One IFCGRID without ObjectPlacement (schema-optional) throws in getAbsolutePlacement — the catch in GridReader.read then drops EVERY grid in the file, silently

[stokuewe]

Versions

@thatopen/fragments 3.4.7 (verified against the published dist/index.mjs; source on main: packages/fragments/src/Importers/IfcImporter/src/geometry/grid-reader.ts, packages/fragments/src/Utils/ifc-utils.ts).

What happens

ObjectPlacement is OPTIONAL on IfcProduct, so IFCGRID('…',#15,'Achsraster',$,$,$,$,(…),(…),$) is schema-valid IFC. Convert a file that contains one such grid and:

• IfcImporter.process(...) completes normally — no thrown error, no rejected promise, no return-value signal;
• the resulting fragments buffer contains no grid data at all — the ThatOpenGrid category is absent from getCategories(), and model.getGrids() yields an empty group;
• every other grid in the same file is lost too, including fully placed ones;
• element geometry is unaffected, so the file looks perfectly converted.

The only trace is a console.error inside the importer, which in a worker-based conversion nobody reads.

Why

FragmentsIfcUtils.getAbsolutePlacement dereferences the optional attribute unconditionally:

// packages/fragments/src/Utils/ifc-utils.ts
static getAbsolutePlacement(webIfc, item, unitsFactor = this.getUnitsFactor(webIfc)) {
  const placementId = item.ObjectPlacement.value;   // ← TypeError when the IFC omits it
  …
}

→ TypeError: Cannot read properties of null (reading 'value').

GridReader.read wraps the whole grid loop in one try/catch that swallows the error and returns an empty list, so the failure of a single grid destroys the result for the entire file:

// packages/fragments/src/Importers/IfcImporter/src/geometry/grid-reader.ts
read(webIfc: WEBIFC.IfcAPI) {
  try {
    …
    for (let i = 0; i < size; i++) {
      const grid = webIfc.GetLine(0, id);
      const transform = FragmentsIfcUtils.getAbsolutePlacement(webIfc, grid, units);   // throws here
      …
      result.push(data);
    }
    return result;
  } catch (error) {
    console.error(error);
    return [] as GridData[];       // ← all-or-nothing
  }
}

Note that this is the same all-or-nothing failure that was already fixed one line below, for the other optional attribute on this path. The comment sitting in the current source says so:

// AxisTag is an optional IfcLabel. web-ifc returns it as null when the
// IFC omits it (IFCGRIDAXIS($,...)), so read it defensively; otherwise
// a single tagless axis threw and the outer catch dropped every grid.
tag: axisCurve.AxisTag?.value ?? "",

ObjectPlacement has exactly that shape (optional attribute, null from web-ifc, dereferenced without ?.) and exactly that blast radius, and it is still open. The defensive read fixed the symptom for one attribute; the structural cause — one catch around the whole loop — is untouched.

Reproduction sketch

Hand-authored IFC2X3, measured on fragments 3.4.7:

1. Start from a file with a working grid (IFCGRID with an IFCLOCALPLACEMENT, two UAxes, two VAxes, all IFCPOLYLINE). Convert → getCategories() contains ThatOpenGrid, the grid data reads back with four tagged axes.
2. Add a second IFCGRID whose ObjectPlacement is $, changing nothing else.
3. Convert the same way:

const importer = new IfcImporter();
const bytes = await importer.process({ bytes: ifcBytes });   // resolves normally

4. Load the result. Observed: getCategories() still lists IFCGRID (the entity is there) but not ThatOpenGrid; the grid item count is 0; the wall in the same file keeps its correct bounding box.

Suggested fix

Both halves are worth fixing, and neither alone is sufficient:

1. Tolerate the omitted placement. An IfcProduct without ObjectPlacement is not "broken", it is a product with no placement — i.e. the identity placement. So:

   const placementId = item.ObjectPlacement?.value;
   if (placementId === undefined || placementId === null) {
     // no placement → identity, still with the IFC→three axis conversion applied
     const ifcResult = new THREE.Matrix4();
     ifcResult.premultiply(new THREE.Matrix4().makeRotationX(-Math.PI / 2));
     return ifcResult;
   }

2. Move the try/catch inside the per-grid loop in GridReader.read, so one unreadable grid costs one grid instead of all of them — and report it with the express ID (console.warn(\Fragments: skipping grid #${id}`, error)`) rather than a bare stack for the whole file.

With (1) alone a differently-malformed grid would still wipe the file; with (2) alone a schema-valid grid would still be silently dropped.

Related: #244 (the AxisTag variant of the same failure, fixed defensively), #184 / #161 (grid reading against real exporters).