[Jscheme-user] RE: Fwd: Re: Best Java Scheme implementation?

[Anton v. S. <an...@ap...>]

Ken Anderson wrote:
> Anton, could you say more about what your liked about
> SISC's generics over java dot notation?

Joe Marshall also asked me about why I preferred SISC's generics.  I gave
some more detail in this post on c.l.s.:

http://groups.google.com/groups?selm=YVkcb.6092%24ai7.2351%40newsread1.news.
atl.earthlink.net

I think there are pros and cons to both approaches (see the above post for
more detail).  The main advantage I found with the SISC approach is very
much one of the "feel" of working with Java objects in Scheme, in terms of
the Schemeliness of procedure names, so it is certainly a subjective matter.

My comment re Javadot on c.l.s. came from having first felt that I needed
something like Javadot to work with Java objects, but ultimately finding
that (a) I didn't really need it, and (b) that I preferred not having all
those Java-esque names in my code.  Conceptually, I take the position that
any Java functionality could in theory be replaced by Scheme functionality,
so I'd rather have a consistent interface to both - a sort of variation on
Bertrand Meyer's principle of uniform access.

However, I should make clear that the reason I began using SISC over JScheme
was more to do with R5RS-compliance.  The biggest single issue for me was
that I use syntax-rules macros in code on other Schemes, and wanted to have
as much portability as possible between the implementations I use.  At the
time, JScheme didn't support syntax-rules, at least not in the default
distribution.  This seemed like an indicator that R5RS-compliance was not a
priority - combined, of course, with the R4RS mention on the web pages.

> In the first version of JScheme (SILK) you had to
> mention each method you wanted to use from a
> Java class and that became pretty tedious quickly.

I imagine that this would have a lot to do with the kinds of applications
being written.  In most cases, I've ended up implementing a module to wrap
some Java functionality, and declared the needed generic procedures there,
so the main body of code doesn't need to worry about declaring them.  I see
the need for these explicit declarations as a tradeoff for the ability to
have Scheme-like names for Java methods, as well as having generic methods
that can work on either Java or Scheme objects.

> It doesn't look like SISC much with its Java side.  For example,
> there isn't a gui library like JLIB, or Tim's collaboration stuff.

I haven't looked into that, since I've been working on server-side web
applications.  The kind of things I've been doing have involved interfacing
to Java business objects, web framework code, and persistence code.  I'll
clarify that on c.l.s.

Anton