[cclan-list] Another defsystem proposal

[Daniel B. <da...@te...>]

[ This mail is sent to cclan-list and to clocc-devel.  Most of the
impetus I've seen lately for defsystem changes has been driven by
cCLan, and most of the work on creating a new one by CLOCC.  When
replying, please consider which of the lists your reply is appropriate
to, and adjust your headers appropriately.  Cheers. ]

I have spent some time over the last couple of days designing Another
System Definition Facility.  This is supposed to 

(a) do the things that mk-defsystem presently doesn't that would be
good for cCLan packages (mostly version identifiers)

(b) be implementable in a fairly portable fashion (requires CLOS and
ANSI CL pathnames) without a zillion #+impl tokens.

(c) be extensible - allow new operations and new forms of module to be
defined easily and reasonably hygenically by the user

(d) be small.  For example, there is no distinction between a 'module'
and a 'system' except in terms of external visibility or lack thereof.

A number of sources were used for inspiration: mk-defsystem, the
defsystem 4 proposal, kmp's "Design of large Systems" paper, and the
ACL defsystem utility.  Ideas that I thought useful were shamelessly
stolen from all of them - thanks, guys.

My emphasis in design was "inside out", whereas Marco and Peter's
defsystem 4 proposal is more "outside in".  There are a number of
things in defsystem 4 that I don't presently support, but for the most
part it would be good (and I think also feasible) if we could meet in
the middle.

There is work yet to do on case issues (LPNs make that fun) and the
external syntax, but I'd appreciate any feedback meanwhile.
Especially on 

- whether we can reconcile it with defsystem 4, because I strongly
believe it would be beneficial to have one specification rather than
two.

- whether you're already using or have a need for any of the 'missing'
features.  I'm not but if demand for them is great, I'm amenable to
replacing them.

- how much more information you would need before you could implement
something that satisfies this spec (accessor names and gf names are
the obvious gaps)

Take a look: it's at 

   http://ww.telent.net/lisp/defsystem-proposal


-dan

-- 

  http://ww.telent.net/cliki/ - Link farm for free CL-on-Unix resources