@google-cloud/kcmd 0.0.0-stage → 0.1.0

package/build/ts/kcmd/semantic/converters/owl/parse.js

@@ -0,0 +1,634 @@
+"use strict";
+// Turtle (.ttl) -> OwlModel parser.
+//
+// A mechanical triples-to-structs step: it reads RDF with `n3`, recognizes the
+// OWL/RDFS/SKOS/Dublin-Core terms the converter supports, and fills the staging
+// OwlModel (model.ts). All OWL -> OSI mapping policy lives in to_ir.ts, not
+// here; this file only decides "which triples matter and where do their values
+// go".
+//
+// Declaration order is preserved: classes and properties appear in the order
+// their `rdf:type` triple is first seen in the document, so the generated OSI
+// is stable across runs (and matches its golden).
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.localName = localName;
+exports.namespace = namespace;
+exports.parseOwl = parseOwl;
+const n3_1 = require("n3");
+// RDF/RDFS/OWL/SKOS/Dublin-Core term IRIs we recognize. Only these; anything
+// else is carried by neither the parser nor the model (see the user guide's
+// "not covered yet").
+const RDF = 'http://www.w3.org/1999/02/22-rdf-syntax-ns#';
+const RDFS = 'http://www.w3.org/2000/01/rdf-schema#';
+const OWL = 'http://www.w3.org/2002/07/owl#';
+const SKOS = 'http://www.w3.org/2004/02/skos/core#';
+const DCTERMS = 'http://purl.org/dc/terms/';
+const DC = 'http://purl.org/dc/elements/1.1/';
+const XSD = 'http://www.w3.org/2001/XMLSchema#';
+const XSD_BOOLEAN = `${XSD}boolean`;
+const XSD_STRING = `${XSD}string`;
+// The two lexical forms of xsd:boolean `true` (XSD admits both `true` and `1`);
+// `false`/`0` restate the default and are not carried (see the OWL_DEPRECATED
+// case).
+const XSD_BOOLEAN_TRUE = new Set(['true', '1']);
+const RDF_TYPE = `${RDF}type`;
+// RDF collection (list) terms, used to walk an owl:hasKey list.
+const RDF_FIRST = `${RDF}first`;
+const RDF_REST = `${RDF}rest`;
+const RDF_NIL = `${RDF}nil`;
+const OWL_CLASS = `${OWL}Class`;
+const OWL_DATATYPE_PROPERTY = `${OWL}DatatypeProperty`;
+const OWL_OBJECT_PROPERTY = `${OWL}ObjectProperty`;
+const OWL_INVERSE_FUNCTIONAL_PROPERTY = `${OWL}InverseFunctionalProperty`;
+// Property characteristics carried verbatim (no native OSI home). Recognized by
+// their rdf:type, like owl:InverseFunctionalProperty, and collected as
+// typed-subject sets.
+const OWL_SYMMETRIC_PROPERTY = `${OWL}SymmetricProperty`;
+const OWL_TRANSITIVE_PROPERTY = `${OWL}TransitiveProperty`;
+const OWL_FUNCTIONAL_PROPERTY = `${OWL}FunctionalProperty`;
+const OWL_REFLEXIVE_PROPERTY = `${OWL}ReflexiveProperty`;
+const OWL_IRREFLEXIVE_PROPERTY = `${OWL}IrreflexiveProperty`;
+const OWL_ASYMMETRIC_PROPERTY = `${OWL}AsymmetricProperty`;
+const OWL_ONTOLOGY = `${OWL}Ontology`;
+const OWL_HAS_KEY = `${OWL}hasKey`;
+// List-valued constructs whose object is an RDF collection head (a blank node),
+// resolved with the same list walker as owl:hasKey. owl:oneOf enumerates a
+// class's members; owl:propertyChainAxiom composes a property from a sequence
+// of others. Both are carried verbatim (no native OSI home).
