Ken, why is this patch correct? It seems like you avoid initializing the address member at all some of the time, not initialize it to the address the krb524d receives from the KDC.