dolist vs. while-loop (was Re: epg and expired recipient keys)

[Katsumi Yamaoka <yamaoka@jpl.org>]

>>>>> In <87k5xs26v2.fsf@catnip.gol.com> Miles Bader wrote:

> Miles Bader <miles@gnu.org> writes:

>> That isn't true in general; it may be true if you're using cl.el's
>> craptastic version of dolist.  Unfortunately, gnus uses cl, so I guess
>> it loses there...
>>
>> [The non-cl version of dolist is a macro that expands to basically the
>> same code as your hand-written while loop.]

> BTW, to be fair to cl.el, I think it tries to do additional tricks at
> compilation time to get rid of the additional overhead of its version of
> dolist (by removing unnecessary calls to catch/throw that its macro
> previously inserted to emulate common-lisp blocks).  I don't know how
> effective this is though.

Yes, there's no difference between the byte codes of the subr.el
version and the cl-macs.el version except for variable names.
But I don't know in what case and for what purpose the byte code
of the cl-macs.el version will have the catch-throw form.  That
is a mystery.

BTW, the results of the benchmark tests I wrote were done in
XEmacs 21.4.20 (<b4modn7q9jr.fsf@jpl.org>).  In Emacs 22 (in the
same host), I got:

[while loop]
1.0685009956359863
1.074789047241211
1.0869169235229492
1.1011860370635986
1.0744349956512451

[dolist]
1.122694969177246
1.1409659385681152
1.140753984451294
1.1298248767852783
1.1383061408996582

I also used Lars' benchmark.el.  Please note that the later test
has nested two dolist forms like the one that Wolfram Fenske wrote.
I guess the reason it is slower than the while-loop version is that
the `let' form in the inside dolist is used many times.

Regards,