+const OWL_ONE_OF = `${OWL}oneOf`;
+const OWL_PROPERTY_CHAIN_AXIOM = `${OWL}propertyChainAxiom`;
+const OWL_INVERSE_OF = `${OWL}inverseOf`;
+const OWL_EQUIVALENT_PROPERTY = `${OWL}equivalentProperty`;
+const OWL_PROPERTY_DISJOINT_WITH = `${OWL}propertyDisjointWith`;
+const OWL_DISJOINT_WITH = `${OWL}disjointWith`;
+const OWL_DEPRECATED = `${OWL}deprecated`;
+const OWL_VERSION_INFO = `${OWL}versionInfo`;
+const OWL_THING = `${OWL}Thing`;
+// Set-level axioms. Each is asserted on an ANONYMOUS node (not a named term)
+// typed with one of these, whose `owl:members` list names the set the axiom
+// constrains. owl:AllDifferent also accepts the legacy OWL 1
+// `owl:distinctMembers` spelling. Carried verbatim at the model level (no
+// native OSI home).
+const OWL_ALL_DISJOINT_CLASSES = `${OWL}AllDisjointClasses`;
+const OWL_ALL_DISJOINT_PROPERTIES = `${OWL}AllDisjointProperties`;
+const OWL_ALL_DIFFERENT = `${OWL}AllDifferent`;
+const OWL_MEMBERS = `${OWL}members`;
+const OWL_DISTINCT_MEMBERS = `${OWL}distinctMembers`;
+const RDFS_LABEL = `${RDFS}label`;
+const RDFS_COMMENT = `${RDFS}comment`;
+const RDFS_DOMAIN = `${RDFS}domain`;
+const RDFS_RANGE = `${RDFS}range`;
+const RDFS_SUBCLASS_OF = `${RDFS}subClassOf`;
+const RDFS_SUBPROPERTY_OF = `${RDFS}subPropertyOf`;
+const RDFS_SEE_ALSO = `${RDFS}seeAlso`;
+const RDFS_IS_DEFINED_BY = `${RDFS}isDefinedBy`;
+const OWL_EQUIVALENT_CLASS = `${OWL}equivalentClass`;
+const RDFS_RESOURCE = `${RDFS}Resource`;
+// The implicit universal superclasses: every class is trivially a subclass of
+// owl:Thing / rdfs:Resource, so an explicit `rdfs:subClassOf` naming one
+// carries no inheritance information and is not recorded as `extends` (nor
+// warned about).
+const TOP_CLASS_IRIS = new Set([OWL_THING, RDFS_RESOURCE]);
+const SKOS_ALT_LABEL = `${SKOS}altLabel`;
+const SKOS_PREF_LABEL = `${SKOS}prefLabel`;
+const SKOS_HIDDEN_LABEL = `${SKOS}hiddenLabel`;
+const SKOS_DEFINITION = `${SKOS}definition`;
+const SKOS_EXAMPLE = `${SKOS}example`;
+const DCTERMS_DESCRIPTION = `${DCTERMS}description`;
+const DC_DESCRIPTION = `${DC}description`;
+const KIND_BY_TYPE = {
+    [OWL_CLASS]: 'class',
+    [OWL_DATATYPE_PROPERTY]: 'datatypeProperty',
+    [OWL_OBJECT_PROPERTY]: 'objectProperty',
+};
+// The namespace-stripped local name of a term IRI: the part after the last `#`
+// or `/`. This is the OSI-facing identity of an in-namespace term (e.g. an
+// entity/field name); the mapper also uses it to shorten a carried in-namespace
+// cross-reference (see to_ir.refValue). Exported for that mapper use.
+function localName(iri) {
+    const cut = Math.max(iri.lastIndexOf('#'), iri.lastIndexOf('/'));
+    return cut >= 0 ? iri.slice(cut + 1) : iri;
+}
+// The namespace portion of a term IRI (everything up to and including the last
+// `#`/`/`); the ontology's base IRI for provenance, and the key the mapper
+// compares against to decide whether a cross-reference is in-namespace (see
+// to_ir.refValue). Exported for that mapper use.
