[jdsika]

Mirror of linkml#4170, same head. Review there; this PR tracks the fork stack.

Proposed upstream description (2026-10-30), not yet applied to linkml#4170; it is copied there after review.

Merge order of this series: linkml#3716 → linkml#4168 → linkml#3993 → linkml#4144 → linkml#4171 → linkml#3989 → linkml#3990 → linkml#4156 → linkml#4157 → linkml#4158 → linkml#4028 → linkml#4169 → linkml#4170 → linkml#3446. A stack is rebased when its base lands.

Stacked on linkml#4028 and linkml#4169, which GitHub shows too; this PR's own diff is #34.

Read first: shaclgen.py (193), jsonschemagen.py (106) and class_expression.py (24); 323 lines of source; the rest is tests, snapshots and docs.

Summary

has_member asks for "at least one member satisfying the condition" of a multivalued slot. gen-shacl and gen-json-schema now read it the same way: a member is one value, which must satisfy the value operators, range and range_expression of the expression.

slots:
  tags:
    range: string
    multivalued: true
    has_member: {equals_string: reviewed}

tags	main: SHACL	main: JSON Schema	this PR: both
absent	valid	valid	invalid
[]	valid	invalid	invalid
[pending]	valid	invalid	invalid
[reviewed, pending]	valid	valid	valid

• gen-shacl ignored has_member (SHACL generator: has_member is ignored linkml/linkml#2465). It now emits sh:qualifiedValueShape with sh:qualifiedMinCount 1 (SHACL §4.7.3), on slots and in slot conditions of class-level expressions.
• gen-json-schema applied only the member's value operators, accepted an absent slot, and ignored has_member in class-level expressions. It now emits contains on a required slot, and for a slot inlined as a dict not: {additionalProperties: {not: <member>}}, since contains applies only to arrays.
• In a slot condition has_member decides presence, and the member takes the form of the condition.

Decisions

• A slot without a value has no member, so an absent slot and [] both fail: the metamodel says "at least one member", value_presence treats an empty list as absent, the compliance suite rejects [], and gen-owl emits owl:someValuesFrom. gen-json-schema accepted an absent slot but rejected [], a distinction RDF can't express.
• This changes gen-json-schema: an optional slot with has_member is now required.
• has_member on a single-valued slot contradicts the metamodel ("the value of the slot is multivalued"); both generators ignore it with a warning (gen-json-schema ignored it silently).
• gen-shacl skips, with a warning, a member that uses presence or slot-level operators, and equals_string on a reference.

Other generators

gen-owl keeps owl:someValuesFrom; ShEx, Python and Pydantic don't support has_member. linkml#3495 asks for a slot constraint that maps to JSON Schema contains: has_member is that constraint, and now holds for absent slots and in class-level expressions too.

How was this tested?

• test_has_member.py (55 cases) validates the same instance with both generated artifacts, plus SHACL-only and JSON-only cases: inheritance through slot_usage, dict inlining, member range and range_expression, negation, null.
• jsonschema_multivalued_has_member.yaml adds the absent, empty and null cases; the compliance test test_membership now expects SHACL to implement has_member, except on references.
• Rebased with feat(generators): translate class-level boolean expressions alike to SHACL and JSON Schema linkml/linkml#4028 and fix(generators): enforce range_expression on each value alike in SHACL and JSON Schema linkml/linkml#4169 onto main ae3df7a45 (2026-10-30), no conflict: tests/linkml 12,824 passed locally; CI on the head.

Areas of uncertainty

• The behaviour change for gen-json-schema users who rely on an absent slot passing has_member.
• Found, not addressed: an optional multivalued slot with minimum_cardinality: 1 and no value is valid for gen-json-schema (minItems without required) and invalid for gen-shacl (sh:minCount 1), on main too.

Fixes linkml#2465. Refs linkml#3495.

Checklist

• My code follows the contributor guidelines
• I have added tests that prove my fix/feature works
• Existing tests pass locally with my changes

AI Assistance

If you used AI tools while preparing this PR, you are still the author and responsible for understanding, verifying, and defending your submission. Please engage with reviewers personally rather than through your agent during feedback and revisions. See our AI Covenant for details.

Fork review (kept from the earlier mirror description)

2026-10-23, fourth push: re-stacked on the re-stacked #30 (baff32593); no conflict, no code change. test_has_member.py + test_range_expression.py + test_shaclgen.py + test_jsonschemagen.py + test_boolean_slot_compliance.py + test_validation_context.py: 4191 passed, 631 skipped, 2 xfailed.

2026-10-23, third push (adversarial review): has_member on a slot that isn't multivalued is now ignored with a warning by gen-json-schema too (it was silent; gen-shacl already warned), with test_has_member_on_a_single_valued_slot_is_ignored_with_a_warning; test_has_member_is_inherited covers slot_usage on a parent class for both generators; linkml#3495 and the pre-existing minimum_cardinality divergence recorded below. test_has_member.py: 55 passed; test_shaclgen.py + test_jsonschemagen.py + test_boolean_slot_compliance.py + test_validation_context.py + test_range_expression.py + test_has_member.py: 4146 passed, 663 skipped, 2 xfailed, 264 subtests passed.

2026-10-23, second push: rebased onto the reviewed #30 (c1527ebe7); the only conflict was the json-schema.rst paragraph both PRs extend, both sentences kept. No code change. test_shaclgen.py + test_jsonschemagen.py + test_boolean_slot_compliance.py + test_validation_context.py + test_has_member.py + test_range_expression.py: 4139 passed, 663 skipped, 2 xfailed.

