How to use the jedi.inference.base_value.NO_VALUES function in jedi

def wrapper(function_value, param_index):
        inf = function_value.inference_state
        with recursion.execution_allowed(inf, function_value.tree_node) as allowed:
            # We need to catch recursions that may occur, because an
            # anonymous functions can create an anonymous parameter that is
            # more or less self referencing.
            if allowed:
                inf.dynamic_params_depth += 1
                try:
                    return func(function_value, param_index)
                finally:
                    inf.dynamic_params_depth -= 1
            return NO_VALUES
    return wrapper

if len(param_list) == 2 and len(arglist.children) == 3:
            (key1, _), (key2, lazy_value_cls) = param_list
            if key1 is None and key2 is None:
                call = _get_call_string(search_name)
                is_instance_call = _get_call_string(arglist.children[0])
                # Do a simple get_code comparison of the strings . They should
                # just have the same code, and everything will be all right.
                # There are ways that this is not correct, if some stuff is
                # redefined in between. However here we don't care, because
                # it's a heuristic that works pretty well.
                if call == is_instance_call:
                    lazy_cls = lazy_value_cls
    if lazy_cls is None:
        return None

    value_set = NO_VALUES
    for cls_or_tup in lazy_cls.infer():
        if isinstance(cls_or_tup, iterable.Sequence) and cls_or_tup.array_type == 'tuple':
            for lazy_value in cls_or_tup.py__iter__():
                value_set |= lazy_value.infer().execute_with_values()
        else:
            value_set |= cls_or_tup.execute_with_values()
    return value_set

def execution_recursion_decorator(default=NO_VALUES):
    def decorator(func):
        def wrapper(self, **kwargs):
            detector = self.inference_state.execution_recursion_detector
            limit_reached = detector.push_execution(self)
            try:
                if limit_reached:
                    result = default
                else:
                    result = func(self, **kwargs)
            finally:
                detector.pop_execution()
            return result
        return wrapper
    return decorator

def infer(inference_state, context, leaf):
    if leaf.type == 'name':
        return inference_state.infer(context, leaf)

    parent = leaf.parent
    definitions = NO_VALUES
    if parent.type == 'atom':
        # e.g. `(a + b)`
        definitions = context.infer_node(leaf.parent)
    elif parent.type == 'trailer':
        # e.g. `a()`
        definitions = infer_call_of_leaf(context, leaf)
    elif isinstance(leaf, tree.Literal):
        # e.g. `"foo"` or `1.0`
        return infer_atom(context, leaf)
    elif leaf.type in ('fstring_string', 'fstring_start', 'fstring_end'):
        return get_string_value_set(inference_state)
    return definitions

def _follow_param(inference_state, arguments, index):
    try:
        key, lazy_value = list(arguments.unpack())[index]
    except IndexError:
        return NO_VALUES
    else:
        return lazy_value.infer()

def import_module(inference_state, import_names, parent_module_value, sys_path):
    """
    This method is very similar to importlib's `_gcd_import`.
    """
    if import_names[0] in settings.auto_import_modules:
        module = _load_builtin_module(inference_state, import_names, sys_path)
        if module is None:
            return NO_VALUES
        return ValueSet([module])

    module_name = '.'.join(import_names)
    if parent_module_value is None:
        # Override the sys.path. It works only good that way.
        # Injecting the path directly into `find_module` did not work.
        file_io_or_ns, is_pkg = inference_state.compiled_subprocess.get_module_info(
            string=import_names[-1],
            full_name=module_name,
            sys_path=sys_path,
            is_global_search=True,
        )
        if is_pkg is None:
            return NO_VALUES
    else:
        paths = parent_module_value.py__path__()

def infer_param(function_value, param):
    def infer_docstring(docstring):
        return ValueSet(
            p
            for param_str in _search_param_in_docstr(docstring, param.name.value)
            for p in _infer_for_statement_string(module_context, param_str)
        )
    module_context = function_value.get_root_context()
    func = param.get_parent_function()
    if func.type == 'lambdef':
        return NO_VALUES

    types = infer_docstring(function_value.py__doc__())
    if function_value.is_bound_method() \
            and function_value.py__name__() == '__init__':
        types |= infer_docstring(function_value.class_context.py__doc__())

    debug.dbg('Found param types for docstring: %s', types, color='BLUE')
    return types

def py__call__(self, item_value_set):
        value_set = NO_VALUES
        for args_value in self._args_value_set:
            lazy_values = list(args_value.py__iter__())
            if len(lazy_values) == 1:
                # TODO we need to add the contextualized value.
                value_set |= item_value_set.get_item(lazy_values[0].infer(), None)
            else:
                value_set |= ValueSet([iterable.FakeList(
                    self._wrapped_value.inference_state,
                    [
                        LazyKnownValues(item_value_set.get_item(lazy_value.infer(), None))
                        for lazy_value in lazy_values
                    ],
                )])
        return value_set

def py__call__(self, arguments):
        debug.dbg("Execute overloaded function %s", self._wrapped_value, color='BLUE')
        function_executions = []
        for signature in self.get_signatures():
            function_execution = signature.value.as_context(arguments)
            function_executions.append(function_execution)
            if signature.matches_signature(arguments):
                return function_execution.infer()

        if self.inference_state.is_analysis:
            # In this case we want precision.
            return NO_VALUES
        return ValueSet.from_sets(fe.infer() for fe in function_executions)

def py__getattribute__alternatives(self, string_name):
        '''
        Since nothing was inferred, now check the __getattr__ and
        __getattribute__ methods. Stubs don't need to be checked, because
        they don't contain any logic.
        '''
        if self.is_stub():
            return NO_VALUES

        name = compiled.create_simple_object(self.inference_state, string_name)

        # This is a little bit special. `__getattribute__` is in Python
        # executed before `__getattr__`. But: I know no use case, where
        # this could be practical and where Jedi would return wrong types.
        # If you ever find something, let me know!
        # We are inversing this, because a hand-crafted `__getattribute__`
        # could still call another hand-crafted `__getattr__`, but not the
        # other way around.
        names = (self.get_function_slot_names(u'__getattr__')
                 or self.get_function_slot_names(u'__getattribute__'))
        return self.execute_function_slots(names, name)