+function namespace(iri) {
+    const cut = Math.max(iri.lastIndexOf('#'), iri.lastIndexOf('/'));
+    return cut >= 0 ? iri.slice(0, cut + 1) : undefined;
+}
+// Renders a literal term in the N-Triples form used to carry an rdfs:seeAlso
+// literal (see the RDFS_SEE_ALSO case): the lexical value wrapped in double
+// quotes, followed by an optional `@lang` language tag or `^^<datatype>` type
+// IRI so a language-tagged or typed literal round-trips rather than collapsing
+// to a bare string. A plain xsd:string carries no suffix (it is the implicit
+// default). Backslash and double quote in the value are escaped, the two
+// characters that would otherwise break the wrapping. Reversed by parsing the
+// N-Triples literal grammar.
+function ntriplesLiteral(value, language, datatype) {
+    const quoted = `"${value.replace(/\\/g, '\\\\').replace(/"/g, '\\"')}"`;
+    if (language)
+        return `${quoted}@${language}`;
+    if (datatype && datatype !== XSD_STRING)
+        return `${quoted}^^<${datatype}>`;
+    return quoted;
+}
+// The most common namespace among a set of term IRIs -- the ontology's own
+// namespace. A tie (or a single term) resolves to the first in document order,
+// since `best` is only replaced on a STRICTLY greater count. Returns undefined
+// for an empty set (a document with no typed terms). Delimiter-exact, unlike
+// deriving the base from the ontology IRI (which often omits the trailing
+// `#`/`/`).
+function dominantNamespace(iris) {
+    const counts = new Map();
+    let best;
+    let bestCount = 0;
+    for (const iri of iris) {
+        const ns = namespace(iri);
+        if (ns === undefined)
+            continue;
+        const n = (counts.get(ns) ?? 0) + 1;
+        counts.set(ns, n);
+        if (n > bestCount) {
+            bestCount = n;
+            best = ns;
+        }
+    }
+    return best;
+}
+function emptyAnnotations() {
+    return {
+        synonyms: [],
+        examples: [],
+        domains: [],
+        ranges: [],
+        keyListHeads: [],
+        oneOfListHeads: [],
+        chainListHeads: [],
+        subClassOf: [],
+        subPropertyOf: [],
+        equivalentClass: [],
+        inverseOf: [],
+        disjointWith: [],
+        equivalentProperty: [],
+        propertyDisjointWith: [],
+        seeAlso: [],
+        isDefinedBy: [],
+        deprecated: false
+    };
+}
+// The effective description: the first present of rdfs:comment,
+// skos:definition, dcterms:description, dc:description -- a fixed precedence so
+// a term with more than one carries the most specific.
+function descriptionOf(a) {
+    return a.rdfsComment ?? a.skosDefinition ?? a.dctermsDescription ??
+        a.dcDescription;
+}
+// The per-term carried annotations (rdfs:seeAlso / isDefinedBy, owl:deprecated
+// / versionInfo), shared by classes and both property kinds, projected from the
+// accumulator into the model's common-annotation shape.
+function commonAnnotations(a) {
+    return {
+        seeAlso: a.seeAlso,
+        isDefinedBy: a.isDefinedBy,
+        deprecated: a.deprecated,
+        versionInfo: a.versionInfo,
+    };
+}
+/**
+ * Parses a Turtle document into an OwlModel.
+ *
+ * Synchronous: `n3`'s Parser returns the full quad array when called without a
+ * callback, which suits a one-shot file import. Throws on malformed Turtle (the
+ * parser's own error), which the CLI surfaces.
