On 17 Jun 2004, at 12:00, Jochen Wiedmann wrote:
On Do, 2004-06-17 at 12:35, John Wilson wrote:
"MinML-RPC is a minimal XML-RPC implementation. This is very interesting, but I have some pushback. There is no such thing as "the <nil/> extension." If we added it, it would break at least one implementation. Please appreciate the bigger picture. XML-RPC is what it is, not something to be endlessly debated and "extended" -- it's already totally extensible at the next level up."
ref: http://essaysfromexodus.scripting.com/backIssues/2001/05/13
The discussion on the mailing list following that post lead to the removal of <nil> from the Helma XML-RPC implementation.
Well, you got me here. He used to think different at some point, see
http://www.xmlrpc.com/discuss/msgReader$7?mode=topic
However, I still see a difference between an extension, which is *very* clearly declared as violating the SPEC and turned on the users behalf only and the addition of a "nil" element without further notice.
Others would disagree - Sun's reaction to Microsoft's Java extensions would be a good example.
One of the problems is that people start publishing APIs which requires these extensions. Then interoperability breaks down. I use software every day which requires Java/Perl/C XML-RPC interoperability. I don't control any of the APIs but the stuff just works. That's because they don't do anything cleaver with the API because they can't - that's *good*.
Vendor specific extensions is another way of spelling vendor lock in. Apache isn't in that business :)
John Wilson The Wilson Partnership http://www.wilson.co.uk