diff --git a/jedi/evaluate/base_context.py b/jedi/evaluate/base_context.py
index ca2e074c..19df52ea 100644
--- a/jedi/evaluate/base_context.py
+++ b/jedi/evaluate/base_context.py
@@ -33,26 +33,14 @@ class Context(BaseContext):
         # overwritten.
         return self.__class__.__name__.lower()
 
-    @debug.increase_indent
-    def execute(self, arguments):
-        """
-        In contrast to py__call__ this function is always available.
-
-        `hasattr(x, py__call__)` can also be checked to see if a context is
-        executable.
-        """
-        if self.evaluator.is_analysis:
-            arguments.eval_all()
-
-        return self.evaluator.execute(self, arguments)
-
     def execute_evaluated(self, *value_list):
         """
         Execute a function with already executed arguments.
         """
+        # TODO move this out of here to the evaluator.
         from jedi.evaluate.arguments import ValuesArguments
         arguments = ValuesArguments([ContextSet(value) for value in value_list])
-        return self.execute(arguments)
+        return self.evaluator.execute(self, arguments)
 
     def iterate(self, contextualized_node=None, is_async=False):
         debug.dbg('iterate %s', self)
@@ -248,6 +236,9 @@ class ContextSet(BaseContextSet):
                 [l for l in lazy_contexts if l is not None]
             )
 
+    def execute(self, arguments):
+        return ContextSet.from_sets(c.evaluator.execute(c, arguments) for c in self._set)
+
 
 NO_CONTEXTS = ContextSet()
 
diff --git a/jedi/evaluate/compiled/context.py b/jedi/evaluate/compiled/context.py
index 2b87e3b1..3950aff8 100644
--- a/jedi/evaluate/compiled/context.py
+++ b/jedi/evaluate/compiled/context.py
@@ -187,7 +187,7 @@ class CompiledObject(Context):
                 continue
             else:
                 bltn_obj = builtin_from_name(self.evaluator, name)
-                for result in bltn_obj.execute(params):
+                for result in self.evaluator.execute(bltn_obj, params):
                     yield result
         for type_ in docstrings.infer_return_types(self):
             yield type_
diff --git a/jedi/evaluate/context/instance.py b/jedi/evaluate/context/instance.py
index def5e19a..f6fc344a 100644
--- a/jedi/evaluate/context/instance.py
+++ b/jedi/evaluate/context/instance.py
@@ -63,7 +63,7 @@ class AbstractInstanceContext(Context):
             raise AttributeError
 
         def execute(arguments):
-            return ContextSet.from_sets(name.execute(arguments) for name in names)
+            return ContextSet.from_sets(name.infer().execute(arguments) for name in names)
 
         return execute
 
diff --git a/jedi/plugins/typeshed.py b/jedi/plugins/typeshed.py
index 0126e7e4..6091642a 100644
--- a/jedi/plugins/typeshed.py
+++ b/jedi/plugins/typeshed.py
@@ -5,7 +5,7 @@ from pkg_resources import resource_filename
 from jedi._compatibility import FileNotFoundError
 from jedi.plugins.base import BasePlugin
 from jedi.evaluate.cache import evaluator_as_method_param_cache
-from jedi.evaluate.base_context import Context, ContextSet
+from jedi.evaluate.base_context import Context, ContextSet, NO_CONTEXTS
 from jedi.evaluate.context import ModuleContext
 
 
@@ -19,7 +19,8 @@ def _create_stub_map(directory):
     def generate():
         try:
             listed = os.listdir(directory)
-        except FileNotFoundError:
+        except (FileNotFoundError, OSError):
+            # OSError is Python 2
             return
 
         for entry in listed:
@@ -85,7 +86,7 @@ class TypeshedPlugin(BasePlugin):
             # ``os``.
             mapped = self._cache_stub_file_map(evaluator.grammar.version_info)
             context_set = callback(evaluator, import_names, module_context, sys_path)
-            if len(import_names) == 1:
+            if len(import_names) == 1 and import_names[0] != 'typing':
                 path = mapped.get(import_names[0])
                 if path is not None:
                     try:
@@ -119,15 +120,25 @@ class StubProxy(object):
         context_results = self._context.py__getattribute__(
             *args, **kwargs
         )
-        typeshed_results = self._stub_context.py__getattribute__(
+        typeshed_results = list(self._stub_context.py__getattribute__(
             *args, **kwargs
+        ))
+        if not typeshed_results:
+            return NO_CONTEXTS
+
+        return ContextSet.from_iterable(
+            StubProxy(c.parent_context, c, typeshed_results[0]) for c in context_results
         )
-        print()
-        print(context_results, typeshed_results)
-        return context_results
+
+    @property
+    def py__call__(self):
+        def py__call__(arguments):
+            return self._stub_context.py__call__(arguments)
+
+        return py__call__
 
     def __getattr__(self, name):
         return getattr(self._context, name)
 
     def __repr__(self):
-        return '<%s: %s>' % (type(self).__name__, self.access_handle.get_repr())
+        return '<%s: %s %s>' % (type(self).__name__, self._context, self._stub_context)