Lists Home |
Date Index |
Bullard, Claude L (Len) <firstname.lastname@example.org> writes:
> With an ammendment: if one is using XML for process communication,
> one can want to validate the XML against a schema if there is any
> chance that another process can touch that file. This takes some
> wisdom and smarts about class and message design. The example
> that made me add this is the MS configuration file handler. Note
> we are inside a framework now, not in an very large open environment
> but the concept of schema as a message contract is the same.
> so now, irrespective of scale, we have to ask the same questions
> about boundaries. A fundamental value is scaling across
> the view dimensions, for you chaos/complexity theorists, and
> I guess it is the concept of the schema-as-contract. The good
> news is that the schema-as-test is useful at different scales
> but following that, not the same one, so the notion that we
> should fit the schema to the scope is empirically right.
> I wonder about that one given correct-by-construction
> techniques using
> components that have not been altered. Agghh... version
> control rears
> its ugly head again. Do I apply it to the component or to the
> contract or both? Can I prove it or should I just run it and wait
> for the exception to be thrown?
> This stuff must drive the framework.xml and application class
> designers nuts.
If I'm following you here (and I'm not sure that I am), I think the
normal attack is recursive boot-strapping: start with open security/no
audit/no validation, inject the fundamental constraints. Now add the
base security, make the base constraints required, validate the base
against the base, point the audit data at the results of the validation,
etc. (Possibly at this point you add in database constraints that didn't
previously exist, turn certain fields non-null, etc.) Now you can inject
the next layer and go around again. For our system, if we were starting
from scratch we'd have to fake the creation of 2, maybe 3 layers before
we'd have the complete bootstrap in place. Is it provable? Not in our
case, but I can't see why it couldn't be in theory...