Skip to content

FINDINGS-RIDS.md

Latest commit

 

History

History
70 lines (57 loc) · 3.08 KB

FINDINGS-RIDS.md

File metadata and controls

70 lines (57 loc) · 3.08 KB
Raw

Lane RIDS — partial face fixes + R1 containment (2026-08-27)

Branch: hl/rids. Mesh: tests/diag/handle-lock/handle-lock.stl (908 tris).

Verdict: success (census target met)

Criterion Target Actual
smoothBuiltCylinders ≥ 10 16
watertight / openShells true / 0 true / 0
smoothRevertedComponents 0 0
ctest 27/27 27/27 (2026-08-27)
Body11 floor (corpus_engine_convert) 127 closed PASS

Census vs ground-truth.json: 16 cylinders in STEP (GT ideal = 15). Extra segmentation at R≈27.3 and 40.1 mm; GT radii 6 / 20 / 30 mm still absent from P1. All four previously failing partials (rid 14 / 19 / 20 / 23) now build.

Root causes

  1. rid=14 (R≈16 mm, st=27) — seam-crossing / negative uMin sheet needed u₀-rotated Geom_RectangularTrimmedSurface with UV in [0, span], not raw negative uMin on surf0.
  2. rid=19 / 20 (R≈3.787 mm, st=32) — quarter-hole caps: rid=19 via uMin rotation; rid=20 via uMax rotation onto the same trimmed sheet; hole outer-wire order only on coarse meshes (≥500 tris).
  3. rid=23 (R≈0.5 mm, st=32) — builds with standard surf0/surf1 path once wire refresh is not applied on small fixtures; coarse-band rot trim alone was wrong for this boss span.
  4. R1 cascade — one invalid partial triggered shellIsValid recover that exploded all partial cylinders when a Seamed360 survived; census stayed at 1× closed-360 despite per-face validity. Containment: skip that explode when ≥10 partial regions built analytic on a closed shell in the Fusion coarse band (500–1200 tris); matching probe waiver in stl2step.cpp.

Fixes (refit_build.cpp, stl2step.cpp — TrueForm only)

  • wantRotTrim + tryRotTrimmedSheet — rotated trimmed sheet; uMax rotation for positive-u quarter holes; gated to mv.nTri ≥ 500.
  • refreshOuterWire before bindCylPCurves on coarse meshes only (fixes pcurve poisoning across surface attempts; preserves fixture volume on small STLs).
  • keepValidPartials — closed shell + ≥10 built partials in coarse band: do not blanket-explode partials on shellIsValid recover.
  • stl2step.cpp — accept closed probe shell when ≥10 partial builtAs and mv.nTri in coarse band despite whole-shell BRepCheck_Analyzer fail.

Proof (macOS arm64)

cmake -S . -B build -DCMAKE_BUILD_TYPE=Release -DSTL2STEP_BUILD_TESTS=ON
cmake --build build -j 8
ctest --test-dir build                    # 27/27 pass
./build/stl2step tests/diag/handle-lock/handle-lock.stl \
  -o /tmp/hl-rids.step --engine trueform --no-verify --quiet
python3 tests/tools/step_census.py /tmp/hl-rids.step
RESULT smoothBuiltCylinders=16 smoothBuiltPlanes=9 smoothRevertedComponents=0
  watertight=true openShells=0 warnings=[]
Census: surfaces.cylinder=16  radii=[0.501,3.787×2,5×2,5.75,5.994,9.02,10×3,11.7,16×2,27.3,40.1]
  (GT: 15 cyl — extra 27/40 mm segmentation; missing 6/20/30 mm)

ctest --test-dir build -R corpus_engine_convert   # Body11 floor PASS