update instancing

README.md

@@ -16,7 +16,7 @@ The exporter receives a Speckle model version, walks its object tree, and produc
 
 - Correct IFC entity classification (IfcWall, IfcSlab, IfcColumn, etc.)
 - Tessellated geometry (IfcPolygonalFaceSet)
-- Curve geometry for Lines and Arcs (IfcGeometricCurveSet with IfcPolyline)
+- Curve geometry for Lines, Arcs, and Polycurves (IfcIndexedPolyCurve with IfcLineIndex/IfcArcIndex)
 - Material colours from Speckle render materials
 - Revit property sets (Common psets, instance/type parameters, material quantities)
 - IFC type objects (IfcWallType, IfcSlabType, etc.) shared across instances
@@ -42,7 +42,7 @@ Speckle Model
 4. Traverse object tree
     │   For each leaf element:
     │   ├── Classify → IFC entity class (skip analytical categories)
-    │   ├── Convert geometry → IfcPolygonalFaceSet or IfcGeometricCurveSet
+    │   ├── Convert geometry → IfcPolygonalFaceSet or IfcIndexedPolyCurve
     │   ├── Create IFC element + placement
     │   ├── Write property sets & quantities
     │   └── Assign IFC type object
@@ -61,8 +61,10 @@ Speckle Model
 | `main.py` | Entry point, orchestrates the full pipeline |
 | `utils/traversal.py` | Walks the Speckle Collection tree (Project > Level > Category > Element) |
 | `utils/mapper.py` | Classifies Speckle objects into IFC entity types |
-| `utils/geometry.py` | Converts Speckle meshes to IfcPolygonalFaceSet and Lines/Arcs to IfcGeometricCurveSet geometry |
-| `utils/instances.py` | Handles InstanceProxy objects with shared geometry (IfcMappedItem) |
+| `utils/helpers.py` | Shared utilities (`_get` safe accessor, `MM_SCALES` unit conversion) |
+| `utils/geometry.py` | Converts Speckle meshes to IfcPolygonalFaceSet geometry (handles nested BrepX) |
+| `utils/curves.py` | Converts Lines, Arcs, and Polycurves to IfcIndexedPolyCurve geometry |
+| `utils/instances.py` | Handles InstanceProxy objects with shared geometry (IfcMappedItem), content-based deduplication |
 | `utils/properties.py` | Writes IFC property sets and quantities from Revit parameters |
 | `utils/type_manager.py` | Creates and caches IfcTypeObjects (IfcWallType, etc.) |
 | `utils/materials.py` | Maps Speckle render materials to IfcSurfaceStyle colours |
@@ -134,11 +136,12 @@ If none of the above match, the object is classified as `IfcBuildingElementProxy
 
 Objects with `displayValue` containing Mesh objects are converted directly:
 
-1. Extract vertices and faces from each mesh in `displayValue`
+1. Extract vertices and faces from each mesh in `displayValue` (recursively handles nested BrepX/Brep objects)
 2. Scale vertices to millimetres based on the mesh's unit declaration
 3. Deduplicate vertices via snap grid (0.01mm tolerance) to avoid IFC GEM111 errors
-4. Build `IfcPolygonalFaceSet` with `IfcCartesianPointList3D` + `IfcIndexedPolygonalFace`
-5. Compute bounding box origin for `IfcLocalPlacement`, offset vertices relative to it
+4. Round vertex coordinates to 0.001mm precision for smaller IFC file output
+5. Build `IfcPolygonalFaceSet` with `IfcCartesianPointList3D` + `IfcIndexedPolygonalFace`
+6. Compute bounding box origin incrementally for `IfcLocalPlacement`, offset vertices relative to it
 
 ### Instance Objects (Path A / B2)
 
@@ -147,16 +150,17 @@ Speckle `InstanceProxy` objects reference shared definition geometry via `defini
 - **Revit format**: `definitionId` is a 64-char hex hash; geometry is found by walking the object tree
 - **IFC format**: `definitionId` starts with `DEFINITION:`; geometry is in `definitionGeometry` collection
 
-Performance optimisation: geometry is built once as an `IfcRepresentationMap`, then each instance references it via `IfcMappedItem` + `IfcCartesianTransformationOperator3DnonUniform`. This avoids duplicating vertex data across hundreds of identical elements.
+Performance optimisation: geometry is built once as an `IfcRepresentationMap`, then each instance references it via `IfcMappedItem` + `IfcCartesianTransformationOperator3DnonUniform`. This avoids duplicating vertex data across hundreds of identical elements. Content-based hashing further deduplicates definitions that share identical geometry.
 
 ### Curve Geometry (Path B3)
 
-Objects whose `displayValue` contains `Objects.Geometry.Line` or `Objects.Geometry.Arc` items (and no meshes or instances) are exported as curve geometry:
+Objects whose `displayValue` contains `Objects.Geometry.Line`, `Objects.Geometry.Arc`, or `Objects.Geometry.Polycurve` items (and no meshes or instances) are exported as curve geometry using native IFC curve types:
 
-- **Lines** → `IfcPolyline` with start and end points
-- **Arcs** → `IfcPolyline` approximated with 8 segments, sampled parametrically from the arc's plane origin, radius, and domain angles. Falls back to start/mid/end points if plane data is unavailable.
+- **Lines** → `IfcLineIndex` segments (start/end points)
+- **Arcs** → `IfcArcIndex` segments (start/mid/end points)
+- **Polycurves** → Mixed `IfcLineIndex` and `IfcArcIndex` segments from the polycurve's segment list (supports Line, Arc, and Polyline sub-segments)
 
-All curves are wrapped in an `IfcGeometricCurveSet` inside an `IfcShapeRepresentation` with `RepresentationType="GeometricCurveSet"`.
+All curves use `IfcIndexedPolyCurve` with `IfcCartesianPointList3D` for compact, deduplicated point storage. The representation uses `RepresentationType="Curve3D"`.
 
 ### Composite Objects (Path B2 — merged instances)
 

main.py

@@ -1,3 +1,4 @@
+import zipfile
 from datetime import datetime
 
 import ifcopenshell.api
@@ -5,8 +6,9 @@ import ifcopenshell.api
 from utils.materials import MaterialManager
 from utils.traversal import traverse, print_tree
 from utils.mapper import classify, reset_caches as reset_mapper_caches
-from utils.geometry import mesh_to_ifc, get_display_instances, curves_to_ifc, _make_placement
-from utils.instances import is_instance, instance_to_ifc, build_definition_map, print_instance_stats, get_definition_object
+from utils.geometry import mesh_to_ifc, get_display_instances, _make_placement
+from utils.curves import curve_to_ifc
+from utils.instances import is_instance, instance_to_ifc, build_definition_map, print_instance_stats, get_definition_object, reset_caches as reset_instance_caches
 from utils.properties import write_properties, write_common_properties, build_element_name, get_element_tag, get_ifc_guid, reset_caches as reset_props_caches
 from utils.writer import create_ifc_scaffold, StoreyManager
 from utils.type_manager import TypeManager
@@ -61,6 +63,7 @@ def automate_function(
     # Reset caches from any previous run
     reset_props_caches()
     reset_mapper_caches()
+    reset_instance_caches()
 
     # ------------------------------------------------------------------ #
     # 1. Receive
@@ -199,7 +202,7 @@ def automate_function(
 
             # B3: Curve geometry (Lines, Arcs in displayValue)
             if not rep and not nested_instances:
-                curve_rep, curve_placement = curves_to_ifc(ifc, body_context, obj, scale=scale, material_manager=material_manager)
+                curve_rep, curve_placement = curve_to_ifc(ifc, body_context, obj, scale=scale, material_manager=material_manager)
                 if curve_rep:
                     element = _create_element(ifc, ifc_class, name, curve_rep, curve_placement, storey,
                                                  storey_manager=storey_manager,
@@ -230,11 +233,17 @@ def automate_function(
 
     ifc.write(ifc_filename)
     print(f"\n💾 IFC file written: {ifc_filename}")
+
+    zip_filename = f"{file_name}_{timestamp}.zip"
+    with zipfile.ZipFile(zip_filename, "w", zipfile.ZIP_DEFLATED) as zf:
+        zf.write(ifc_filename)
+    print(f"Zipped: {zip_filename}")
+
     try:
-        automate_context.mark_run_success("Success! You can download the IF file below.")
-        automate_context.store_file_result(f"./{ifc_filename}")
+        automate_context.mark_run_success("Success! You can download the IFC file below.")
+        automate_context.store_file_result(f"./{zip_filename}")
     except Exception as e:
-        print(f"  ⚠️  Could not upload file result (network issue?): {e}")
+        print(f"  Could not upload file result (network issue?): {e}")
         automate_context.mark_run_failed(f"Something went wrong when storing file result. Exception detail: {e}") 
 
     print(f"\n{'=' * 60}")

