I've been working with Eric Bouck on a project of this sort. We are
moving over into ILU (hopefully!). Anyway, it currently works like this...
User requests a URL of the form:
http://www.com/broker/object/method?field1=value1&field2=value2
This goes to a small Python cgi script which packs up the request and
sends it to a persistent object server, using Guido's persistent objects.
The way Eric wrote it was fantastic, and really highlights the difference
between objects and relational databases. Now we are looking at
replacing the Python-written object server with ILU.
Anyway, I would discourage writing a Python-based httpd. The next
several months is going to bring a lot of changes on the server side, and
they are already pretty complex. What would be interesting is to take
Bill's idea, and _embed_ it in the CERN httpd (or NCSA) code. Then, a
URL of that matched /object would call an embedded Python interpreter
that was connected to ILU. No resource hit for (a) cranking up a CGI
shell and putting env variables in it, and (b) cranking up a Python
interpreter.
Might scale pretty well.
I would love to see this discussion continue. There are those that might
not see the endpoint of removing the Unix filesystem as the source of
answering HTTP GETs (or POSTs!). However, there are some seriously
interesting interactive applications, such as writing a multi-user
project manager, that would be laughably simple with an object based system.
--Paul
Paul Everitt V 703.371.6909 Email Paul.Everitt@cminds.com
Connecting Minds, Inc. F 703.371.1201 WWW http://www.cminds.com/