[Ecls-list] cl-bench problem/oddity with latest ecls-cvs

[Michael Abshoff]

Juan Jose Garcia-Ripoll wrote:

Hello Juanjo,

> BTW, there is something bogus in this benchmark -- probably a high
> level of recursion in some function -- both both ECL interpreted and
> Clisp fail to finish. The compiled version using ECL works just fine.
>

Okay, good to know.

The reason I ran the benchmark was/is to get a feel how ecls and clisp
compare, what are their respective strength and so on.

My initial interest in ecls is due to the Sage project. I do a lot of
their porting and clisp has proven to be difficult to get to compile on a
wide range of platforms, i.e.

 Linux: x86, x86-64, Itanium (PPC, PPC 64 later)
 Solaris 9/10: Sparc/x86-64
 OSX: PPC/Intel (32 bit for now, 64 bit later)
 (later: Windows: x86/x86-64 with MSVC, the Cygwin port has been dropped)

So as you can imagine it isn't easy to find a lisp implementation that
covers all those and runs Axiom as well as Maxima and compiles from
scratch without the need to bootstrap itself via another lisp compiler.
There doesn't seem to be official ecls support for Maxima yet, but I hope
that it can be implemented. FriCAS supports ecls, OpenAxiom will with the
next 1.1.0 relase (it works now, but needs cleanup).

What makes ecls especially attractive for Sage is that it can be embedded
(hence its name), and that opens up the potential the integrate Axiom (and
Maxime) as a library into Sage instead of relying on pexcept and parsing
of text. The fact that the codebase is pretty clean and I could get it to
work on Solaris in less than an hours just speaks for it. I also checked
cvs and it seems that the MSVC makefiles have been updated recently, so I
am hopeful that when I have some time that ecls will build with MSVC
compilers without too much pain.

So ecls looks very promising, and I am willing to do testing on a wide
range of platforms (see above ;)) and also try my best to fix bugs in the
code as best as I can, the only thing I need to do is to be able to sell
ecls to Sage as the solution to our lisp problem. You win in portability
hands down, it build in about 6 minutes of my Opteron which is also pretty
good, but I needed to get a feeling for the relative performance. That is
why I tried cl-bench and was surprised it didn't work. I hope that the
issue I described in the other email can be resolved, I certainly can work
around it.

> Juanjo
>

Cheers,

Michael
> --
> Facultad de Fisicas, Universidad Complutense,
> Ciudad Universitaria s/n Madrid 28040 (Spain)
> http://juanjose.garciaripoll.googlepages.com
>