2026-10-23: rebased onto the rebased #30 (38285fa89); no conflict, no code change. test_shaclgen.py + test_jsonschemagen.py + test_boolean_slot_compliance.py + test_validation_context.py: 3973 passed, 264 subtests passed. test_range_expression.py + test_has_member.py: 166 passed.

2026-10-30: rebased with #29 and #30 onto upstream main (ae3df7a45); no conflict, no code change. tests/linkml: 12,824 passed.

Stacked on #30. Covers upstream linkml#2465 (gen-shacl ignores has_member) and linkml#3495, which asks for a slot constraint that maps to JSON Schema contains: has_member is that constraint, and with this PR it holds for absent and empty slots and in class-level expressions too.

Change

has_member asks for "at least one member satisfying the condition" of a multivalued slot. Both generators now read it the same way: a member is one value, which must satisfy the value operators, range and range_expression of the expression.

slots:
  tags:
    range: string
    multivalued: true
    has_member: {equals_string: reviewed}

tags	before: SHACL	before: JSON Schema	now: both
absent	valid	valid	invalid
[]	valid	invalid	invalid
[pending]	valid	invalid	invalid
[reviewed, pending]	valid	valid	valid

• gen-shacl ignored has_member. It now emits sh:qualifiedValueShape with sh:qualifiedMinCount 1 (SHACL §4.7.3), on slots and in slot conditions of class-level expressions.
• gen-json-schema applied only the member's value operators, accepted an absent slot, and ignored has_member in class-level expressions. It now emits contains on a required slot. For a slot inlined as a dict it emits not: {additionalProperties: {not: <member>}}, because contains doesn't apply to objects.
• In a slot condition, has_member decides presence (an absent slot fails it), and the member takes the form of the condition: "definitely true" under none_of.

Decisions

• A slot without a value has no member, so an absent slot and [] both fail. The evidence:

  • the metamodel: "at least one member"; and value_presence says that for lists, ABSENT means empty;
  • the compliance suite: has_member on [] is invalid;
  • gen-owl's owl:someValuesFrom.

  gen-json-schema accepted an absent slot but rejected [], a distinction RDF can't express.

• This changes gen-json-schema: an optional slot with has_member is now required. The has_member cases of jsonschema_multivalued_element_constraints.yaml move to jsonschema_multivalued_has_member.yaml, which adds the absent, empty and null cases.

• has_member on a single-valued slot is contradictory input: the metamodel says "the value of the slot is multivalued". Both generators ignore it and log a warning (gen-json-schema used to ignore it silently; SHACL generator: has_member is ignored linkml/linkml#2465's own example is single-valued). In a class-level condition gen-shacl drops the operator, as for any untranslatable member, while gen-json-schema drops only the has_member.

• Skipped by SHACL, with a warning:

  • a member with presence or slot operators (minimum_cardinality, any_of, …);
  • equals_string on a reference; that is fork issue gen-shacl rules: translate equals_string on IRI- and typed-literal-valued ranges (follow-up to linkml/linkml#3989) #37.

• Dropped from the earlier version: comparing equals_string with class references by IRI (issue gen-shacl rules: translate equals_string on IRI- and typed-literal-valued ranges (follow-up to linkml/linkml#3989) #37), equals_expression (see fix(generators): enforce range_expression on each value alike in SHACL and JSON Schema #30), raising on unsupported input, and the changes to rules, which keep their own reading (feat(gen-shacl): translate presence-implies-value rules to SHACL-SPARQL linkml/linkml#3989).

Other generators

• gen-owl: owl:someValuesFrom under open-world semantics; unchanged.
• ShEx, Python, Pydantic: no has_member support; unchanged.
• Rules: on this base, neither generator's rule translation handles has_member; the rules stack (feat(gen-shacl): translate presence-implies-value rules to SHACL-SPARQL #19/feat(gen-shacl): add a compositional fallback for rule-to-SPARQL conversion #20) adds it to gen-shacl's composed rule translation.

Validation

• test_has_member.py: 48 cases. They validate the same instance with both generated artifacts (RDF through the generated JSON-LD context), plus SHACL-only and JSON-only cases. 34 fail on fix(generators): enforce range_expression on each value alike in SHACL and JSON Schema #30.
• Out-of-tree check during development: 480 random schemas and 5,760 instances (has_member on slots and in class-level conditions, with member range_expression). gen-shacl, gen-json-schema and an independent three-valued model agreed on all of them. That harness is not part of this PR.
• Mutation: 21 mutants of the new code, all killed; one by the doctest of states_presence.
• Sweep: across 387 schemas in the repository, SHACL and JSON Schema output is unchanged, with identical errors and warnings.
• Compliance: test_membership now expects SHACL to implement has_member, except on references.
• Tests: the agreement harness of fix(generators): enforce range_expression on each value alike in SHACL and JSON Schema #30's tests moves to tests/linkml/utils/generator_agreement.py, shared with these tests.
• CI: see the checks on the current head.

Found, not addressed here (it predates this PR): an optional multivalued slot with minimum_cardinality: 1 and no value is valid for gen-json-schema (minItems without required) and invalid for gen-shacl (sh:minCount 1), on main too. The same reasoning as for has_member would make the slot required in JSON Schema.