Re: Alternative algorithm to restrict use of data/string in patterns.

[Murata Makoto <mura034@attglobal.net>]

Kohsuke KAWAGUCHI wrote:

> One of my concern to the current restriction is that it's probably not
> closed to boolean operations(union/intersection/difference).
> 
> Mine is made to satisfy closedness property, and it exactly exclude
> those patterns which cause the problem to validators. No more, no less.

#18. restrictStringUse is still open.

Let me propose another possibility.  Allow me to write in a formal 
manner.

We denote the class of regular languages over a finite set X by REC(X).  
Given a normalized grammar G, let N be the (finite) set of <ref> 
in G, A be the (finite) set of <attribute> in G, and D be the (finite) 
set of <data> in G.

Any content model (i.e., the children of some <element>) in G certainly 
denotes a regular set L over the union of N, A, and D.  That is, 
L is in REC(N \cup A \cup D).

I would propose the following restriction.
 
	L - D  is in REC(N \cup A)

This restriction is closed under intersection and union, but is not 
closed under difference.  This is closer to James' proposal, but is 
more restrictive.

Cheers,

Makoto