Add Base IRI to objects with relative IRIs

[marioscrock]

Hi everyone,
the R2RML specification says the base IRI is added to a term if the term map's term type is rr:IRI and the IRI is relative. As far as I understand, in the current version, this behavior is implemented only with respect to subjects so I modified the Executor to behave similarly also for named objects. Please let me know if the behavior described is correct and, eventually if the proposed changes can be the right way to introduce the modification. I tested it and it seems to behave as expected.

[andimou]

@marioscrock thank you for your remark. I will let one of the other guys look into the implementation but I will just add a clarification following your comment:

There are actually multiple ways of generating an IRI in RML and we need to make sure that the base IRI is properly handled in all cases. I think the table in this chapter: http://ebooks.iospress.nl/volumearticle/56006 has a very detailed table with all possible cases.

[DylanVanAssche]

Hi @marioscrock

Apologies for the long response time!
I agree with your approach but we need to extend some parts of it to fully comply with the [R2]RML spec.

When generating an absolute IRI from a relative one, a string concat is used, but this may not always be correct as
the R2RML spec says that we can only concatenate the base IRI with the relative one if the following conditions are met:

As @andimou mentioned, we need to take into account all possible cases for this:

Your patch only does this for objects from Object Maps, while the vanilla RMLMapper does it only for subjects of Subject Maps.
We may need to extend this so it happens for any R2RML Term Map as mentioned in the R2RML spec:

In short:

• We should check these conditions first before doing that, if there's one that fails, we can print a warning.
• We need to handle all possible cases for IRI generation.
• We need to create a set of tests to verify this behavior.

Please let us know what you think :)

[marioscrock]

Thank you @DylanVanAssche and @andimou for your reply and very detailed explanation. I think this pull request is associated with a quite old version of the source code, so if you prefer we can close it and maybe add an issue on this topic.
Some comments on the proposed patch:

• I assumed isValidIRI performed the checks mentioned
• To avoid the problem of dealing with different Term Maps and term types, I added the patch before writing the triples in the output store and checking only for NamedNodes. However, I agree that a set of tests should be defined to check that all the possible cases are handled correctly (relevant also for https://rml.io/test-cases/?)

[DylanVanAssche]

@marioscrock
Yes, the validation of IRIs is verified using isValidIRI but does not enforce the R2RML conditions mentioned in the spec to easily concat them.

Indeed, we need a set of test cases first, we can discuss that and add them at https://github.com/kg-construct/rml-test-cases-support :)