@ifc-lite/parser 5.0.0 → 5.2.0

package/dist/columnar-parser.js

@@ -4,591 +4,412 @@
 import { SpatialHierarchyBuilder } from './spatial-hierarchy-builder.js';
 import { EntityExtractor } from './entity-extractor.js';
 import { extractLengthUnitScale } from './unit-extractor.js';
-import { getAttributeNames, getAttributeNamesAcrossSchemas, getInheritanceChain } from './ifc-schema.js';
 import { parsePropertyValueWithComplex } from './on-demand-extractors.js';
 import { readQuantitySet } from './quantity-collect.js';
-import { buildCompactEntityIndexAsync } from './compact-entity-index.js';
+import { prepareColumnarEntities } from './columnar-entity-preparation.js';
 import { yieldToEventLoop } from './yield-to-event-loop.js';
 import { StringTable, EntityTableBuilder, PropertyTableBuilder, QuantityTableBuilder, RelationshipGraphBuilder, RelationshipType, } from '@ifc-lite/data';
 import { BufferEntitySource } from './entity-source.js';
 import { batchExtractGlobalIdAndName } from './columnar-parser-attributes.js';
-import { GEOMETRY_TYPES, REL_TYPE_MAP, SPATIAL_TYPES, HIERARCHY_REL_TYPES, PROPERTY_REL_TYPES, ASSOCIATION_REL_TYPES, SKIP_DISPLAY_ATTRS, PROPERTY_ENTITY_TYPES, PROPERTY_CONTAINER_TYPES, isIfcTypeLikeEntity, } from './columnar-parser-indexes.js';
+import { REL_TYPE_MAP, } from './columnar-parser-indexes.js';
 import { extractRelFast, extractPropertyRelFast } from './columnar-parser-relationships.js';
 import { detectSchemaVersion, parseSourceHeader } from './source-header.js';
 import { contiguousSourceBytes } from './source-bytes.js';
+import { getEntityRefFromStore, extractRootAttributesFromEntity, pickLongName } from './columnar-parser-root-attributes.js';
+// Re-exported: part of the package's public on-demand-extraction surface
+// (see packages/parser/src/index.ts).
+export { extractEntityAttributesOnDemand, extractAllEntityAttributes, getRawNamedAttributes, extractRootAttributesFromEntity, } from './columnar-parser-root-attributes.js';
 /** Appends `ref` to `map.get(objId)`, skipping a repeat — these maps win over the deduped graph (#3760/#3782). */
 function addOnDemandRef(map, objId, ref) {
     const list = map.get(objId) ?? map.set(objId, []).get(objId);
     if (!list.includes(ref))
         list.push(ref);
 }
-export class ColumnarParser {
-    /**
-     * Parse IFC file into columnar data store
-     *
-     * Uses fast semicolon-based scanning with on-demand property extraction.
-     * Properties are parsed lazily when accessed, not upfront.
-     * This provides instant UI responsiveness even for very large files.
-     */
-    async parseLite(buffer, entityRefs, options = {}) {
-        const startTime = performance.now();
-        const uint8Buffer = new Uint8Array(buffer);
-        const totalEntities = entityRefs.length;
-        // Phase timing for performance telemetry
-        let phaseStart = startTime;
-        const emitDiagnostic = (message) => {
-            options.onDiagnostic?.(message);
-        };
-        const logPhase = (name) => {
-            const now = performance.now();
-            const elapsed = Math.round(now - phaseStart);
-            console.log(`[parseLite] ${name}: ${elapsed}ms`);
-            emitDiagnostic(`${name}: ${elapsed}ms`);
-            phaseStart = now;
-        };
-        options.onProgress?.({ phase: 'building', percent: 0 });
-        // Capture verbatim HEADER fields so a round-trip export can reproduce
-        // the source FILE_DESCRIPTION items + exact FILE_SCHEMA token instead
-        // of regenerating a fresh ifc-lite header. Cheap: the scan stops at the
-        // header's ENDSEC, so the DATA section is never touched.
-        const sourceHeader = parseSourceHeader(uint8Buffer);
-        // Schema version comes from that header's FILE_SCHEMA declaration, not
-        // from a substring scan of the raw bytes — exporter product names in
-        // FILE_NAME free text carry schema tokens too (issue #3278).
-        const schemaVersion = detectSchemaVersion(uint8Buffer, sourceHeader);
-        // Initialize builders (entity table capacity set after categorization below)
-        const strings = new StringTable();
-        const propertyTableBuilder = new PropertyTableBuilder(strings);
-        const quantityTableBuilder = new QuantityTableBuilder(strings);
-        const relationshipGraphBuilder = new RelationshipGraphBuilder();
-        logPhase('init builders');
-        // Single pass: build byType index AND categorize entities simultaneously.
-        // Uses a type-name cache to avoid calling .toUpperCase() on 4.4M refs
-        // (only ~776 unique type names in IFC4).
-        const byType = new Map();
-        const deferPropertyAtomIndex = options.deferPropertyAtomIndex === true;
-        const typeUpperCache = new Map();
-        const getTypeUpper = (type) => {
-            let upper = typeUpperCache.get(type);
-            if (upper === undefined) {
-                upper = type.toUpperCase();
-                typeUpperCache.set(type, upper);
-            }
-            return upper;
-        };
-        // Non-product helper entities that on-demand extraction / StepExporter
-        // need addressable in `byId`. These are not IfcProduct subtypes so the
-        // schema-driven IFCPRODUCT subtype check below cannot capture them.
-        // Without them, findPreferredGeometricRepresentationContextId() and
-        // findLengthUnitReference() fail because the entities are missing from
-        // the compact entity index.
-        const RELEVANT_NON_PRODUCT_HELPERS = new Set([
-            'IFCGEOMETRICREPRESENTATIONCONTEXT', 'IFCGEOMETRICREPRESENTATIONSUBCONTEXT',
-            'IFCUNITASSIGNMENT', 'IFCSIUNIT', 'IFCCONVERSIONBASEDUNIT',
-            'IFCDERIVEDUNIT', 'IFCDERIVEDUNITELEMENT', 'IFCMEASUREWITHUNIT',
-            'IFCDIMENSIONALEXPONENTS',
-            'IFCMAPCONVERSION', 'IFCPROJECTEDCRS',
-            'IFCMATERIALLAYER', 'IFCMATERIALLAYERSET', 'IFCMATERIALLAYERSETUSAGE',
-            'IFCMATERIALCONSTITUENTSET', 'IFCMATERIALCONSTITUENT',
-            'IFCMATERIALPROFILESET', 'IFCMATERIALPROFILE', 'IFCMATERIAL',
-            'IFCCLASSIFICATION', 'IFCCLASSIFICATIONREFERENCE',
-            'IFCDOCUMENTINFORMATION', 'IFCDOCUMENTREFERENCE',
-        ]);
-        // Schema-driven inclusion: every IfcProduct subtype belongs in the
-        // EntityTable. The previous hardcoded enumeration of IFC4 building-
-        // element leaves (IFCWALL, IFCSLAB, …) and IFC4x3 infrastructure
-        // leaves (IFCREFERENT, IFCSIGNAL, IFCALIGNMENT, IFCPAVEMENT, …) drifted
-        // with every schema bump — new entities silently became CAT_SKIP and
-        // disappeared from the hierarchy panel. The generated schema registry
-        // already knows the full inheritance chain, so use it.
-        const RELEVANT_PRODUCT_ROOTS = new Set(['IFCPRODUCT']);
-        // IfcGroup family (IfcZone, IfcSystem, IfcDistributionSystem,
-        // IfcBuildingSystem, IfcDistributionCircuit, …). These are NOT
-        // IfcProduct subtypes, so without an explicit branch they fall through
-        // to CAT_SKIP and never enter the EntityTable — leaving their Name
-        // unresolvable (`getName` → '') and making them invisible to
-        // `getByType`. The Relationships card then shows "Group #<id>" and the
-        // lens/lists can't surface them. Route them into their own bucket so we
-        // can extract Name/LongName/ObjectType for the group label (#1075).
