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

[James Clark <jjc@jclark.com>]

Your algorithm is a *lot* more complicated than the current (JTREX)
algorithm, which in your notation is simply:

function void check( pattern p ) {
    switch( p ) {
    case: p == <oneOrMore> p1 </oneOrMore>
      if (p1 contains <data> or <string>)
         this pattern is illegal!

    case: p == <interleave> p1 p2 </interleave>
    case: p == <group> p1 p2 </group>
        check( p1 )
        check( p2 )
        if( p1 contains <data/> or <string/>,
        &&  p2 contains <data/>, <string/>, <anyString/> or <element />, or
viceversa)
            this pattern is illegal!

    case: p == <choice> p1 p2 </choice>
        check( p1 )
        check( p2 )

    case: p == <anyString/>
    case: p == <string> ... </string> or <data> ... </data>
    case: p == <attribute /> or <element />
        do nothing;
}

It's nice that your algorithm finds exactly the cases that cause a validator
problems, but is that useful?  Are there are cases that your algorithm
allows and that the JTREX algorithm disallows that users might reasonably
want?

James