WLAnalyzer()  # PolishAnalyzer(Version.LUCENE_34)
         # self.analyzer = WLAnalyzer()
-        self.searcher = IndexSearcher(self.store, True)
+        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()
+        print "got signal to reopen index"
+        if not rdr.equals(reader):
+            print "will reopen index"
+            oldsearch = self.searcher
+            self.searcher = IndexSearcher(rdr)
+            oldsearch.close()
+            reader.close()
 
     def query(self, query):
         """Parse query in default Lucene Syntax. (for humans)
@@ -1222,9 +1330,11 @@ class Search(IndexStore):
         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.
         book_id = int(stored.get('book_id'))
-        snippets = Snippets(book_id).open()
+        snippets = Snippets(book_id, revision=revision).open()
         try:
             try:
                 text = snippets.get((int(position),
@@ -1274,19 +1384,24 @@ class Search(IndexStore):
             doc = self.searcher.doc(found.doc)
             is_pdcounter = doc.get('is_pdcounter')
             category = doc.get('tag_category')
-            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.
+            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:
-                    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
-            if not (is_pdcounter and filter(lambda t: tag.slug == t.slug, tags)):
-                tags.append(tag)
+                    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)
+            except catalogue.models.Tag.DoesNotExist: pass
+            except PDCounterAuthor.DoesNotExist: pass
+            except PDCounterBook.DoesNotExist: pass
+
                 #            print "%s (%d) -> %f" % (tag, tag.id, found.score)
         print 'returning %s' % tags
         return tags
@@ -1299,7 +1414,9 @@ class Search(IndexStore):
         tops = self.searcher.search(query, filter, 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 make_prefix_phrase(self, toks, field):
index 2945c42..dcd80b2 100644 (file)
@@ -54,6 +54,7 @@ def did_you_mean(query, tokens):
 JVM.attachCurrentThread()
 search = Search()
 
+
 def hint(request):
     prefix = request.GET.get('term', '')
     if len(prefix) < 2:
@@ -77,7 +78,7 @@ def hint(request):
 
     def category_name(c):
         if c.startswith('pd_'):
-            c=c[len('pd_'):]
+            c = c[len('pd_'):]
         return _(c)
 
     return JSONResponse(
@@ -101,109 +102,117 @@ def main(request):
     query = None
     fuzzy = False #0.8
 
