findByIri/find Cartesian-products independent multi-valued properties in one BGP

[ddeboer]

Summary

When a schema has more than one independent multi-valued (@array) property, the query generated for findByIri/find places them all as OPTIONALs in a single BGP. The solution set becomes the Cartesian product of those properties' values. The parsed entity is correct, but on a store that does not deduplicate CONSTRUCT output (e.g. ad-freiburg/qlever#2062) the product is materialised on the wire, so the response grows as the product of the cardinalities instead of their sum.

Reproduction

A schema with three independent multi-valued properties (VoID here, but the shape is generic):

const Summary = {
  "@type": "http://rdfs.org/ns/void#Dataset",
  propertyPartition: { "@id": ".../propertyPartition", "@array": true, "@optional": true,
    "@schema": { property: { "@id": ".../property" }, entities: { "@id": ".../entities" } } },
  vocabulary: { "@id": ".../vocabulary", "@array": true, "@optional": true },
  dataDump:   { "@id": ".../dataDump",   "@array": true, "@optional": true },
};
await createLens(Summary, { sources: [endpoint] }).findByIri(iri);

On a real entity with 273 property partitions × 10 vocabularies, QLever returned ~3.2 MB / ~11k triples, of which only ~175 KB / ~820 are distinct — a ~13× inflation and ~10× slower to fetch/parse.

Workaround

Hand-write a CONSTRUCT for lens.query() giving each independent multi-valued property its own UNION branch. Same parsed result, ~13× smaller wire.

Possible directions (deferring to you on the generator)

• Emit each independent multi-valued property in its own UNION branch (single-valued ones stay grouped), so results are proportional to sum(cardinalities) not product(cardinalities).
• Or fetch each multi-valued property separately and merge client-side.

Happy to help if a direction is chosen.

Environment

• ldkit 2.7.1; backing store QLever (does not deduplicate CONSTRUCT output).