Re: 7 logical-xor implementations in source tree

[Juri Linkov]

> That depends on the preferred semantics, right?  What semantics are you
> suggesting?  Judging from online discussions I've come across, there's
> more than one way to skin this cat.

Shouldn't xor have an associative property:

  (xor a b c) ≡ (xor a (xor b c)) ≡ (xor (xor a b) c)

> OTOH, a definition like the following:
>
>   (defsubst xor (cond1 cond2)
>     (cond ((not cond1) cond2)
>           ((not cond2) cond1)))
>
> not only performs simple logical xor more efficiently, but also returns
> whichever of the two arguments is non-nil (though I can't imagine when
> this last detail would ever be useful).

Returning a non-nil argument as-is is a useful property.
Most implementations lack this property, so better to use:

  (unless (and a b) (or a b))

But the implementation with 'cond' is nice too.

Another useful property is to evaluate arguments only once
that currently is not the case:

  (defvar a 0)
  (defvar b 0)
  (defmacro xor (a b) `(cond ((not ,a) ,b) ((not ,b) ,a)))
  (xor (setq a (1+ a)) nil)
  ⇒ 2

But I don't know if an implementation with such property
is possible at all (without using temporary variables).