Discussed in virtual F2F

The rules what datatypes are needed in a PublishedDataSet are only known at the time the PublishedDataSet is configured. One reason could be DataTypes that allow subtypes or fields that allow all DataTypes.

If the information is compressed in the file header, there is no way to put the right information back into the PublishedDataSets.

Further discussions and conclusions during UAFX prototyping and plug-fest:

(1) Scope namespaces and DataTypeDescriptions
We need both the DataTypeDescriptions and the Namespace array in each affected DataSetMetaData that uses DataTypes not defined by OPC UA
The DataTypeDescription part is already stated this way in the current Part 14 but the namespace part needs futher clarification
This requires a namespace index mapping for messages (e.g. ExtensionObject EncodingId) potentially on sender and receiver side.

(2) Mapping to target (and source) variables
The DataType NodeId in the DataSetMetaData may not match the DataType NodeId of the target variable (maybe also on the source variable).
In this case the application must check if the DataTypeDescription content matches and need to decide if it would accept the difference or not.
Mapping would only be possible if the structure is the same.

6.2.3.2.2 DataTypeSchemaHeader

Added following additional text:

A Publisher should keep the namespaces array unchanged even if the oder of namespaces get changed in the Publisher application. A change of the namespace array requires a change to the MajorVersion in the DataSetMetaData.
The Subscriber must map namespace indices in received messages if the data is processed on the context of an OPC UA information model e.g. if the values are written to target Variables. This affects encoding NodeIds in ExtensionObjects but also all other namespace indices in NodeIds and BrowseNames contained in the messages. If the Subscriber receives Structure DataTypes where the target Variables DataTypes have the same structure but different DataType NodeIds, the Subscriber must verify the consistency of the layout at start-up and must map the complete encoding NodeId when receiving the data.
If the DataSetMetadData is created outside the Publisher, the same mapping may be required in the Publisher.

Agreed to changes edited in Dallas F2F.