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
Blend: Refusals: size the remaining classes; refuse the collapsed offset #2401
Carried over from tangency-labs/smlib#91: SMLib construction ended on 2026-11-02, so this continues in solidcore. The generic blend's design is docs/generic-blend.md; coordination is on #2243.
Prerequisite: Opus 5.5 at effort xhigh or max.
Part of the generic rolling-ball blend: one construction for every pair of faces, built on SmBlendCorner for corners. Each piece deletes the old code it replaces. Tests come first, every commit builds and passes on its own, and fixes are root causes only.
In progress. These are the refusals without a size through smlib_blend_fit_size:
the open-end causes;
a polygon face taken whole, where a trace failure comes first;
the collapsed offset (middle_edge_of_an_arc_tangent_chain_r2), to be told apart from the tray by its geometry.
Claimed in #2431 (draft): sizing the refusals that have no size (the open-end causes, a polygon face taken whole, the collapsed offset told apart from the tray by its geometry), with tests first. The fit's shared margin is also in scope, per cause, as #2415 found.
Written by Claude Code (Anthropic) on behalf of @brett19.
Draft #2431 now sizes the remaining classes. Each is decided from its own geometry, under fillet_gn_section_generator, with tests in crates/solidcore/tests/issue_2401_refusal_sizes.rs. Closed forms, check_brep Level2.
Polygon taken whole: judged before the corners and the trace (the polygon fit's depth reached by the least ball, to the edge's tolerance).
The square top chamfered 5 is refused FaceConsumed naming 4.99, which builds.
jyviko's 10 x 6 top at 3.5 named 0.094 (BandsOverlap); it now names 2.99, which builds.
Collapsed offset: a ball as wide as a face (QuadricKind::Line/Point offset) has its centre fixed. It is refused (OffsetCollapses) where the plane through an end of the edge, square to it, misses that centre.
The arc chain's middle edge at r2 misses it by 0.19 at its top end and names 2 - 0.04. At 2.5 the ball-fit refusal now names 1.96, where it named 2.
Open ends:CliffEndFaceNotPlanar/CliffEndNotCut, and a trimmed blend no corner takes, name LargestSizeOnItsFaces (open). A box with a concave cylindrical end, rounded 12, names 9.99, which builds.
Activity
brett19 commentedon Nov 2, 2026
Claimed in #2431 (draft): sizing the refusals that have no size (the open-end causes, a polygon face taken whole, the collapsed offset told apart from the tray by its geometry), with tests first. The fit's shared margin is also in scope, per cause, as #2415 found.
Written by Claude Code (Anthropic) on behalf of @brett19.
fillet: the remaining refusal classes name a size from their own geom…
brett19 commentedon Nov 2, 2026
Draft #2431 now sizes the remaining classes. Each is decided from its own geometry, under
fillet_gn_section_generator, with tests incrates/solidcore/tests/issue_2401_refusal_sizes.rs. Closed forms,check_brepLevel2.FaceConsumednaming 4.99, which builds.BandsOverlap); it now names 2.99, which builds.QuadricKind::Line/Pointoffset) has its centre fixed. It is refused (OffsetCollapses) where the plane through an end of the edge, square to it, misses that centre.CliffEndFaceNotPlanar/CliffEndNotCut, and a trimmed blend no corner takes, nameLargestSizeOnItsFaces(open). A box with a concave cylindrical end, rounded 12, names 9.99, which builds.LargestSizeOnItsFacesdoes not tell a footprint the rail runs along from one it touches at a point, and the bossed can builds at 4.3 only with fix: keep features sweeps a boss's walls apart, so a round just under its footprint builds (part of #2404) #2415.It stays draft for SolidCraft's rig.
Written by Claude Code (Anthropic) on behalf of @brett19.
fillet: a collapsed offset's chain end that runs on into another blen…
fix: blend refusals size the remaining classes and refuse the collaps…
Merge pull request #2431 from brett19/fix/2401-refusal-sizes
Blend outline: review fixes
test: the fork's number back wherever the fit comes from the port, as…
test: the fork's number back wherever the fit comes from the port, as…
fix: a face as wide as the ball along part of the edge is refused wit…
test: the fork's number back wherever the fit comes from the port, as…
Fillet: a round as large as the round it runs along is refused, and a…