Boost on exact matches in all locales (bug 972039)

Author: Rob Hudson

mkt/search/filters.py

@@ -50,10 +50,10 @@ def filter_queryset(self, request, queryset, view):
                 should.append(k(**{field: v}))
 
         # Exact matches need to be queried against a non-analyzed field. Let's
-        # do a term query on `name_sort` for an exact match against the app
+        # do a term query on `name.raw` for an exact match against the app
         # name and give it a good boost since this is likely what the user
         # wants.
-        should.append(query.Term(name_sort={'value': q, 'boost': 10}))
+        should.append(query.Term(**{'name.raw': {'value': q, 'boost': 10}}))
 
         if analyzer:
             should.append(

mkt/search/tests/test_views.py

@@ -536,6 +536,28 @@ def test_name_localized_to_default_locale(self):
         eq_(obj['slug'], self.webapp.app_slug)
         eq_(obj['name'], u'Algo Algo Steamcube!')
 
+    def test_name_localized_exact_match(self):
+        other_apps = [app_factory(), app_factory()]
+        other_apps[0].name = {'en-US': 'Spanish Tests',
+                              'es': 'Pruebas de Español'}
+        other_apps[1].name = {'en-US': 'Tests in Spanish',
+                              'es': 'Pruebas en Español'}
+        for app in other_apps:
+            app.save()
+        self.refresh('webapp')
+
+        res = self.anon.get(self.url, data={'q': 'Pruebas en Español',
+                                            'lang': 'es'})
+        eq_(res.status_code, 200)
+
+        # Ensure the exact matched name is first.
+        obj = res.json['objects'][0]
+        eq_(obj['id'], other_apps[1].id)
+
+        for app in other_apps:
+            app.delete()
+        unindex_webapps([app.pk for app in other_apps])
+
     def test_author(self):
         res = self.anon.get(self.url,
                             data={'author': self.webapp.developer_name})

mkt/webapps/indexers.py

@@ -133,8 +133,23 @@ def _locale_field_mapping(field, analyzer):
                     'manifest_url': cls.string_not_analyzed(),
                     'modified': {'format': 'dateOptionalTime',
                                  'type': 'date'},
-                    # Name for searching.
-                    'name': {'type': 'string', 'analyzer': 'default_icu'},
+                    # Name for searching. This is a list of all the localized
+                    # names for the app. We add "position_offset_gap" to work
+                    # around the fact that ES stores the same list of tokens as
+                    # if this were a single string. The offset gap adds 100
+                    # positions between each name and ensures one string from
+                    # one name and one string from another name won't both
+                    # match with a phrase match query.
+                    'name': {
+                        'type': 'string',
+                        'analyzer': 'default_icu',
+                        'position_offset_gap': 100,
+                        # For exact matches. Referenced as `name.raw`.
+                        'fields': {
+                            'raw': cls.string_not_analyzed(
+                                position_offset_gap=100)
+                        },
+                    },
                     # Name for sorting.
                     'name_sort': cls.string_not_analyzed(doc_values=True),
                     # Name for suggestions.

settings_test.py

@@ -53,7 +53,9 @@ def _polite_tmpdir():
 DEBUG_PROPAGATE_EXCEPTIONS = False
 EMAIL_BACKEND = 'django.core.mail.backends.locmem.EmailBackend'
 ES_DEFAULT_NUM_REPLICAS = 0
-ES_DEFAULT_NUM_SHARDS = 3
+# See the following URL on why we set num_shards to 1 for tests:
+# http://www.elasticsearch.org/guide/en/elasticsearch/guide/current/relevance-is-broken.html
+ES_DEFAULT_NUM_SHARDS = 1
 IARC_MOCK = True
 IN_TEST_SUITE = True
 INSTALLED_APPS += ('mkt.translations.tests.testapp',)