[XML-SIG] WSDL library ?
Mon, 11 Feb 2002 11:23:50 -0800
Rich Salz wrote:
> > I will be extremely disappointed if WSDL 1.1 is made SOAP only.
> Don't worry, I expect to lose that battle.
> > WSDL's biggest problem, IMO, is that people feel it is not a language
> > that should be authored by hand.
> What's preventing it? In my view, it's the cut/paste repetition you
> have to keep doing.
That's part of it. Plus there are so many levels (see below). Often you
see an element type with the same name as a part, with the same name as
an operation, with the same name as a port type. You've got four levels
to do one thing.
Another part is using XML Schema as an "abstract type description
language". It's a little ridiculous to use a schema language without
standard OO notions of "property", "array", "struct" etc. But I'm sure
I'd get voted down.
Of course it should be possible to refer to an XML schema for
*document-oriented* SOAP. But for RPC-oriented SOAP, older IDLs are
actually much simpler.
> ... The WG is starting to collect requirements, I'll
> forward anything you can suggest.
Here are some of my suggestions:
1. Interfaces and implementations are two very different classes of
specification object. It confuses people to combine them and also
encourages the poor design habit of combining them in a manner that
makes re-use difficult. This in turn hurts interoperability of diverse
WSDL 1.2 should be formally split into two languages: e.g. WS Interface
Description L and Service Implementation Definition Language. Perhaps
the Service Implementation Definition Language could be entirely
2. The interface description language should be as easy to read and
write (semantically, not syntactically) as standard IDLs like CORBA and
COM IDLs. There is a sense that WSDL is much, much more complicated and
this is going to hurt its adoption on the open Internet.
3. If the interface definition language is to be used both for RPC
*and* document-style interfaces then it needs first-class concepts of
array and struct, which XML Schema lacks, for RPC-style interfaces.
Arguably, using schema for both kinds of interfaces is more confusing