diff --git a/jedi/evaluate/context/function.py b/jedi/evaluate/context/function.py
index c9735a3d..bbfa752d 100644
--- a/jedi/evaluate/context/function.py
+++ b/jedi/evaluate/context/function.py
@@ -328,13 +328,16 @@ def signature_matches(function_context, arguments):
         key, argument = next(unpacked_arguments, (None, None))
         if argument is None:
             # This signature has an parameter more than arguments were given.
-            return False
+            return bool(param_node.star_count)
         if key is not None:
             # TODO this is obviously wrong, we cannot just ignore keyword
             # arguments, but it's easier for now.
             return False
 
         if param_node.annotation is not None:
+            if param_node.star_count == 2:
+                return False  # TODO allow this
+
             annotation_contexts = function_context.evaluator.eval_element(
                 function_context.parent_context,
                 param_node.annotation
diff --git a/jedi/evaluate/context/instance.py b/jedi/evaluate/context/instance.py
index 957364c9..bbc538b1 100644
--- a/jedi/evaluate/context/instance.py
+++ b/jedi/evaluate/context/instance.py
@@ -487,31 +487,31 @@ class SelfAttributeFilter(ClassFilter):
 
 
 class InstanceArguments(AbstractArguments):
-    def __init__(self, instance, var_args):
+    def __init__(self, instance, arguments):
         self.instance = instance
-        self._var_args = var_args
+        self._arguments = arguments
 
     @property
     def argument_node(self):
-        return self._var_args.argument_node
+        return self._arguments.argument_node
 
     @property
     def trailer(self):
-        return self._var_args.trailer
+        return self._arguments.trailer
 
     def unpack(self, func=None):
         yield None, LazyKnownContext(self.instance)
-        for values in self._var_args.unpack(func):
+        for values in self._arguments.unpack(func):
             yield values
 
     def get_calling_nodes(self):
-        return self._var_args.get_calling_nodes()
+        return self._arguments.get_calling_nodes()
 
     def get_executed_params(self, execution_context):
-        if isinstance(self._var_args, AnonymousInstanceArguments):
-            return self._var_args.get_executed_params(execution_context)
+        if isinstance(self._arguments, AnonymousInstanceArguments):
+            return self._arguments.get_executed_params(execution_context)
 
         return super(InstanceArguments, self).get_executed_params(execution_context)
 
     def __repr__(self):
-        return '<%s: %s>' % (self.__class__.__name__, self._var_args)
+        return '<%s: %s>' % (self.__class__.__name__, self._arguments)
diff --git a/jedi/evaluate/context/iterable.py b/jedi/evaluate/context/iterable.py
index 0090676c..4064d59b 100644
--- a/jedi/evaluate/context/iterable.py
+++ b/jedi/evaluate/context/iterable.py
@@ -208,15 +208,19 @@ class Sequence(BuiltinOverwrite, IterableMixin):
         assert "Should never land here, probably an issue with typeshed changes"
 
     def _get_init_functions(self, instance):
-        from jedi.evaluate.context.function import OverloadedFunctionContext
         from jedi.evaluate import arguments
+        from jedi.evaluate.context.instance import InstanceArguments
         for init in instance.py__getattribute__('__init__'):
             try:
                 method = init.get_matching_functions
             except AttributeError:
                 continue
             else:
-                for x in method(arguments.ValuesArguments([ContextSet(self)])):
+                arguments = InstanceArguments(
+                    instance,
+                    arguments.ValuesArguments([ContextSet(self)])
+                )
+                for x in method(arguments):
                     yield x
 
     def py__bool__(self):
diff --git a/jedi/evaluate/context/klass.py b/jedi/evaluate/context/klass.py
index 05d3b736..ec818dfe 100644
--- a/jedi/evaluate/context/klass.py
+++ b/jedi/evaluate/context/klass.py
@@ -179,7 +179,10 @@ class ClassContext(use_metaclass(CachedMetaClass, TreeContext)):
                 continue  # These are not relevant for this search.
 
             from jedi.evaluate.pep0484 import find_unknown_type_vars
-            found += find_unknown_type_vars(self.parent_context, node)
+            for type_var in find_unknown_type_vars(self.parent_context, node):
+                if type_var not in found:
+                    # The order matters and it's therefore a list.
+                    found.append(type_var)
         return found
 
     @evaluator_method_cache(default=())
diff --git a/jedi/evaluate/context/typing.py b/jedi/evaluate/context/typing.py
index 40b97a09..d23d48d4 100644
--- a/jedi/evaluate/context/typing.py
+++ b/jedi/evaluate/context/typing.py
@@ -506,9 +506,9 @@ class _AbstractAnnotatedClass(ClassContext):
         if self.tree_node != other.tree_node:
             # TODO not sure if this is nice.
             return False
-
         given_params1 = self.get_given_types()
         given_params2 = other.get_given_types()
+
         if len(given_params1) != len(given_params2):
             # If the amount of type vars doesn't match, the class doesn't
             # match.
diff --git a/jedi/evaluate/pep0484.py b/jedi/evaluate/pep0484.py
index 42146310..69474bfe 100644
--- a/jedi/evaluate/pep0484.py
+++ b/jedi/evaluate/pep0484.py
@@ -21,7 +21,6 @@ x support for type hint comments for functions, `# type: (int, str) -> int`.
 
 import os
 import re
-from collections import OrderedDict
 
 from parso import ParserSyntaxError, parse, split_lines
 from parso.python import tree
@@ -178,7 +177,7 @@ def infer_param(execution_context, param):
 
 
 def py__annotations__(funcdef):
-    dct = OrderedDict()
+    dct = {}
     for function_param in funcdef.get_params():
         param_annotation = function_param.annotation
         if param_annotation is not None:
@@ -456,10 +455,10 @@ def find_unknown_type_vars(context, node):
         else:
             type_var_set = context.eval_node(node)
             for type_var in type_var_set:
-                if isinstance(type_var, TypeVar):
-                    found.add(type_var)
+                if isinstance(type_var, TypeVar) and type_var not in found:
+                    found.append(type_var)
 
-    found = set()
+    found = []  # We're not using a set, because the order matters.
     check_node(node)
     return found