tighten regex on objectify

There are some strange and wonderous work arounds in the json loader
for samples templates to do with templated values which are ints. This
was always kind of a hack, but when attempting to add a replacement
value in the middle of a string it was clear that the anchoring for
this regex was wrong.

Update the begin anchor to be ': ' instead of [^"], which would match
anywhere inside a string.

Change-Id: I9811b34d58a4d4cf7f181689b25212893880dcf4

nova/tests/functional/api_samples_test_base.py

@@ -43,8 +43,27 @@ class ApiSampleTestBase(integrated_helpers._IntegratedTestBase):
     def _objectify(self, data):
         if not data:
             return {}
-        # NOTE(vish): allow non-quoted replacements to survive json
-        data = re.sub(r'([^"])%\((.+)\)s([^"])', r'\1"%(int:\2)s"\3', data)
+        # NOTE(sdague): templates will contain values like %(foo)s
+        # throughout them. If these are inside of double quoted
+        # strings, life is good, and we can treat it just like valid
+        # json to load it to python.
+        #
+        # However we've got some fields which are ints, like
+        # aggregate_id. This means we've got a snippet in the sample
+        # that looks like:
+        #
+        #     "id": %(aggregate_id)s,
+        #
+        # which is not valid json, and will explode. We do a quick and
+        # dirty transform of this to:
+        #
+        #     "id": "%(int:aggregate_id)s",
+        #
+        # That makes it valid data to convert to json, but keeps
+        # around the information that we need to drop those strings
+        # later. The regex anchors from the ': ', as all of these will
+        # be top rooted keys.
+        data = re.sub(r'(\: )%\((.+)\)s([^"])', r'\1"%(int:\2)s"\3', data)
         return jsonutils.loads(data)
 
     @classmethod