-    if 'q' in request.GET:
-        # tags = request.GET.get('tags', '')
-        query = request.GET['q']
-        # book_id = request.GET.get('book', None)
-        # book = None
-        # if book_id is not None:
-        #     book = get_object_or_404(Book, id=book_id)
-
-        # hint = search.hint()
-        # try:
-        #     tag_list = Tag.get_tag_list(tags)
-        # except:
-        #     tag_list = []
-
-        if len(query) < 2:
-            return render_to_response('catalogue/search_too_short.html', {'prefix': query},
-                                      context_instance=RequestContext(request))
-
-        # hint.tags(tag_list)
-        # if book:
-        #     hint.books(book)
-        tags = search.hint_tags(query, pdcounter=True, prefix=False, fuzzy=fuzzy)
-        tags = split_tags(tags)
-
-        toks = StringReader(query)
-        tokens_cache = {}
-
-        author_results = search.search_phrase(toks, 'authors', fuzzy=fuzzy, tokens_cache=tokens_cache)
-        title_results = search.search_phrase(toks, 'title', fuzzy=fuzzy, tokens_cache=tokens_cache)
-
-        # Boost main author/title results with mixed search, and save some of its results for end of list.
-        # boost author, title results
-        author_title_mixed = search.search_some(toks, ['authors', 'title', 'tags'], fuzzy=fuzzy, tokens_cache=tokens_cache)
-        author_title_rest = []
-        for b in author_title_mixed:
-            bks = filter(lambda ba: ba.book_id == b.book_id, author_results + title_results)
-            for b2 in bks:
-                b2.boost *= 1.1
-            if bks is []:
-                author_title_rest.append(b)
-
-        # Do a phrase search but a term search as well - this can give us better snippets then search_everywhere,
-        # Because the query is using only one field.
-        text_phrase = SearchResult.aggregate(
-            search.search_phrase(toks, 'content', fuzzy=fuzzy, tokens_cache=tokens_cache, snippets=True, book=False, slop=4),
-            search.search_some(toks, ['content'], tokens_cache=tokens_cache, snippets=True, book=False))
-
-        everywhere = search.search_everywhere(toks, fuzzy=fuzzy, tokens_cache=tokens_cache)
-
-        def already_found(results):
-            def f(e):
-                for r in results:
-                    if e.book_id == r.book_id:
-                        e.boost = 0.9
-                        results.append(e)
-                        return True
-                return False
-            return f
-        f = already_found(author_results + title_results + text_phrase)
-        everywhere = filter(lambda x: not f(x), everywhere)
-
-        author_results = SearchResult.aggregate(author_results)
-        title_results = SearchResult.aggregate(title_results)
-
-        everywhere = SearchResult.aggregate(everywhere, author_title_rest)
-
-        for res in [author_results, title_results, text_phrase, everywhere]:
-            res.sort(reverse=True)
-            for r in res:
-                for h in r.hits:
-                    h['snippets'] = map(lambda s:
-                                        re.subn(r"(^[ \t\n]+|[ \t\n]+$)", u"",
-                                                re.subn(r"[ \t\n]*\n[ \t\n]*", u"\n", s)[0])[0], h['snippets'])
-
-        suggestion = did_you_mean(query, search.get_tokens(toks, field="SIMPLE"))
-        print "dym? %s" % repr(suggestion).encode('utf-8')
-
-        results = author_results + title_results + text_phrase + everywhere
-        results.sort(reverse=True)
-
-        if len(results) == 1:
-            fragment_hits = filter(lambda h: 'fragment' in h, results[0].hits)
-            if len(fragment_hits) == 1:
-                #anchor = fragment_hits[0]['fragment']
-                #frag = Fragment.objects.get(anchor=anchor)
-                return HttpResponseRedirect(fragment_hits[0]['fragment'].get_absolute_url())
-            return HttpResponseRedirect(results[0].book.get_absolute_url())
-        elif len(results) == 0:
-            form = PublishingSuggestForm(initial={"books": query + ", "})
-            return render_to_response('catalogue/search_no_hits.html',
-                                      {'tags': tags,
-                                       'prefix': query,
-                                       "form": form,
-                                       'did_you_mean': suggestion},
-                context_instance=RequestContext(request))
-
-        print "TAGS: %s" % tags
-        return render_to_response('catalogue/search_multiple_hits.html',
+    query = request.GET.get('q','')
+    # book_id = request.GET.get('book', None)
+    # book = None
+    # if book_id is not None:
+    #     book = get_object_or_404(Book, id=book_id)
+
+    # hint = search.hint()
+    # try:
+    #     tag_list = Tag.get_tag_list(tags)
+    # except:
+    #     tag_list = []
+
+    if len(query) < 2:
+        return render_to_response('catalogue/search_too_short.html', {'prefix': query},
+                                  context_instance=RequestContext(request))
+
+    # hint.tags(tag_list)
+    # if book:
+    #     hint.books(book)
+    tags = search.hint_tags(query, pdcounter=True, prefix=False, fuzzy=fuzzy)