utils/curves.py

@@ -0,0 +1,357 @@
+# =============================================================================
+# curves.py
+# Converts Speckle 2D curve geometry (Polycurve, Line, Arc, Circle, Polyline)
+# into IFC IfcIndexedPolyCurve representations.
+#
+# Curve types in segments:
+#   - Objects.Geometry.Line     → start/end Points → IfcLineIndex
+#   - Objects.Geometry.Arc      → startPoint/midPoint/endPoint → IfcArcIndex
+#   - Objects.Geometry.Circle   → converted to arc segments
+#   - Objects.Geometry.Polyline → point sequence → IfcLineIndex chains
+#
+# The result is an IfcIndexedPolyCurve with IfcCartesianPointList3D.
+# =============================================================================
+
+import ifcopenshell
+import ifcopenshell.api
+from specklepy.objects.base import Base
+from utils.helpers import _get, MM_SCALES
+from utils.geometry import _get_shared, _make_placement
+
+
+# Speckle types that are curve geometry
+_CURVE_TYPES = {"Line", "Arc", "Circle", "Ellipse", "Polycurve", "Polyline", "Curve"}
+
+
+def is_curve(obj) -> bool:
+    """Return True if this object is a Speckle curve type."""
+    speckle_type = _get(obj, "speckle_type") or ""
+    return any(ct in speckle_type for ct in _CURVE_TYPES)
+
+
+def _resolve_scale(obj, fallback: float) -> float:
+    """Resolve unit scale for a curve object."""
+    units = _get(obj, "units")
+    if units and isinstance(units, str):
+        return MM_SCALES.get(units.lower().strip(), fallback)
+    return fallback
+
+
+def _point_coords(pt, scale: float) -> tuple:
+    """Extract (x, y, z) from a Speckle Point, scaled to mm and rounded."""
+    x = round(float(_get(pt, "x") or 0) * scale, 3)
+    y = round(float(_get(pt, "y") or 0) * scale, 3)
+    z = round(float(_get(pt, "z") or 0) * scale, 3)
+    return x, y, z
+
+
+def _extract_polycurve(obj, scale: float) -> tuple:
+    """
+    Extract points and segment indices from a Polycurve.
+
+    Returns (points_3d, segments) where:
+      points_3d: list of [x, y, z] coordinate lists
+      segments:  list of IfcLineIndex/IfcArcIndex-compatible tuples
+                 ("line", [i, j]) or ("arc", [i, mid, j])  (1-based)
+    """
+    segments_raw = _get(obj, "segments") or []
+    if not isinstance(segments_raw, list):
+        segments_raw = list(segments_raw)
+    if not segments_raw:
+        return [], []
+
+    obj_scale = _resolve_scale(obj, scale)
+    points = []       # list of [x, y, z]
+    point_map = {}    # (rounded_x, rounded_y, rounded_z) -> 1-based index
+    ifc_segments = []
+
+    def _add_point(pt, seg_scale: float) -> int:
+        """Add a point and return its 1-based index (deduplicating nearby points)."""
+        x, y, z = _point_coords(pt, seg_scale)
+        # Snap to 0.01mm grid for deduplication
+        key = (round(x * 100), round(y * 100), round(z * 100))
+        if key in point_map:
+            return point_map[key]
+        idx = len(points) + 1  # 1-based for IFC
+        points.append([x, y, z])
+        point_map[key] = idx
+        return idx
+
+    for seg in segments_raw:
+        if seg is None:
+            continue
+        seg_type = (_get(seg, "speckle_type") or "").split(".")[-1]
+        seg_scale = _resolve_scale(seg, obj_scale)
+
+        if seg_type == "Line":
+            start_pt = _get(seg, "start")
+            end_pt = _get(seg, "end")
+            if start_pt is None or end_pt is None:
+                continue
+            i = _add_point(start_pt, seg_scale)
+            j = _add_point(end_pt, seg_scale)
+            if i != j:
+                ifc_segments.append(("line", [i, j]))
+
+        elif seg_type == "Arc":
+            start_pt = _get(seg, "startPoint")
+            mid_pt = _get(seg, "midPoint")
+            end_pt = _get(seg, "endPoint")
+            if start_pt is None or mid_pt is None or end_pt is None:
+                continue
+            i = _add_point(start_pt, seg_scale)
+            m = _add_point(mid_pt, seg_scale)
+            j = _add_point(end_pt, seg_scale)
+            if i != j and i != m and m != j:
+                ifc_segments.append(("arc", [i, m, j]))
+
+        elif seg_type == "Polyline":
+            raw_value = _get(seg, "value") or []
+            if not raw_value:
+                continue
+            values = list(raw_value) if not isinstance(raw_value, list) else raw_value
+            indices = []
+            for vi in range(0, len(values) - 2, 3):
+                x = round(float(values[vi]) * seg_scale, 3)
+                y = round(float(values[vi + 1]) * seg_scale, 3)
+                z = round(float(values[vi + 2]) * seg_scale, 3)
+                key = (round(x * 100), round(y * 100), round(z * 100))
+                if key in point_map:
+                    idx = point_map[key]
+                else:
+                    idx = len(points) + 1
+                    points.append([x, y, z])
+                    point_map[key] = idx
+                indices.append(idx)
+            if len(indices) >= 2:
+                ifc_segments.append(("line", indices))
+
+    return points, ifc_segments
+
+
+def _extract_single_line(obj, scale: float) -> tuple:
+    """Extract a single Line as points + segment."""
+    obj_scale = _resolve_scale(obj, scale)
+    start_pt = _get(obj, "start")
+    end_pt = _get(obj, "end")
+    if start_pt is None or end_pt is None:
+        return [], []
+    sx, sy, sz = _point_coords(start_pt, obj_scale)
+    ex, ey, ez = _point_coords(end_pt, obj_scale)
+    return [[sx, sy, sz], [ex, ey, ez]], [("line", [1, 2])]
+
+
+def _extract_single_arc(obj, scale: float) -> tuple:
+    """Extract a single Arc as points + segment."""
+    obj_scale = _resolve_scale(obj, scale)
+    start_pt = _get(obj, "startPoint")
+    mid_pt = _get(obj, "midPoint")
+    end_pt = _get(obj, "endPoint")
+    if start_pt is None or mid_pt is None or end_pt is None:
+        return [], []
+    sx, sy, sz = _point_coords(start_pt, obj_scale)
+    mx, my, mz = _point_coords(mid_pt, obj_scale)
+    ex, ey, ez = _point_coords(end_pt, obj_scale)
+    return [[sx, sy, sz], [mx, my, mz], [ex, ey, ez]], [("arc", [1, 2, 3])]
+
+
+def extract_curve_data(obj, scale: float = 1.0) -> tuple:
+    """
+    Extract curve points and segments from any supported curve type.
+    Returns (points_3d, segments) or ([], []) if not a curve.
+    """
+    speckle_type = (_get(obj, "speckle_type") or "").split(".")[-1]
+
+    if speckle_type == "Polycurve":
+        return _extract_polycurve(obj, scale)
+    elif speckle_type == "Line":
+        return _extract_single_line(obj, scale)
+    elif speckle_type == "Arc":
+        return _extract_single_arc(obj, scale)
+    return [], []
+
+
+def build_ifc_curve(ifc, points: list, segments: list):
+    """
+    Build an IfcIndexedPolyCurve from points and segment descriptors.
+
+    points:   list of [x, y, z] coordinates
+    segments: list of ("line", [indices]) or ("arc", [indices])
+
+    Returns IfcIndexedPolyCurve or None.
+    """
+    if not points or not segments:
+        return None
+
+    point_list = ifc.createIfcCartesianPointList3D(points)
+
+    ifc_segments = []
+    for seg_type, indices in segments:
+        if seg_type == "arc":
+            ifc_segments.append(ifc.create_entity("IfcArcIndex", indices))
+        else:
+            ifc_segments.append(ifc.create_entity("IfcLineIndex", indices))
+
+    if not ifc_segments:
+        return None
+
+    return ifc.createIfcIndexedPolyCurve(
+        Points=point_list,
+        Segments=ifc_segments,
+        SelfIntersect=False,
+    )
+
+
+def get_display_curves(obj) -> list:
+    """
+    Collect curve objects from an object's displayValue, or the object itself.
+    Returns a list of curve objects (Polycurve, Line, Arc, etc.).
+    """
+    curves = []
+    for key in ["displayValue", "@displayValue", "_displayValue"]:
+        display = _get(obj, key)
+        if display is None:
+            continue
+        items = display if isinstance(display, list) else [display]
+        for item in items:
+            if item is not None and is_curve(item):
+                curves.append(item)
+        if curves:
+            break
+
+    # Fallback: the object itself is a curve
+    if not curves and is_curve(obj):
+        curves.append(obj)
+
+    return curves
+
+
+def curve_to_ifc(
+    ifc: ifcopenshell.file,
+    body_context,
+    obj: Base,
+    scale: float = 1.0,
+    material_manager=None,
+) -> tuple:
+    """
+    Convert a Speckle object with curve geometry -> (IfcShapeRepresentation, IfcLocalPlacement).
+    Looks for curves in displayValue first, then checks the object itself.
+    Creates one IfcIndexedPolyCurve per curve item.
+    Returns (None, None) if no usable curve geometry.
+    """
+    curves = get_display_curves(obj)
+    if not curves:
+        return None, None
+
+    obj_app_id = _get(obj, "applicationId")
+    obj_scale = _resolve_scale(obj, scale)
+
+    # Collect curve data and compute origin incrementally
+    curve_cache = []
+    xmin = ymin = zmin = float("inf")
+    xmax = ymax = float("-inf")
+    has_points = False
+
+    for curve_obj in curves:
+        points, segments = extract_curve_data(curve_obj, obj_scale)
+        if points and segments:
+            curve_cache.append((points, segments))
+            has_points = True
+            for p in points:
+                x, y, z = p[0], p[1], p[2]
+                if x < xmin: xmin = x
+                if x > xmax: xmax = x
+                if y < ymin: ymin = y
+                if y > ymax: ymax = y
+                if z < zmin: zmin = z
+        else:
+            curve_cache.append(None)
+
+    if not has_points:
+        return None, None
+
+    ox = (xmin + xmax) / 2.0
+    oy = (ymin + ymax) / 2.0
+    oz = zmin
+
+    # Build IFC curve entities
+    geom_items = []
+    for i, cached in enumerate(curve_cache):
+        if cached is None:
+            continue
+        points, segments = cached
+
+        offset_points = [
+            [p[0] - ox, p[1] - oy, p[2] - oz] for p in points
+        ]
+
+        curve_entity = build_ifc_curve(ifc, offset_points, segments)
+        if curve_entity is None:
+            continue
+
+        # Apply material
+        if material_manager:
+            curve_app_id = _get(curves[i], "applicationId") or obj_app_id
+            if curve_app_id:
+                material_manager.apply_to_item(curve_entity, str(curve_app_id))
+
+        geom_items.append(curve_entity)
+
+    if not geom_items:
+        return None, None
+
+    rep = ifc.createIfcShapeRepresentation(
+        ContextOfItems=body_context,
+        RepresentationIdentifier="Body",
+        RepresentationType="Curve3D",
+        Items=geom_items,
+    )
+    placement = _make_placement(ifc, ox, oy, oz)
+    return rep, placement
+
+
+def build_curve_rep_map(ifc, body_context, obj, scale: float = 1.0,
+                        material_manager=None, fallback_app_ids: list = None,
+                        definition_id: str = None):
+    """
+    Build an IfcRepresentationMap from a curve definition object.
+    Used for instance-based curve geometry (shared across instances).
+    Returns IfcRepresentationMap or None.
+    """
+    points, segments = extract_curve_data(obj, scale)
+    if not points or not segments:
+        return None
+
+    curve_entity = build_ifc_curve(ifc, points, segments)
+    if curve_entity is None:
+        return None
+
+    # Apply material (3-tier: object app_id -> fallbacks -> definition)
+    if material_manager:
+        app_id = _get(obj, "applicationId")
+        style = material_manager.get_style_with_fallbacks(
+            primary_app_id=str(app_id) if app_id else None,
+            fallback_app_ids=fallback_app_ids,
+            definition_id=definition_id,
+        )
+        if style:
+            try:
+                ifcopenshell.api.run(
+                    "style.assign_item_style", ifc,
+                    item=curve_entity, style=style,
+                )
+                material_manager._apply_count += 1
+            except Exception:
+                pass
+
+    shared = _get_shared(ifc)
+    a2p = ifc.createIfcAxis2Placement3D(shared["origin_0"], None, None)
+
+    mapped_rep = ifc.createIfcShapeRepresentation(
+        ContextOfItems=body_context,
+        RepresentationIdentifier="Body",
+        RepresentationType="Curve3D",
+        Items=[curve_entity],
+    )
+
+    return ifc.createIfcRepresentationMap(a2p, mapped_rep)