-        const GROUP_ROOTS = new Set(['IFCGROUP']);
-        // Category constants for the lookup cache
-        const CAT_SKIP = 0, CAT_SPATIAL = 1, CAT_GEOMETRY = 2, CAT_HIERARCHY_REL = 3, CAT_PROPERTY_REL = 4, CAT_PROPERTY_ENTITY = 5, CAT_ASSOCIATION_REL = 6, CAT_TYPE_OBJECT = 7, CAT_RELEVANT = 8, CAT_GROUP = 9;
-        /** Returns true if `upper` (already uppercased) is a subtype of any type in `set`. */
-        function isSubtypeOfAny(upper, set) {
-            const chain = getInheritanceChain(upper);
-            return chain.some(ancestor => set.has(ancestor.toUpperCase()));
-        }
-        // Cache: type name → category (avoids 4.4M .toUpperCase() calls)
-        const typeCategoryCache = new Map();
-        function getCategory(type) {
-            let cat = typeCategoryCache.get(type);
-            if (cat !== undefined)
-                return cat;
-            const upper = getTypeUpper(type);
-            if (SPATIAL_TYPES.has(upper) || isSubtypeOfAny(upper, SPATIAL_TYPES))
-                cat = CAT_SPATIAL;
-            else if (GEOMETRY_TYPES.has(upper) || isSubtypeOfAny(upper, GEOMETRY_TYPES))
-                cat = CAT_GEOMETRY;
-            else if (HIERARCHY_REL_TYPES.has(upper))
-                cat = CAT_HIERARCHY_REL;
-            else if (PROPERTY_REL_TYPES.has(upper))
-                cat = CAT_PROPERTY_REL;
-            else if (PROPERTY_ENTITY_TYPES.has(upper))
-                cat = CAT_PROPERTY_ENTITY;
-            else if (ASSOCIATION_REL_TYPES.has(upper))
-                cat = CAT_ASSOCIATION_REL;
-            else if (isIfcTypeLikeEntity(upper))
-                cat = CAT_TYPE_OBJECT;
-            else if (isSubtypeOfAny(upper, GROUP_ROOTS))
-                cat = CAT_GROUP;
-            else if (RELEVANT_NON_PRODUCT_HELPERS.has(upper)
-                || isSubtypeOfAny(upper, RELEVANT_PRODUCT_ROOTS)
-                || upper.startsWith('IFCREL'))
-                cat = CAT_RELEVANT;
-            else
-                cat = CAT_SKIP;
-            typeCategoryCache.set(type, cat);
-            return cat;
-        }
-        // Time-based yielding: yield to the main thread every ~80ms so geometry
-        // streaming callbacks can fire. This limits main-thread blocking to short
-        // bursts that don't starve geometry, while adding minimal overhead (~15 yields
-        // × ~1ms each ≈ 15ms total over the full parse).
-        const YIELD_INTERVAL_MS = Math.max(16, options.yieldIntervalMs ?? 80);
-        let lastYieldTime = performance.now();
-        const yieldIfNeeded = async () => {
-            const now = performance.now();
-            if (now - lastYieldTime >= YIELD_INTERVAL_MS) {
-                await yieldToEventLoop();
-                lastYieldTime = performance.now();
-            }
-        };
-        emitDiagnostic(`parseLite start: totalEntities=${totalEntities} yieldInterval=${YIELD_INTERVAL_MS}ms`);
-        const spatialRefs = [];
-        const geometryRefs = [];
-        const relationshipRefs = [];
-        const propertyRelRefs = [];
-        const propertyContainerRefs = [];
-        const propertyAtomRefs = [];
-        const associationRelRefs = [];
-        const typeObjectRefs = [];
-        const otherRelevantRefs = [];
-        const groupRefs = [];
-        for (let i = 0; i < entityRefs.length; i++) {
-            if ((i & 0x3FF) === 0)
-                await yieldIfNeeded();
-            const ref = entityRefs[i];
-            // Categorize (cached — .toUpperCase() called once per unique type)
-            const cat = getCategory(ref.type);
-            const typeUpper = cat === CAT_PROPERTY_ENTITY ? getTypeUpper(ref.type) : '';
-            // ALL entities must be indexed in byType for on-demand extraction
-            // (e.g. IfcGeometricRepresentationContext, IfcSiUnit, IfcMaterialLayer).
-            // Only property atoms are optionally deferred for huge-file lazy loading.
-            const includeInPrimaryIndex = !deferPropertyAtomIndex || cat !== CAT_PROPERTY_ENTITY || PROPERTY_CONTAINER_TYPES.has(typeUpper);
-            if (includeInPrimaryIndex) {
-                // STEP convention is uppercase entity type names and every
-                // downstream consumer (schedule-extractor, property readers,
-                // test helpers) keys on uppercase. The tokenizer preserves
-                // original case though, so if a STEP writer ever emits
-                // mixed-case or lowercase types the index would miss on
-                // canonical lookups. Normalise once here — `getTypeUpper`
-                // is already cached by type name so the cost is ~0.
-                const typeKey = getTypeUpper(ref.type);
-                let typeList = byType.get(typeKey);
-                if (!typeList) {
-                    typeList = [];
-                    byType.set(typeKey, typeList);
-                }
-                typeList.push(ref.expressId);
-            }
-            if (cat === CAT_SPATIAL)
-                spatialRefs.push(ref);
-            else if (cat === CAT_GEOMETRY)
-                geometryRefs.push(ref);
-            else if (cat === CAT_HIERARCHY_REL)
-                relationshipRefs.push(ref);
-            else if (cat === CAT_PROPERTY_REL)
-                propertyRelRefs.push(ref);
-            else if (cat === CAT_PROPERTY_ENTITY) {
-                if (PROPERTY_CONTAINER_TYPES.has(typeUpper))
-                    propertyContainerRefs.push(ref);
-                else
-                    propertyAtomRefs.push(ref);
-            }
-            else if (cat === CAT_ASSOCIATION_REL)
-                associationRelRefs.push(ref);
-            else if (cat === CAT_TYPE_OBJECT)
-                typeObjectRefs.push(ref);
-            else if (cat === CAT_GROUP)
-                groupRefs.push(ref);
-            else if (cat === CAT_RELEVANT)
-                otherRelevantRefs.push(ref);
-        }
-        logPhase(`categorize ${totalEntities} → spatial:${spatialRefs.length} geom:${geometryRefs.length} rel:${relationshipRefs.length} propRel:${propertyRelRefs.length} propContainers:${propertyContainerRefs.length} propAtoms:${propertyAtomRefs.length} assocRel:${associationRelRefs.length} type:${typeObjectRefs.length} group:${groupRefs.length} other:${otherRelevantRefs.length}`);
-        // Pre-scan association rels to discover relatingRef target IDs (e.g.
-        // IfcClassificationReference, IfcMaterial, IfcDocumentReference).  These
-        // entities are typically categorised as CAT_SKIP and would otherwise be
-        // missing from the compact index, making on-demand extraction fail.
-        const associationTargetIds = new Set();
-        for (const ref of associationRelRefs) {
-            const result = extractPropertyRelFast(uint8Buffer, ref.byteOffset, ref.byteLength);
-            if (result)
-                for (const id of result.relatingDefs)
-                    associationTargetIds.add(id);
-        }
-        // Collect EntityRefs for association targets that aren't already categorised.
-        // Single O(n) pass over entityRefs filtered to the (small) target ID set.
-        const alreadyIndexedIds = new Set();
-        for (const arr of [spatialRefs, geometryRefs, relationshipRefs, propertyRelRefs,
-            propertyContainerRefs, associationRelRefs, typeObjectRefs, groupRefs, otherRelevantRefs,
-            ...(deferPropertyAtomIndex ? [] : [propertyAtomRefs])]) {
-            for (const r of arr)
-                alreadyIndexedIds.add(r.expressId);
-        }
-        const extraAssocRefs = [];
-        for (const ref of entityRefs) {
-            if (associationTargetIds.has(ref.expressId) && !alreadyIndexedIds.has(ref.expressId)) {
-                extraAssocRefs.push(ref);
-            }
-        }
-        logPhase(`association target pre-scan: ${associationTargetIds.size} targets, ${extraAssocRefs.length} extra refs`);
-        // ALL entity refs must be indexed in byId so that on-demand extraction
-        // can look up any entity by expressId (e.g. IfcUnitAssignment,
-        // IfcGeometricRepresentationContext, IfcSiUnit, IfcLocalPlacement, etc.).
-        // Only property atoms are optionally deferred for huge-file lazy loading.
