Using the 'reduce_max' function along a zero size dimesion

Hi there,
I am to use ‘tf.reduce_max’ on tensor of arbitrary shape including possible zero size axes. As I can see, it results in ‘-inf’ for a regular tensor and ‘tf.float32.min’ for a ragged tensor (tensor’s ‘dtype’ is ‘float32’).
Is this a documented behavior? Can I be sure if it will remain unchanged?