utils/geometry.py

@@ -1,6 +1,6 @@
 # =============================================================================
 # geometry.py
-# Converts Speckle DataObject geometry → IFC IfcPolygonalFaceSet + IfcLocalPlacement
+# Converts Speckle DataObject geometry -> IFC IfcPolygonalFaceSet + IfcLocalPlacement
 #
 # Key facts:
 #   - After specklepy receive(), vertices and faces are FLAT Python lists
@@ -8,23 +8,12 @@
 #   - Units are in mm (for Revit), scale to metres for IFC
 #   - Vertices are in absolute world coordinates
 #   - Uses IfcPolygonalFaceSet (indexed vertices) instead of IfcFacetedBrep
-#     for compact output — each vertex stored once, not once per face.
+#     for compact output -- each vertex stored once, not once per face.
 # =============================================================================
 
-import math
-
 import ifcopenshell
 from specklepy.objects.base import Base
-
-
-# Scale factors → MILLIMETRES (IFC file is declared as mm)
-_UNIT_SCALES = {
-    "mm": 1.0,    "millimeter": 1.0,    "millimeters": 1.0,
-    "cm": 10.0,   "centimeter": 10.0,   "centimeters": 10.0,
-    "m":  1000.0, "meter": 1000.0,      "meters": 1000.0,
-    "ft": 304.8,  "foot": 304.8,        "feet": 304.8,
-    "in": 25.4,   "inch": 25.4,         "inches": 25.4,
-}
+from utils.helpers import _get, MM_SCALES as _UNIT_SCALES
 
 
 # --------------------------------------------------------------------------- #
@@ -49,8 +38,8 @@ def build_ifc_facesets(ifc, verts_scaled: list, face_groups: list) -> list:
     face_groups:  list of index lists [[i,j,k], [i,j,k,l], ...]
     Returns: list of IfcPolygonalFaceSet (typically one, empty on failure).
     """
-    snap_to_idx = {}   # snap_key → 0-based index in deduped_verts
-    deduped_verts = [] # [[x, y, z], ...] — lists for direct IFC use
+    snap_to_idx = {}   # snap_key -> 0-based index in deduped_verts
+    deduped_verts = [] # [[x, y, z], ...] -- lists for direct IFC use
     inv_tol = _INV_TOL
 
     # Validate faces and remap indices to deduplicated vertex list
@@ -87,6 +76,13 @@ def build_ifc_facesets(ifc, verts_scaled: list, face_groups: list) -> list:
     if not valid_faces or not deduped_verts:
         return []
 
+    # Round vertex coordinates to reduce IFC text file size
+    # 3 decimal places = 0.001mm precision (more than sufficient)
+    for v in deduped_verts:
+        v[0] = round(v[0], 3)
+        v[1] = round(v[1], 3)
+        v[2] = round(v[2], 3)
+
     # Build IFC entities
     try:
         point_list = ifc.createIfcCartesianPointList3D(deduped_verts)
@@ -99,39 +95,15 @@ def build_ifc_facesets(ifc, verts_scaled: list, face_groups: list) -> list:
         return []
 
 
-# --------------------------------------------------------------------------- #
-# Safe data access helpers
-# --------------------------------------------------------------------------- #
-
-def _get(obj, key, default=None):
-    """
-    Safe access for specklepy Base objects.
-    Tries attribute access first, then bracket access.
-    """
-    try:
-        val = getattr(obj, key, None)
-        if val is not None:
-            return val
-    except Exception:
-        pass
-    try:
-        val = obj[key]
-        if val is not None:
-            return val
-    except Exception:
-        pass
-    return default
-
-
 def unwrap_chunks(raw) -> list:
     """
     Flatten a Speckle data array into a plain Python list of numbers.
 
     Handles two cases:
       1. Already flat list of numbers (after specklepy receive deserializes)
-         → returned as-is (fast path)
+         -> returned as-is (fast path)
       2. List of DataChunk objects (raw from server before deserialization)
-         → each chunk's .data list is concatenated
+         -> each chunk's .data list is concatenated
     """
     if not raw:
         return []
@@ -163,7 +135,7 @@ def unwrap_chunks(raw) -> list:
 
 
 def _resolve_scale(obj, stream_scale: float) -> float:
-    """Resolve unit scale: obj.units → stream fallback."""
+    """Resolve unit scale: obj.units -> stream fallback."""
     units = _get(obj, "units")
     if units and isinstance(units, str):
         return _UNIT_SCALES.get(units.lower().strip(), stream_scale)
@@ -177,7 +149,7 @@ def _resolve_scale(obj, stream_scale: float) -> float:
 def _is_mesh(item) -> bool:
     """
     Detect if a specklepy object is a Mesh.
-    Uses speckle_type string — more reliable than hasattr on Base objects.
+    Uses speckle_type string -- more reliable than hasattr on Base objects.
     """
     if item is None:
         return False
@@ -190,23 +162,39 @@ def _is_mesh(item) -> bool:
     return verts is not None and faces is not None
 
 
-def get_display_meshes(obj: Base) -> list:
+def _collect_meshes_from_display(obj) -> list:
     """
-    Extract all Mesh objects from a DataObject's displayValue.
-    displayValue is always an array per the Speckle schema docs.
+    Collect Mesh objects from an object's displayValue.
+    If an item is not a Mesh (e.g. BrepX, Brep), recursively check
+    its own displayValue for nested meshes.
     """
     meshes = []
-
-    for key in ["displayValue", "@displayValue"]:
+    for key in ["displayValue", "@displayValue", "_displayValue"]:
         display = _get(obj, key)
         if display is None:
             continue
         items = display if isinstance(display, list) else [display]
         for item in items:
+            if item is None:
+                continue
             if _is_mesh(item):
                 meshes.append(item)
+            else:
+                # BrepX / Brep / other geometry types may carry a nested
+                # displayValue with the tessellated mesh representation
+                meshes.extend(_collect_meshes_from_display(item))
         if meshes:
-            break  # found meshes, don't check @displayValue too
+            break
+    return meshes
+
+
+def get_display_meshes(obj: Base) -> list:
+    """
+    Extract all Mesh objects from a DataObject's displayValue.
+    Handles nested geometry types (BrepX, Brep) that wrap meshes
+    inside their own displayValue.
+    """
+    meshes = _collect_meshes_from_display(obj)
 
