is xml schema bad wrt schema extensions?

[Erik Wilde <net.dret@dret.net>]

hi.

recently (http://lists.xml.org/archives/xml-dev/200209/msg00032.html), 
paul prescod brought up the interesting issue of whether xml schema 
extensibility is a bad thing because naively designed software might break.

i think this is a very interesting and important issue. paul is right 
that schema's derivation methods make it possible for software to fail 
when processing instances of derived schemas. so one basic question is: 
is this the fault of xml schema of or sloppy software authors? and, 
regardless of that, how can this be made more safe? i think so far there 
has been remarkably little effort to point out how to *safely and 
reasonably use* xml schema, in contrast to the large amount of effort in 
rather emotional schema bashing that can be seen on xml-dev...

one point i would be very interested in is to look at this a bit more 
formally. for example, what kind of xpath can i safely apply to 
instances of schemas and derived schemas? only child paths? anything 
else? and if i restrict extension capabilities by some of schema's 
features, restricting type restriction and/or extension or type 
substitution, does this change the "safe set" of xpath? i have no answer 
to this question, but maybe this could lead to some framework for how to 
design schemas for extensibility, and how to restrict the possibilities 
for extensions to make it easier for software to deal with schema 
extensions.

in a next step, one could start to think about how software authors 
could be supported to not write naively designed software. writing 
software which is forward compatible never has been an easy thing to do, 
but maybe it could be made easier, for example by disallowing certain 
xpath expressions which could break with extended instances. in a very 
futuristic view of the world of xml, i could even imagine a very smart 
dom module which only allows access to instances in a robust manner, 
making it impossible to write code that breaks.

maybe some people will find this too complicated to be of any use, but 
there are a lot of people out there trying to use xml schema in a 
reasonable and robust way, and if all xml experts can come up with is 
advice like "it's complicated and dangerous", then we do not provide 
them a lot of help.

cheers,

erik wilde  -  tel:+41-1-6325132  -  fax:+41-1-6321035
           mailto:net.dret@dret.net -  http://dret.net/
           computer engineering and networks laboratory
           swiss federal institute of technology  (eth)
           * try not. do, or do not. there is no try. *