Re: Section 5 vs Schema

 Gudge, 
 I finally got to compose my reply...
 I'll start from the current situation: we do indeed have XML 
Schema now and we can describe messages only with XML Schema and 
the mapping is an implementation detail. This is being done with
the WSDL constructs in document style and literal use of schemas.
 Nevertheless, I have to argue that the simplified data model and
encoding is useful. An important part of the spec is the RPC
section (although many see traditional RPC as a suboptimal use of
the Web Services paradigm). RPC comes from programming languages
environments which usually have simple data models close to ours.
 It would IMHO be strange to have RPC with parameters being "any 
XML data" rather than data in an RPCish data model.
 The mapping of the data from such a data model into XML is quite
straightforward but it can be done in a number of ways. The
problem is that in languages like WSDL we do the opposite
mapping, from XML Schema (the language of choice in WSDL)
schemata into programming data model schemata. I would see it 
more appropriate if there was a SOAP Data Model Schema Language 
(SODAMOSL) which would describe data in terms of structs, arrays 
and primitives (which may be very well the XML Schema datatypes, 
that is why the XML Schema parts are separate, isn't it?), and 
which would be used in WSDL instead of XML Schema for "encoded" 
use and RPC style. But I have grown accustomed to using XML 
Schema. 
 There are problems with XML Schema in such use though because
the mapping from XML Schema to our data model is sometimes not
straightforward or even discutable (e.g. <choice>s).
 But this might be handled adequately in WSDL. As might SOAP 
Encoding (and Data Model) be a part of WSDL or something other, 
rather than our spec. Both have, in fact, the status of adjuncts, 
which means that "this particular piece of work that could be 
separated from SOAP the XMLP WG has taken on itself anyway."
 We might think about adding the mapping of XML Schema to and 
from SOAP Data Model Schema, so far an implied language. Since we 
already have the Encoding and the Data Model and we seem to want 
to support and use XML Schema as much as (reasonably) possible, 
this would be a natural addition. The set of adjuncts related to 
data encoding would be more complete.
 If we decided it's a viable course, I'd volunteer to draft the 
first cut of this mapping.
 We can, of course, keep the mapping unspecified and depend on 
WSDL to specify it (which it does not do) or on the agreement 
resulting from some kind of WSDL interops (planned, I think).

                   Jacek Kopecky

                   Senior Architect, Systinet (formerly Idoox)
                   http://www.systinet.com/



On Mon, 3 Dec 2001, Martin Gudgin wrote:

 > SOAP 1.2 Part 2 Section 4[1] ( old section 5 ) defines a set of encoding
 > rules for mapping from programmatic type systems to XML.
 > 
 > There was some discussion on the last editors conference about how to deal
 > with issue 17[2] regarding the schemas that appear in section 5. I took an
 > action to start discussion about this on this list. Please note I will be on
 > holiday from today and will not be back until the New Year so will not be
 > able to actively participate until then, hopefully you'll all have nailed
 > the issue by then!
 > 
 > One suggestion was that section 5 actually defines an implicit schema so
 > each mapping from some programmatic type essentially defines a schema type.
 > This seems reasonable but at the same time feels a little odd. We have
 > section 5 because when SOAP 0.9, 1.0 and 1.1 were written XML Schema was not
 > done, we didn't have an XML based type system. So we had to start from a
 > type system we did have. So Section 5 defines a set of rules for mapping
 > from programmatic type systems iuntNow that XML Schema is done it is
 > possible to define the messages being exchanged entirely in XML Schema
 > without reference to any programmatic type system. Mapping to the
 > programmatic type system ( if any ) at either end of the exchange is an
 > implementation detail.
 > 
 > So, given that we have XML Schema, does it make sense to infer a schema from
 > some other type system?
 > 
 > And if it does, what do we do about examples in the spec. It seems very
 > strange to say 'we start from a programmatic type system' and then only show
 > schemas! We are defining a language binding, even if we never show a Java
 > class or a C struct or whatever.
 > 
 > OK, that's it. I hope the discussion is fruitful, I'll read through it when
 > I get back from holiday.
 > 
 > Regards
 > 
 > Martin Gudgin
 > DevelopMentor
 > 
 > 
 > 
 > [1] http://www.w3.org/TR/2001/WD-soap12-part2-20011002/#soapenc
 > [2] http://www.w3.org/2000/xp/Group/xmlp-issues.html#x17
 >

Received on Thursday, 10 January 2002 10:27:32 UTC