     # Fallback: object itself is a Mesh
     if not meshes and _is_mesh(obj):
@@ -227,10 +215,10 @@ def get_display_instances(obj: Base) -> list:
       - definitionId: "DEFINITION:{meshAppId}" string
       - units:        "m"
 
-    Raw meshes do NOT appear in displayValue in IFC→Speckle exports.
+    Raw meshes do NOT appear in displayValue in IFC->Speckle exports.
     """
     instances = []
-    for key in ["displayValue", "@displayValue"]:
+    for key in ["displayValue", "@displayValue", "_displayValue"]:
         display = _get(obj, key)
         if display is None:
             continue
@@ -247,181 +235,6 @@ def get_display_instances(obj: Base) -> list:
     return instances
 
 
-# --------------------------------------------------------------------------- #
-# Curve detection & extraction (Lines, Arcs)
-# --------------------------------------------------------------------------- #
-
-def _is_line(item) -> bool:
-    """Detect Objects.Geometry.Line (but not Polyline)."""
-    if item is None:
-        return False
-    st = _get(item, "speckle_type") or ""
-    return "Line" in st and "Polyline" not in st
-
-
-def _is_arc(item) -> bool:
-    """Detect Objects.Geometry.Arc."""
-    if item is None:
-        return False
-    st = _get(item, "speckle_type") or ""
-    return "Arc" in st
-
-
-def get_display_curves(obj: Base) -> list:
-    """Extract Line and Arc objects from a DataObject's displayValue."""
-    curves = []
-    for key in ["displayValue", "@displayValue"]:
-        display = _get(obj, key)
-        if display is None:
-            continue
-        items = display if isinstance(display, list) else [display]
-        for item in items:
-            if _is_line(item) or _is_arc(item):
-                curves.append(item)
-        if curves:
-            break
-    return curves
-
-
-def _point_coords(pt, fallback_scale: float) -> tuple:
-    """Extract (x, y, z) from a Speckle Point, scaled to mm."""
-    scale = _resolve_scale(pt, fallback_scale)
-    x = float(_get(pt, "x") or 0.0) * scale
-    y = float(_get(pt, "y") or 0.0) * scale
-    z = float(_get(pt, "z") or 0.0) * scale
-    return x, y, z
-
-
-def _arc_to_points(arc, scale: float, num_segments: int = 8) -> list:
-    """
-    Approximate a Speckle Arc as a list of (x, y, z) points in mm.
-    Uses plane origin (center), radius, and domain angles for parametric sampling.
-    Falls back to start/mid/end points if plane data is missing.
-    """
-    plane = _get(arc, "plane")
-    radius = _get(arc, "radius")
-    domain = _get(arc, "domain")
-
-    if not plane or not radius or not domain:
-        points = []
-        for key in ["startPoint", "midPoint", "endPoint"]:
-            pt = _get(arc, key)
-            if pt:
-                points.append(_point_coords(pt, scale))
-        return points if len(points) >= 2 else []
-
-    origin = _get(plane, "origin")
-    xdir = _get(plane, "xdir")
-    ydir = _get(plane, "ydir")
-
-    if not origin or not xdir or not ydir:
-        points = []
-        for key in ["startPoint", "midPoint", "endPoint"]:
-            pt = _get(arc, key)
-            if pt:
-                points.append(_point_coords(pt, scale))
-        return points if len(points) >= 2 else []
-
-    cx, cy, cz = _point_coords(origin, scale)
-    # Direction vectors are unitless — do not scale
-    dxx = float(_get(xdir, "x") or 0.0)
-    dxy = float(_get(xdir, "y") or 0.0)
-    dxz = float(_get(xdir, "z") or 0.0)
-    dyx = float(_get(ydir, "x") or 0.0)
-    dyy = float(_get(ydir, "y") or 0.0)
-    dyz = float(_get(ydir, "z") or 0.0)
-
-    r = float(radius) * scale
-    t_start = float(_get(domain, "start") or 0.0)
-    t_end = float(_get(domain, "end") or 0.0)
-
-    points = []
-    for i in range(num_segments + 1):
-        t = t_start + (t_end - t_start) * i / num_segments
-        cos_t = math.cos(t)
-        sin_t = math.sin(t)
-        x = cx + r * (cos_t * dxx + sin_t * dyx)
-        y = cy + r * (cos_t * dxy + sin_t * dyy)
-        z = cz + r * (cos_t * dxz + sin_t * dyz)
-        points.append((x, y, z))
-    return points
-
-
-def curves_to_ifc(
-    ifc: ifcopenshell.file,
-    body_context,
-    obj: Base,
-    scale: float = 0.001,
-    material_manager=None,
-) -> tuple:
-    """
-    Convert Speckle Line/Arc objects in displayValue to IFC curve geometry.
-    Lines → IfcPolyline (2 points), Arcs → IfcPolyline (sampled points).
-    Wrapped in IfcGeometricCurveSet.
-    Returns (IfcShapeRepresentation, IfcLocalPlacement) or (None, None).
-    """
-    curves = get_display_curves(obj)
-    if not curves:
-        return None, None
-
-    obj_scale = _resolve_scale(obj, scale)
-    polylines = []
-    all_points = []
-
-    for curve in curves:
-        cs = _resolve_scale(curve, obj_scale)
-
-        if _is_line(curve):
-            start = _get(curve, "start")
-            end = _get(curve, "end")
-            if not start or not end:
-                continue
-            p1 = _point_coords(start, cs)
-            p2 = _point_coords(end, cs)
-            all_points.extend([p1, p2])
-            polylines.append([p1, p2])
-
-        elif _is_arc(curve):
-            pts = _arc_to_points(curve, cs)
-            if len(pts) >= 2:
-                all_points.extend(pts)
-                polylines.append(pts)
-
-    if not polylines or not all_points:
-        return None, None
-
-    # Compute origin from all curve points
-    xs = [p[0] for p in all_points]
-    ys = [p[1] for p in all_points]
-    zs = [p[2] for p in all_points]
-    ox = (min(xs) + max(xs)) / 2.0
-    oy = (min(ys) + max(ys)) / 2.0
-    oz = min(zs)
-
-    # Build IfcPolylines offset from origin
-    ifc_polylines = []
-    for pts in polylines:
-        ifc_points = [
-            ifc.createIfcCartesianPoint([p[0] - ox, p[1] - oy, p[2] - oz])
-            for p in pts
-        ]
-        ifc_polylines.append(ifc.createIfcPolyline(ifc_points))
-
-    if not ifc_polylines:
-        return None, None
-
-    curve_set = ifc.createIfcGeometricCurveSet(ifc_polylines)
-
-    rep = ifc.createIfcShapeRepresentation(
-        ContextOfItems=body_context,
-        RepresentationIdentifier="Body",
-        RepresentationType="GeometricCurveSet",
-        Items=[curve_set],
-    )
-    placement = _make_placement(ifc, ox, oy, oz)
-    return rep, placement
-
-
 # --------------------------------------------------------------------------- #
 # Face decoding
 # --------------------------------------------------------------------------- #
@@ -430,7 +243,7 @@ def decode_faces(faces_raw: list) -> list:
     """
     Decode Speckle's run-length encoded face list into vertex index groups.
     Format: [n, i0, i1, ..., n, i0, i1, ...]
-      n=0 → triangle (legacy), n=1 → quad (legacy), n≥3 → n-gon
+      n=0 -> triangle (legacy), n=1 -> quad (legacy), n>=3 -> n-gon
     """
     decoded = []
     i = 0
@@ -462,7 +275,7 @@ def compute_origin(flat_verts: list) -> tuple:
     """
     Compute placement origin from scaled vertex list (mm).
     X, Y = bounding box centroid
-    Z = minimum Z (bottom face of element — more natural for IFC)
+    Z = minimum Z (bottom face of element -- more natural for IFC)
     Single-pass to avoid creating 3 sliced copies of a large list.
     """
     x0 = flat_verts[0]
@@ -505,9 +318,9 @@ def _get_shared(ifc):
 
 
 def _make_placement(ifc, x: float, y: float, z: float):
-    """Create an IfcLocalPlacement at absolute world coordinates (metres)."""
+    """Create an IfcLocalPlacement at absolute world coordinates (mm)."""
     shared = _get_shared(ifc)
-    origin = ifc.createIfcCartesianPoint([x, y, z])
+    origin = ifc.createIfcCartesianPoint([round(x, 3), round(y, 3), round(z, 3)])
     a2p    = ifc.createIfcAxis2Placement3D(origin, shared["z_axis"], shared["x_axis"])
     return ifc.createIfcLocalPlacement(PlacementRelTo=None, RelativePlacement=a2p)
 
