Re: [xml] xmllint complains about non-determinist content model

[Robin Munn <rmunn pobox com>]

On Tue, Dec 23, 2003 at 03:28:47PM -0500, Marcel Jemio wrote:
> Robin - 
>
> This is the "famed" non-deterministic or ambiguous content model
> scenario...  How is the parser supposed to know which content grouping
> to validate element "name"  to?  It can't in your scenario.  Even if you
> were to use the schmea "group" to abstract your "name/email" grouping,
> that would still qualify as non-deterministic.
>
> Is this an inclusive-or example?  If so the best way I have found to do
> this, is to represent the:
> >                 <xsd:sequence>
> >                     <xsd:element ref="name"/>
> >                     <xsd:element ref="email"/>
> >                 </xsd:sequence>
>
> Under another element so your final would like similar to this:
> >     <xsd:element name="person">
> >         <xsd:complexType>
> >             <xsd:choice>
> >                 <xsd:element ref="name"/>
> >                 <xsd:element ref="email"/>
> >                 <xsd:element name="group">    
>                      <xsd:sequence>
> >                        <xsd:element ref="name"/>
> >                        <xsd:element ref="email"/>
> >                    </xsd:sequence>
>                   </xsd:element>
> >             </xsd:choice>
> >         </xsd:complexType>
> >     </xsd:element>
>
> Yes this adds another layer to your scenario but there is no error.

Thanks for the suggestion. Yes, this is inclusive-or: a <person> element
can have a <name> element or an <email> element, or both; the only thing
forbidden is a <person> element with neither a <name> or an <email>.

(In the actual project I'mn working on, the offending tag occurs about
five or six levels deep in our XML structure; I abstracted it down to
the simplest possible example of the problem for my question).

Unfortunately, at this stage in the project we have far too much code
already written to make it feasible to change the XML spec by adding a
<group> tag to group the "name+email" combination. What I think I will
end up doing is something like this:

    <xsd:element name="person">
        <xsd:complexType>
            <xsd:sequence>
                <xsd:element ref="name" minOccurs="0">
                <xsd:element ref="email" minOccurs="0">
            </xsd:sequence>
        </xsd:complexType>
    </xsd:element>

and write special-case code to check for the invalid case of a <person>
element with no <name> or <email> (which would be considered valid by
that schema fragment). I hate to be writing special-case code when I
thought the <xsd:choice> element was going to be the exact thing I
needed, though.

This still doesn't seem right. Is there really no way to express in XML
Schema the concept "A alone, or B alone, or A and B, but not neither"?

-- 
Robin Munn
rmunn pobox com