[Maya-Kid]

For #923, as suggested there.

UCO states its constraints as SHACL shapes rather than as rdfs:domain / rdfs:range, so its properties arrive at import with almost no domains. A datatype property without a domain is skipped at import. This change reads the shapes.

Problem

• The projection reads a signature from rdfs:domain / rdfs:range, and from schema.org's domainIncludes / rangeIncludes.
• UCO 1.5.0 declares 6 domains for 756 properties. Its constraints live in node shapes, for example:

  observable:AccountAuthenticationFacet a owl:Class, sh:NodeShape ;
      sh:property [ sh:path observable:passwordType ; sh:datatype xsd:string ] ;
      sh:targetClass observable:AccountAuthenticationFacet .

• An attribute without a domain is skipped at import (no_domain). On the merged UCO, 561 of the 579 datatype properties are skipped for this reason, and only 2 would be imported.
• The property shapes are blank nodes, and read_triples drops every triple whose subject is a blank node.

Change

Keeping the shape triples. In crates/utopia-ingest/src/ontology_rdf.rs, read_triples now keeps the three triples a property shape needs (sh:path, sh:class, sh:datatype) when their subject is a blank node. It records a blank-node object in a new field that only the shape pass reads. Every other blank-node triple is dropped as before, and triples still counts only triples whose subject is an IRI.

Reading the shapes. shape_signatures reads node shape sh:property property shape:

• the property is the property shape's sh:path, when that path is an IRI;
• its domain is the class the node shape targets: sh:targetClass, or the node itself when it is both a class and a sh:NodeShape (SHACL's implicit class target);
• its range is the property shape's sh:class or sh:datatype.

Declarations win. project uses these only where the property declares no domain or range of its own.

• A property constrained on several classes gets all of them as its domain, read the way domainIncludes is.
• Ranges that come from different shapes are alternatives (ranges_union), as with rangeIncludes.

Not read. Each of these is noted in the code:

• sh:targetSubjectsOf and sh:targetObjectsOf, which do not name a class;
• inverse and sequence paths;
• sh:or and sh:node.

The report. sh:targetClass, sh:property, sh:path, sh:class and sh:datatype are no longer reported as not projected. The other SHACL predicates still are.

The record. In docs/decisions/0001-ontology-import-and-governance.md, decision 13 now says where a signature comes from when a property declares none, and a revision entry records the change.

None of the five bundled packs mentions SHACL, so their projection is unchanged.

Verification

Run in containers on a Linux host: Rust 1.98.1 and PostgreSQL 16 with pgvector.

New tests (ontology_rdf::shape_tests)

One Turtle snippet covers each form:

• a class that is its own shape, with an explicit sh:targetClass and blank-node property shapes carrying sh:class and sh:datatype;
• a class that is a sh:NodeShape without sh:targetClass;
• a named node shape that points at a named property shape;
• a property constrained on two classes, which gets two domains and two ranges read as alternatives;
• a declared rdfs:domain / rdfs:range, which wins over the shape;
• forms that are not read: an inverse path, sh:or, sh:targetSubjectsOf, and a path to a property the file does not declare;
• the consumed SHACL predicates, which are not reported as not projected; the triple count still counts only IRI subjects.

On dev with only the new tests, three of the four fail. The one that passes is the test that a declared domain and range win: it pins that a shape never overrides a declaration.

This branch

• cargo fmt --all --check and cargo clippy --locked --workspace --all-targets -- -D warnings are clean.
• UTOPIA_TEST_REQUIRE_PDFTOTEXT=1 cargo test --locked --workspace passes: 1297 passed, 0 failed. This includes the projection tests against the bundled packs.

UCO 1.5.0 through the import preview

The 18 Turtle files under ontology/ were merged by concatenation; they have no labelled blank nodes and no relative IRIs, so that is a valid merge. The import plan that the preview returns was run into an empty base, on dev and on this branch:

	dev	this branch
Properties (object / datatype)	756 (177 / 579)	756 (177 / 579)
With a domain	6	753
With a range	722	748
With both	3	748
Datatype properties skipped for having no domain	561	1

• Imported. 549 datatype properties are imported instead of 2. The 30 still skipped are:
  • the same 16 key collisions;
  • 13 whose range is xsd:hexBinary or xsd:base64Binary, which the importer does not map today;
  • core:informalType, a super-property that no shape constrains.
• Examples.
  • observable:passwordType gets AccountAuthenticationFacet → xsd:string.
  • core:hasFacet gets UcoObject → Facet.
  • core:name gets all eight classes that constrain it.
• Still without a domain or range. Three properties have no domain and eight have no range. Their shapes use forms this change does not read:
  • sh:or over datatypes;
  • a literal with no datatype;
  • a property shape targeting the subjects of its own path.

[WaylandYang]

Approving. Thanks @Maya-Kid.

Reviewed at df75d62. What I checked:

• Only the gaps are filled. A declared rdfs:domain / rdfs:range wins, side by side, and a test pins it.
• Blank nodes stay contained. Only sh:path, sh:class and sh:datatype are kept for a blank subject, the blank object sits in a field nothing else reads, and the triple count is unchanged. The members of an sh:or list are kept too but never reached, since no node shape lists them under sh:property.
• What is not read is said, in the code and in the tests, and still shows up in the report.

The UCO table answers what I was going to ask: 549 datatype properties imported where there were 2, and the 30 left over each have a reason. The 13 with xsd:hexBinary / xsd:base64Binary look like a small follow-up of their own if you want it.