@@ -524,7 +337,7 @@ def mesh_to_ifc(
     material_manager=None,
 ) -> tuple:
     """
-    Convert a Speckle DataObject → (IfcShapeRepresentation, IfcLocalPlacement).
+    Convert a Speckle DataObject -> (IfcShapeRepresentation, IfcLocalPlacement).
     Creates one IfcPolygonalFaceSet per mesh so each can carry its own material style.
     Returns (None, None) if no usable geometry is found.
     """
@@ -532,14 +345,21 @@ def mesh_to_ifc(
     if not meshes:
         return None, None
 
+    # Parent object's applicationId -- used as fallback for material lookup
+    # when inner meshes (e.g. from BrepX) don't have their own applicationId
+    obj_app_id = _get(obj, "applicationId")
+
     obj_scale = _resolve_scale(obj, scale)
 
     # ------------------------------------------------------------------ #
-    # Pass 1: unpack vertices once per mesh, collect all scaled coords
-    #         to compute world origin. Cache (verts, ms) for Pass 2.
+    # Pass 1: unpack and scale vertices once per mesh, compute origin
+    #         incrementally without accumulating all vertices in memory.
     # ------------------------------------------------------------------ #
-    mesh_cache = []   # [(verts_list, ms, scaled)] or None per mesh
-    all_scaled = []
+    mesh_cache = []   # [scaled_verts_list] or None per mesh
+    xmin = ymin = zmin = float("inf")
+    xmax = ymax = float("-inf")
+    has_verts = False
+
     for mesh in meshes:
         raw_verts = _get(mesh, "vertices") or []
         verts = unwrap_chunks(raw_verts if isinstance(raw_verts, list) else list(raw_verts))
@@ -547,25 +367,34 @@ def mesh_to_ifc(
             mesh_cache.append(None)
             continue
         ms = _resolve_scale(mesh, obj_scale)
-        # Pre-scale vertices once, reuse in Pass 2
         scaled = [float(v) * ms for v in verts]
-        mesh_cache.append((verts, ms, scaled))
-        all_scaled.extend(scaled)
+        mesh_cache.append(scaled)
+        has_verts = True
 
-    if not all_scaled:
+        # Update bounding box from this mesh's scaled vertices
+        for i in range(0, len(scaled) - 2, 3):
+            x, y, z = scaled[i], scaled[i + 1], scaled[i + 2]
+            if x < xmin: xmin = x
+            if x > xmax: xmax = x
+            if y < ymin: ymin = y
+            if y > ymax: ymax = y
+            if z < zmin: zmin = z
+
+    if not has_verts:
         return None, None
 
-    ox, oy, oz = compute_origin(all_scaled)
+    ox = (xmin + xmax) / 2.0
+    oy = (ymin + ymax) / 2.0
+    oz = zmin
 
     # ------------------------------------------------------------------ #
-    # Pass 2: one faceset per mesh — reuse cached verts, only unpack faces
+    # Pass 2: one faceset per mesh -- reuse cached verts, only unpack faces
     # ------------------------------------------------------------------ #
     geom_items = []
 
-    for mesh, cached in zip(meshes, mesh_cache):
-        if cached is None:
+    for mesh, scaled in zip(meshes, mesh_cache):
+        if scaled is None:
             continue
-        verts, ms, scaled = cached
         raw_faces = _get(mesh, "faces") or []
         faces_raw = unwrap_chunks(raw_faces if isinstance(raw_faces, list) else list(raw_faces))
 
@@ -575,7 +404,7 @@ def mesh_to_ifc(
         try:
             face_groups = decode_faces(faces_raw)
         except Exception as e:
-            print(f"  ⚠️  Face decode error: {e}")
+            print(f"  Warning: Face decode error: {e}")
             continue
 
         # Offset pre-scaled vertices relative to origin (flat list, no tuples)
@@ -592,8 +421,9 @@ def mesh_to_ifc(
             continue
 
         # Apply material style to every faceset of this mesh
+        # Inner meshes (from BrepX) may lack applicationId -- fall back to parent's
         if material_manager:
-            mesh_app_id = _get(mesh, "applicationId")
+            mesh_app_id = _get(mesh, "applicationId") or obj_app_id
             if mesh_app_id:
                 for fs in mesh_facesets:
                     material_manager.apply_to_item(fs, str(mesh_app_id))
@@ -614,4 +444,4 @@ def mesh_to_ifc(
     )
     placement = _make_placement(ifc, ox, oy, oz)
 
-    return rep, placement
+    return rep, placement

utils/helpers.py

@@ -0,0 +1,38 @@
+# =============================================================================
+# helpers.py
+# Shared utilities used across the exporter modules.
+# =============================================================================
+
+
+def _get(obj, key, default=None):
+    """
+    Safe access for specklepy Base objects, dicts, or any hybrid.
+    Tries attribute access first, then bracket access.
+    """
+    if obj is None:
+        return default
+    if isinstance(obj, dict):
+        return obj.get(key, default)
+    try:
+        val = getattr(obj, key, None)
+        if val is not None:
+            return val
+    except Exception:
+        pass
+    try:
+        val = obj[key]
+        if val is not None:
+            return val
+    except Exception:
+        pass
+    return default
+
+
+# Scale factors → MILLIMETRES (IFC file is declared as mm)
+MM_SCALES = {
+    "mm": 1.0,    "millimeter": 1.0,    "millimeters": 1.0,
+    "cm": 10.0,   "centimeter": 10.0,   "centimeters": 10.0,
+    "m":  1000.0, "meter": 1000.0,      "meters": 1000.0,
+    "ft": 304.8,  "foot": 304.8,        "feet": 304.8,
+    "in": 25.4,   "inch": 25.4,         "inches": 25.4,
+}

utils/instances.py

@@ -2,17 +2,17 @@
 # instances.py
 # Handles Speckle InstanceProxy objects from both:
 #
-# FORMAT A — Revit connector (our actual use case):
+# FORMAT A -- Revit connector (our actual use case):
 #   _units      = "mm"
 #   transform   = 16 floats, row-major, translation in MM
 #   definitionId = 64-char uppercase hex hash (matches object id[:32] in tree)
 #   The definition object lives somewhere in the object tree.
 #
-# FORMAT B — speckleifc IFC→Speckle converter:
+# FORMAT B -- speckleifc IFC->Speckle converter:
 #   units       = "m"
 #   transform   = 16 floats, row-major, translation in METRES
 #   definitionId = "DEFINITION:{meshAppId}"
-#   Definition geometry lives in root → Collection("definitionGeometry")
+#   Definition geometry lives in root -> Collection("definitionGeometry")
 #
 # We detect the format by the definitionId prefix.
 #
@@ -20,9 +20,14 @@
 # sharing the same definition reference a single copy of the geometry.
 # =============================================================================
 
+import hashlib
 import math
+import struct
+import ifcopenshell.api
 from specklepy.objects.base import Base
-from utils.geometry import _get, unwrap_chunks, decode_faces, _UNIT_SCALES, build_ifc_facesets, _get_shared
+from utils.helpers import _get, MM_SCALES
+from utils.geometry import unwrap_chunks, decode_faces, build_ifc_facesets, _get_shared, _is_mesh
+from utils.curves import is_curve, build_curve_rep_map
 
 
 def is_instance(obj) -> bool:
@@ -40,15 +45,18 @@ def build_definition_map(root: Base) -> dict:
     Build a unified definition map that handles both formats.
 
     Returns dict with keys:
-      "by_id"       : {obj_id_lower[:32] → object}   for Revit format
-      "by_app_id"   : {applicationId_lower → object}  for Revit format
-      "ifc_proxies" : {"DEFINITION:xxx" → proxy}      for IFC format
-      "ifc_meshes"  : {meshAppId → Mesh}               for IFC format
+      "by_id"             : {obj_id_lower[:32] -> object}   for Revit format
+      "by_app_id"         : {applicationId_lower -> object}  for Revit format
+      "ifc_proxies"       : {"DEFINITION:xxx" -> proxy}      for IFC format
+      "ifc_meshes"        : {meshAppId -> Mesh}               for IFC format
+      "definition_sources": set of applicationId (lowercase) that are definition
+                            geometry sources -- these should be skipped during export
     """
     by_id      = {}
     by_app_id  = {}
     ifc_proxies = {}
     ifc_meshes  = {}
+    definition_sources = set()
 