-        const indexedRefs = deferPropertyAtomIndex
-            ? entityRefs.filter(ref => {
-                const cat = getCategory(ref.type);
-                return cat !== CAT_PROPERTY_ENTITY || PROPERTY_CONTAINER_TYPES.has(getTypeUpper(ref.type));
-            })
-            : entityRefs;
-        emitDiagnostic(`index input: indexedRefs=${indexedRefs.length} deferredPropertyAtoms=${deferPropertyAtomIndex ? propertyAtomRefs.length : 0} extraAssocTargets=${extraAssocRefs.length}`);
-        // Build compact entity index from only the refs that survive lite parsing.
-        // This avoids spending huge-file startup time indexing millions of skipped
-        // representation/helper entities that the viewer never queries.
-        const compactByIdIndex = await buildCompactEntityIndexAsync(indexedRefs);
-        logPhase('compact entity index');
-        // Create entity table builder with EXACT capacity (not totalEntities which
-        // includes millions of geometry-representation entities we don't store).
-        // For a 14M entity file, this reduces allocation from ~546MB to ~20MB.
-        const relevantCount = spatialRefs.length + geometryRefs.length + typeObjectRefs.length
-            + relationshipRefs.length + groupRefs.length + otherRelevantRefs.length;
-        const entityTableBuilder = new EntityTableBuilder(relevantCount, strings);
-        const entityIndex = {
-            byId: compactByIdIndex,
-            byType,
-        };
-        // === TARGETED PARSING using batch byte-level extraction ===
-        // Uses 2 TextDecoder.decode() calls total for ALL entity GlobalIds/Names
-        // (instead of per-entity calls), and pure byte scanning for relationships.
-        options.onProgress?.({ phase: 'parsing entities', percent: 10 });
-        const extractor = new EntityExtractor(uint8Buffer);
-        // Spatial entities: small count, use extractEntity for full accuracy
-        const parsedEntityData = new Map();
-        for (const ref of spatialRefs) {
-            const entity = extractor.extractEntity(ref);
-            if (entity) {
-                const attrs = entity.attributes || [];
-                parsedEntityData.set(ref.expressId, {
-                    globalId: typeof attrs[0] === 'string' ? attrs[0] : '',
-                    name: typeof attrs[2] === 'string' ? attrs[2] : '',
-                });
-            }
+/** Internal implementation shared by public refs and parser-worker columns. */
+export async function parseColumnarInput(buffer, input, options = {}) {
+    const startTime = performance.now();
+    const uint8Buffer = new Uint8Array(buffer);
+    const totalEntities = Array.isArray(input) ? input.length : input.expressIds.length;
+    // Phase timing for performance telemetry
+    let phaseStart = startTime;
+    const emitDiagnostic = (message) => {
+        options.onDiagnostic?.(message);
+    };
+    const logPhase = (name) => {
+        const now = performance.now();
+        const elapsed = Math.round(now - phaseStart);
+        console.log(`[parseLite] ${name}: ${elapsed}ms`);
+        emitDiagnostic(`${name}: ${elapsed}ms`);
+        phaseStart = now;
+    };
+    options.onProgress?.({ phase: 'building', percent: 0 });
+    // Capture verbatim HEADER fields so a round-trip export can reproduce
+    // the source FILE_DESCRIPTION items + exact FILE_SCHEMA token instead
+    // of regenerating a fresh ifc-lite header. Cheap: the scan stops at the
+    // header's ENDSEC, so the DATA section is never touched.
+    const sourceHeader = parseSourceHeader(uint8Buffer);
+    // Schema version comes from that header's FILE_SCHEMA declaration, not
+    // from a substring scan of the raw bytes — exporter product names in
+    // FILE_NAME free text carry schema tokens too (issue #3278).
+    const schemaVersion = detectSchemaVersion(uint8Buffer, sourceHeader);
+    // Initialize builders (entity table capacity set after categorization below)
+    const strings = new StringTable();
+    const propertyTableBuilder = new PropertyTableBuilder(strings);
+    const quantityTableBuilder = new QuantityTableBuilder(strings);
+    const relationshipGraphBuilder = new RelationshipGraphBuilder();
+    logPhase('init builders');
+    // Time-based yielding: yield to the main thread every ~80ms so geometry
+    // streaming callbacks can fire. This limits main-thread blocking to short
+    // bursts that don't starve geometry, while adding minimal overhead (~15 yields
+    // × ~1ms each ≈ 15ms total over the full parse).
+    const YIELD_INTERVAL_MS = Math.max(16, options.yieldIntervalMs ?? 80);
+    let lastYieldTime = performance.now();
+    const yieldIfNeeded = async () => {
+        const now = performance.now();
+        if (now - lastYieldTime >= YIELD_INTERVAL_MS) {
+            await yieldToEventLoop();
+            lastYieldTime = performance.now();
         }
-        logPhase('spatial entities');
-        await yieldIfNeeded();
-        // Group family (IfcZone / IfcSystem / IfcDistributionSystem / …): small
-        // count, full extract so we can resolve Name + LongName + ObjectType for
-        // the group label. Name is often empty on these — the human label lives
-        // in LongName (IfcZone/IfcDistributionSystem) — so fall back to it, and
-        // keep ObjectType (e.g. a system designation) for richer display. (#1075)
-        const groupExtra = new Map();
-        for (const ref of groupRefs) {
-            const entity = extractor.extractEntity(ref);
-            if (!entity)
-                continue;
-            const root = extractRootAttributesFromEntity(entity);
-            const longName = pickLongName(entity);
+    };
+    emitDiagnostic(`parseLite start: totalEntities=${totalEntities} yieldInterval=${YIELD_INTERVAL_MS}ms`);
+    const prepared = await prepareColumnarEntities(input, options.deferPropertyAtomIndex === true, yieldIfNeeded);
+    const { byType, getTypeUpper, spatialRefs, geometryRefs, relationshipRefs, propertyRelRefs, propertyContainerRefs, associationRelRefs, typeObjectRefs, otherRelevantRefs, groupRefs } = prepared;
+    logPhase(`categorize ${totalEntities} → spatial:${spatialRefs.length} geom:${geometryRefs.length} rel:${relationshipRefs.length} propRel:${propertyRelRefs.length} propContainers:${propertyContainerRefs.length} propAtoms:${prepared.propertyAtomCount} assocRel:${associationRelRefs.length} type:${typeObjectRefs.length} group:${groupRefs.length} other:${otherRelevantRefs.length}`);
+    emitDiagnostic(`index input: indexedRefs=${prepared.indexedCount} deferredPropertyAtoms=${options.deferPropertyAtomIndex ? prepared.propertyAtomCount : 0}`);
+    const compactByIdIndex = await prepared.buildPrimaryIndex();
+    logPhase('compact entity index');
+    // Create entity table builder with EXACT capacity (not totalEntities which
+    // includes millions of geometry-representation entities we don't store).
+    // For a 14M entity file, this reduces allocation from ~546MB to ~20MB.
+    const relevantCount = spatialRefs.length + geometryRefs.length + typeObjectRefs.length
+        + relationshipRefs.length + groupRefs.length + otherRelevantRefs.length;
+    const entityTableBuilder = new EntityTableBuilder(relevantCount, strings);
+    const entityIndex = {
+        byId: compactByIdIndex,
+        byType,
+    };
+    // === TARGETED PARSING using batch byte-level extraction ===
+    // Uses 2 TextDecoder.decode() calls total for ALL entity GlobalIds/Names
+    // (instead of per-entity calls), and pure byte scanning for relationships.