+    tags = split_tags(tags)
+
+    toks = StringReader(query)
+    tokens_cache = {}
+
+    author_results = search.search_phrase(toks, 'authors', fuzzy=fuzzy, tokens_cache=tokens_cache)
+    title_results = search.search_phrase(toks, 'title', fuzzy=fuzzy, tokens_cache=tokens_cache)
+
+    # Boost main author/title results with mixed search, and save some of its results for end of list.
+    # boost author, title results
+    author_title_mixed = search.search_some(toks, ['authors', 'title', 'tags'], fuzzy=fuzzy, tokens_cache=tokens_cache)
+    author_title_rest = []
+    for b in author_title_mixed:
+        bks = filter(lambda ba: ba.book_id == b.book_id, author_results + title_results)
+        for b2 in bks:
+            b2.boost *= 1.1
+        if bks is []:
+            author_title_rest.append(b)
+
+    # Do a phrase search but a term search as well - this can give us better snippets then search_everywhere,
+    # Because the query is using only one field.
+    text_phrase = SearchResult.aggregate(
+        search.search_phrase(toks, 'content', fuzzy=fuzzy, tokens_cache=tokens_cache, snippets=True, book=False, slop=4),
+        search.search_some(toks, ['content'], tokens_cache=tokens_cache, snippets=True, book=False))
+
+    everywhere = search.search_everywhere(toks, fuzzy=fuzzy, tokens_cache=tokens_cache)
+
+    def already_found(results):
+        def f(e):
+            for r in results:
+                if e.book_id == r.book_id:
+                    e.boost = 0.9
+                    results.append(e)
+                    return True
+            return False
+        return f
+    f = already_found(author_results + title_results + text_phrase)
+    everywhere = filter(lambda x: not f(x), everywhere)
+
+    author_results = SearchResult.aggregate(author_results)
+    title_results = SearchResult.aggregate(title_results)
+
+    everywhere = SearchResult.aggregate(everywhere, author_title_rest)
+
+    for res in [author_results, title_results, text_phrase, everywhere]:
+        res.sort(reverse=True)
+        for r in res:
+            for h in r.hits:
+                h['snippets'] = map(lambda s:
+                                    re.subn(r"(^[ \t\n]+|[ \t\n]+$)", u"",
+                                            re.subn(r"[ \t\n]*\n[ \t\n]*", u"\n", s)[0])[0], h['snippets'])
+
+    suggestion = did_you_mean(query, search.get_tokens(toks, field="SIMPLE"))
+
+    def ensure_exists(r):
+        try:
+            return r.book
+        except Book.DoesNotExist:
+            return False
+
+    author_results = filter(ensure_exists, author_results)
+    title_results = filter(ensure_exists, title_results)
+    text_phrase = filter(ensure_exists, text_phrase)
+    everywhere = filter(ensure_exists, everywhere)
+
+    results = author_results + title_results + text_phrase + everywhere
+    # ensure books do exists & sort them
+    results.sort(reverse=True)
+
+    if len(results) == 1:
+        fragment_hits = filter(lambda h: 'fragment' in h, results[0].hits)
+        if len(fragment_hits) == 1:
+            #anchor = fragment_hits[0]['fragment']
+            #frag = Fragment.objects.get(anchor=anchor)
+            return HttpResponseRedirect(fragment_hits[0]['fragment'].get_absolute_url())
+        return HttpResponseRedirect(results[0].book.get_absolute_url())
+    elif len(results) == 0:
+        form = PublishingSuggestForm(initial={"books": query + ", "})
+        return render_to_response('catalogue/search_no_hits.html',
                                   {'tags': tags,
                                    'prefix': query,
-                                   'results': { 'author': author_results,
-                                                'title': title_results,
-                                                'content': text_phrase,
-                                                'other': everywhere},
+                                   "form": form,
                                    'did_you_mean': suggestion},
             context_instance=RequestContext(request))
+
+    return render_to_response('catalogue/search_multiple_hits.html',
+                              {'tags': tags,
+                               'prefix': query,
+                               'results': { 'author': author_results,
+                                            'title': title_results,
+                                            'content': text_phrase,
+                                            'other': everywhere},
+                               'did_you_mean': suggestion},
+        context_instance=RequestContext(request))
index cc90240..61d3bb5 100644 (file)
@@ -23,4 +23,3 @@ CACHES = {
     },
 }
 CACHE_MIDDLEWARE_ANONYMOUS_ONLY=True
-SEARCH_INDEX = path.join(PROJECT_DIR, '../search_index/')
index 797433d..188400d 100644 (file)
@@ -5,6 +5,7 @@ from settings.paths import PROJECT_DIR
 # Example: "/home/media/media.lawrence.com/"
 MEDIA_ROOT = path.join(PROJECT_DIR, '../media/')
 STATIC_ROOT = path.join(PROJECT_DIR, 'static/')
+SEARCH_INDEX = path.join(PROJECT_DIR, '../search_index/')
 
 # URL that handles the media served from MEDIA_ROOT. Make sure to use a
 # trailing slash if there is a path component (optional in other cases).