     # --- Walk entire tree for Revit format ---
     _collect_all(root, by_id, by_app_id, depth=0)
@@ -61,6 +69,11 @@ def build_definition_map(root: Base) -> dict:
             if app_id:
                 ifc_proxies[app_id] = proxy            # original case (for IFC format)
                 ifc_proxies[app_id.lower()] = proxy    # lowercase (for Revit format)
+            # Collect all objects referenced by this proxy as definition sources
+            object_ids = _get(proxy, "objects") or []
+            for oid in (object_ids if isinstance(object_ids, list) else [object_ids]):
+                if oid:
+                    definition_sources.add(str(oid).lower())
 
     elements = _get(root, "elements") or _get(root, "@elements") or []
     for child in (elements if isinstance(elements, list) else []):
@@ -75,12 +88,14 @@ def build_definition_map(root: Base) -> dict:
     print(f"   Objects indexed by appId:  {len(by_app_id)}")
     print(f"   IFC definition proxies:    {len(ifc_proxies)}")
     print(f"   IFC definition meshes:     {len(ifc_meshes)}")
+    print(f"   Definition sources:        {len(definition_sources)}")
 
     return {
-        "by_id":       by_id,
-        "by_app_id":   by_app_id,
-        "ifc_proxies": ifc_proxies,
-        "ifc_meshes":  ifc_meshes,
+        "by_id":              by_id,
+        "by_app_id":          by_app_id,
+        "ifc_proxies":        ifc_proxies,
+        "ifc_meshes":         ifc_meshes,
+        "definition_sources": definition_sources,
     }
 
 
@@ -92,7 +107,7 @@ def _collect_all(obj, by_id: dict, by_app_id: dict, depth: int):
     if obj_id and isinstance(obj_id, str):
         key = obj_id.lower()
         by_id[key] = obj
-        # Also store truncated — definitionId (64 chars) matches id (32 chars)
+        # Also store truncated -- definitionId (64 chars) matches id (32 chars)
         if len(key) == 32:
             by_id[key] = obj
         elif len(key) > 32:
@@ -102,7 +117,8 @@ def _collect_all(obj, by_id: dict, by_app_id: dict, depth: int):
     if app_id and isinstance(app_id, str):
         by_app_id[app_id.lower()] = obj
 
-    for key in ["elements", "@elements", "displayValue", "@displayValue",
+    for key in ["elements", "@elements", "_elements",
+                "displayValue", "@displayValue", "_displayValue",
                 "objects", "@objects", "definition", "@definition"]:
         try:
             children = obj[key]
@@ -116,11 +132,29 @@ def _collect_all(obj, by_id: dict, by_app_id: dict, depth: int):
             continue
 
 
-def _get_revit_meshes(definition_id: str, definition_map: dict) -> list:
+def _get_definition_source_object(definition_id: str, definition_map: dict):
+    """Resolve the first source object referenced by a definition proxy."""
+    ifc_proxies = definition_map.get("ifc_proxies", {})
+    proxy = ifc_proxies.get(definition_id) or ifc_proxies.get(definition_id.lower())
+    if proxy is None:
+        return None
+    object_ids = _get(proxy, "objects") or []
+    if not isinstance(object_ids, list):
+        object_ids = list(object_ids)
+    if not object_ids:
+        return None
+    by_app_id = definition_map.get("by_app_id", {})
+    return by_app_id.get(str(object_ids[0]).lower())
+
+
+def _get_revit_meshes(definition_id: str, definition_map: dict) -> tuple:
     """
     Revit format:
-      definitionId (64-char hex) → InstanceDefinitionProxy.applicationId
-      proxy.objects[0] is a UUID applicationId → find mesh by applicationId
+      definitionId (64-char hex) -> InstanceDefinitionProxy.applicationId
+      proxy.objects[0] is a UUID applicationId -> find mesh by applicationId
+
+    Returns (meshes, app_ids) where app_ids are all applicationIds encountered
+    in the resolution chain (definition objects, geometry objects) for material fallback.
     """
     from utils.geometry import get_display_meshes
 
@@ -128,40 +162,55 @@ def _get_revit_meshes(definition_id: str, definition_map: dict) -> list:
     ifc_proxies = definition_map.get("ifc_proxies", {})
     proxy = ifc_proxies.get(definition_id) or ifc_proxies.get(definition_id.lower())
     if proxy is None:
-        return []
+        return [], []
 
     # Step 2: get the mesh applicationIds from proxy.objects
     object_ids = _get(proxy, "objects") or []
     if not isinstance(object_ids, list):
         object_ids = list(object_ids)
 
-    # Step 3: look up each mesh by applicationId
+    # Step 3: look up each mesh by applicationId, collecting all encountered app IDs
     by_app_id = definition_map.get("by_app_id", {})
     meshes = []
+    encountered_app_ids = []
     for oid in object_ids:
         obj = by_app_id.get(str(oid).lower())
         if obj is not None:
+            # Collect this object's applicationId
+            obj_aid = _get(obj, "applicationId")
+            if obj_aid:
+                encountered_app_ids.append(str(obj_aid))
+            # Also collect applicationIds from displayValue items (BrepX, etc.)
+            for key in ["displayValue", "@displayValue", "_displayValue"]:
+                display = _get(obj, key)
+                if display:
+                    items = display if isinstance(display, list) else [display]
+                    for item in items:
+                        item_aid = _get(item, "applicationId")
+                        if item_aid:
+                            encountered_app_ids.append(str(item_aid))
+                    break
             # The found object may itself be a mesh, or contain displayValue meshes
             found_meshes = get_display_meshes(obj)
             if found_meshes:
                 meshes.extend(found_meshes)
-            else:
-                # It IS the mesh directly
+            elif _is_mesh(obj):
                 meshes.append(obj)
-    return meshes
+    return meshes, encountered_app_ids
 
 
-def _get_ifc_meshes(definition_id: str, definition_map: dict) -> list:
+def _get_ifc_meshes(definition_id: str, definition_map: dict) -> tuple:
     """
     IFC format: definitionId = "DEFINITION:224058_mat0"
-    Look up proxy → objects list → meshes from ifc_meshes dict.
+    Look up proxy -> objects list -> meshes from ifc_meshes dict.
+    Returns (meshes, []) -- no extra app_ids needed, mesh applicationIds match directly.
     """
     ifc_proxies = definition_map.get("ifc_proxies", {})
     ifc_meshes  = definition_map.get("ifc_meshes", {})
 
     proxy = ifc_proxies.get(definition_id)
     if proxy is None:
-        return []
+        return [], []
 
     object_ids = _get(proxy, "objects") or []
     result = []
@@ -169,20 +218,20 @@ def _get_ifc_meshes(definition_id: str, definition_map: dict) -> list:
         mesh = ifc_meshes.get(str(oid))
         if mesh is not None:
             result.append(mesh)
-    return result
+    return result, []
 
 
 def _resolve_instance_scale(obj, stream_scale: float) -> float:
     """
     Resolve scale for the transform translation.
     Tries bracket access for '_units' (Revit uses underscore).
-    IFC format instances have units="m" → scale=1.0 (no scaling).
+    IFC format instances have units="m" -> scale=1.0 (no scaling).
     """
     for key in ["units", "_units"]:
         try:
             units = obj[key]
             if units and isinstance(units, str):
-                s = _UNIT_SCALES.get(units.lower().strip())
+                s = MM_SCALES.get(units.lower().strip())
                 if s is not None:
                     return s
         except Exception:
@@ -193,39 +242,62 @@ def _resolve_instance_scale(obj, stream_scale: float) -> float:
 # Stats
 _stats = {"found": 0, "not_found": 0}
 
-# Cache: mesh id → (verts_scaled, face_groups) to avoid re-unpacking
+# Cache: mesh id -> (verts_scaled, face_groups) to avoid re-unpacking
 # AND re-scaling the same definition mesh across many instances that share it.
 _mesh_data_cache: dict = {}
 
-# Cache: definition_id → IfcRepresentationMap (or None if no geometry)
+# Cache: definition_id -> IfcRepresentationMap (or None if no geometry)
 # All instances sharing the same definition reuse one geometry copy.
 _rep_map_cache: dict = {}
 
+# Cache: geometry content hash -> IfcRepresentationMap
+# Enables sharing across different definitionIds that have identical geometry.
+_geometry_hash_cache: dict = {}
+
 # Shared identity placement for all instances (keyed by ifc file id)
 _identity_placement_cache: dict[int, object] = {}
 
 
-_MM_SCALES = {
-    "mm": 1.0, "millimeter": 1.0, "millimeters": 1.0,
-    "cm": 10.0, "centimeter": 10.0,
-    "m": 1000.0, "meter": 1000.0, "meters": 1000.0,
-    "ft": 304.8, "in": 25.4,
-}
+# --------------------------------------------------------------------------- #
+# Geometry content hashing
+# --------------------------------------------------------------------------- #
+
+def _hash_mesh_data(mesh_data_list: list, material_key: str = "") -> str:
+    """Compute a content hash from mesh geometry data for deduplication.
+
+    mesh_data_list: list of (verts_local, face_groups) tuples
+    material_key:   string identifying the material (included in hash)
+    Returns: hex digest string
+    """
+    h = hashlib.md5(usedforsecurity=False)
+    for verts_local, face_groups in mesh_data_list:
+        # Hash rounded vertices as packed floats (faster than str conversion)
+        for i in range(0, len(verts_local), 3):
+            h.update(struct.pack("3f",
+                round(verts_local[i], 3),
+                round(verts_local[i+1], 3),
+                round(verts_local[i+2], 3),
+            ))
+        # Hash face indices
+        for face in face_groups:
+            h.update(struct.pack(f"{len(face)}i", *face))
+        # Separator between meshes
+        h.update(b"|")
+    if material_key:
+        h.update(material_key.encode())
+    return h.hexdigest()
 
 
 # --------------------------------------------------------------------------- #
-# IfcRepresentationMap builder — geometry created once per definition
+# IfcRepresentationMap builder -- geometry created once per definition
 # --------------------------------------------------------------------------- #
 
-def _build_rep_map(ifc, body_context, meshes: list, ifc_format: bool,
-                   material_manager=None):
-    """
-    Build an IfcRepresentationMap from definition meshes.
-    Geometry is in local coordinates (mm, no instance transform applied).
-    Returns IfcRepresentationMap or None if no valid geometry.
-    """
-    geom_items = []
+def _collect_mesh_data(meshes: list, ifc_format: bool) -> list:
+    """Unpack, scale, and cache mesh vertex/face data.
 
