Only first page of Hydra paged collection returned

[ddeboer]

Only first page of Hydra paged collection returned

[rubensworks]

Not sure yet, will look into it next week.

[github-actions]

Thanks for reporting!

[ddeboer]

@rubensworks Is this an easy fix on your side? If not, I will implement a workaround on ours.

[rubensworks]

Support for paged collections (non-TPF/QPF) in Comunica is not well-tested at the moment, so issues like these are not surprising.
In any case, this is something we want to properly support, so this must be fixed.
But we won't be able to fix this in the very short term, so a workaround is preferred.

Some notes to self:

Simpler query that also fails to produce results:

SELECT *
WHERE {
  <https://opendata.picturae.com/dataset/dre_a2a_webservice> <http://purl.org/dc/terms/identifier> ?i.
  <https://opendata.picturae.com/dataset/dre_a2a_webservice> <http://purl.org/dc/terms/issued> ?issued.
}

The problem is that the linked hypermedia iterator is overwriting metadata per new page. In this case, each page is defaulting to the none-source-type, which provides exact cardinalities for matches in that page (while TPF falls uses Hydra cardinality). This causes the empty-join actor to be used, which returns an empty result stream.

One solution would be to merge (and test) the feature/adaptive-join branch. In that branch, we prefer dataset-level cardinalities, which will contain the Hydra cardinality, so that we don't use the page-specific cardinality.

Also, if comunica/comunica-feature-link-traversal#102 is the same problem, we will want to change the none-source-type to not emit exact cardinalities, but only lowerLimit cardinalities. Furthermore, the empty-join can then not be used for lowerLimit's.