+    options.onProgress?.({ phase: 'parsing entities', percent: 10 });
+    const extractor = new EntityExtractor(uint8Buffer);
+    // Spatial entities: small count, use extractEntity for full accuracy
+    const parsedEntityData = new Map();
+    for (const ref of spatialRefs) {
+        const entity = extractor.extractEntity(ref);
+        if (entity) {
+            const attrs = entity.attributes || [];
             parsedEntityData.set(ref.expressId, {
-                globalId: root.globalId,
-                name: root.name || longName,
+                globalId: typeof attrs[0] === 'string' ? attrs[0] : '',
+                name: typeof attrs[2] === 'string' ? attrs[2] : '',
             });
-            groupExtra.set(ref.expressId, { description: root.description, objectType: root.objectType });
         }
-        logPhase('group entities');
-        await yieldIfNeeded();
-        // Geometry + type object entities: batch extract GlobalId+Name with 2 TextDecoder calls
-        options.onProgress?.({ phase: 'parsing geometry names', percent: 12 });
-        const geomData = await batchExtractGlobalIdAndName(uint8Buffer, geometryRefs, yieldIfNeeded);
-        for (const [id, data] of geomData)
-            parsedEntityData.set(id, data);
-        await yieldIfNeeded();
-        const typeData = await batchExtractGlobalIdAndName(uint8Buffer, typeObjectRefs, yieldIfNeeded);
-        for (const [id, data] of typeData)
-            parsedEntityData.set(id, data);
-        logPhase('batch geom GlobalId+Name');
-        await yieldIfNeeded();
-        // Other relevant products (IfcSpatialZone, IfcVirtualElement, IfcGrid,
-        // IfcAnnotation, …): batch GlobalId+Name so they show their Name in the
-        // hierarchy / lists instead of nothing. Previously these were added to
-        // the EntityTable with an empty name (parsedEntityData never covered
-        // this bucket), so e.g. IfcSpatialZone listed without a label. (#1075)
-        const otherData = await batchExtractGlobalIdAndName(uint8Buffer, otherRelevantRefs, yieldIfNeeded);
-        for (const [id, data] of otherData)
-            parsedEntityData.set(id, data);
-        logPhase('batch other-relevant GlobalId+Name');
-        await yieldIfNeeded();
-        // Relationships: byte-level scanning (numbers only, no TextDecoder)
-        options.onProgress?.({ phase: 'parsing relationships', percent: 20 });
-        for (let i = 0; i < relationshipRefs.length; i++) {
-            if ((i & 0x3FF) === 0)
-                await yieldIfNeeded();
-            const ref = relationshipRefs[i];
-            const typeUpper = getTypeUpper(ref.type);
-            const rel = extractRelFast(uint8Buffer, ref.byteOffset, ref.byteLength, typeUpper);
-            if (rel) {
-                const relType = REL_TYPE_MAP[typeUpper];
-                if (relType) {
-                    for (const targetId of rel.relatedObjects) {
-                        relationshipGraphBuilder.addEdge(rel.relatingObject, targetId, relType, ref.expressId);
-                    }
+    }
+    logPhase('spatial entities');
+    await yieldIfNeeded();
+    // Group family (IfcZone / IfcSystem / IfcDistributionSystem / …): small
+    // count, full extract so we can resolve Name + LongName + ObjectType for
+    // the group label. Name is often empty on these — the human label lives
+    // in LongName (IfcZone/IfcDistributionSystem) — so fall back to it, and
+    // keep ObjectType (e.g. a system designation) for richer display. (#1075)
+    const groupExtra = new Map();
+    for (const ref of groupRefs) {
+        const entity = extractor.extractEntity(ref);
+        if (!entity)
+            continue;
+        const root = extractRootAttributesFromEntity(entity);
+        const longName = pickLongName(entity);
+        parsedEntityData.set(ref.expressId, {
+            globalId: root.globalId,
+            name: root.name || longName,
+        });
+        groupExtra.set(ref.expressId, { description: root.description, objectType: root.objectType });
+    }
+    logPhase('group entities');
+    await yieldIfNeeded();
+    // Geometry + type object entities: batch extract GlobalId+Name with 2 TextDecoder calls
+    options.onProgress?.({ phase: 'parsing geometry names', percent: 12 });
+    const geomData = await batchExtractGlobalIdAndName(uint8Buffer, geometryRefs, yieldIfNeeded);
+    for (const [id, data] of geomData)
+        parsedEntityData.set(id, data);
+    await yieldIfNeeded();
+    const typeData = await batchExtractGlobalIdAndName(uint8Buffer, typeObjectRefs, yieldIfNeeded);
+    for (const [id, data] of typeData)
+        parsedEntityData.set(id, data);
+    logPhase('batch geom GlobalId+Name');
+    await yieldIfNeeded();
+    // Other relevant products (IfcSpatialZone, IfcVirtualElement, IfcGrid,
+    // IfcAnnotation, …): batch GlobalId+Name so they show their Name in the
+    // hierarchy / lists instead of nothing. Previously these were added to
+    // the EntityTable with an empty name (parsedEntityData never covered
+    // this bucket), so e.g. IfcSpatialZone listed without a label. (#1075)
+    const otherData = await batchExtractGlobalIdAndName(uint8Buffer, otherRelevantRefs, yieldIfNeeded);
+    for (const [id, data] of otherData)
+        parsedEntityData.set(id, data);
+    logPhase('batch other-relevant GlobalId+Name');
+    await yieldIfNeeded();
+    // Relationships: byte-level scanning (numbers only, no TextDecoder)
+    options.onProgress?.({ phase: 'parsing relationships', percent: 20 });
+    for (let i = 0; i < relationshipRefs.length; i++) {
+        if ((i & 0x3FF) === 0)
+            await yieldIfNeeded();
+        const ref = relationshipRefs[i];
+        const typeUpper = getTypeUpper(ref.type);
+        const rel = extractRelFast(uint8Buffer, ref.byteOffset, ref.byteLength, typeUpper);
+        if (rel) {
+            const relType = REL_TYPE_MAP[typeUpper];
+            if (relType) {
+                for (const targetId of rel.relatedObjects) {
+                    relationshipGraphBuilder.addEdge(rel.relatingObject, targetId, relType, ref.expressId);
                 }
             }
         }
-        logPhase('byte-level relationships');
-        // === BUILD ENTITY TABLE from categorized arrays ===
-        // Instead of iterating ALL 4.4M entityRefs, iterate only categorized arrays
-        // (~100K-200K total). This eliminates a 200-300ms loop over 4.4M items.
-        options.onProgress?.({ phase: 'building entities', percent: 30 });
-        // Helper to add entities with pre-parsed data
-        const addEntityBatch = (refs, hasGeometry, isType) => {
-            for (const ref of refs) {
-                const entityData = parsedEntityData.get(ref.expressId);
-                entityTableBuilder.add(ref.expressId, ref.type, entityData?.globalId || '', entityData?.name || '', '', // description
-                '', // objectType
-                hasGeometry, isType);
-            }
-        };
-        addEntityBatch(spatialRefs, false, false);
-        addEntityBatch(geometryRefs, true, false);
-        addEntityBatch(typeObjectRefs, false, true);
-        addEntityBatch(relationshipRefs, false, false);
-        addEntityBatch(otherRelevantRefs, false, false);
-        // Groups carry Description + ObjectType (the system designation), which
-        // the shared addEntityBatch drops as ''. Add them with the extra fields
-        // so the Properties title / lists / lens can surface them. (#1075)
-        for (const ref of groupRefs) {
+    }
+    logPhase('byte-level relationships');
+    // === BUILD ENTITY TABLE from categorized arrays ===
+    // Instead of iterating ALL 4.4M entityRefs, iterate only categorized arrays
+    // (~100K-200K total). This eliminates a 200-300ms loop over 4.4M items.
+    options.onProgress?.({ phase: 'building entities', percent: 30 });
+    // Helper to add entities with pre-parsed data
+    const addEntityBatch = (refs, hasGeometry, isType) => {
+        for (const ref of refs) {
             const entityData = parsedEntityData.get(ref.expressId);
-            const extra = groupExtra.get(ref.expressId);
-            entityTableBuilder.add(ref.expressId, ref.type, entityData?.globalId || '', entityData?.name || '', extra?.description || '', extra?.objectType || '', false, false);
-        }
-        logPhase('add entity batches');
-        const entityTable = entityTableBuilder.build();
-        logPhase('entity table build()');
-        // Empty property/quantity tables - use on-demand extraction instead
-        const propertyTable = propertyTableBuilder.build();
-        const quantityTable = quantityTableBuilder.build();
-        // Build intermediate relationship graph (spatial/hierarchy edges only).
-        // Property/association edges are added later; final graph is rebuilt at the end.
