GeoSampler boundary¶

GeoSampler owns CAD loading, face/edge selection, geometry-aware sampling and normal construction. Margam consumes explicitly prepared point sets; it does not reimplement those tools or add a 3D viewport.

The implemented bridge accepts GeoSampler schema-v2 NPZ exports for datasets on a common axis-aligned plane. It reads _gs_metadata, labelled array references, optional entity tags and normal-array references without importing GeoSampler, Gmsh, SciPy or xarray. The user explicitly selects source axes and labels. A dropped coordinate must be constant across selected sets; normal components along that coordinate must be zero. Non-planar data is rejected rather than projected silently.

python -m examples.prepare_geosampler_2d \
  --input /absolute/path/samples.npz \
  --output /absolute/path/new-prepared-directory \
  --axes x,y --labels interior wall

The destination must be new. It contains dataset-000.npz, etc., plus mapping.json with original labels, source hash, source axes, plane coordinate and unchanged units metadata. Map each prepared dataset explicitly in Margam: points for coordinates, normals only when present, and no target values until physics is assigned. The built-in scalar/flow builder uses named side slots; Prepared 2D scalar models accept user-named sets. Neither naming scheme infers CAD topology. Arbitrary Dirichlet point sets do not need normals. Normal-dependent operators on off-box sets require explicit unit normals with outward orientation supplied by the data provider. No units or normal sign is inferred.

For imported non-rectangular clouds the rectangle is only a normalisation/evaluation bounding box. The optional sample preview says so. Its regular output grid includes all bounding-box points; the application does not infer a fluid/solid mask. Choose an appropriate explicit evaluation workflow before interpreting values outside a complex domain. The CSV export is not a CAD or ParaView integration.

Full 3D coordinate/field support, non-axis-aligned reductions, NetCDF manifests, automatic region-to-BC linking and live CAD selection remain deferred extensions. The implemented planar bridge is tested against the upstream schema contract; no claim of validating every CAD model or upstream release is made.

Use prepared labels in a non-rectangular scalar model¶

The Interior data and Boundary data blocks now accept the prepared coordinate/normal arrays. Choose Prepared 2D points, map the interior file, add each selected boundary label, then assign scalar conditions downstream. Labels have no inferred normal direction; normal-dependent conditions require explicit outward unit arrays. The existing preparation helper remains separate from graph editing; automatic mapping.json-to-graph wiring is not implemented.