Re: [xml-dev] Caution using XML Schema backward- or forward-compatibilit

["Fraser Goffin" <goffinf@googlemail.com>]

Greg,

yes I agree that structural validation is important, and I further
agree that the various checks that are made on the data are
cummulative and go to the heart of data integrity. So yes, in some
cases you *may* choose to delegate responsibility for certain aspects
of validation to specific layers in your application. However,
fundementally a core business application ought really to protect
itself from data integrity threats rather than rely on someone else
ensuring that no bad data gets through to it (it just depends where
you draw your application boundary). So, thats why sometimes I see
validation done early as optimisation (i.e. it may run twice)rather
than being a critical protection mechanism (but different applications
and designers choose different layering so your mileage may vary :-).

In my case, business services from core applications can be accessed
internally as well as externally. I may choose a different approach
for each.

To add to my point about optimisation, it isn't that we can
necessarily choose not to perform validation processing, only that we
may choose to perform it earlier in the processing chain to provide a
better experience for the caller. For example, if I implemented a
service where I can receive insurance quotes 24/7, but my rating
engine (where most of my business data validaion and rules live) is
only available in [extended] office hours, I may choose to move *some*
of the validation to occur earlier such that a caller doesn't have to
wait until tommorrow only to find that the format of the DoB field was
invalid in their request. My rating engine would probably have spotted
that also so doing it earlier on is not actually critical (and comes
with the maintenance cost of having logic in 2 places) but as far as
the usefulness of my service is concerned it may be a good choice
(i.e. an optimisation).

Sometimes I think about data validation implemented as above as trying
to ensure that most of the obvious client-side format and content
errors are spotted straight away before chewing processing cycles
(sometimes even more costly human cycles ;-) further upstream. So a
message that gets through is not necessarily valid, it still might
fail other more rigourous checks, but it has a reasonable chance of
being so.

Fraser.

On 27/12/2007, Greg Hunt <greg@firmansyah.com> wrote:
> Fraser,
> Your point 5: particular types of validation should occur in clearly defined
> places and should be regarded as cumulative.  Turning off structural
> validation (as suggested in the earlier email) damages the effectiveness of
> the later, more business oriented validations and can lead to integrity
> problems.  Structural validation is hardly an optimisation, its an integral
> part of a process whose cumulative aim is to protect the integrity of the
> consumer business process.
>
> Greg
>
>
> On 12/28/07, Fraser Goffin <goffinf@googlemail.com> wrote:
> >
> > Some general thoughts :-
> >
> > 1. XML Schema, Relax, or any other vocabulary for decribing data do
> > not provide any mechanism for preventing people from abusing those
> > structures for purposes that differ from their original intent, in the
> > (imo) mis-guided belief that doing so will provide a less intrusive
> > opportunity to accomodate change (particularly in the case of semantic
> > drift that Micheal refers to above). More often than not this leds to
> > complexity and assumed knowledge which, when taking Rogers' scenario
> > of a business service with 'unknown' consumers, will typically just be
> > storing up a more difficult and possibly intractable problem for later
> > on (i.e. once users of the service have got that 'bad' design baked
> > into their applications - that is, applications that you as the
> > service provider have no control over - they are not going to be
> > motivated to make any changes whilst they are still working - only
> > solution is to introduce a breaking change and as a provider, support
> > multiple versions).
> >
> > 2. Using information items for purposes other than their original
> > intent is usually a very bad idea. It just provides a short-term fix
> > and a long term pain (for providers and consumers alike - vocabulary
> > owner and vocabulary user.).
> >
> > 3. In nearly all cases sematic changs are ALWAYS breaking changes
> > (notwithstanding that you might think you can get away without because
> > the syntax doesn't change  -  in some ways this very lack of
> > visibility is the problem). Clearly this is even more the case when
> > talking about business services offered to 'unknown' consumers (or at
> > least consumers with applications whose development resource is not
> > under the control or influence of the service provider). Therefore
> > semantic changes are ALWAYS INcompatible so
> > forwards/backwards/whatever doesn't apply.
> >
> > 4. A minor (excuse the pun) point. Many people use a pattern of
> > <Major>.<Minor> for indicating the version of something (schema,
> > service, whatever). An increment of the MINOR value can be used to
> > identify a [possible] compatible change, depending on your
> > [advertised] versioning strategy. Whereas, a change to the MAJOR
> > version number is an EXPLICIT SIGNAL of an INcompatible (breaking)
> > change.
> >
> > 5. Schema validation is typically NOT the same as validation in the
> > context of a business process (i.e. data can be valid according to a
> > data definition, but invalid when processing, since other contextual
> > rules apply). We may often use schema validation as an optimisation to
> > provide more immediate feedback to a caller about the possibility that
> > their request may be fulfilled (or not), but not IMO as a way of
> > guaranteeing that any upstream business process will be successful,
> > and definately not as a way of protecting the integrity of a core
> > business application (now that *would* be a dangerous approach).
> >
> > We also appreciate that most vocabularies have limitations in terms of
> > the data and business rules that they can express, so naturally we
> > often apply multiple techniques and apply them at different stages of
> > the processing chain (e.g. a bit of structural and data content
> > checking early on to weed out obviously invalid messages, up to state
> > machines and complex business rules deeper into the process). What is
> > more, the outcomes of validation failures often differ depending on
> > what stage they are detected. Of course we are loathed to spread our
> > business data validation rules across the potentially many layers of
> > our application infrastructure, but, we may choose to do this in a
> > controlled way to provide process optimisation, so even using a
> > solution such as DataPower might only extend to a certain category of
> > validation checking.
> >
> > 6. There are occasions where we would either not locate ALL validation
> > processing inside the receiving application logic or where we would
> > choose to duplicate parts of it further out. In those cases we
> > wouldn't remove it after testing (lets face it, testing is often a
> > problematic area for complex integration scenarios and often is not as
> > comprehensive as it might be). I find that this is particularly the
> > case for EXTERNAL integration (i.e. you only control one side of the
> > exchange) and where business services (external interface) needs an
> > SLA of 24/7 availability even when the actual processing application
> > is not.
> >
> > Fraser
> >
> >
> _______________________________________________________________________
> >
> > XML-DEV is a publicly archived, unmoderated list hosted by OASIS
> > to support XML implementation and development. To minimize
> > spam in the archives, you must subscribe before posting.
> >
> > [Un]Subscribe/change address:
> http://www.oasis-open.org/mlmanage/
> > Or unsubscribe: xml-dev-unsubscribe@lists.xml.org
> > subscribe: xml-dev-subscribe@lists.xml.org
> > List archive: http://lists.xml.org/archives/xml-dev/
> > List Guidelines:
> http://www.oasis-open.org/maillists/guidelines.php
> >
> >
>
>