Yes I understand that XPATH allows all the nodes to be selected but you're assuming that I already know there will be multiple Attribute nodes. My use case is streaming with a rather complicated schema so I'm concerned about the above example in that Databricks doesn't rescue data when the schema is initially misconfigured.

Converting to JSON would have the same problem - if there aren't files with multiple nodes then the converted JSON wouldn't use an array to store the resulting field which would lead to issues when multiple nodes start appearing.

Does the Maven library rescue data that wasn't captured in the schema like in the above example?