!!Obsolete!! See: WWW and OOP
A very interesting area it the transition of HTTP to a full distributed object-oriented system. This was always intended that the HTTP method should become extendable according to the capabilities of the object.
Mark Tucker <mct@philabs.Philips.COM> pointed out a similarity between WWW and OMG goals. OMG has its "IDL" for describing interfaces. Mark says Gary Adams gra@zeppo-sa.East.Sun.COM and ec@strange.pms.ford.com are also interested, and gives pointers:
The (old-ish) Corba spec is in: ftp://ftp.ifi.unizh.ch/pub/standards/corba/spec/corba.ps A general introduction http://www.acl.lanl.gov/sunrise/DistComp/Objects/objects.html A nice little tutorial http://www.acl.lanl.gov/sunrise/DistComp/Objects/CORBAtalk/CORBAtalk.html
The message stream should be ascii. Its easier to debug, and gives the appearance of being friendly. On top of the ascii stream, we should implement something that looks like Smalltalk
[object selector]
or
(send object :selector)
There seem to be two kinds of ``ascii'' protocols out there.
The first, NNTP, HTTP, is command line:
CMD .....
response
451 comment
.... optional other text
This can be used with mime-like separators
351 body of file follows 2498afkajl290x092
.....<body here>
--2498afkajl290x092--
and we don't have to worry about quoting anything (nor ascii-fying the response).
The second obvious candidate is a lisp like syntax
(methods-of (url "some/object/that_looks_like_a.file"))
where the return value is a either
(some-lisp sexpr as determined by the signature of the 'methods-of command)
or an error value
(orb-error .... some sexpr encoding error values, where orb-error is reserved)
I prefer the lisp syntax,
As to the semantics of the communication, I think that's easy. We just define a binding from IDL to our Lisp syntax, and then we're done! (... after we implement a server that obey's all these commands!)
The advantage of using HTTP is that we can bring the WWW community around into the full OO picture. It doesn'treally matter what formatwe use, as we can map trivially between them. In the end, we will maybe make a very compact binaryform of hTTP, but in the mean time the ASCII style (just as LISP) allows easy experimentation.
Of course LISP s-expressions would have to be checked very carefully before being evaluated of they have come over the net!
Tim BL