-        const hierarchyRelGraph = relationshipGraphBuilder.build();
-        logPhase('hierarchy rel graph build()');
-        await yieldIfNeeded();
-        // === EXTRACT LENGTH UNIT SCALE ===
-        options.onProgress?.({ phase: 'extracting units', percent: 85 });
-        const lengthUnitScale = extractLengthUnitScale(uint8Buffer, entityIndex);
-        // === BUILD SPATIAL HIERARCHY ===
-        options.onProgress?.({ phase: 'building hierarchy', percent: 90 });
-        let spatialHierarchy;
-        try {
-            const hierarchyBuilder = new SpatialHierarchyBuilder();
-            spatialHierarchy = hierarchyBuilder.build(entityTable, hierarchyRelGraph, strings, uint8Buffer, entityIndex, lengthUnitScale);
-            logPhase('spatial hierarchy');
+            entityTableBuilder.add(ref.expressId, ref.type, entityData?.globalId || '', entityData?.name || '', '', // description
+            '', // objectType
+            hasGeometry, isType);
         }
-        catch (error) {
-            console.warn('[ColumnarParser] Failed to build spatial hierarchy:', error);
-        }
-        // === EMIT SPATIAL HIERARCHY EARLY ===
-        // The hierarchy panel can render immediately while property/association
-        // parsing continues. This lets the panel appear at the same time as
-        // geometry streaming completes.
-        // ONE accessor, shared by the store field and the entity source.
-        //
-        // These used to be built separately -- `BufferEntitySource` got the raw
-        // `uint8Buffer` and wrapped it in its own accessor, while `source` got
-        // another. That is invisible while both are resident views over the
-        // same bytes, and fatal once the source can switch to compressed
-        // storage in place (#2183): `getEntity` would keep reading its private
-        // resident accessor, so the original buffer would never be released and
-        // the store would serve entities from one representation and properties
-        // from the other. Measured: with two accessors the buffer survives GC
-        // after the swap; with one it is collected.
-        const source = contiguousSourceBytes(uint8Buffer);
-        const entitySource = new BufferEntitySource(source, entityIndex);
-        const earlyStore = {
-            fileSize: buffer.byteLength,
-            schemaVersion,
-            sourceHeader,
-            entityCount: totalEntities,
-            parseTime: performance.now() - startTime,
-            source,
-            entityIndex,
-            strings,
-            entities: entityTable,
-            properties: propertyTable,
-            quantities: quantityTable,
-            relationships: hierarchyRelGraph,
-            spatialHierarchy,
-            lengthUnitScale,
-            getEntity(expressId) { return entitySource.getEntity(expressId); },
-            getEntitiesByType(typeName) { return entitySource.getEntitiesByType(typeName); },
-            getProperties(expressId) { return this.properties.getForEntity(expressId); },
-            getQuantities(expressId) { return this.quantities.getForEntity(expressId); },
-        };
-        options.onSpatialReady?.(earlyStore);
-        await yieldIfNeeded(); // Let geometry process after hierarchy emission
-        // === DEFERRED: Parse property and association relationships ===
-        // These are NOT needed for the spatial hierarchy panel.
-        options.onProgress?.({ phase: 'parsing property refs', percent: 92 });
-        const onDemandPropertyMap = new Map();
-        const onDemandQuantityMap = new Map();
-        // Pre-build Sets of property set / quantity set IDs from already-categorized refs.
-        // This replaces 252K binary searches on the 14M compact entity index with O(1) Set lookups.
-        const propertySetIds = new Set();
-        const quantitySetIds = new Set();
-        for (const ref of propertyContainerRefs) {
-            const tu = getTypeUpper(ref.type);
-            if (tu === 'IFCPROPERTYSET')
-                propertySetIds.add(ref.expressId);
-            else if (tu === 'IFCELEMENTQUANTITY')
-                quantitySetIds.add(ref.expressId);
-        }
-        // Property rels: byte-level scanning + addEdge (now fast with SoA builder).
-        let totalPropRelObjects = 0;
-        for (let pi = 0; pi < propertyRelRefs.length; pi++) {
-            if ((pi & 0x3FF) === 0)
-                await yieldIfNeeded();
-            const ref = propertyRelRefs[pi];
-            const result = extractPropertyRelFast(uint8Buffer, ref.byteOffset, ref.byteLength);
-            if (result) {
-                const { relatedObjects, relatingDefs } = result;
-                totalPropRelObjects += relatedObjects.length;
-                // RelatingPropertyDefinition is IfcPropertySetDefinitionSelect, whose
-                // second alternative (IfcPropertySetDefinitionSet) is a SET of pset/
-                // qset definitions — `relatingDefs` has more than one entry for that
-                // case. Every entry in the group applies to every related object.
-                for (const relatingDef of relatingDefs) {
+    };
+    addEntityBatch(spatialRefs, false, false);
+    addEntityBatch(geometryRefs, true, false);
+    addEntityBatch(typeObjectRefs, false, true);
+    addEntityBatch(relationshipRefs, false, false);
+    addEntityBatch(otherRelevantRefs, false, false);
+    // Groups carry Description + ObjectType (the system designation), which
+    // the shared addEntityBatch drops as ''. Add them with the extra fields
+    // so the Properties title / lists / lens can surface them. (#1075)
+    for (const ref of groupRefs) {
+        const entityData = parsedEntityData.get(ref.expressId);
+        const extra = groupExtra.get(ref.expressId);
+        entityTableBuilder.add(ref.expressId, ref.type, entityData?.globalId || '', entityData?.name || '', extra?.description || '', extra?.objectType || '', false, false);
+    }
+    logPhase('add entity batches');
+    const entityTable = entityTableBuilder.build();
+    logPhase('entity table build()');
+    // Empty property/quantity tables - use on-demand extraction instead
+    const propertyTable = propertyTableBuilder.build();
+    const quantityTable = quantityTableBuilder.build();
+    // Build intermediate relationship graph (spatial/hierarchy edges only).
+    // Property/association edges are added later; final graph is rebuilt at the end.
+    const hierarchyRelGraph = relationshipGraphBuilder.build();
+    logPhase('hierarchy rel graph build()');
+    await yieldIfNeeded();
+    // === EXTRACT LENGTH UNIT SCALE ===
+    options.onProgress?.({ phase: 'extracting units', percent: 85 });
+    const lengthUnitScale = extractLengthUnitScale(uint8Buffer, entityIndex);
+    // === BUILD SPATIAL HIERARCHY ===
+    options.onProgress?.({ phase: 'building hierarchy', percent: 90 });
+    let spatialHierarchy;
+    try {
+        const hierarchyBuilder = new SpatialHierarchyBuilder();
+        spatialHierarchy = hierarchyBuilder.build(entityTable, hierarchyRelGraph, strings, uint8Buffer, entityIndex, lengthUnitScale);
+        logPhase('spatial hierarchy');
+    }
+    catch (error) {
+        console.warn('[ColumnarParser] Failed to build spatial hierarchy:', error);
+    }
+    // === EMIT SPATIAL HIERARCHY EARLY ===
+    // The hierarchy panel can render immediately while property/association
+    // parsing continues. This lets the panel appear at the same time as
+    // geometry streaming completes.
+    // ONE accessor, shared by the store field and the entity source.
+    //
+    // These used to be built separately -- `BufferEntitySource` got the raw
+    // `uint8Buffer` and wrapped it in its own accessor, while `source` got
+    // another. That is invisible while both are resident views over the
+    // same bytes, and fatal once the source can switch to compressed
+    // storage in place (#2183): `getEntity` would keep reading its private
+    // resident accessor, so the original buffer would never be released and
+    // the store would serve entities from one representation and properties
+    // from the other. Measured: with two accessors the buffer survives GC
+    // after the swap; with one it is collected.
+    const source = contiguousSourceBytes(uint8Buffer);
+    const entitySource = new BufferEntitySource(source, entityIndex);
+    const earlyStore = {
+        fileSize: buffer.byteLength,
+        schemaVersion,
+        sourceHeader,
+        entityCount: totalEntities,
+        parseTime: performance.now() - startTime,
+        source,
+        entityIndex,
+        strings,
+        entities: entityTable,
+        properties: propertyTable,
+        quantities: quantityTable,
+        relationships: hierarchyRelGraph,
+        spatialHierarchy,
+        lengthUnitScale,
+        getEntity(expressId) { return entitySource.getEntity(expressId); },
+        getEntitiesByType(typeName) { return entitySource.getEntitiesByType(typeName); },
+        getProperties(expressId) { return this.properties.getForEntity(expressId); },
+        getQuantities(expressId) { return this.quantities.getForEntity(expressId); },
+    };
+    options.onSpatialReady?.(earlyStore);
+    await yieldIfNeeded(); // Let geometry process after hierarchy emission
+    // === DEFERRED: Parse property and association relationships ===
+    // These are NOT needed for the spatial hierarchy panel.
+    options.onProgress?.({ phase: 'parsing property refs', percent: 92 });
+    const onDemandPropertyMap = new Map();
+    const onDemandQuantityMap = new Map();
+    // Pre-build Sets of property set / quantity set IDs from already-categorized refs.
