extensions/views/public.py

@@ -2,7 +2,7 @@ import logging
 
 from django.conf import settings
 from django.contrib.auth import get_user_model
-from django.contrib.postgres.search import SearchQuery, SearchRank, SearchVector
+from django.db import connection
 from django.db.models import Q
 from django.shortcuts import get_object_or_404, redirect
 from django.views.generic.list import ListView
@@ -24,7 +24,15 @@ log = logging.getLogger(__name__)
 
 class ListedExtensionsView(ListView):
     model = Extension
-    queryset = Extension.objects.listed
+    queryset = Extension.objects.listed.prefetch_related(
+        'authors',
+        'latest_version__file',
+        'latest_version__tags',
+        'preview_set',
+        'preview_set__file',
+        'ratings',
+        'team',
+    )
     context_object_name = 'extensions'
 
 
@@ -34,19 +42,7 @@ class HomeView(ListedExtensionsView):
 
     def get_context_data(self, **kwargs):
         context = super().get_context_data(**kwargs)
-        q = (
-            super()
-            .get_queryset()
-            .prefetch_related(
-                'authors',
-                'latest_version__file',
-                'latest_version__tags',
-                'preview_set',
-                'preview_set__file',
-                'ratings',
-                'team',
-            )
-        )
+        q = super().get_queryset()
         context['addons'] = q.filter(type=EXTENSION_TYPE_CHOICES.BPY)[:8]
         context['themes'] = q.filter(type=EXTENSION_TYPE_CHOICES.THEME)[:8]
         return context
@@ -90,39 +86,68 @@ class SearchView(ListedExtensionsView):
         if self.kwargs.get('type_slug'):
             _type = self._get_type_id_by_slug()
             queryset = queryset.filter(type=_type)
-        if 'q' in self.request.GET:
-            # using DEBUG as a shortcut for checking if we run on postgres vs sqlite
-            if settings.DEBUG:
-                qs = self.request.GET['q'].split()
-                search_query = Q()
-                for token in qs:
-                    search_query &= (
-                        Q(slug__icontains=token)
-                        | Q(name__icontains=token)
-                        | Q(description__icontains=token)
-                        | Q(latest_version__tags__name__icontains=token)
-                    )
-                queryset = queryset.filter(search_query).distinct()
-            else:
-                query = SearchQuery(self.request.GET['q'], search_type='websearch')
-                vector = (
-                    SearchVector('name', weight='A')
-                    + SearchVector('description', weight='B')
-                    + SearchVector('latest_version__tags__name', weight='C')
+
+        search_query = self.request.GET.get('q')
+        if not search_query:
+            return queryset
+
+        # WARNING: full-text search support only on postgres
+        # using DEBUG as a shortcut for checking if we run on postgres vs sqlite
+        if settings.DEBUG and 0:
+            filter = Q()
+            for token in search_query.split():
+                filter &= (
+                    Q(slug__icontains=token)
+                    | Q(name__icontains=token)
+                    | Q(description__icontains=token)
+                    | Q(latest_version__tags__name__icontains=token)
                 )
-                rank = SearchRank(vector, query)
-                queryset = (
-                    queryset.annotate(rank=rank).filter(rank__gte=0.3).distinct().order_by('-rank')
-                )
-        return queryset.prefetch_related(
-            'authors',
-            'latest_version__file',
-            'latest_version__tags',
-            'preview_set',
-            'preview_set__file',
-            'ratings',
-            'team',
-        )
+            queryset = queryset.filter(filter).distinct()
+        else:
+            queryset = self.postgres_fts(queryset, search_query)
+        return queryset
+
+    def postgres_fts(self, queryset, search_query):
+        """Postgres full text search (fast) and a fuzzy trigram search (slow) as a fallback.
+
+        Searches Extension name and description only. If we need to extend the functionality,
+        it's better to consider using a different approach, e.g. introduce meilisearch.
+
+        Limits the results size to 32 items (2 pages), assuming that nobody will click through many
+        pages if we failed to present the vital results on the first page.
+
+        Relies on indexed expressions:
+            CREATE INDEX extensions_fts ON extensions_extension USING
+            gin ((to_tsvector('english', name) || ' ' || to_tsvector('english', description)));
+            CREATE INDEX extensions_trgm_gin ON extensions_extension USING
+            gin((((name)::text || ' '::text) || description) gin_trgm_ops);
+
+        """
+        with connection.cursor() as cursor:
+            sql = """
+                select id
+                from extensions_extension
+                where (
+                    (to_tsvector('english', name) || ' ' || to_tsvector('english', description))
+                    @@ websearch_to_tsquery('english', %(query)s)
+                ) and is_listed
+                limit 32"""
+            cursor.execute(sql, {'query': search_query})
+            pks = [row[0] for row in cursor.fetchall()]
+            if not pks:
+                # fallback to fuzzy trigram search
+                sql = """
+                    select id
+                    from extensions_extension
+                    where ((name || ' ' || description) %%> %(query)s)
+                    and is_listed
+                    order by %(query)s <<<-> (name || ' ' || description)
+                    limit 32"""
+                cursor.execute(sql, {'query': search_query})
+                pks = [row[0] for row in cursor.fetchall()]
+            # pks are ordered by ranking, keep that order
+            # this approach is fine under the assumption that the list is small
+            return sorted(queryset.filter(pk__in=pks).order_by(), key=lambda x: pks.index(x.pk))
 
     def get_context_data(self, **kwargs):
         context = super().get_context_data(**kwargs)