[Ecls-list] [ANN] ECL 9.7.1

[Gabriel Dos Reis]

On Mon, Jul 13, 2009 at 9:54 AM, Juan Jose
Garcia-Ripoll<juanjose.garciaripoll at googlemail.com> wrote:
> Since nobody complained, I took the liberty to produce this release,
> which is needed by other projects. If there are further problems, a
> 9.7.2 can be done.
>
> A couple of remarks. With respect to the last week, the current
> sources contain a important fix in the code that inlines functions
> using C expressions. More precisely, there was a bug in the matcher
> from argument types to inline forms. This actually made almost all
> inline forms unused, even the safe ones. In other words, instead of
> using the fast C macro CONS, we were calling cl_cons(), and instead of
> the C library function sqrt(), we were using the generic cl_sqrt(),
> etc, etc

so, I guess we should be expecting 'faster CONS', or at least see some
benefit of it?


>
> On the ugly side, this means that if some of those inactive inline
> forms now turn out to be not so safe and stable, ECL might reveal new
> problems. However, on a cursory inspection, and after the usual stress
> tests, I can see no significant problems ahead and instead a
> noticeable speed up...

I'll report after rebuild from ECL CVS as of today.

>
> ... ok, perhaps not for you Gabriel ;-)

:-)

> Actually, I am beginning to
> think that the problems in OpenAxion are related to inefficiencies in
> the compiler, not on ECL itself.

you mean OpenAxiom compiler or ECL compiler?


There certainly are room for improvements in the OpenAxiom compiler.  However,
on the same machine where it takes 2h30min ECL-based build, I have
25-28min for SBCL and GCL, and 1h30min for CLISP (as of yesterday).

> 2 1/2 hours seems like a lot of time to build...

Yup.  :-(

The OpenAxiom compiler does use the LISP compiler a lot -- COMPILE-FILE
for the build, and COMPILE for interactive use.


Anyways, many thanks for your outstanding work.

-- Gaby