Re: [Texmacs-dev] More Scheme benchmarks

[Joris van der Hoeven]

> I have added several Schemes to Doug benchmark. Right know, I have only
> added compute intensive ones because they only use standardised
> Scheme. I intend to port other benches only on Schemes we find
> interesting.

Great; this is really helpful job.

> >From those first results, I find RScheme, STklos and gauche the fastest
> ones. Of course, they are far behind compiled Scheme, Bigloo and Stalin
> (stalin is the fastest, but it requires long compilation time). And of
> course, no Scheme stands above the crowd on all tests.

We probably should consider using a compiled scheme. Maybe someone can
investigate whether there exists a Scheme implementation which can
smoothly mix compiled and interpreted code. Also: for compiled schemes,
are there any restrictions on 1) the module system 2) the macro system.

> Licenses:
>
>  - RScheme: unusual one. Seems BSD-like.
>
>  - STklos: GNU GPL
>
>  - Gauche: BSD-like (many sub license, would need to check but I trust
>    the Debian guys)
>
>  - Bigloo: GNU GPL
>
>  - Stalin: GNU GPL

Seems OK.

> I would avoid MZScheme, it is regularly slower than Guile.

OK, nice to know, because I was considering that implementation...

> As choice criteria, one should also consider if the Scheme implements
> tail recursion. For example, it seems that Gauche does not implement it.

Do interpreters implement that (maybe byte code interpreters;
by the way, which Scheme interpreters use byte code?).

Another question: would it be possible for you to send us the benchmarks
for different *versions* of Guile. I am very interested in differences
between versions 1.3.4 (Unix), 1.4 (Windows), 1.6.* (Last).

Thanks for your sunday afternoon ;^)