- From: Jacek Kopecky <jacek@systinet.com>
- Date: Tue, 4 Dec 2001 17:11:06 +0100 (CET)
- To: <noah_mendelsohn@us.ibm.com>
- cc: <xml-dist-app@w3.org>
Noah, of course the fault would be generated when a processor actually encounters a reference it cannot resolve. I don't want to mandate that all data be deserialized even when unnecessary. Anyway, I'm very nervous about you tying the term weak reference with an optionally handled argument. Weak references, as I understand them (according to Java), are references to data that can disappear and the reference can still be handled graciously. Optionaly handled arguments, as in LISP's (if a b c) form, require some implementation support but don't require weak references, on the other hand weak references don't always (nor usually IMO) mean optionally handled arguments. I think that we don't need to put weak references into our data model as they can be handled by the application equally well and the slight optimization of weak references being inside our data model and encoding is outweighed by the disadvantage of the additional complexity, while weak references are used only in very few special applications. Actually, the same argument could be used on partially transmitted arrays. 8-) BTW, I think the current encoding does not forbid lazy deserialization, IMHO, which can be used for implementing optionally handled parameters. Best regards, Jacek Kopecky Senior Architect, Systinet (formerly Idoox) http://www.systinet.com/ On Tue, 4 Dec 2001 noah_mendelsohn@us.ibm.com wrote: > Whenever we mandate a fault, we must be very careful about ordering. I > would be reluctant to do anything that forced processsing, particularly of > a large message, to proceed in any particular order. > > Consider the case where there are several problems with a message, some > related to encoding and some not. If we mandate the generartion of an > encoding fault, then we might be forcing processors to do more decoding > than the application needed to discover other problems. Furthermore, > there's a general issue of weak references. What if you're doing RPC and > by looking at argument #1 you decide that you don't even need to mess with > argument #3. Should we still mandate that you check for the fault in a > bad reference out of #3? > > I think not. Therefore I prefer Henrik's suggestion: offer a standard > fault that MAY be used, and indicate that failure to resolve a reference > indicates that there is no value available for the node in the graph (a > new state we haven't had before) and that processors MAY generate the > standard fault when this situation arises. > > ------------------------------------------------------------------------ > Noah Mendelsohn Voice: 1-617-693-4036 > Lotus Development Corp. Fax: 1-617-693-8676 > One Rogers Street > Cambridge, MA 02142 > ------------------------------------------------------------------------ > > >
Received on Tuesday, 4 December 2001 11:11:08 UTC