Here, the calling line np.pad(a, 2, pad_with) works magically!. Not able to find the corresponding python PEP document.

    >>> def pad_with(vector, pad_width, iaxis, kwargs):
    ...     pad_value = kwargs.get('padder', 10)
    ...     vector[:pad_width[0]] = pad_value
    ...     vector[-pad_width[1]:] = pad_value
    >>> a = np.arange(6)
    >>> a = a.reshape((2, 3))
    >>> np.pad(a, 2, pad_with)
    array([[10, 10, 10, 10, 10, 10, 10],
           [10, 10, 10, 10, 10, 10, 10],
           [10, 10,  0,  1,  2, 10, 10],
           [10, 10,  3,  4,  5, 10, 10],
           [10, 10, 10, 10, 10, 10, 10],
           [10, 10, 10, 10, 10, 10, 10]])

What is the hidden syntax in details? When should we use or avoid its abuse?

I knew how to use default arguments and *args and **kwargs for optional arguments, but not familiar with such a case where the call np.pad(a, 2, pad_with) does not use () either and still works as expected.

NOTE: It might be partially answered by this post, but the procedures when stepping inside help with understanding its usage.