start caching some parts of the imports to significantly speed up numpy completions.

jedi/api/classes.py

@@ -493,7 +493,8 @@ class Completion(BaseDefinition):
         # separately in Completion.
         if self._definition.isinstance(pr.Import) and self._definition.alias is None:
             i = imports.ImportPath(self._evaluator, self._definition, True)
-            return imports.Importer(i.import_path + [unicode(self._name)],
+            import_path = tuple(i.import_path + [unicode(self._name)])
+            return imports.Importer(self._evaluator, import_path,
                                     i._importer.module).follow(self._evaluator)
         return super(Completion, self)._follow_statements_imports()