Parsing of commented code throws 'misplaced type annotation'

[Flamefire]

Steps to reproduce

1. Try to parse e.g. tensorflow_core/python/ops/gen_resource_variable_ops.py

Alternatively the code that triggers this:

def _InitOpDefLibrary(op_list_proto_bytes):
  op_list = _op_def_pb2.OpList()
  op_list.ParseFromString(op_list_proto_bytes)
  _op_def_registry.register_op_list(op_list)
  op_def_lib = _op_def_library.OpDefLibrary()
  op_def_lib.add_op_list(op_list)
  return op_def_lib
# op {
#   name: "AssignAddVariableOp"
#   input_arg {
#     name: "resource"
#     type: DT_RESOURCE
#   }
#   input_arg {
#     name: "value"
#     type_attr: "dtype"
#   }
#   attr {
#     name: "dtype"
#     type: "type"
#   }
#   is_stateful: true
# }

Current behavior

Exception thrown (SyntaxError)

Expected behavior

Exception/Error ignored (as there is none)

python -c "from astroid import __pkginfo__; print(__pkginfo__.version)" output

2.3.1

Note that this happens when parsing tensorflow e.g. for pylint, so this is a real use case, although with presumptly generated code.

[PCManticore]

Thanks for submitting an issue! The problem was caused by typed_ast which we use to parse the source code for retrieving type comments. Unfortunately whenever it finds an invalid type comment, it fails with a SyntaxError. In your example it was treating the # type: comments as type comments.