+    // This replaces 252K binary searches on the 14M compact entity index with O(1) Set lookups.
+    const propertySetIds = new Set();
+    const quantitySetIds = new Set();
+    for (const ref of propertyContainerRefs) {
+        const tu = getTypeUpper(ref.type);
+        if (tu === 'IFCPROPERTYSET')
+            propertySetIds.add(ref.expressId);
+        else if (tu === 'IFCELEMENTQUANTITY')
+            quantitySetIds.add(ref.expressId);
+    }
+    // Property rels: byte-level scanning + addEdge (now fast with SoA builder).
+    let totalPropRelObjects = 0;
+    for (let pi = 0; pi < propertyRelRefs.length; pi++) {
+        if ((pi & 0x3FF) === 0)
+            await yieldIfNeeded();
+        const ref = propertyRelRefs[pi];
+        const result = extractPropertyRelFast(uint8Buffer, ref.byteOffset, ref.byteLength);
+        if (result) {
+            const { relatedObjects, relatingDefs } = result;
+            totalPropRelObjects += relatedObjects.length;
+            // RelatingPropertyDefinition is IfcPropertySetDefinitionSelect, whose
+            // second alternative (IfcPropertySetDefinitionSet) is a SET of pset/
+            // qset definitions — `relatingDefs` has more than one entry for that
+            // case. Every entry in the group applies to every related object.
+            for (const relatingDef of relatingDefs) {
+                for (const objId of relatedObjects) {
+                    relationshipGraphBuilder.addEdge(relatingDef, objId, RelationshipType.DefinesByProperties, ref.expressId);
+                }
+                // Build on-demand property/quantity maps using pre-built Sets (O(1) vs binary search)
+                const isPropSet = propertySetIds.has(relatingDef);
+                const isQtySet = !isPropSet && quantitySetIds.has(relatingDef);
+                if (isPropSet || isQtySet) {
+                    const targetMap = isPropSet ? onDemandPropertyMap : onDemandQuantityMap;
                     for (const objId of relatedObjects) {
-                        relationshipGraphBuilder.addEdge(relatingDef, objId, RelationshipType.DefinesByProperties, ref.expressId);
-                    }
-                    // Build on-demand property/quantity maps using pre-built Sets (O(1) vs binary search)
-                    const isPropSet = propertySetIds.has(relatingDef);
-                    const isQtySet = !isPropSet && quantitySetIds.has(relatingDef);
-                    if (isPropSet || isQtySet) {
-                        const targetMap = isPropSet ? onDemandPropertyMap : onDemandQuantityMap;
-                        for (const objId of relatedObjects) {
-                            addOnDemandRef(targetMap, objId, relatingDef);
-                        }
+                        addOnDemandRef(targetMap, objId, relatingDef);
                     }
                 }
             }
         }
-        await yieldIfNeeded();
-        // Association rels: byte-level scanning, no addEdge (same reasoning as property rels)
-        options.onProgress?.({ phase: 'parsing associations', percent: 95 });
-        const onDemandClassificationMap = new Map();
-        const onDemandMaterialMap = new Map();
-        const onDemandDocumentMap = new Map();
-        // Determinism rule for elements with MULTIPLE IfcRelAssociatesMaterial:
-        // the map keeps EVERY association (list-valued), ordered by rel express
-        // id, so list[0] — the "primary" — is the RelatingMaterial of the
-        // LOWEST rel express id regardless of file order. The cache rebuild
-        // (viewer spatialHierarchy rebuildOnDemandMaps) applies the same rule
-        // via edge relationshipIds, so fresh-parse and cache-load agree.
-        const materialRelIds = new Map();
-        for (let i = 0; i < associationRelRefs.length; i++) {
-            if ((i & 0x3FF) === 0)
-                await yieldIfNeeded();
-            const ref = associationRelRefs[i];
-            const result = extractPropertyRelFast(uint8Buffer, ref.byteOffset, ref.byteLength);
-            if (result) {
-                // IfcRelAssociates{Material,Classification,Document}'s Relating*
-                // selects are always single refs (unlike RelatingPropertyDefinition
-                // above, none of IfcMaterialSelect/IfcClassificationSelect/
-                // IfcDocumentSelect admit a SET alternative) — take the one entry.
-                const { relatedObjects, relatingDefs } = result;
-                const relatingRef = relatingDefs[0];
-                const typeUpper = getTypeUpper(ref.type);
-                if (typeUpper === 'IFCRELASSOCIATESCLASSIFICATION') {
-                    for (const objId of relatedObjects) {
-                        addOnDemandRef(onDemandClassificationMap, objId, relatingRef);
-                        relationshipGraphBuilder.addEdge(relatingRef, objId, RelationshipType.AssociatesClassification, ref.expressId);
-                    }
+    }
+    await yieldIfNeeded();
+    // Association rels: byte-level scanning, no addEdge (same reasoning as property rels)
+    options.onProgress?.({ phase: 'parsing associations', percent: 95 });
+    const onDemandClassificationMap = new Map();
+    const onDemandMaterialMap = new Map();
+    const onDemandDocumentMap = new Map();
+    // Determinism rule for elements with MULTIPLE IfcRelAssociatesMaterial:
+    // the map keeps EVERY association (list-valued), ordered by rel express
+    // id, so list[0] — the "primary" — is the RelatingMaterial of the
+    // LOWEST rel express id regardless of file order. The cache rebuild
+    // (viewer spatialHierarchy rebuildOnDemandMaps) applies the same rule
+    // via edge relationshipIds, so fresh-parse and cache-load agree.
+    const materialRelIds = new Map();
+    for (let i = 0; i < associationRelRefs.length; i++) {
+        if ((i & 0x3FF) === 0)
+            await yieldIfNeeded();
+        const ref = associationRelRefs[i];
+        const result = extractPropertyRelFast(uint8Buffer, ref.byteOffset, ref.byteLength);
+        if (result) {
+            // IfcRelAssociates{Material,Classification,Document}'s Relating*
+            // selects are always single refs (unlike RelatingPropertyDefinition
+            // above, none of IfcMaterialSelect/IfcClassificationSelect/
+            // IfcDocumentSelect admit a SET alternative) — take the one entry.
+            const { relatedObjects, relatingDefs } = result;
+            const relatingRef = relatingDefs[0];
+            const typeUpper = getTypeUpper(ref.type);
+            if (typeUpper === 'IFCRELASSOCIATESCLASSIFICATION') {
+                for (const objId of relatedObjects) {
+                    addOnDemandRef(onDemandClassificationMap, objId, relatingRef);
+                    relationshipGraphBuilder.addEdge(relatingRef, objId, RelationshipType.AssociatesClassification, ref.expressId);
                 }
-                else if (typeUpper === 'IFCRELASSOCIATESMATERIAL') {
-                    for (const objId of relatedObjects) {
-                        let list = onDemandMaterialMap.get(objId);
-                        let relIds = materialRelIds.get(objId);
-                        if (!list || !relIds) {
-                            list = [];
-                            onDemandMaterialMap.set(objId, list);
-                            relIds = [];
-                            materialRelIds.set(objId, relIds);
-                        }
-                        // Deliberately NOT deduped (#3782 review): buildMaterialUsageIndex
-                        // already dedupes downstream (seenPerMaterial), per material-fraction-and-associations.test.ts.
-                        let at = relIds.length;
-                        while (at > 0 && relIds[at - 1] > ref.expressId)
-                            at--;
-                        relIds.splice(at, 0, ref.expressId);
-                        list.splice(at, 0, relatingRef);
-                        relationshipGraphBuilder.addEdge(relatingRef, objId, RelationshipType.AssociatesMaterial, ref.expressId);
+            }
+            else if (typeUpper === 'IFCRELASSOCIATESMATERIAL') {
+                for (const objId of relatedObjects) {
+                    let list = onDemandMaterialMap.get(objId);
+                    let relIds = materialRelIds.get(objId);
+                    if (!list || !relIds) {
+                        list = [];
+                        onDemandMaterialMap.set(objId, list);
+                        relIds = [];
+                        materialRelIds.set(objId, relIds);
                     }
+                    // Deliberately NOT deduped (#3782 review): buildMaterialUsageIndex
+                    // already dedupes downstream (seenPerMaterial), per material-fraction-and-associations.test.ts.
