tf_executor.fetch %outputs : tensor<8x1xf32>
  }
  func.return %0 : tensor<8x1xf32>
}

// Check that we generate _handle_dtypes and _handle_shapes for the resource
// argument.

// CHECK:      op: "_Arg"

// CHECK:        key: "_handle_dtypes"
// CHECK-NEXT:   value {
// CHECK-NEXT:     list {
// CHECK-NEXT:       type: DT_FLOAT
// CHECK-NEXT:     }
// CHECK-NEXT:   }
// CHECK-NEXT: }

  op: "_Arg"
  device: "/job:localhost/replica:0/task:0/device:GPU:0"
  attr {
    key: "T"
    value {
      type: DT_RESOURCE
    }
  }
  attr {
    key: "_handle_dtypes"
    value {
      list {
        type: DT_FLOAT
      }
    }
  }
  attr {
    key: "_handle_shapes"
    value {
      list {
        shape {
          dim {
            size: 1001

    key: "index"
    value {
      i: 0
    }
  }
}
node {
  name: "arg1"
  op: "_Arg"
  attr {
    key: "T"
    value {
      type: DT_RESOURCE
    }
  }
  attr {
    key: "_handle_dtypes"
    value {
      list {
        type: DT_FLOAT
      }
    }
  }
  attr {
    key: "_handle_shapes"
    value {
      list {
        shape {
        }
      }
    }
  }

    key: "index"
    value {
      i: 0
    }
  }
}
node {
  name: "args_1"
  op: "_Arg"
  attr {
    key: "T"
    value {
      type: DT_RESOURCE
    }
  }
  attr {
    key: "_handle_dtypes"
    value {
      list {
        type: DT_FLOAT
      }
    }
  }
  attr {
    key: "_handle_shapes"
    value {
      list {
        shape {
          dim {
            size: 3

        SetTensorShapeProto(subtype,
                            handle_shapes_attr.mutable_list()->add_shape());
      }

      (*node_def->mutable_attr())["_handle_dtypes"] = handle_dtypes_attr;
      (*node_def->mutable_attr())["_handle_shapes"] = handle_shapes_attr;
    }
  }

  TF_RETURN_IF_ERROR(
      SetShapeAttribute("_output_shapes", arg_type, node_def->mutable_attr()));

        SetTensorShapeProto(subtype,
                            handle_shapes_attr.mutable_list()->add_shape());
      }

      (*node_def->mutable_attr())["_handle_dtypes"] = handle_dtypes_attr;
      (*node_def->mutable_attr())["_handle_shapes"] = handle_shapes_attr;
    }
  }

  TF_RETURN_IF_ERROR(
      SetShapeAttribute("_output_shapes", arg_type, node_def->mutable_attr()));

    if (dtype == DT_RESOURCE) {
      const AttrValue* dtype_attr = node.attrs().Find("_handle_dtypes");
      const AttrValue* shape_attr = node.attrs().Find("_handle_shapes");
      if (dtype_attr && shape_attr) {
        if (dtype_attr->list().type().empty()) {
          return errors::InvalidArgument(
              "Invalid \"_handle_dtypes\" attribute value for _Arg node: ",
              shape_attr->DebugString());
        }

}

Type GetTensorTypeForTensorList(Type element_type, TF::VariantType handle_dtype,
                                PatternRewriter *rewriter) {
  // If the variant type in the output handle has item shape available, use it
  // to derive the output shape by setting unknown leading dimension.
  // Otherwise, result type will be of unranked type.
  if (handle_dtype.getSubtypes().empty()) {
    return UnrankedTensorType::get(element_type);
  }

  let extraClassDeclaration = [{
    // Returns type of the TensorList element produced by this op.
    TensorType element_type() { return handle_dtype().getSubtypes()[0]; }

    // Returns data type of the result handle. Returned type contains type of
    // the TensorList element as a subtype.
    VariantType handle_dtype() {
      return getElementTypeOrSelf(getHandle().getType()).cast<TF::VariantType>();
    }
  }];
}

  // This is required to populate derived attributes during export in a
  // meaningful way. Else during export to GraphDef element_type() query
  // will result in out of bounds access/assert.
  if (handle_dtype().getSubtypes().size() != 1) {
    return emitOpError(
        "must have exactly one subtype in the result variant type");
  }
  if (!IsOfRankOrUnranked(op.getElementShape(), 0) &&