Files
bherbruck b0270f41dc Tests: G2 corners measure smoothness away from both corners, and a 3.5 setback is refused (#13539, #13522) (#13896)
* claim: #13539 and #13522, the G2 corner's tests after solidcore #2988

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* kernel tests: the G2 corner's seams and the set-back 3.5 case follow solidcore #2988's Setback (#13539, #13522)

Fixes tangency-labs/solidcraft#13539
Fixes tangency-labs/solidcraft#13522

Both regressed in the solidcore bump a445bdaac6 -> c0d3cc0124 for one cause: tangency-labs/solidcore#2988 (e9a5ded95c, "a vertex marked set back between planes is the blend corner's Setback"). The same cause as tangency-labs/solidcraft#13895 (#13540 and #13542). solidcore's change is deliberate and certified: the corner's fill meets the rounds tangentially (4.71 degrees -> 0.00 on the golden box), where the N x N patch it replaces was uncertified. So the tests that encoded the old corner change, not the kernel.

**#13539: corner_g2 `a_concave_corner_takes_the_g2_patch_too` and `set_back_8_the_g2_corner_lies_between_the_hull_and_the_prisms`.**
Both failed only at their curvature-jump-clear-of-the-corners assert (0.0611 and 0.0554 against CURVATURE_SHARE / 5 = 1e-3); faces, volumes and validity were right.

The cause is the test helper `corner_seams`, not the fill:
- Its doc says a station counts when it is at least `share` of its side from the side's end at a corner. Its code took one corner end per seam and counted stations at least 2 x share of the edge from it, assuming "SMLib splits each side in two".
- Since #2988 each round's end section is one edge with a corner at both ends. The code excluded only the first, so stations right at the far corner, where no patch is G2 beside G1 rounds, counted as clear. The 405 stations counted are what that predicts: 6 face-curve halves and 3 whole sections.
- The helper now counts a station when it is at least `share` of its side from each end of the side at a corner. The side is the seam plus the seam continuing it smoothly at an end that doesn't turn.

This is the same rule on the old topology. Take a half-side of length L whose start is the corner, continued by a half of length L. A station at fraction f is fL from the corner and L + (1 - f)L >= L >= 0.15 x 2L from the side's other end, so it counts iff fL >= 0.15 x 2L, i.e. f >= 0.3: the old `|f - 0| >= 2 x share`, and the same from the other end. `CURVATURE_SHARE` and `share` (0.15) are unchanged. The set-back-8 test now prints its `clear`, as the concave one does.

Run (sc-build, solidcore c0d3cc0124 + batch):
- set back 8: clear 2.954e-4 per mm (405 stations), pass
- concave corner: clear 4.136e-4 per mm (the ball's 0.2), pass

**#13522: fillets/every-box-corner-three-edges-r3-curvature-setback-3-5** (r3, set back 3.5).
The case expected the corner to build. Since #2988, solidcore refuses it: at 3.5 the Setback's fill doesn't stay convex, and the least setback that builds is 3.5595. The N x N patch built there before was uncertified. The case now expects that refusal, as its sibling `...-setback-3-refused` expects its own.

SolidCraft's refusal doesn't name the size yet: "the fillet with open setback corners is larger than the faces beside its edges allow". solidcore reports the least setback only in its error comment, and `blend_failure` sizes only radius, scale and chord fits. Carrying it as typed data and naming it: size named in tangency-labs/solidcraft#13913.

KNOWN_FAILING: the two corner_g2 lines and the case's fillets line come out.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-09 15:03:59 +00:00

21 lines
3.2 KiB
Plaintext

# The fillets cases that fail today on solidcore, one a line: `<case> <why it fails today>`. Sorted.
# The kernel rig gate fails on a case that fails and is not listed, and on a listed case that passes.
# Baseline: solidcore pin 8ddce9016, tests/kernel-cases run by `cargo xtask kernel-cases`, x86_64 Linux.
1080-fillet-equal-slope-basin-1-mm-rim-loop-4 [fail] continuity_order: got {'body': 'Slab', 'edges': 'blend', 'order': 'g0', want >= g1 at every blend seam
1080-fillet-sloped-basin-1-mm-rim-r4-2-edges [fail] continuity_order: got {'body': 'Slab', 'edges': 'blend', 'order': 'g0', want >= g1 at every blend seam
1080-fillet-sloped-basin-1-mm-rim-r4-4-edges [fail] continuity_order: got {'body': 'Slab', 'edges': 'blend', 'order': 'g0', want >= g1 at every blend seam
4782-fillet-bar-edge-wider-than-both-faces [fail] continuity_order: got Box: no null seams want >= g1 at every blend seam
9-fillet-corner-curvature-with-reduce-fit [unsupported] solid.fillet builds `corner: g2` with `fit: exact` only
9-fillet-corner-curvature-with-taper-fit [unsupported] solid.fillet builds `corner: g2` with `fit: exact` only
cov-fillet-g2-corner-beside-a-cylinder [fail] step 6 (solid.fillet): the open setback corner is not a closed solid, as on SMLib d01d7b0
cov-fillet-g2-five-edge-apex-corner [fail] step 12 (solid.fillet): invalid input: the setback must be more than 3.5056 mm at the corner (-0.000, 0.000, 30.000): the rounds' contact lines meet 3
cov-fillet-g2-four-edge-apex-corner [fail] step 11 (solid.fillet): invalid input: the setback must be more than 3.9027 mm at the corner (0.000, 0.000, 30.000): the rounds' contact lines meet 3.
cov-fillet-g2-mixed-corner [unsupported] refused as on SMLib d01d7b0: a G2 corner is set back only where every edge meeting it is rounded and all are convex or all concave
cov-fillet-g2-oblique-three-edge-corner [unsupported] refused on solidcore: its three edges meet at different wedge angles, and one curvature fillet takes the first edge's equal-area thumbweight for every edge (solidcore#2250, issuecomment-6563419897); SMLib d01d7b0 fails it too
cov-fillet-g2-pocket-corner [fail] continuity_order: G1 at one seam where the corner patch meets a contact face (g2 wanted), as on SMLib d01d7b0
cov-junction-boss-tilt-20-root-round [fail] fillet status 1001 on a cylinder tilted 20 degrees into the plate, as on SMLib
cov-junction-y-junction-120-round [unsupported] refused as on SMLib: the faces between the three intersection curves narrow to nothing at the two triple points, so the room check trips (the corner step)
report-fillet-curvature-three-edges-at-bar-corner [fail] continuity_order: G1 at one seam where the corner patch meets a contact face (g2 wanted), as on SMLib d01d7b0 (see KNOWN_FAILING)
report-fillet-curvature-three-edges-at-bar-corner-solver-residues [fail] continuity_order: G1 at one seam where the corner patch meets a contact face (g2 wanted), as on SMLib d01d7b0 (see KNOWN_FAILING)
report-fillet-curvature-three-edges-at-bar-corner-solver-residues-outside-setback [unsupported] solidcore#2697: solidcore's own filled corner builds this to its closed form (7e-11); SolidCraft's least-squares fit misses curvature by 4.02e-3 against 2.5e-3 at this pin