Fix generator issues for typing

jedi/evaluate/context/iterable.py

@@ -76,7 +76,7 @@ class GeneratorBase(LazyAttributeOverwrite, IterableMixin):
 
     @property
     def name(self):
-        return compiled.CompiledContextName(self, 'generator')
+        return compiled.CompiledContextName(self, 'Generator')
 
 
 class Generator(GeneratorBase):

jedi/evaluate/gradual/conversion.py

@@ -1,3 +1,4 @@
+from jedi import debug
 from jedi.evaluate.base_context import ContextSet, \
     NO_CONTEXTS
 from jedi.evaluate.utils import to_list
@@ -49,11 +50,13 @@ def _infer_from_stub(stub_module, qualified_names, ignore_compiled):
 
 
 def try_stubs_to_actual_context_set(stub_contexts, prefer_stub_to_compiled=False):
-    return ContextSet.from_sets(
+    contexts = ContextSet.from_sets(
         stub_to_actual_context_set(stub_context, ignore_compiled=prefer_stub_to_compiled)
         or ContextSet([stub_context])
         for stub_context in stub_contexts
     )
+    debug.dbg('Stubs to actual: %s to %s', stub_contexts, contexts)
+    return contexts
 
 
 @to_list

jedi/plugins/stdlib.py

@@ -575,5 +575,11 @@ _implemented = {
         # Not sure if this is necessary, but it's used a lot in typeshed and
         # it's for now easier to just pass the function.
         'abstractmethod': _return_first_param,
-    }
+    },
+    'typing': {
+        # The _alias function just leads to some annoying type inference.
+        # Therefore, just make it return nothing, which leads to the stubs
+        # being used instead. This only matters for 3.7+.
+        '_alias': lambda obj, arguments: NO_CONTEXTS,
+    },
 }

test/test_api/test_api.py

@@ -128,7 +128,7 @@ def test_goto_definitions_on_non_name(Script):
 
 def test_goto_definitions_on_generator(Script):
     def_, = Script('def x(): yield 1\ny=x()\ny').goto_definitions()
-    assert def_.name == 'generator'
+    assert def_.name == 'Generator'
 
 
 def test_goto_definition_not_multiple(Script):