+ */
+function parseOwl(turtle) {
+    const quads = new n3_1.Parser().parse(turtle);
+    // Pass 1: types. An ordered list of typed subjects (first `rdf:type`
+    // occurrence wins the position) plus a kind index; the set of
+    // inverse-functional subjects and the ontology-header subject, which are
+    // recognized by their type but are not themselves classes/properties.
+    // Scanning `quads` directly (not the Store) keeps document order.
+    const order = [];
+    const kind = new Map();
+    const inverseFunctional = new Set();
+    // Property characteristics carried verbatim, each a set of the subjects typed
+    // with it. A property is commonly typed with several types (e.g.
+    // `a owl:ObjectProperty, owl:SymmetricProperty`), so a characteristic type is
+    // recorded here and does not compete with the kind that routes it to a
+    // class/property.
+    const symmetric = new Set();
+    const transitive = new Set();
+    const functional = new Set();
+    const reflexive = new Set();
+    const irreflexive = new Set();
+    const asymmetric = new Set();
+    // Anonymous set-level-axiom nodes, in document order, deduped by node id
+    // within each kind (a duplicate type triple must not list the node twice). A
+    // node typed as two different kinds -- legal, if unusual, RDF -- is recorded
+    // under each, so the dedup is per-kind rather than one shared seen-set. Their
+    // `owl:members` list is resolved after the list structure is known.
+    const allDisjointClassesNodes = [];
+    const allDisjointPropertiesNodes = [];
+    const allDifferentNodes = [];
+    const seenDisjointClasses = new Set();
+    const seenDisjointProperties = new Set();
+    const seenDifferent = new Set();
+    let ontologyIri;
+    for (const q of quads) {
+        if (q.predicate.value !== RDF_TYPE)
+            continue;
+        const type = q.object.value;
+        const subject = q.subject.value;
+        if (type === OWL_INVERSE_FUNCTIONAL_PROPERTY) {
+            inverseFunctional.add(subject);
+            continue;
+        }
+        if (type === OWL_SYMMETRIC_PROPERTY) {
+            symmetric.add(subject);
+            continue;
+        }
+        if (type === OWL_TRANSITIVE_PROPERTY) {
+            transitive.add(subject);
+            continue;
+        }
+        if (type === OWL_FUNCTIONAL_PROPERTY) {
+            functional.add(subject);
+            continue;
+        }
+        if (type === OWL_REFLEXIVE_PROPERTY) {
+            reflexive.add(subject);
+            continue;
+        }
+        if (type === OWL_IRREFLEXIVE_PROPERTY) {
+            irreflexive.add(subject);
+            continue;
+        }
+        if (type === OWL_ASYMMETRIC_PROPERTY) {
+            asymmetric.add(subject);
+            continue;
+        }
+        if (type === OWL_ONTOLOGY) {
+            if (ontologyIri === undefined)
+                ontologyIri = subject;
+            continue;
+        }
+        if (type === OWL_ALL_DISJOINT_CLASSES) {
+            if (!seenDisjointClasses.has(subject)) {
+                seenDisjointClasses.add(subject);
+                allDisjointClassesNodes.push(subject);
+            }
+            continue;
+        }
+        if (type === OWL_ALL_DISJOINT_PROPERTIES) {
+            if (!seenDisjointProperties.has(subject)) {
+                seenDisjointProperties.add(subject);
+                allDisjointPropertiesNodes.push(subject);
+            }
+            continue;
+        }
+        if (type === OWL_ALL_DIFFERENT) {
+            if (!seenDifferent.has(subject)) {
+                seenDifferent.add(subject);
+                allDifferentNodes.push(subject);
+            }
+            continue;
+        }
+        const k = KIND_BY_TYPE[type];
+        if (!k)
+            continue;
+        if (!kind.has(subject)) {
+            kind.set(subject, k);
+            order.push(subject);
+        }
+    }
+    // Pass 2: RDF list structure (rdf:first / rdf:rest), so an owl:hasKey list
+    // can be resolved to its member properties. List nodes are blank nodes
+    // unrelated to the typed subjects, so they are collected separately.
