Both [1] and [2] seem to work ,
though on [2], node() still seems to be the slightly faster alternative
(when knowing that elements can only contain at least one non-whitespace
character, or are empty otherwise, as is the case with my data).