+                    let at = relIds.length;
+                    while (at > 0 && relIds[at - 1] > ref.expressId)
+                        at--;
+                    relIds.splice(at, 0, ref.expressId);
+                    list.splice(at, 0, relatingRef);
+                    relationshipGraphBuilder.addEdge(relatingRef, objId, RelationshipType.AssociatesMaterial, ref.expressId);
                 }
-                else if (typeUpper === 'IFCRELASSOCIATESDOCUMENT') {
-                    for (const objId of relatedObjects) {
-                        addOnDemandRef(onDemandDocumentMap, objId, relatingRef);
-                        relationshipGraphBuilder.addEdge(relatingRef, objId, RelationshipType.AssociatesDocument, ref.expressId);
-                    }
+            }
+            else if (typeUpper === 'IFCRELASSOCIATESDOCUMENT') {
+                for (const objId of relatedObjects) {
+                    addOnDemandRef(onDemandDocumentMap, objId, relatingRef);
+                    relationshipGraphBuilder.addEdge(relatingRef, objId, RelationshipType.AssociatesDocument, ref.expressId);
                 }
             }
         }
-        logPhase('property+association rels');
-        // Rebuild relationship graph with ALL edges (hierarchy + property + association)
-        const fullRelationshipGraph = relationshipGraphBuilder.build();
-        logPhase('relationship graph build()');
-        let deferredEntityIndex;
-        if (deferPropertyAtomIndex && propertyAtomRefs.length > 0) {
-            options.onProgress?.({ phase: 'indexing property atoms', percent: 98 });
-            deferredEntityIndex = await buildCompactEntityIndexAsync(propertyAtomRefs, undefined, 1024, 2);
-            logPhase('deferred property atom index');
-        }
-        const parseTime = performance.now() - startTime;
-        options.onProgress?.({ phase: 'complete', percent: 100 });
-        const finalStore = {
-            ...earlyStore,
-            parseTime,
-            relationships: fullRelationshipGraph,
-            deferredEntityIndex,
-            onDemandPropertyMap,
-            onDemandQuantityMap,
-            onDemandClassificationMap,
-            onDemandMaterialMap,
-            onDemandDocumentMap,
-            lengthUnitScale,
-            getEntity(expressId) { return entitySource.getEntity(expressId); },
-            getEntitiesByType(typeName) { return entitySource.getEntitiesByType(typeName); },
-            getProperties(expressId) {
-                if (onDemandPropertyMap.size > 0)
-                    return extractPropertiesOnDemand(this, expressId);
-                return this.properties.getForEntity(expressId);
-            },
-            getQuantities(expressId) {
-                if (onDemandQuantityMap.size > 0)
-                    return extractQuantitiesOnDemand(this, expressId);
-                return this.quantities.getForEntity(expressId);
-            },
-        };
-        return finalStore;
+    }
+    logPhase('property+association rels');
+    // Rebuild relationship graph with ALL edges (hierarchy + property + association)
+    const fullRelationshipGraph = relationshipGraphBuilder.build();
+    logPhase('relationship graph build()');
+    let deferredEntityIndex;
+    if (options.deferPropertyAtomIndex && prepared.propertyAtomCount > 0) {
+        options.onProgress?.({ phase: 'indexing property atoms', percent: 98 });
+        deferredEntityIndex = await prepared.buildDeferredIndex();
+        logPhase('deferred property atom index');
+    }
+    const parseTime = performance.now() - startTime;
+    options.onProgress?.({ phase: 'complete', percent: 100 });
+    const finalStore = {
+        ...earlyStore,
+        parseTime,
+        relationships: fullRelationshipGraph,
+        deferredEntityIndex,
+        onDemandPropertyMap,
+        onDemandQuantityMap,
+        onDemandClassificationMap,
+        onDemandMaterialMap,
+        onDemandDocumentMap,
+        lengthUnitScale,
+        getEntity(expressId) { return entitySource.getEntity(expressId); },
+        getEntitiesByType(typeName) { return entitySource.getEntitiesByType(typeName); },
+        getProperties(expressId) {
+            if (onDemandPropertyMap.size > 0)
+                return extractPropertiesOnDemand(this, expressId);
+            return this.properties.getForEntity(expressId);
+        },
+        getQuantities(expressId) {
+            if (onDemandQuantityMap.size > 0)
+                return extractQuantitiesOnDemand(this, expressId);
+            return this.quantities.getForEntity(expressId);
+        },
+    };
+    return finalStore;
+}
+export class ColumnarParser {
+    /**
+     * Parse IFC file into columnar data store
+     *
+     * Uses fast semicolon-based scanning with on-demand property extraction.
+     * Properties are parsed lazily when accessed, not upfront.
+     * This provides instant UI responsiveness even for very large files.
+     */
+    async parseLite(buffer, entityRefs, options = {}) {
+        return parseColumnarInput(buffer, entityRefs, options);
     }
     /**
      * Extract properties for a single entity ON-DEMAND
@@ -694,235 +515,6 @@ export function extractQuantitiesOnDemand(store, entityId) {
     const parser = new ColumnarParser();
     return parser.extractQuantitiesOnDemand(store, entityId);
 }
-function getEntityRefFromStore(store, expressId) {
-    return store.entityIndex.byId.get(expressId) ?? store.deferredEntityIndex?.get(expressId);
-}
-/**
- * Extract entity attributes on-demand from source buffer.
- * Returns globalId, name, description, objectType, tag mapped by schema name
- * (see {@link extractRootAttributesFromEntity}), so the result stays correct
- * for entity types whose attribute order differs from the IfcElement layout.
- * This is used for entities that weren't fully parsed during initial load.
- */
-export function extractEntityAttributesOnDemand(store, entityId) {
-    const ref = store.entityIndex.byId.get(entityId);
-    if (!ref) {
-        return { globalId: '', name: '', description: '', objectType: '', tag: '' };
-    }
-    const extractor = new EntityExtractor(store.source);
-    const entity = extractor.extractEntity(ref);
-    if (!entity) {
-        return { globalId: '', name: '', description: '', objectType: '', tag: '' };
-    }
-    return extractRootAttributesFromEntity(entity);
-}
-/**
- * Extract ALL named entity attributes on-demand from source buffer.
- * Uses the IFC schema to map attribute indices to names.
- * Returns only string/enum attributes, skipping references and structural attributes.
- */
-export function extractAllEntityAttributes(store, entityId) {
-    const ref = store.entityIndex.byId.get(entityId);
-    if (!ref)
-        return [];
-    const extractor = new EntityExtractor(store.source);
-    const entity = extractor.extractEntity(ref);
-    if (!entity)
-        return [];
-    const attrs = entity.attributes || [];
-    // Use properly-cased type name from entity table (IfcTypeEnumToString)
-    // instead of ref.type which is UPPERCASE from STEP (e.g., IFCWALLSTANDARDCASE)
-    // and breaks multi-word type normalization in getAttributeNames.
-    // For resource-level entities (IfcTask, IfcTaskTime, IfcMaterial,
-    // IfcClassification, ...) the entity table returns 'Unknown';
-    // fall back to ref.type so the schema-driven attribute-name
-    // resolution still works for those types.
-    const tableName = store.entities.getTypeName(entityId);
-    const typeName = tableName && tableName !== 'Unknown' ? tableName : ref.type;
-    // Named across the bundled schema union (2X3 + 4 + 4X3), not through the
-    // IFC4 codegen pin alone. The pin answers an EMPTY list — not a wrong one —
-    // for every class it does not carry, and 251 real classes are outside it:
-    // the IFC2X3 ones IFC4 dropped and the whole IFC4X3 infrastructure
-    // vocabulary (`IfcCourse`, `IfcPavement`, `IfcKerb`, `IfcSignal`, `IfcRoad`,
-    // …). An empty list means this function returns NO attributes for such an
-    // entity, so every caller that looks one up by name silently finds nothing
-    // and cannot tell that apart from an unset slot. The model-diff adapters
-    // read `PredefinedType` this way, which made a cleared `PredefinedType` —
-    // the single most common edit in a real infrastructure revision — compare
-    // equal to itself on exactly the classes IFC4X3 exists for. Same pinned-
-    // registry family as #2001/#2003/#2021.
-    //
-    // Provably additive: `getAttributeNamesAcrossSchemas` returns the pinned
-    // result unchanged whenever the pin has one, so no IFC2X3 or IFC4 entity's
-    // attribute list — or the hash anyone derives from it — moves.
