Simple error recovery in the parser: For now just discard everything that is not a suite or file_input, if we detect an error.

jedi/parser/pgen2/parse.py

@@ -144,6 +144,7 @@ class Parser(object):
                                          type, value, context)
                 else:
                     self.error_recovery(type, value, context)
+                    break
 
     def classify(self, type, value, context):
         """Turn a token into a label.  (Internal)"""
@@ -201,5 +202,16 @@ class Parser(object):
                     index = i
             self.stack[index:] = []
         else:
+            # For now just discard everything that is not a suite or
+            # file_input, if we detect an error.
+            for i, (dfa, state, node) in reversed(list(enumerate(self.stack))):
+                symbol, _, _, _ = node
+
+                # `suite` can sometimes be only simple_stmt, not stmt.
+                if symbol in (self.grammar.symbol2number['file_input'],
+                              self.grammar.symbol2number['suite']):
+                    index = i
+                    break
+            self.stack[index + 1:] = []
             # No success finding a transition
-            raise ParseError("bad input", type, value, context)
+            #raise ParseError("bad input", type, value, context)

jedi/parser/representation.py

@@ -534,7 +534,7 @@ class Scope(Simple, DocstringMixin):
             if is_node(c, 'simple_stmt'):
                 names += chain.from_iterable(
                     [s.get_defined_names() for s in c.children
-                     if isinstance(s, (ExprStmt, Import, KeywordStatement))])
+                     if isinstance(s, (ExprStmt, Import))])
             elif isinstance(c, (Function, Class)):
                 names.append(c.name)
         return names