Blend::Curvature now builds the capi's section, as fork 6148773's capi
does: SM_FSG_BLEND_CURVE of order 2 at the equal-area thumbweight,
passed by solidcore's name GN_THUMBWEIGHT_EQUAL_AREA (solidcore#2274
moves its value from -2 to 2), with no blend-curve-at-thirds setting.
The sync-2 refusal (e9ac0e606) comes off.
One call resolves the equal-area thumbweight once, from the first
blend's wedge, and every edge takes it, in the fork as in solidcore: an
edge of another wedge then takes off the wrong area, depending on the
edges' order (17% short, 21% over;
solidcore#2250 issuecomment-6563419897). A curvature fillet over edges
whose faces meet at different angles (at each edge's midpoint, within
1e-6 rad) is refused as Error::Unported citing that comment; the call
is not split per edge.
Against closed forms on the rig, solidcore within 1e-9 of SMLib
d01d7b0: 862 7845.48670 (7845.48668), 2708 11791.330160 (11791.330159),
every-box-edge-r3 and section box-edge-r3 23922.743338, boss-root-r2
26056.152951, pipe-on-plate-r2 40302.590788 (completes, #2264),
hole-top-rim-r1 15207.56176, crease-r2 13963.15756, l-bracket 7538.62833,
peg front 9383.62724, all within 3e-9 relative. The G2 corner cases
pocket-corner and the bar corner's three edges (two variants) build and
fail continuity G1 at one seam, as on SMLib; the oblique three-edge
corner is refused for its mixed wedges. The kernel's and engine's
curvature tests now pass.
a_curvature_round_takes_its_wedges_equal_area_and_mixed_wedges_are_refused
holds each wedge's r² (cot(α/2) - (π - α)/2) to 1e-6 on a chamfered box
(135° and 90°), both orders refused, two 90° edges built together.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SdcgEu4Lr3moCCisdZShKz