Problem
After #877, vcad export --strict-brep of the rana-60-cnc can.loon (ecto/rana branch 60-cnc, cad/parts-60-cnc/can.loon) produces a 379-face analytic STEP whose shell is not closed:
- STEP declares
CLOSED_SHELL / MANIFOLD_SOLID_BREP, 379 ADVANCED_FACE, 30 cylinders + 39 planes — looks right at the entity-count level
- but of 2 641
EDGE_CURVEs, 1 267 are referenced by exactly one ORIENTED_EDGE (a closed 2-manifold references every edge exactly twice; histogram {1: 1267, 2: 1374})
vcad import of the file reports "Imported 1 solid(s)" but writes a 334-byte STL (empty) — the round-trip silently fails- Shapr3D (Parasolid) opens it as loose sheets: the floor disc floats, the walls are open rings, bosses are unstitched (screenshot from the user attached in ecto/rana if needed)
Also suspicious for a 30-cyl/39-plane solid: 2 199 LINE edges, 15 795 VERTEX_POINTs — the cylindrical faces look split into many strips with unshared edges, i.e. the batched difference emitted per-face edge copies instead of a shared edge table.
Ask
- the writer must emit shared edges (one
EDGE_CURVE per topological edge, two ORIENTED_EDGEs), or the kernel must merge coincident edges after the batched boolean before writing
--strict-brep should fail on a shell whose edge-use histogram isn't all-2s — right now it certifies a non-manifold as B-repvcad import must error (not write an empty STL) when it can't build a solid- regression test: export → import round-trip volume within 0.1 % on the pocketed-disc fixture and on the can, and an assertion that every edge is used exactly twice
Repro: vcad export --strict-brep can.loon can.step && python3 -c "import re,collections;s=open('can.step').read();ec=set(re.findall(r'^#(\d+)\s*=\s*EDGE_CURVE',s,re.M));r=collections.Counter(re.findall(r'ORIENTED_EDGE\s*\(\s*\'[^\']*\'\s*,\s*\*\s*,\s*\*\s*,\s*#(\d+)',s));print(collections.Counter(r[e] for e in ec))"
Follow-up to #872 / #877.
Problem
After #877,
vcad export --strict-brepof the rana-60-cnccan.loon(ecto/rana branch60-cnc,cad/parts-60-cnc/can.loon) produces a 379-face analytic STEP whose shell is not closed:CLOSED_SHELL/MANIFOLD_SOLID_BREP, 379ADVANCED_FACE, 30 cylinders + 39 planes — looks right at the entity-count levelEDGE_CURVEs, 1 267 are referenced by exactly oneORIENTED_EDGE(a closed 2-manifold references every edge exactly twice; histogram{1: 1267, 2: 1374})vcad importof the file reports "Imported 1 solid(s)" but writes a 334-byte STL (empty) — the round-trip silently failsAlso suspicious for a 30-cyl/39-plane solid: 2 199
LINEedges, 15 795VERTEX_POINTs — the cylindrical faces look split into many strips with unshared edges, i.e. the batched difference emitted per-face edge copies instead of a shared edge table.Ask
EDGE_CURVEper topological edge, twoORIENTED_EDGEs), or the kernel must merge coincident edges after the batched boolean before writing--strict-brepshould fail on a shell whose edge-use histogram isn't all-2s — right now it certifies a non-manifold as B-repvcad importmust error (not write an empty STL) when it can't build a solidRepro:
vcad export --strict-brep can.loon can.step && python3 -c "import re,collections;s=open('can.step').read();ec=set(re.findall(r'^#(\d+)\s*=\s*EDGE_CURVE',s,re.M));r=collections.Counter(re.findall(r'ORIENTED_EDGE\s*\(\s*\'[^\']*\'\s*,\s*\*\s*,\s*\*\s*,\s*#(\d+)',s));print(collections.Counter(r[e] for e in ec))"Follow-up to #872 / #877.