Hi,

This is a … somewhat complicated topic.

Like, the actual “format” unless I forget is only the “XML encoding” of NodeId, thus “defined” in https://reference.opcfoundation.org/Core/docs/Part6/5.3.1/#5.3.1.10.

BUT! I would personally say it would be better to say “there is no such format, as no NodeId should ever be referred by the namespace index, but instead by the namespace uri”. The index is basically just an optimization in the binary (numbers are shorter in binary vs. the uri) and xml/json (uri takes more space in plaintxt format). Once it leaves that context it should be the uri, never the index. But pretty much everyone (we including) has used that format in UIs etc. and we have also failed this in the SDK by having APIs that use NodeId (but no-one could maybe have seen this 10-15 years ago).

Anyway, this is why most of the newer SDK APIs such as PubSub instead refer things instead via “UaNodeId” (so basically an “upgraded NodeId”), which uses the “uri form” via UaNamespace objects (that we weak-intern and thus can be reference-equals-compared in pretty same time as an index would be).

Though it should be noted that an index-based representation is maybe… nicer and shorter in UIs, but there is the danger that the mapping to the actual namespaceuri if forgotten (the indexes can freely change between sessions and restarts, there are no guarantees about them). So it is sort of a “known problem” of OPC UA in my mind. We have had some ideas about some “shorthand aliases” that could maybe be displayed instead of an index (e.g. “di” for the DI companion spec), but no actual implementation yet.

So just to be sure, you mean the NamespaceArray would have the same entry more than once? Then that server is not a valid OPC UA server and must be fixed.

I guess you should ask the OPC Foundation for “proof” that the NamespaceUris must be unique. Best would be if the spec would have it somewhere to be easily linked, but I have not found any like direct statement.

It can be inferred by combining multiple points, but not sure if that is .. enough.

For example, https://reference.opcfoundation.org/Core/docs/Part3/8.2.2/

“The namespace is a URI that identifies the naming authority responsible for assigning the identifier element of the NodeId. Naming authorities include the local Server, the underlying system, standards bodies and consortia. It is expected that most Nodes will use the URI of the Server or of the underlying system.

Using a namespace URI allows multiple OPC UA Servers attached to the same underlying system to use the same identifier to identify the same Object. This enables Clients that connect to those Servers to recognise Objects that they have in common.

…
“

Plus then the notes that an index form is used only as an encoding optimization.

“Namespace URIs, like Server names, are identified by numeric values in OPC UA Services to permit more efficient transfer and processing (e.g. table lookups). The numeric values used to identify namespaces correspond to the index into the NamespaceArray. “

It would not be possible to do the mapping in reverse (as one should always search which index to use per the uri, if they are not unique this cannot be done => they must be unique).

Combined with https://reference.opcfoundation.org/v105/Core/docs/Part4/7.30/ and the “remote references” possibility (please do not use them unless you have to, but servers here to prove a point).
” nodeId ExpandedNodeId NodeId of the TargetNode as assigned by the Server identified by the Server index. The ExpandedNodeId type is defined in 7.16.If the serverIndex indicates that the TargetNode is a remote Node, then the nodeId shall contain the absolute namespace URI. If the TargetNode is a local Node the nodeId shall contain the namespace index.”
IF the uris would not have to be unique, that would not work => they must be unique (as the NamespaceArray order can change at will e.g. on a server restart, a reference to other server would not “know” which index would have the uri at a given time, it must use the uri form of ExpandedNodeId, henceworth the uri is the defining part of uniqueness. Since this would be translated to an index for by a client that is connected to the both servers, it must be able to do unique ExpandedNodeId -> NodeId conversion, and that can only happen if there is an unique uri -> index mapping.

This was discussed yesterday at the OPC Foundation UA Working Group, and the consensus was that all entries in the NamespaceTable should be unique.

The following sentence at https://reference.opcfoundation.org/Core/docs/Part3/8.2.2/ defines the uniqueness in practice:

“Namespace URIs, like Server names, are identified by numeric values in OPC UA Services to permit more efficient transfer and processing (e.g. table lookups). The numeric values used to identify namespaces correspond to the index into the NamespaceArray. “