Re: Roadmap and goals?

[Kirill Lisovsky]

Hello!

Working for a long time with a half-dozen of different Schemes in IT and CS 
projects, I've some "comparative impression". 

Tanel Tammet wrote:
> (*) what are the specific benefits of using Guile.
> 
> For example, why exactly should somebody
> use Guile instead of SCM or Bigloo:
> what are the specific advantages and
> what is the downside. Why not use just
> SCM or Bigloo (they are faster, you know :-)
> There have to be answers to this question,
> just that I do not know the answers.
>

Bigloo has nice and fast compiler but its interpreter is slow and 
compiler-incompatible. It makes sense if the application has to be compiled
rather than interpreted.
Bigloo not quite Scheme in proper-tail recursion considerations, but it's
fast and may be useful for practical purposes.
Unix-oriented.
It has a usable JVM backend also.

SCM is a good Scheme, but it has no case sensitive reader, which 
is a big disadvantage for some applications (XML is most important example 
for me). 

IMHO, PLT is most advanced and most promising "big" Scheme.
Version 200 is a great improvement, its design and principles are 
reasonable and clear. 
A lot of libraries, active development, multiple platforms, and so on ...

I like Gambit a lot, but it's not free for commercial projects.
Compiler to C.
Extremely robust.

Chicken is relatively young but already usable.  
It has a solid theoretical basis and good C interface.
Compiler to C.

Guile - it is the slowest one, I'm afraid :-)
Well, it's better _interpreter_ than Bigloo ... 
IMHO, it's main advantage is large installation/users base.
But a mess with versions: 1.7.x , 1.6.x, 1.5.x  while latest Linux distros
are using 1.4.x or even 1.3.4! 
(1.3.4 is used by RedHat 7.2 which makes a lot of installed Guiles pretty
obsolete...)
Using Guile since 1999 I'm still ignoring its latest additions
due to this zoo of versions...

As the result, I mostly use Guile to run Scheme scripts on "just OS installed" 
Linux/BSD boxes. If I'm to install some Scheme, then (usually) it is not 
Guile...

So, this is my point of view:
  1. Guile may have a good future as fast and compact "R5RS SIOD" with
a lot of (optional !) libraries.
  2. Unification of Scheme implementations will be highly desirable from the
practical point of view. Co-existence of ten "major Schemes" is a major 
practical disadvantage.
  3. PLT is best _full-blown_ Scheme implementation now. It's designed as core +
libraries. Porting (best and absent in PLT) Guile's code as PLT collections is 
most realistic way to unification.

Best regards,
         Kirill Lisovsky.

http://pair.com/lisovsky/

P.S. Forward of Nicolas's message to address@hidden
is the reason of this posting. Please, don't consider it as  
destructive :-)

> ------- Start of forwarded message -------
> From: Nicolas Neuss <address@hidden>
> Subject: Re: Roadmap and goals?
> Date: 19 Apr 2002 10:38:06 +0200

...

> P.S.2: I thought about sending this also to address@hidden  I
> refrained from doing so, because Tamel did not do that, and because I
> do not want to be too destructive.  But IMHO, also users should know
> about these problems in Guile's design.  So, it remains for you
> maintainers to inform them...