Re: [Texmacs-dev] First results on Schemes, Lisps & others

[Joris van der Hoeven]

On Sat, 22 May 2004, David MENTRE wrote:
> I have redone most of Doug Bagley's Language Shootout on Schemes, Lisps
> and other languages (python as a well known reference, gcc and g++ for
> max speed reference, ocaml because it is fast and I like it :-).

Thanks a lot; this is really very helpful and exhaustive. Could you please
try some of the other Scheme's which were mentioned by Gilles and others
on this list a month or so ago? It would be interesting to find the best
interpreted (maybe using byte-code) Scheme implementation. We should also
investigate whether we may use compiled Schemes.

> Doug's benches are interesting because there are both compute intensive
> ones (sieve, matrix, moments, ...) and string manipulation ones
> (spellcheck, wc, wordfred, ...).

Yes, yes, very exhaustive.

> All the scripts & results are available at:
> http://www.linux-france.org/~dmentre/texmacs/scheme/scheme-shoot.tar.gz

Thanks.

>  - python and guile are very slow, guile being very very very slow (Oh
>    surprise! ;)

;^(((

>  - cmucl (common lisp) and bigloo (scheme) are close to each other and
>    are both very fast. Both of them are compiled and I don't know how
>    they can be used and their speed in interpreter mode;

We should investigate that.

>  - regarding interpreters, OCaml in bytecode (ocamlb) is the better, but
>    it is not a Lisp-like language;

Would it be possible (easy) to *simulate* scheme from within OCaml
(if necessary without continuations, which we cannot use anyway
currently (unfortunately)). When loading a scheme program,
we might rewrite it as an Ocaml program and execute it.
I think that this mainly supposes that there are equivalents
for the define-macro construct.

> I have not investigated other Joris' criteria (licence, capabilities,
> ...).

We can do that later (except for the license issue; no need
to try proprietary software; we do not want to use it).
Let us first concentrate on raw speed.