Public Comment

[comment-form@oasis-open.org]

Comment from: dms@haplos.com

Name: Dave Scott
Title: President
Organization: Haplos, Inc.
Regarding Specification:  UBL NDR

Rule ATD1 prohibits the use of user-defined attributes, effectively limiting the use of attributes in an NDR-compliant schema to CCT meta-data.  I understand this restriction as an attempt to maximize the extensibility of NDR-compliant schema.  However, this restriction has an adverse affect on stream-processing of very large documents that I have not seen mentioned in discussions of the rule.  Generic XML stream-processing tools usually cache only the ancestor relations -- the current node stack -- for use in determining processing conditions.  To follow the NDR, attributes that are often used as processing 'switches' ('type', 'role', 'method', ...) must be moved down in the hierarchy to become child nodes.  This makes it impossible to use the stack-state-only tools, forcing developers to write custom SAX handlers to stream process documents.  In addition, the only way to avoid caching all the children during stream processing is to add ordering constraints that forces the 'switches' to appear first amongst the children.  Introducing schema constraints which do not have a semantic justification is not a good practice.

I understand that rule ATD1 is making a trade-off.  But I think it gives away too much.  Following the NDR, (1) I lose the ability to use attributes to capture the intended semantics of my model (especially attributes that have a scoping function.)  And, more importantly, (2) I lose the ability to use generic XML stream processing tools to efficiently handle large documents.