+    const listFirst = new Map();
+    const listRest = new Map();
+    // The `owl:members` list head of each set-level-axiom node. owl:members (OWL
+    // 2) wins over the legacy owl:distinctMembers (OWL 1, owl:AllDifferent only)
+    // when a node carries both, regardless of the order the two triples appear.
+    // owl:distinctMembers is honored ONLY on an owl:AllDifferent node, matching
+    // the OWL spec (it has no meaning on the disjointness axioms), so a stray
+    // owl:distinctMembers on an owl:AllDisjoint* node is ignored rather than
+    // carried.
+    const membersHead = new Map();
+    for (const q of quads) {
+        if (q.predicate.value === RDF_FIRST)
+            listFirst.set(q.subject.value, q.object.value);
+        else if (q.predicate.value === RDF_REST)
+            listRest.set(q.subject.value, q.object.value);
+        else if (q.predicate.value === OWL_MEMBERS)
+            membersHead.set(q.subject.value, q.object.value);
+        else if (q.predicate.value === OWL_DISTINCT_MEMBERS &&
+            seenDifferent.has(q.subject.value) && !membersHead.has(q.subject.value))
+            membersHead.set(q.subject.value, q.object.value);
+    }
+    // Resolves an RDF collection to its member IRIs, following rdf:rest to nil.
+    // Defensive against a cycle or a missing link (stops at the first gap).
+    const resolveList = (head) => {
+        const out = [];
+        const seen = new Set();
+        let node = head;
+        while (node && node !== RDF_NIL && !seen.has(node)) {
+            seen.add(node);
+            const value = listFirst.get(node);
+            if (value === undefined)
+                break;
+            out.push(value);
+            node = listRest.get(node);
+        }
+        return out;
+    };
+    // Pass 3: accumulate annotations for each typed subject (and the ontology
+    // header) in document order, so the first label is the primary and the rest
+    // are synonyms.
+    const annotations = new Map();
+    const annotationsFor = (iri) => {
+        let a = annotations.get(iri);
+        if (!a) {
+            a = emptyAnnotations();
+            annotations.set(iri, a);
+        }
+        return a;
+    };
+    const annotated = (iri) => kind.has(iri) || iri === ontologyIri;
+    for (const q of quads) {
+        const subject = q.subject.value;
+        if (!annotated(subject))
+            continue;
+        const a = annotationsFor(subject);
+        switch (q.predicate.value) {
+            case RDFS_LABEL:
+                // First label -> the display label; any further label -> a synonym.
+                if (a.label === undefined)
+                    a.label = q.object.value;
+                else
+                    a.synonyms.push(q.object.value);
+                break;
+            case SKOS_ALT_LABEL:
+            case SKOS_PREF_LABEL:
+            case SKOS_HIDDEN_LABEL:
+                a.synonyms.push(q.object.value);
+                break;
+            case SKOS_EXAMPLE:
+                a.examples.push(q.object.value);
+                break;
+            case RDFS_COMMENT:
+                a.rdfsComment = q.object.value;
+                break;
+            case SKOS_DEFINITION:
+                a.skosDefinition = q.object.value;
+                break;
+            case DCTERMS_DESCRIPTION:
+                a.dctermsDescription = q.object.value;
+                break;
+            case DC_DESCRIPTION:
+                a.dcDescription = q.object.value;
+                break;
+            case RDFS_DOMAIN:
+                a.domains.push(localName(q.object.value));
+                break;
+            case RDFS_RANGE:
+                a.ranges.push(q.object.value); // kept as IRI; mapper maps xsd:*
+                break;
+            case OWL_VERSION_INFO:
+                a.versionInfo = q.object.value;
+                break;
+            case OWL_HAS_KEY:
+                a.keyListHeads.push(q.object.value);
+                break;
+            case OWL_ONE_OF:
+                // Enumerated class -> no native OSI home; carried verbatim. The object
+                // is an RDF collection head; the members (usually individuals) are
+                // resolved with the list walker at build time. This is recorded for any
+                // annotated subject, but only the class build reads it -- if the
+                // subject is typed a property instead, the heads are collected here and
+                // then dropped (an owl:oneOf on a property is not meaningful).
