Ludovic Court?s wrote:
> I think one good rationale for list library names are that (i) they are
> more readable and more easily processed than strings with an ad hoc
> syntax and (ii) hierarchical naming facilitates the mapping of library
> names to file names (FWIW, Guile's module system works like this [0]).
>
> Thanks,
> Ludovic.
>
> [0] http://www.gnu.org/software/guile/manual/html_node/Writing-new-Modules.html
>
Hi.
I designed the Guile module system namespace, at least as it was originally
provided (I don't know how it might have changed since). I think
Miles Bader did most of the design on the module system including a
first cut on names but I do recall putting effort into the details of module
names.
You may be amused to learn then that the definition included a read syntax
based on unix file systems and therefore easily understood as relative URIs.
Thus, the module name (foo bar) could also be usefully written as (something
like, I forget the details) #/foo/bar.
I designed that namespace when URIs were still pretty new to the world.
Web sites were novel things. Netscape was starting up and had
released some software but most people were still using an NCSA
reference implementation of a browser. On my desk, sources who shall
remain anonymous had dropped a copy of a spec for an interesting
language named "Oak". It was still possible to, back then, and with a
straight face, argue that all this new SGML-inspired stuff was off on the
wrong foot and that future netwok service APIs would be better defined
in terms of lisp-ish S-expressions.
Were I redoing that design today, roughly a decade later -- and apparently
that is the path I'm on -- I would just go all out and pick (perhaps
slightly
decorated) URIs for the surface syntax (at least).
-t
Received on Wed Jun 27 2007 - 04:42:27 UTC