+    Returns list of (mesh_obj, verts_local, face_groups) tuples.
+    """
+    result = []
     for mesh in meshes:
         mesh_id = _get(mesh, "id") or _get(mesh, "applicationId")
         if mesh_id and mesh_id in _mesh_data_cache:
@@ -233,48 +305,103 @@ def _build_rep_map(ifc, body_context, meshes: list, ifc_format: bool,
         else:
             raw_verts = _get(mesh, "vertices") or []
             raw_faces = _get(mesh, "faces") or []
-            verts     = unwrap_chunks(list(raw_verts))
-            faces_raw = unwrap_chunks(list(raw_faces))
+            verts     = unwrap_chunks(raw_verts if isinstance(raw_verts, list) else list(raw_verts))
+            faces_raw = unwrap_chunks(raw_faces if isinstance(raw_faces, list) else list(raw_faces))
             if not verts or not faces_raw:
                 continue
 
             mesh_units = _get(mesh, "units") or _get(mesh, "_units") or ("m" if ifc_format else "mm")
-            ms = _MM_SCALES.get(mesh_units.lower().strip(), 1.0)
+            ms = MM_SCALES.get(mesh_units.lower().strip(), 1.0)
 
             try:
                 face_groups = decode_faces(faces_raw)
             except Exception as e:
-                print(f"  ⚠️  Instance face decode: {e}")
+                print(f"  Warning: Instance face decode: {e}")
                 continue
 
-            # Scale vertices once and cache the result
             verts_local = [float(v) * ms for v in verts]
 
             if mesh_id:
                 _mesh_data_cache[mesh_id] = (verts_local, face_groups)
 
-        mesh_facesets = build_ifc_facesets(ifc, verts_local, face_groups)
+        result.append((mesh, verts_local, face_groups))
+    return result
 
+
+def _resolve_material_key(meshes_data: list, material_manager, fallback_app_ids, definition_id) -> str:
+    """Build a material cache key string for geometry hashing."""
+    if not material_manager:
+        return ""
+    parts = []
+    for mesh, _, _ in meshes_data:
+        mesh_app_id = _get(mesh, "applicationId")
+        style = material_manager.get_style_with_fallbacks(
+            primary_app_id=str(mesh_app_id) if mesh_app_id else None,
+            fallback_app_ids=fallback_app_ids,
+            definition_id=definition_id,
+        )
+        parts.append(str(id(style)) if style else "")
+    return "|".join(parts)
+
+
+def _build_rep_map(ifc, body_context, meshes: list, ifc_format: bool,
+                   material_manager=None, fallback_app_ids: list = None,
+                   definition_id: str = None):
+    """
+    Build an IfcRepresentationMap from definition meshes.
+    Uses content-based hashing to reuse identical geometry across different
+    definitionIds. Returns IfcRepresentationMap or None if no valid geometry.
+    """
+    # Step 1: Collect and cache raw mesh data (no IFC entities created yet)
+    meshes_data = _collect_mesh_data(meshes, ifc_format)
+    if not meshes_data:
+        return None
+
+    # Step 2: Compute content hash to check for identical geometry
+    mat_key = _resolve_material_key(meshes_data, material_manager, fallback_app_ids, definition_id)
+    geom_hash = _hash_mesh_data(
+        [(verts, faces) for _, verts, faces in meshes_data],
+        material_key=mat_key,
+    )
+
+    if geom_hash in _geometry_hash_cache:
+        return _geometry_hash_cache[geom_hash]
+
+    # Step 3: No match -- build IFC geometry entities
+    geom_items = []
+
+    for mesh, verts_local, face_groups in meshes_data:
+        mesh_facesets = build_ifc_facesets(ifc, verts_local, face_groups)
         if not mesh_facesets:
             continue
 
-        # Apply material style to each faceset
         if material_manager:
             mesh_app_id = _get(mesh, "applicationId")
-            if mesh_app_id:
+            style = material_manager.get_style_with_fallbacks(
+                primary_app_id=str(mesh_app_id) if mesh_app_id else None,
+                fallback_app_ids=fallback_app_ids,
+                definition_id=definition_id,
+            )
+            if style:
                 for fs in mesh_facesets:
-                    material_manager.apply_to_item(fs, str(mesh_app_id))
+                    try:
+                        ifcopenshell.api.run(
+                            "style.assign_item_style", ifc,
+                            item=fs, style=style,
+                        )
+                        material_manager._apply_count += 1
+                    except Exception:
+                        pass
 
         geom_items.extend(mesh_facesets)
 
     if not geom_items:
+        _geometry_hash_cache[geom_hash] = None
         return None
 
-    # Mapping origin = identity (local coords origin) — reuse shared origin
     shared = _get_shared(ifc)
     a2p    = ifc.createIfcAxis2Placement3D(shared["origin_0"], None, None)
 
-    # The mapped representation holds the actual geometry
     mapped_rep = ifc.createIfcShapeRepresentation(
         ContextOfItems=body_context,
         RepresentationIdentifier="Body",
@@ -282,41 +409,57 @@ def _build_rep_map(ifc, body_context, meshes: list, ifc_format: bool,
         Items=geom_items,
     )
 
-    return ifc.createIfcRepresentationMap(a2p, mapped_rep)
+    rep_map = ifc.createIfcRepresentationMap(a2p, mapped_rep)
+    _geometry_hash_cache[geom_hash] = rep_map
+    return rep_map
 
 
 # --------------------------------------------------------------------------- #
-# Transform → IfcCartesianTransformationOperator3D
+# Transform -> IfcCartesianTransformationOperator3D
 # --------------------------------------------------------------------------- #
 
 def _vec_magnitude(x, y, z):
     return math.sqrt(x*x + y*y + z*z)
 
 
+# Cache: rounded direction tuple -> IfcDirection entity (keyed by ifc file id)
+_direction_cache: dict[int, dict] = {}
+
+def _get_or_create_direction(ifc, dx, dy, dz):
+    """Return a cached IfcDirection or create and cache a new one."""
+    fid = id(ifc)
+    if fid not in _direction_cache:
+        _direction_cache[fid] = {}
+    cache = _direction_cache[fid]
+    # Round to 6 decimals -- sufficient for unit vectors
+    key = (round(dx, 6), round(dy, 6), round(dz, 6))
+    if key not in cache:
+        cache[key] = ifc.createIfcDirection([key[0], key[1], key[2]])
+    return cache[key]
+
+
 def _make_transform_operator(ifc, t: list, ts: float):
     """
     Convert a row-major 4x4 matrix + translation scale into an
     IfcCartesianTransformationOperator3DnonUniform.
 
     t:  16 floats, row-major [r00,r01,r02,tx, r10,r11,r12,ty, r20,r21,r22,tz, 0,0,0,1]
-    ts: scale factor for translation components (e.g. 1000.0 for m→mm)
-
-    The matrix acts as: p' = M * p + translation, where M rows are:
-      row0 = (t[0], t[1], t[2])
-      row1 = (t[4], t[5], t[6])
-      row2 = (t[8], t[9], t[10])
+    ts: scale factor for translation components (e.g. 1000.0 for m->mm)
 
     IfcCartesianTransformationOperator axes represent the COLUMNS of M:
-      Axis1 = column 0 = where local X maps → (t[0], t[4], t[8])
-      Axis2 = column 1 = where local Y maps → (t[1], t[5], t[9])
-      Axis3 = column 2 = where local Z maps → (t[2], t[6], t[10])
+      Axis1 = column 0 = where local X maps -> (t[0], t[4], t[8])
+      Axis2 = column 1 = where local Y maps -> (t[1], t[5], t[9])
+      Axis3 = column 2 = where local Z maps -> (t[2], t[6], t[10])
+
+    Always uses the non-uniform variant with explicit Axis3 to ensure
+    correct orientation for all transform types (mirrors, non-orthogonal, etc.).
 
