Stefan Monnier <monnier@iro.umontreal.ca> schrieb am Di., 25. Juli 2017 um
23:27 Uhr:

> > Why? Its return value clearly only depends on its argument, and it
> doesn't
> > change any global state. It's the poster child of a pure function!
>
>   (let ((s (make-string 5 ?a)))
>     (list (string-to-char s)
>           (progn
>             (aset s 0 ?b)
>             (string-to-char s))))
>
> If string-to-char were a pure function, it would return the same value
> in both calls (since the arguments are `eq').
>
>
Hmm. I get the argument, but wouldn't that mean that only a tiny set of
builtins were pure? If the definition is "A function is pure if applying it
to the same objects (in the sense of `eq') always gives the same result",
then I think only eq, eql, null, and the type predicates (integerp etc.)
are pure. Even the arithmetic functions aren't pure because they can act on
(mutable) markers. The list in byte-opt, however, is noticeably different,
and none of the functions marked pure there are actually pure. The same
applies to third-party libraries such as dash.el.
In any case, the precise definition of "pure" and "side effect free" needs
to be in the manual, and functions that are wrongly declared as pure or
side effect free need to be fixed.