-    const attrNames = getAttributeNamesAcrossSchemas(typeName);
-    const result = [];
-    const len = Math.min(attrs.length, attrNames.length);
-    for (let i = 0; i < len; i++) {
-        const attrName = attrNames[i];
-        if (SKIP_DISPLAY_ATTRS.has(attrName))
-            continue;
-        const raw = attrs[i];
-        // STEP `$` (unset) and `*` (derived) deserialize as null /
-        // undefined and must be skipped. Strings are emitted with
-        // `.ENUM.` markers stripped. Empty strings are preserved so
-        // IDS optional-attribute checks can distinguish "slot truly
-        // absent" from "slot explicitly empty". Numbers and booleans
-        // pass through unchanged so e.g. `CountValue = 0` reads as
-        // present.
-        if (typeof raw === 'string') {
-            // STEP logical-unknown markers (`.U.`, `.X.`) read as
-            // "no value" per IDS spec — they fail any attribute
-            // check, including a bare existence check, so don't
-            // surface them as if the slot were populated.
-            if (raw === '.U.' || raw === '.X.')
-                continue;
-            // Bare boolean tokens (`.T.` / `.F.`) on a schema-typed
-            // IfcBoolean attribute slot — resolve to JS boolean so
-            // IDS checks comparing against `true` / `false` literals
-            // pass without case-sensitive string contortions.
-            if (raw === '.T.') {
-                result.push({ name: attrName, value: true });
-                continue;
-            }
-            if (raw === '.F.') {
-                result.push({ name: attrName, value: false });
-                continue;
-            }
-            const display = raw.startsWith('.') && raw.endsWith('.')
-                ? raw.slice(1, -1)
-                : raw;
-            result.push({ name: attrName, value: display });
-        }
-        else if (typeof raw === 'number' || typeof raw === 'boolean') {
-            result.push({ name: attrName, value: raw });
-        }
-        else if (Array.isArray(raw) && raw.length === 2) {
-            // Typed STEP values like IFCREAL(0.0), IFCBOOLEAN(.T.) —
-            // return the underlying primitive so attribute existence
-            // and value checks can compare it directly.
-            const inner = raw[1];
-            const tag = String(raw[0]).toUpperCase();
-            if (tag.includes('BOOLEAN')) {
-                result.push({ name: attrName, value: inner === '.T.' || inner === true });
-            }
-            else if (tag.includes('LOGICAL')) {
-                if (inner === '.U.' || inner === '.X.') {
-                    // UNKNOWN logical → don't surface (treated as absent)
-                    continue;
-                }
-                result.push({ name: attrName, value: inner === '.T.' || inner === true });
-            }
-            else if (typeof inner === 'number' || typeof inner === 'boolean') {
-                result.push({ name: attrName, value: inner });
-            }
-            else if (typeof inner === 'string' && inner) {
-                const display = inner.startsWith('.') && inner.endsWith('.')
-                    ? inner.slice(1, -1)
-                    : inner;
-                result.push({ name: attrName, value: display });
-            }
-        }
-    }
-    return result;
-}
-/**
- * Returns named raw attribute pairs for an entity, filtered to display-relevant attributes.
- * Skips structural/reference attributes using the IFC schema. Used by query layer for coercion.
- */
-export function getRawNamedAttributes(entity) {
-    const attrs = entity.attributes || [];
-    const attrNames = getAttributeNames(entity.type);
-    const result = [];
-    const len = Math.min(attrs.length, attrNames.length);
-    for (let i = 0; i < len; i++) {
-        const attrName = attrNames[i];
-        if (SKIP_DISPLAY_ATTRS.has(attrName))
-            continue;
-        result.push({ name: attrName, raw: attrs[i] });
-    }
-    return result;
-}
-// getAttributeNames() walks the schema registry (an O(types) scan for the
-// UPPERCASE STEP names entities carry), so memoise the per-type index lookup.
-// There are only a few hundred distinct types but potentially millions of
-// entities, keeping the on-demand path cheap even when called per entity.
-const rootAttrIndexCache = new Map();
-function getRootAttrIndices(type) {
-    let idx = rootAttrIndexCache.get(type);
-    if (!idx) {
-        const names = getAttributeNames(type);
-        idx = {
-            known: names.length > 0,
-            globalId: names.indexOf('GlobalId'),
-            name: names.indexOf('Name'),
-            description: names.indexOf('Description'),
-            objectType: names.indexOf('ObjectType'),
-            tag: names.indexOf('Tag'),
-            longName: names.indexOf('LongName'),
-        };
-        rootAttrIndexCache.set(type, idx);
-    }
-    return idx;
-}
-/**
- * Resolve the common IfcRoot-family display attributes (GlobalId, Name,
- * Description, ObjectType, Tag) from an entity's raw attribute array.
- *
- * These are mapped by schema-derived attribute *name*, not fixed index. The
- * fixed indices `[0],[2],[3],[4],[7]` only hold for the IfcElement layout: for
- * a spatial element `attrs[7]` is LongName (not Tag), and for a resource entity
- * like IfcMaterial `attrs[0]` is Name (not GlobalId). Name-mapping keeps all of
- * these correct for every entity type, returning '' for attributes the type
- * does not declare.
- *
- * For types the schema registry does not recognise (e.g. an IFC4x3 infra leaf
- * outside the codegen pin, or a vendor extension) we fall back to the canonical
- * IfcRoot/IfcElement positions so we never regress vs. the old fixed-index path.
- */
-export function extractRootAttributesFromEntity(entity) {
-    const attrs = entity.attributes || [];
-    const enumIdx = entity.enumAttrIndices;
-    const idx = getRootAttrIndices(entity.type);
-    const pick = (schemaIndex, fallbackIndex) => {
-        const i = idx.known ? schemaIndex : fallbackIndex;
-        const raw = i >= 0 ? attrs[i] : undefined;
-        if (typeof raw !== 'string')
-            return '';
-        // Unknown-type fallback only: a fixed index can land on a STEP bare-enum
-        // token — e.g. a PredefinedType at attr 7 for a non-IfcElement layout —
-        // which must not leak into a Tag/Description cell. Reject by token KIND
-        // via the extractor's side channel (#1799), not by dotted-string shape:
-        // a quoted string that merely looks like an enum ('.USERDEFINED.')
-        // survives, exactly matching the Rust server path (string_at accepts
-        // AttributeValue::String and rejects ::Enum, #1779). Skipped for known
-        // types, whose schema indices point at genuine string slots.
-        if (!idx.known && enumIdx !== undefined && enumIdx.includes(i))
-            return '';
-        return raw;
-    };
-    return {
-        globalId: pick(idx.globalId, 0),
-        name: pick(idx.name, 2),
-        description: pick(idx.description, 3),
-        objectType: pick(idx.objectType, 4),
-        tag: pick(idx.tag, 7),
-    };
-}
-/**
- * Resolve an entity's LongName (IfcZone, IfcSystem subtypes, IfcSpatialZone,
- * IfcBuildingSystem, …) by schema attribute name. Groups frequently leave Name
- * empty and carry their human label in LongName, so consumers fall back to this
- * for the display label. Returns '' when the type does not declare LongName.
- * (#1075)
- */
-export function pickLongName(entity) {
-    const idx = getRootAttrIndices(entity.type);
-    if (!idx.known || idx.longName < 0)
-        return '';
-    const raw = (entity.attributes || [])[idx.longName];
-    return typeof raw === 'string' ? raw : '';
-}
 // Re-export on-demand extraction functions from focused module
 export { extractClassificationsOnDemand, extractClassificationSystemsOnDemand, extractMaterialsOnDemand, extractAllMaterialsOnDemand, extractMaterialPropertiesOnDemand, extractMaterialPropertiesForMaterialId, resolveMaterialDefId, resolveAllMaterialDefIds, collectMaterialLeaves, buildMaterialUsageIndex, getMaterialDisplay, extractTypePropertiesOnDemand, extractTypeEntityOwnProperties, extractTypeQuantitiesOnDemand, extractDocumentsOnDemand, extractRelationshipsOnDemand, extractGroupMembersOnDemand, extractGeoreferencingOnDemand, parsePropertyValue, extractPsetsFromIds, extractQsetsFromIds, } from './on-demand-extractors.js';
 export { mergeInheritedPropertySets, mergeInheritedQuantitySets } from './property-set-merge.js';