UAWG Nov-13-2012:

An interesting CMP/IOP issue has arisen thanks to a CTT test. The test writes a NULL value to all data-types and expects Bad_TypeMismatch. Randy showed in Part 6 the following text which essentially says that the server could/should accept this and simply handle it:

"Variants can be empty. An empty Variant is described as having a null value and should be treated like a null column in a SQL database. A null value in a Variant may not be the same as a null value for data types that support nullssuch as Strings. Some Development Platforms may not be able to preserve the distinction between a null for a DataType and a null for a Variant. Therefore Applications shall not rely on this distinction."

[note the typo "nullssuch"]

... after circulating an email to the greater CMP community, the opinion-at-large was that we need the spec to be clearer in terms of what should be acceptable.

The general agreement at the end of the UA WG call today was that a server should replace the null-variant with a null-string.

Note: the language/platform itself plays a part in this discussion.