Issue: shall we restrict subordinates of <oneOrMore>

[Murata Makoto <mura034@attglobal.net>]

I believe that we have to restrict <oneOrMore> patterns.  If we 
allow an <oneOrMore> element to contain both <attribute> and <ref>, 
I do not think that we can test the equivalence and subset relationship 
of patterns [1].  Furthemore, descriptive power of such patterns is beyond 
the class of regular languages, and (in my opinion) excessive.

I would propose restrictions as below:

After normalization by JamesC [2] , the subordinate of a <oneOrMore> element 
takes one of the three forms as below:

1) a pattern not containing <attribute> directly or indirectly.

2) an <attribute> element.

3) a <choice> element containing <attribute> elements or 
   <choice> elements of the form 3) only.

[1] http://lists.oasis-open.org/archives/trex/200105/msg00015.html

[2] http://lists.oasis-open.org/archives/trex/200104/msg00046.html


Cheers,

Makoto