> Hmm.  I get the argument, but wouldn't that mean that only a tiny set of
> builtins were pure?

Indeed, according to that definition of pure, very few Elisp
functions qualify.

Which is why this is not exactly the definition used for pure-fns.