     Returns the IFC entity, or None if the transform is degenerate.
     """
     # Extract COLUMNS of the 3x3 rotation/scale sub-matrix
-    ax1 = (float(t[0]), float(t[4]), float(t[8]))    # column 0: X-axis direction
-    ax2 = (float(t[1]), float(t[5]), float(t[9]))    # column 1: Y-axis direction
-    ax3 = (float(t[2]), float(t[6]), float(t[10]))   # column 2: Z-axis direction
+    ax1 = (float(t[0]), float(t[4]), float(t[8]))
+    ax2 = (float(t[1]), float(t[5]), float(t[9]))
+    ax3 = (float(t[2]), float(t[6]), float(t[10]))
 
     s1 = _vec_magnitude(*ax1)
     s2 = _vec_magnitude(*ax2)
@@ -325,37 +468,41 @@ def _make_transform_operator(ifc, t: list, ts: float):
     if s1 < 1e-10 or s2 < 1e-10 or s3 < 1e-10:
         return None  # degenerate transform
 
-    # Normalized direction vectors
-    d1 = ifc.createIfcDirection([ax1[0]/s1, ax1[1]/s1, ax1[2]/s1])
-    d2 = ifc.createIfcDirection([ax2[0]/s2, ax2[1]/s2, ax2[2]/s2])
-    d3 = ifc.createIfcDirection([ax3[0]/s3, ax3[1]/s3, ax3[2]/s3])
+    # Normalized direction vectors -- reuse cached IfcDirection entities
+    d1 = _get_or_create_direction(ifc, ax1[0]/s1, ax1[1]/s1, ax1[2]/s1)
+    d2 = _get_or_create_direction(ifc, ax2[0]/s2, ax2[1]/s2, ax2[2]/s2)
+    d3 = _get_or_create_direction(ifc, ax3[0]/s3, ax3[1]/s3, ax3[2]/s3)
 
-    # Translation, scaled to mm
-    tx = float(t[3])  * ts
-    ty = float(t[7])  * ts
-    tz = float(t[11]) * ts
+    # Translation, scaled and rounded to mm
+    tx = round(float(t[3])  * ts, 3)
+    ty = round(float(t[7])  * ts, 3)
+    tz = round(float(t[11]) * ts, 3)
     origin = ifc.createIfcCartesianPoint([tx, ty, tz])
 
-    # Use non-uniform variant to handle mirrors and non-uniform scale
+    # Round scales for cleaner output
+    s1 = round(s1, 6)
+    s2 = round(s2, 6)
+    s3 = round(s3, 6)
+
     return ifc.createIfcCartesianTransformationOperator3DnonUniform(
         d1,      # Axis1
         d2,      # Axis2
         origin,  # LocalOrigin
         s1,      # Scale
-        d3,      # Axis3
+        d3,      # Axis3 (explicit -- never derived)
         s2,      # Scale2
         s3,      # Scale3
     )
 
 
 # --------------------------------------------------------------------------- #
-# Main conversion — IfcMappedItem approach
+# Main conversion -- IfcMappedItem approach
 # --------------------------------------------------------------------------- #
 
 def instance_to_ifc(ifc, body_context, obj: Base, definition_map: dict,
                     scale: float = 1.0, material_manager=None):
     """
-    Convert a Speckle InstanceProxy → (IfcShapeRepresentation, IfcLocalPlacement).
+    Convert a Speckle InstanceProxy -> (IfcShapeRepresentation, IfcLocalPlacement).
 
     Strategy: create geometry once per definition as an IfcRepresentationMap,
     then reference it via IfcMappedItem + IfcCartesianTransformationOperator3D
@@ -371,11 +518,11 @@ def instance_to_ifc(ifc, body_context, obj: Base, definition_map: dict,
     definition_id = _get(obj, "definitionId") or ""
     ifc_format    = _is_ifc_format(definition_id)
 
-    # Translation scale: IFC format transform is in metres → convert to mm
+    # Translation scale: IFC format transform is in metres -> convert to mm
     # Revit format transform is already in mm (same as IFC file units)
     ts = 1000.0 if ifc_format else _resolve_instance_scale(obj, scale)
 
-    # Identity placement (transform is encoded in the MappedItem) — shared across all instances
+    # Identity placement (transform is encoded in the MappedItem) -- shared across all instances
     fid = id(ifc)
     if fid not in _identity_placement_cache:
         shared = _get_shared(ifc)
@@ -386,19 +533,42 @@ def instance_to_ifc(ifc, body_context, obj: Base, definition_map: dict,
     # --- Get or build IfcRepresentationMap (cached per definition_id) ---
     if definition_id not in _rep_map_cache:
         if ifc_format:
-            meshes = _get_ifc_meshes(definition_id, definition_map)
+            meshes, extra_app_ids = _get_ifc_meshes(definition_id, definition_map)
         else:
-            meshes = _get_revit_meshes(definition_id, definition_map)
+            meshes, extra_app_ids = _get_revit_meshes(definition_id, definition_map)
 
-        if not meshes:
+        # Build fallback app_id list: instance's own + definition chain IDs
+        instance_app_id = _get(obj, "applicationId")
+        fallback_ids = []
+        if instance_app_id:
+            fallback_ids.append(str(instance_app_id))
+        fallback_ids.extend(extra_app_ids)
+
+        rep_map_result = None
+        if meshes:
+            rep_map_result = _build_rep_map(
+                ifc, body_context, meshes, ifc_format, material_manager,
+                fallback_app_ids=fallback_ids,
+                definition_id=definition_id,
+            )
+
+        # If no mesh geometry produced, try curve geometry from the definition object
+        if rep_map_result is None:
+            curve_obj = _get_definition_source_object(definition_id, definition_map)
+            if curve_obj and is_curve(curve_obj):
+                curve_scale = _resolve_instance_scale(curve_obj, 1.0)
+                rep_map_result = build_curve_rep_map(
+                    ifc, body_context, curve_obj, scale=curve_scale,
+                    material_manager=material_manager,
+                    fallback_app_ids=fallback_ids,
+                    definition_id=definition_id,
+                )
+
+        _rep_map_cache[definition_id] = rep_map_result
+        if rep_map_result is not None:
+            _stats["found"] += 1
+        else:
             _stats["not_found"] += 1
-            _rep_map_cache[definition_id] = None
-            return None, placement
-
-        _stats["found"] += 1
-        _rep_map_cache[definition_id] = _build_rep_map(
-            ifc, body_context, meshes, ifc_format, material_manager
-        )
     else:
         # Track stats even for cached definitions
         if _rep_map_cache[definition_id] is not None:
@@ -436,34 +606,34 @@ def get_definition_object(obj: Base, definition_map: dict):
     definition_id = _get(obj, "definitionId") or ""
     if not definition_id:
         return None
+    return _get_definition_source_object(definition_id, definition_map)
 
-    ifc_proxies = definition_map.get("ifc_proxies", {})
-    proxy = ifc_proxies.get(definition_id) or ifc_proxies.get(definition_id.lower())
-    if proxy is None:
-        return None
 
-    object_ids = _get(proxy, "objects") or []
-    if not isinstance(object_ids, list):
-        object_ids = list(object_ids)
-    if not object_ids:
-        return None
-
-    by_app_id = definition_map.get("by_app_id", {})
-    source = by_app_id.get(str(object_ids[0]).lower())
-    return source
+def is_definition_source(obj, definition_map: dict) -> bool:
+    """Return True if this object is a definition geometry source (should not be exported standalone)."""
+    app_id = _get(obj, "applicationId")
+    if not app_id:
+        return False
+    return str(app_id).lower() in definition_map.get("definition_sources", set())
 
 
 def print_instance_stats():
     total = _stats["found"] + _stats["not_found"]
     print(f"  Instance resolution: {_stats['found']}/{total} definitions found")
     if _stats["not_found"] > 0:
-        print(f"  ⚠️  {_stats['not_found']} instances had no definition geometry")
+        print(f"  Warning: {_stats['not_found']} instances had no definition geometry")
+    unique_defs = len(_rep_map_cache)
+    unique_geom = len([v for v in _geometry_hash_cache.values() if v is not None])
+    if unique_defs > unique_geom:
+        print(f"  Geometry dedup: {unique_defs} definitions -> {unique_geom} unique geometries")
 
 
 def reset_caches():
     """Reset module-level caches (call at start of each export run)."""
     _mesh_data_cache.clear()
     _rep_map_cache.clear()
+    _geometry_hash_cache.clear()
     _identity_placement_cache.clear()
+    _direction_cache.clear()
     _stats["found"] = 0
     _stats["not_found"] = 0

utils/materials.py

@@ -64,6 +64,7 @@ class MaterialManager:
         self._style_map: dict[str, object] = {}
         # name → IfcSurfaceStyle (cache to avoid duplicates)
         self._style_cache: dict[str, object] = {}
+        self._apply_count: int = 0
         self._build(root)
 
     def _build(self, root: Base):
@@ -135,6 +136,24 @@ class MaterialManager:
         self._style_map[key] = style
         return style
 
+    def get_style_with_fallbacks(self, primary_app_id: str = None,
+                                 fallback_app_ids: list = None,
+                                 definition_id: str = None):
+        """Try primary app_id first, then each fallback, then definition_id. Return style or None."""
+        if primary_app_id:
+            style = self.get_style(primary_app_id)
+            if style:
+                return style
+        for aid in (fallback_app_ids or []):
+            style = self.get_style(aid)
+            if style:
+                return style
+        if definition_id:
+            style = self.get_style(definition_id)
+            if style:
+                return style
+        return None
+
     def apply_to_item(self, item, mesh_app_id: str):
         """Assign the material style to a single IFC geometry item (e.g. IfcPolygonalFaceSet)."""
         style = self.get_style(mesh_app_id)