SubjectMaps with BN, no reference data involved in join conditions

[dachafra]

Currently, it is possible to define Subject Maps using blank nodes without reference to the input data. In such cases, subjects are generated randomly, and there is no relationship with any input data. Thus, if a Triples Map configured in this way is used in a join condition, an RML processor must internally preserve the correspondence between the generated subjects and the data references used in the join conditions. This could introduce a significant performance overhead.

Therefore, an important question: should join conditions be allowed between Triples Maps that use blank nodes without reference data?

[chrdebru]

Imho yes. We discussed this while at USC (the discussion was also related to views). The solution here is that iterations keep track of their iteration (and their order). The work on LV by @elsdvlee and @pmaria makes that more explicit for their logical views, but the idea is similar. If you keep track of the randomly generated unused BN identifiers for each iteration, then you can reuse them. The only caveat is: what if you have two deep copies of the same logical source? Imho: that will yield different iterations as you will enter the rabbit hole of determining what is "equivalent", such as the equivalence of SQL queries.