Re: RNC modularisation redux

Robert Leftwich wrote:

> I have a similar problem to Sean McGrath as described in
> http://relaxng.org/pipermail/relaxng-user/2004-July/000544.html and 
> also Adrian
> Robert in
> http://lists.oasis-open.org/archives/relax-ng-comment/200407/msg00005.html 
> but
> there was no follow up to Sean's last email and there was only the one 
> answer to
> Adrian's. Given that it seems to work as required in my (limited) 
> testing, is
> specifying 'combine="choice" for every <define> in the base schemas (as
> suggested in
> http://lists.oasis-open.org/archives/relax-ng-comment/200407/msg00007.html) 
> the
> best way to go?

Robert,

An update from my own work in this area. Yes, using combine = "choice" 
gets over the multiple defines error but, unfortunately trang generates 
an error producing XSD if any file is included more than once.

There appear to be good reasons for this :-) I tweaked the trang souce 
to warn rather than generate an error just to see what would happen. The 
result was XSDs that where missing element declarations and other such 
things.

So, I switched tack. I wrote a Python program that "de-includes" 
includes in the compact syntax generating a single output RNC file. As 
well as resolving the includes, a couple of other things were necessary:

1. Hoisting the datatype declarations to the top
2. Hoisting the namespace declarations to the top (including any default 
namespace)
3. Generating an error if multiple default namespace declarations found
4. Comment out any "start" definitions in any included RNCs

The result monolithic RNCs work great. I treat them as compiled, freeze 
dried snapshots of a schema at a point in time. Great for knowing 
exactly what iteration of each module was used in a schema at any one 
moment.

As a bonus prize, trang generates modularised XSDs - one for each 
namespace in the monolithic RNC.

Sean
http://seanmcgrath.blogspot.com