[Openmcl-devel] Re: Standardization, ho!
helink at sandia.gov
Thu Sep 12 01:37:06 UTC 2002
Wait wait wait! Let's get organized while we start talking about what
should and shouldn't be standardized. It's great to see people want
this, that's promising, but if we go too far down this path we'll have
no systematic way of making (and recording) decisions. Gary, do you or
someone want to set up or a few initial such organizational things
separate from the openmcl project (unless people object)? Flood control
dam #3 seems to be opening.
I had figured standardizing an FFI syntax might well be trivial as a
wrapper in some systems while simultaneously imposing a huge burden on
other implementations. If that's not the case, that'd be great.
One thing Gary and I have been considering is the feasability of
defining "the" standard as a standard set of optional APIs. Things along
the lines of "If you add the feature :OSSCL-THREADS to *features*, you
must support the following API this feature implies..." for a bunch of
things, plus a dependancy graph that "If you have this feature, you must
supply that feature" etc. so we can have an "OSSCL" feature that implies
all of them are included or go piecemeal.
This would allow a standard to develop and people to depend upon it for
portability without forcing certain (hopefully rare) lisp systems to
provide things that were contrary to the implementation -- essentially,
provide a standard that may exist and that is reasonable, but isn't
forced upon lisps that feel strongly about or need to implement certain
things a different way. Ultimately it will make portable applications
have far fewer exceptional conditionalized bits to make XYZ work in this
or that odd lisp.
John DeSoi wrote:
> At 5:17 PM -0600 9/11/02, Hamilton Link wrote:
> >Fwiw, I'm the aforementioned colleague. Personally I think it would be
> >good for not only the open-source lisps to have a set of common APIs for
> >certain things, but for lisp in general to be able to say we have
> >standard libraries for XYZ commonly needed modern functionality.
> Yes, excellent. And I think one of the key starting points is a
> standardized FFI. A lot of great libraries already exist (SSL, Expat,
> etc.) but each Lisp implementation has to start from scratch to use
> them. UFFI (http://uffi.med-info.com/) seems to fit the bill and
> already supports quite a few implementations (ACL, CMU, LispWorks,
> MCL). I'm just about finished with the first shot at OpenMCL. I hope
> to spend some time with OpenMCL's header parser and see if I can
> adapt it to spit out UFFI declarations.
> John DeSoi, Ph.D.
> Openmcl-devel mailing list
> Openmcl-devel at clozure.com
Openmcl-devel mailing list
Openmcl-devel at clozure.com
More information about the Openmcl-devel