You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Round on a reopened round fails with 1006 since #2364 (smooth-vertex End); the plain B-spline builds #2550
A round on a reopened round fails on main with merge_curve_on_surfaces: upstream status 1006 since #2364 (an End at a smooth vertex). It's a regression from #2364.
Test:wr698_gn_round_is_its_definition::a_round_on_a_reopened_round_lies_on_it, #[ignore]d on main as slow. Run with --ignored on main 819a5ad, it fails in 49 s: definition true: merge_curve_on_surfaces: upstream status 1006.
What fails: only the run on the reopened definition (the round rebuilt from the record as its definition). The plain B-spline run builds.
Expected: the round builds on the reopened round as it does on the plain B-spline, with the rail within the reopened round's bound of the plain rail. Cause still to find: which of #2364's changes meets a definition face differently from its net.
Taking this one, since it's a regression from our #2364. #2364's builder is starting a bisect of its changes on the reopened definition face. A draft PR will follow and link here.
Written by Claude Code (Anthropic) on behalf of @brett19.
@jyviko#2555 is ready for your rig, at head c05497e. It changes the test only; there is no library change.
Cause:
reopen in wr698_gn_round_is_its_definition put each rebuilt ProceduralBlend into its face without an owner, so the face's surface had no SmSurface::GetFace().
For SolidCraft's reopen: a face's rebuilt surface must be owned by that face. Otherwise any blend whose rail merges onto that face fails with 1006 under fillet_blend_corner.
Written by Claude Code (Anthropic) on behalf of @brett19.
Thanks brett19, that's a clean bisect, and good to know it was the fixture, not #2364. #2555 changes no library code, so a SolidCraft rig run can't move and isn't needed; it's fine to merge on its CI. The ownership rule you found also binds the real record→kernel paths (#2480's rebuild and the layer split's conversion); that's being checked there with its own test.
Activity
brett19 commentedon Nov 3, 2026
Taking this one, since it's a regression from our #2364. #2364's builder is starting a bisect of its changes on the reopened definition face. A draft PR will follow and link here.
Written by Claude Code (Anthropic) on behalf of @brett19.
test: a reopened round goes into its face through SetSurface (tangenc…
test: the round on a reopened round stays ignored: it fails on main s…
test: the reopened round is owned without dropping its trims; run it …
brett19 commentedon Nov 3, 2026
@jyviko #2555 is ready for your rig, at head c05497e. It changes the test only; there is no library change.
Cause:
reopeninwr698_gn_round_is_its_definitionput each rebuiltProceduralBlendinto its face without an owner, so the face's surface had noSmSurface::GetFace().MergeCurveOnSurfaces). With no owning face, that merge returnsSM_ERR_NULL_POINTER(1006), as upstream would.Fix:
reopennow sets the owner and the face tolerance on the rebuilt surface, and keeps the UV trims. It does not go throughset_surface: that call's change-of-geometry notification drops the trims, and the fillet then re-drops them onto S outside its net scope (1271 evaluations against A round on a definition face is ~200x slower than on its B-spline: the trace solves S point by point (68 s vs 0.34 s) #2510's pin of 64).For SolidCraft's reopen: a face's rebuilt surface must be owned by that face. Otherwise any blend whose rail merges onto that face fails with 1006 under
fillet_blend_corner.Written by Claude Code (Anthropic) on behalf of @brett19.
jyviko commentedon Nov 3, 2026
Thanks brett19, that's a clean bisect, and good to know it was the fixture, not #2364. #2555 changes no library code, so a SolidCraft rig run can't move and isn't needed; it's fine to merge on its CI. The ownership rule you found also binds the real record→kernel paths (#2480's rebuild and the layer split's conversion); that's being checked there with its own test.
Claim #2550: round on a reopened round after the smooth-vertex End
Merge pull request #2555 from brett19/fix/2550-reopened-round-end