Re: [regrep] Proposed change to Classification Infomodel

[Farrukh Najmi <farrukh@wellfleetsoftware.com>]

Copying regrep-comment list to keep an archive of complex discussion:

The nested syntax is only syntactical sugar used to conveniently 
represent a single-inheritence ClassificationScheme
without having to explicitly wire Child ClassificationNodes to their 
parents. Servers are required to flatten out such nested
representation and replace the nested structure with explicit 
child-to-parent references via the parent attribute in RIM 3.

As we move to support Multiple Inheritance (MI)  of ClassificationNodes  
the nested syntax is no longer useful when there is MI involved.

Ideally we would just remove the ability for a ClassificationNode to 
have a set of child ClassificationNodes nested inside it as 
sub-elements. The last time we tried this we had a public out-cry from 
users that missed that convenience. Then there is the issue of migrating 
existing single-inheritence schemes to
not use nesting if we removed the convenience syntax.

Thus what we are saying is that we can continue to support nested syntax 
as convenience but it should only be used in single inheritance schemes.
If they are abused in MI schemes then the effect would be single 
inheritance semantics with last scheme published determining the parent 
of node that is child of multiple schemes.

The proposal does not compromise the need for supporting MI in 
ClassificationNodes. It simply states that we will continue to allow 
nested syntax for backward compatibility of pre-existing nested schemes 
which are single inheritence. MI schemes should not use nested syntax.

Questions for you are:

a) Do you think the proposal as it stands is good?
b) Should we removed nested syntax (ClassificationNode can no longer 
have child ClassificationNodes - only parent ClassificationNodes)
c) Any other changes on current proposal?

Thanks in advance for your valuable input.

Evan Wallace wrote:
>> Here is a summary of what we suggest refining:
>>
>>    * Use nested Parent elements inside a ClassificationNode to
>>      reference parent TaxonomyElement instead of using HasChild
>>      Associations from parent to child.
>>          o It is more similar to when we had a single parent attribute
>>            in ClassificationNode
>>          o It is more obvious a schema
>>          o It may (or may not) be a more efficient schema to implement
>>    * Add statement that nested ClassificationNode syntax would be
>>      supported as a convenience but only with single inheritance 
>> semantics
>>    * Retract suggestion to make Classification a sub-class of 
>> Association.
>>          o Impact of breaking backward compatibility in this area is
>>            not worth the gain. 
>
> Can you explain what this will mean in terms of expressivity?  The
> comment about single inheritance semantics has me worried.  Isn't part of
> the point to this to allow multiple classification?


-- 
Regards,
Farrukh Najmi

Web: http://www.wellfleetsoftware.com