[Texmacs-dev] Embedded Lisp / Scheme

[Karl M. Hegbloom]

On Thu, 2003-05-08 at 02:29, Joris van der Hoeven wrote:
> On Wed, 7 May 2003, Leo wrote:
> > On Wednesday May 07 2003 06:46, Joris van der Hoeven wrote:
> > > Hi,
> > > [...snip...]
> > >
> > > Unfortunately, Guile is a *very bad* Scheme implementation,
> > > so the use of DRDs is slow.
> >
> > I second your opinion that guile is one of the worst
> > scheme implementations available.
> > That said, do you plan "to change the horse" before it is too late.
> > TeXmacs is becoming dependent on more and more guile specific constructs
> > like guile module system which makes porting effort ever harder??

I think that the reason that Guile is so slow is that it uses a
tree-code interpreter rather than a byte code virtual machine.  A long
time ago, I noted that Scwm was a lot slower than Sawfish, and sent mail
to the guile team telling them about sawfish and the librep virtual
machine.  Soon after that, the "qscheme" (iirc) virtual machine showed
up in the Guile repository -- they found a virtual machine or wrote one
and at that point planned to incorporate it into guile.  I don't know
what happened to that.  It seems to me that in the guile repository are
a compiler, that VM, etc.

> First of all, in the next couple of months I will mainly be
> reorganizing C++ code, so the Guile interface won't change a lot,
> except for additions by David. I asked David not to depend on
> new Scheme features and even less on Guile specific features.
> 
> Secondly, even if we use another Scheme implementation,
> we want to remain compatible with Guile too,
> because it *is* the implementation which is most available.
> Notice also that we want to remain compatible with
> all older versions of Guile.
> 
> Third, I have understood that the MzScheme guys try to
> provide a Guile compatability library (for Guile 1.3.4,
> but that is OK for us, since we keep the compatability).
> This does not work yet, but TeXmacs might be a good
> Cobaye for testing such a library.

The guile 1.6 manual says that the gh layer is deprecated, and I doubt
that the mzscheme people will support it.

> 
> Finally, after one year of silence, the development of
> Guile seems to have resumed. It may be worth it to wait
> six months and see whether they manage to speed things up.

Don't just sit idle and wait.  Ask them a lot of questions, and make a
presence with them so that you are percieved as active and important
Guile users.  Express your needs and concerns.  Perhaps you can get what
you want?

> As a conclusion, I prefer to wait a while (three or six months)
> and reconsider the question. Either Guile improved a lot
> (and the question disappears), or MzScheme provides a good
> Guile compatability library (so we can switch without much pain),
> or we will need to really spend a few weeks on this issue.
> In the last case, a good solution might be to contact
> the MzScheme developers and ask *them* whether they are
> willing to do the porting, so that they can get an opinion
> on how easy future portings might be.

Any thoughts about using a Common Lisp, rather than a Scheme for
TeXmacs?  What about ECoLisp? (ECL)  I wonder if it's any good?

http://sourceforge.net/projects/ecls/

With this in it's own thread or something, Maxima could run right inside
of TeXmacs.