• Validation on input and output? Means conforming to the schema defined (e.g. integer between 0 and 100). dfdl assert could be used for asserting correctness of the data. Further complex validation would be via an external step (eg, Schematron invocation).
• Using DFDL to model data structures generated by OO code, particularly use of inheritance? Could be a use case for allowing complex type inheritance in DFDL in the future.
• Data often comes with units, infoset would want to reflect this? Could be achieved using expressions, dfdl:inputValueCalc and dfdl:hidden to apply a scaling factor based on units.
• What's the efficiency of DFDL? Particularly, can tests be turned off for fast read/write? The spec defines the behavior of the parser but not how to implement it. For instance, validation is not mandatory.
• Open source reference implementation would be good to have. (Lots of nodding).
• Scientific floating point data compresses badly. Knowing data structure can allow a more intelligent compression. DFDL is not intending to do transformations but that could be done on top.
• Encryption/compression could also be a use case for multi-layers, or for additional functions in the expression language.
• Need to reach out to DAIS-WG and DR-WG for (public) comments and to see whether DFDL will be actively used by those groups. Open source implementation would obviously help here.
• Comment on the spec: Should be a proposed recommendation (GFD-P-R) not informational (GFD-I).

Unless stated otherwise above:
IBM United Kingdom Limited - Registered in England and Wales with number 741598.
Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU