Re: Specification of a transform.

On Mon, 2013-09-09 at 08:42 +0100, davep wrote:
> Given schema A as input XML. Schema B as XML output.
>    Assume no hierarchical simple relationship.
>    Assume mapping of values needed from input values to output values.
>    Assume literals are needed.
> How do (might) you specify the required transform?

I have been struggling with a problem I think is similar to the one you
have posed, except that it involves RDF graphs instead of XML schemas
and instances. But since a tree is a specialized graph, the techniques
for graphs should work for XML.

I was led to a solution whereby I specified the expected outcomes in a
generic rule language. I selected ISO Common Logic. Other options would
have been RuleML or RIF.

The rules are expressed as quantified logical formulas, mainly of the
form "for all things like this in the world of A, there must exist
things like that in the world of B". Or conversely "If something like
this exists in B, then something like that must exist in A."

I'm sure there must be a large literature out there that hasn't filtered
down to general industrial usage--particularly in XML practice--, which
would specify the types of rules and transformations that are amenable
to this approach. I'm also pretty sure there are limitations, in that
some types of transformations cannot be specified in this manner. But I
believe there is a large subset of practical problems that can be
tackled this way.

Rules specified in a generic language can be put to different purposes:
for generating test cases; for generating code for certain
well-understood transformations; for generating queries over the
datasets to check for rule violations; and for generating human-readable
documentation.

I have not had the opportunity to do much of this in XML, although I
developed something similar using schematron for an XML validation
problem.

Regards,
--Paul

> How do (might) you validate that instance A has been correctly 
> transformed to instance B, assuming input and output are both valid to 
> the schemas.
> 
> Not something I've seen mentioned on this list?
> 
> regards