+                a.oneOfListHeads.push(q.object.value);
+                break;
+            case OWL_PROPERTY_CHAIN_AXIOM:
+                // Property chain -> no native OSI home; carried verbatim. The object is
+                // an RDF collection head naming the composed properties, in order.
+                a.chainListHeads.push(q.object.value);
+                break;
+            case RDFS_SUBCLASS_OF:
+                // Class hierarchy -> the entity's `extends`. Named superclasses only:
+                // a blank-node object (an owl:Restriction and similar axioms) is not a
+                // named class, so it is ignored here (out of scope). The implicit
+                // universal superclasses (owl:Thing / rdfs:Resource) are ignored too --
+                // every class subclasses them, so they carry no inheritance
+                // information.
+                if (q.object.termType === 'NamedNode' &&
+                    !TOP_CLASS_IRIS.has(q.object.value))
+                    a.subClassOf.push(localName(q.object.value));
+                break;
+            case RDFS_SUBPROPERTY_OF:
+                // Property hierarchy -> no native OSI home; carried verbatim by the
+                // mapper as a custom extension. Named superproperties only (a
+                // blank-node property expression is out of scope). The full IRI is
+                // kept; the mapper shortens an in-namespace one (see to_ir.refValue).
+                if (q.object.termType === 'NamedNode')
+                    a.subPropertyOf.push(q.object.value);
+                break;
+            case OWL_EQUIVALENT_CLASS:
+                // Class equivalence -> no native OSI home; carried verbatim. Named
+                // classes only; a blank-node class expression (owl:intersectionOf, ...)
+                // is a class definition, not a plain cross-reference, so it is ignored
+                // (out of scope, like a blank-node subClassOf).
+                if (q.object.termType === 'NamedNode')
+                    a.equivalentClass.push(q.object.value);
+                break;
+            case OWL_DISJOINT_WITH:
+                // Class disjointness -> no native OSI home; carried verbatim. Named
+                // classes only (a blank-node class expression is out of scope).
+                if (q.object.termType === 'NamedNode')
+                    a.disjointWith.push(q.object.value);
+                break;
+            case OWL_INVERSE_OF:
+                // Inverse edge -> no native OSI home; carried verbatim. Named object
+                // properties only.
+                if (q.object.termType === 'NamedNode')
+                    a.inverseOf.push(q.object.value);
+                break;
+            case OWL_EQUIVALENT_PROPERTY:
+                // Property equivalence -> no native OSI home; carried verbatim. Named
+                // properties only.
+                if (q.object.termType === 'NamedNode')
+                    a.equivalentProperty.push(q.object.value);
+                break;
+            case OWL_PROPERTY_DISJOINT_WITH:
+                // Property disjointness -> no native OSI home; carried verbatim. Named
+                // properties only.
+                if (q.object.termType === 'NamedNode')
+                    a.propertyDisjointWith.push(q.object.value);
+                break;
+            case RDFS_SEE_ALSO:
+                // A pointer to further information -> carried verbatim as an N-Triples
+                // object term, so an IRI stays distinguishable from a literal on
+                // round-trip: an IRI as `<iri>`, a literal as `"text"` with any
+                // language tag (`@en`) or datatype (`^^<iri>`) preserved (see
+                // ntriplesLiteral). seeAlso is the only carried slot that admits either
+                // kind; every other carries a bare IRI or local name. A blank node has
+                // no stable identity to carry, so it is skipped.
