Re: The reason for roots?

 Noah,
 since SOAP 1.1 has a special case for strings (and byte arrays,
IIRC) allowing these to be serialized inline (as opposed to the
other stuff serialized "on top of serialization"), the
appropriate code for finding references in all the data must be
there already. Disallowing out-of-line serialization removes the
"out-of-line-ing" code from serializers and removes the
"search-on-top-of-serialization" code from deserializers, if it
was there in the first place.
 And, BTW, SOAP 1.1 doesn't define "top level of serialization" -
is it "siblings of serialization root" or "in the Body" or "after
Body as so called trailers" or somewhere else? I've seen all
three interpretations. And if we go to describing the top level
of serialization, we might end up reallowing trailers, too, if we
accept the first or the third interpretation.
 Best regards,

                   Jacek Kopecky

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



On Tue, 26 Mar 2002 noah_mendelsohn@us.ibm.com wrote:

 > Jacek Kopecky asks (re indep elements):
 > 
 > >> Do we want to reintroduce this complexity?
 > 
 > I think it's harder for deserializers, perhaps easier when serializing an 
 > arbitrary graph.  Not sure which way I'd prefer to go, but it's not 
 > clearly a complication in all cases.   One way to write a serializer is 
 > (a) dump all nodes as independents, with an ID (b) dump all edges as 
 > hrefs.  We still have to settle the question of "sourceless edges" on the 
 > indendents, I think, but that's there for the tree root anyway I think (or 
 > was all this settled in a manner I didn't notice?)
 > 
 > ------------------------------------------------------------------
 > Noah Mendelsohn                              Voice: 1-617-693-4036
 > IBM Corporation                                Fax: 1-617-693-8676
 > One Rogers Street
 > Cambridge, MA 02142
 > ------------------------------------------------------------------
 > 
 > 
 > 
 > 
 > 
 > 
 > 
 > Jacek Kopecky <jacek@systinet.com>
 > Sent by: xml-dist-app-request@w3.org
 > 03/22/2002 10:25 AM
 > 
 >  
 >         To:     Martin Gudgin <marting@develop.com>
 >         cc:     XML Protocol Discussion <xml-dist-app@w3.org>, (bcc: Noah 
 > Mendelsohn/Cambridge/IBM)
 >         Subject:        Re: The reason for roots?
 > 
 > 
 >  Gudge, replies inside. 8-)
 > 
 >                    Jacek Kopecky
 > 
 >                    Senior Architect, Systinet (formerly Idoox)
 >                    http://www.systinet.com/
 > 
 > 
 > 
 > On Fri, 22 Mar 2002, Martin Gudgin wrote:
 > 
 >  > ----- Original Message -----
 >  > From: "Jacek Kopecky" <jacek@systinet.com>
 >  > Subject: Re: The reason for roots?
 >  > 
 >  > >  the problem is that in SOAP 1.1 serialization rules would say
 >  > > that C must be serialized "as an independent element on top level
 >  > > of serialization" because it has multiple references to it.
 >  > 
 >  > MUST be or MAY be?
 > 
 > In SOAP 1.1, MUST (except for strings and arrays of bytes, I 
 > think).
 > 
 >  > >  In SOAP 1.2 we haven't forbidden this, although we don't talk
 >  > > about this any more (so if somebody started from reading SOAP
 >  > > 1.2, they would not even think of serializing something
 >  > > out-of-line).
 >  > 
 >  > Agreed, although I could add a clause into section 3.1.1 stating how
 >  > out-of-line serialization would work
 > 
 > Yes, which would reintroduce the explicit statements from SOAP
 > 1.1, maybe lessening the MUST above to MAY. Do we want to
 > reintroduce this complexity?
 > 
 >  > >  Now if non-roots (non-serialization-roots, that is) can be
 >  > > anywhere in the message, not just as descendant EIIs of a
 >  > > serialization root, we have to mark some of them. SOAP 1.1 took
 >  > > the approach of marking the non-roots that appear somewhere
 >  > > funky, but this was not crisp enough. So we can either mandate
 >  > > marking the roots or the non-roots. We chose roots.
 >  > >  Oh, BTW, I thought my graph below has two roots (according to
 >  > > your original definition), not zero.
 >  > 
 >  > No. It has no root because of rule 2
 >  > There is no way to get from A to B or B to A. Remember it is a 
 > *directed*
 >  > graph.
 > 
 > OK, I understand.
 > 
 >  > Still not convinced we need the notion of root at all in the 
 > encoding...
 > 
 > If we allow some people to serialize stuff out-of-line, we must 
 > say where to put these out-of-line non-serialization-roots and 
 > how to know which is which.
 > 
 > I'm quite OK with the current rewrite's version which disallows 
 > that out-of-line serialization, thus obviating the root 
 > attribute. Oh, I think that should go from the rewrite if we 
 > don't reintroduce the out-of-line serialization.
 > 
 > 
 > 
 >

Received on Tuesday, 26 March 2002 15:38:22 UTC