Re: [RFC] caar/cadr/cdar/cddr

[Tom Tromey <tromey@redhat.com>]

>>>>> "Dmitry" == Dmitry Antipov <dmantipov@yandex.ru> writes:

Dmitry> These are widely used, so why not doing them as C primitives and
Dmitry> corresponding bytecodes?  It shouldn't be too hard to tweak
Dmitry> byte-opt.el to optimize (car (cdr X)) -> (cadr X) etc., and so
Dmitry> emit new bytecodes.

Did you try to measure the performance benefit?

Dmitry> +(byte-defop 48   0 byte-caar)
Dmitry> +(byte-defop 49   0 byte-cadr)
Dmitry> +(byte-defop 50   0 byte-cdar)
Dmitry> +(byte-defop 51   0 byte-cddr)

Occasionally I wish we were sharing these definitions with bytecode.c;
or more generally that parts of the Emacs core could be written in some
form of Lisp.

Dmitry> + loop:
Dmitry> +  if (CONSP (tem))
Dmitry> +    {
Dmitry> +      if (!EQ (tem, list))
Dmitry> +	return XCAR (tem);
Dmitry> +      tem = XCDR (tem);
Dmitry> +      goto loop;
Dmitry> +    }
Dmitry> +  else return NILP (tem) ? Qnil : wrong_type_argument (Qlistp, tem);

All these functions are implemented in what seems, to me, a very weird
way.  Why not just write the obvious straight-line code?

Tom