+                if (q.object.termType === 'NamedNode')
+                    a.seeAlso.push(`<${q.object.value}>`);
+                else if (q.object.termType === 'Literal')
+                    a.seeAlso.push(ntriplesLiteral(q.object.value, q.object.language, q.object.datatype?.value ?? ''));
+                break;
+            case RDFS_IS_DEFINED_BY:
+                // The resource (usually the defining ontology) that defines this term
+                // -> carried verbatim. Named resources only.
+                if (q.object.termType === 'NamedNode')
+                    a.isDefinedBy.push(q.object.value);
+                break;
+            case OWL_DEPRECATED:
+                // Lifecycle flag -> carried only for an xsd:boolean literal that is
+                // `true`. That is the one meaningful assertion; an explicit `false`/`0`
+                // restates the default, and a non-boolean value (e.g. the string
+                // "true") is malformed -- neither is carried. Both XSD lexical forms of
+                // true (`true` and `1`) are accepted.
+                if (q.object.termType === 'Literal' &&
+                    q.object.datatype?.value === XSD_BOOLEAN &&
+                    XSD_BOOLEAN_TRUE.has(q.object.value))
+                    a.deprecated = true;
+                break;
+            default:
+                break;
+        }
+    }
+    const classes = [];
+    const datatypeProperties = [];
+    const objectProperties = [];
+    // The ontology's own namespace: the one shared by MOST of its typed terms.
+    // Taking the most common namespace (not merely the first typed term's) is
+    // robust to a document that also types a handful of foreign-namespace terms
+    // -- e.g. an imported class it annotates -- which would otherwise hijack the
+    // base if one happened to appear first and silently invert the mapper's
+    // in-namespace/cross-namespace shortening (see to_ir.refValue).
+    let baseIri = dominantNamespace(order);
+    for (const iri of order) {
+        const a = annotations.get(iri) ?? emptyAnnotations();
+        switch (kind.get(iri)) {
+            case 'class':
+                classes.push({
+                    localName: localName(iri),
+                    label: a.label,
+                    comment: descriptionOf(a),
+                    synonyms: a.synonyms,
+                    examples: a.examples,
+                    keys: a.keyListHeads.flatMap(resolveList).map(localName),
+                    subClassOf: a.subClassOf,
+                    equivalentClass: a.equivalentClass,
+                    disjointWith: a.disjointWith,
+                    // Full member IRIs of the enumeration set (a carried cross-reference;
+                    // the mapper shortens an in-namespace one). flatMap unions the
+                    // members across every oneOf head -- an enumeration is a set, so the
+                    // mapper dedupes and order does not matter (contrast propertyChain).
+                    oneOf: a.oneOfListHeads.flatMap(resolveList),
+                    ...commonAnnotations(a),
+                });
+                break;
+            case 'datatypeProperty':
+                datatypeProperties.push({
+                    localName: localName(iri),
+                    domains: a.domains,
+                    rangeIri: a.ranges[0],
+                    label: a.label,
+                    comment: descriptionOf(a),
+                    synonyms: a.synonyms,
+                    examples: a.examples,
+                    inverseFunctional: inverseFunctional.has(iri),
+                    functional: functional.has(iri),
+                    subPropertyOf: a.subPropertyOf,
+                    equivalentProperty: a.equivalentProperty,
+                    propertyDisjointWith: a.propertyDisjointWith,
+                    ...commonAnnotations(a),
+                });
+                break;
+            case 'objectProperty':
+                objectProperties.push({
+                    localName: localName(iri),
+                    // a.domains are already local names; a.ranges are raw IRIs. Carry all
+                    // of each so the mapper can warn about (and drop) the extras rather
+                    // than silently keeping only the first.
