Re: Elisp manual, node "Comparison of Numbers"

[Richard Stallman]

    The problem here is that fuzz is a _relative_ measure of equality, and
    you employ it as an absolute measure here.

That is true -- but, in the case where one of the numbers is zero,
we have no way to apply a relative criterion to the other number.
There is no possible standard for it to be relative to.

Perhaps the idea of approximate equality can't be applied
to the case where either of the numbers is zero.
Perhaps it should always return nil in that case.

If so, the current definition is correct,
though not written very clearly for the result it gives.