Re: Issue 4 Proposed Resolution (was: why no doc type declaration and PIs in SOAP)

Andrew Layman wrote:
> 
> My motivation to permit PIs is primarily that the XML Protocol should be
> able to carry XML documents as body content without rewriting whenever
> possible.

Thank you for addressing this - this is precisely my concern. I was
baffled by the absence of any explicit coverage of this in SOAP 1.1, and
I'd like to feel that the issues have at least been addressed in the
design of SOAP 1.2. I think that re-writing - or at least parsing and
re-serialization - may be necessary, for instance to ensure that the
payload(s) don't have their own XML declaration or XML encodings. But I
would really like to see data content passed straight through without
requiring a new SOAP encoding / decoding. 

> The cost/benefit analysis is very different for DTDs. A DTD is
> substantially more complicated to parse.  It may make Infoset
> contributions, making skipping one unsafe.  Permitting DTDs in messages
> is either pointless (if no infoset contributions are made) or expensive
> (if Infoset contributions appear).

We're in danger of close-coupling the XML protocol and the general issue
of implementing XML parsers. Surely the simnplest, cleanest and most
productive solution is to take XSLT 1.0 as our model? DTDs may appear in
the input documents, but parsing business such as DTDs and encodings is
firmly excluded from the XPath 1.0 infoset mapping (appendix B), the
input data model and from the output.

I think we should re-factorize this by distinguishing between XML
content, which can be nested, and XML serialization, which cannot.
Simply state that SOAP, like XSLT, views its input and output as
ready-parsed and DTD-free. We could give this infoset profile a fancy
name if it helps ("normalized XML" or "XML dataset" might give the right
impression) but it would still just be XPath 1.0 appendix B. 

Francis.

Received on Monday, 1 October 2001 18:32:17 UTC