+                    domains: a.domains,
+                    ranges: a.ranges.map(localName),
+                    label: a.label,
+                    comment: descriptionOf(a),
+                    synonyms: a.synonyms,
+                    examples: a.examples,
+                    subPropertyOf: a.subPropertyOf,
+                    inverseOf: a.inverseOf,
+                    equivalentProperty: a.equivalentProperty,
+                    propertyDisjointWith: a.propertyDisjointWith,
+                    // One resolved chain per owl:propertyChainAxiom (map, NOT flatMap):
+                    // OWL 2 permits several chain axioms on one property, and fusing them
+                    // would be indistinguishable from a single longer chain. Full
+                    // property IRIs in chain order (the mapper shortens in-namespace
+                    // ones); within a chain order and repetition are preserved, so no
+                    // dedupe. An empty chain (a broken/gapped list) is dropped rather
+                    // than carried as [].
+                    propertyChain: a.chainListHeads.map(resolveList).filter(c => c.length),
+                    symmetric: symmetric.has(iri),
+                    transitive: transitive.has(iri),
+                    functional: functional.has(iri),
+                    reflexive: reflexive.has(iri),
+                    irreflexive: irreflexive.has(iri),
+                    asymmetric: asymmetric.has(iri),
+                    ...commonAnnotations(a),
+                });
+                break;
+            default:
+                break;
+        }
+    }
+    // The ontology header, if present, becomes model-level metadata. The base IRI
+    // is best derived from an actual term's namespace (exact, delimiter and all);
+    // the ontology IRI is only a heuristic fallback (its own IRI often omits the
+    // trailing `#`/`/`), so use it only when no term supplied a namespace.
+    let ontology;
+    if (ontologyIri !== undefined) {
+        const a = annotations.get(ontologyIri) ?? emptyAnnotations();
+        const labels = a.label !== undefined ? [a.label, ...a.synonyms] : a.synonyms;
+        ontology = {
+            comment: descriptionOf(a),
+            synonyms: labels,
+            examples: a.examples,
+            version: a.versionInfo,
+        };
+        baseIri = baseIri ?? ontologyBaseIri(ontologyIri);
+    }
+    // Resolve each set-level-axiom node's `owl:members` list to its member IRIs.
+    // A node with no members list, or one whose list is broken/empty, contributes
+    // nothing (an axiom over no members is meaningless). Member names are kept as
+    // full IRIs; the mapper shortens an in-namespace one.
+    const resolveMembers = (nodes) => nodes.map(n => membersHead.get(n))
+        .filter((h) => h !== undefined)
+        .map(resolveList)
+        .filter(members => members.length > 0);
+    return {
+        baseIri,
+        ontology,
+        classes,
+        datatypeProperties,
+        objectProperties,
+        allDisjointClasses: resolveMembers(allDisjointClassesNodes),
+        allDisjointProperties: resolveMembers(allDisjointPropertiesNodes),
+        allDifferent: resolveMembers(allDifferentNodes),
+    };
+}
+// The base namespace an ontology IRI stands for. An owl:Ontology is commonly
+// named by the namespace root without the trailing `#`/`/` (e.g.
+// `http://example.com/sales`), so append `#` when the IRI has no delimiter of
+// its own; otherwise keep it as given.
+function ontologyBaseIri(iri) {
+    if (!iri)
+        return undefined;
+    const last = iri[iri.length - 1];
+    if (last === '#' || last === '/')
+        return iri;
+    return `${iri}#`;
+}

package/build/ts/kcmd/semantic/converters/owl/to_ir.d.ts

@@ -0,0 +1,19 @@
+import { SemanticModel } from '../../ir';
+import { OwlModel } from './model';
+export interface ToIrResult {
+    model: SemanticModel;
+    warnings: string[];
+    stats: {
+        classes: number;
+        datatypeProperties: number;
+        objectProperties: number;
+    };
+}
+/**
+ * Maps a parsed OwlModel to a Semantic Model IR (see ../../ir.ts).
+ *
+ * `modelName` names the resulting semantic model (the CLI derives it from the
+ * source filename). The IR it returns serializes, via osi_converter, to the
+ * OSI YAML shown in the user guide.
+ */
+export declare function owlToIr(owl: OwlModel, modelName: string): ToIrResult;