Repository navigation
Blend step 4: the End at a smooth vertex, and rails past a side edge refused naming the radius that reaches it - #124
Merged
Merged
Conversation
…ound ends at the smooth vertex (failing) The constant radius law on the top back edge of a 60 x 40 x 20 box whose vertical edge at (60, 40) is rounded R5 moves from main's body into the scenario block, its expectation re-derived. The top back edge runs on at x = 55, tangentially and along the top face, into the R5 round's top arc, which is not selected: a variable round is never chain-expanded by its caller (SolidCraft's fillet passes a variable or spline style's edges as picked), so this is a partial selection, and with tangent propagation off a round ends at the smooth vertex, cut by the plane square to its edge there (the owner's ruling, as other systems do). So the round lies over 0 <= x <= 55 only, its section the r x r square less the quarter disc: removed 55 r^2 (1 - pi/4), the end at x = 55 one plane face of area r^2 (1 - pi/4), r 1, 2, 3. The old expectation (the round running on to the R5 cylinder: 55 r^2 (1 - pi/4) plus the integral of (r - sqrt(r^2 - u^2)) sqrt(25 - (5 - r + u)^2) over 0 < u < r) was that run-on reading, which the ruling drops. Listed failing: today 12.087853554, 48.797265494 and 110.521227189 removed against 11.803101013, 47.212404053 and 106.227909118, no end face. Refs #81 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01NvoxKMbMFSQU8tzRtoJhtT
…SolidCraft sends (§27) Every SolidCraft fillet/chamfer case now rounds with tangent propagation (decision register 11931, §27), and the capi copies take the same inputs (tangent_chain_points): - 016 and 018 (r7, r2): the R3 on the top right edge with its tangent chain (3 edges: the top edge, the R5 round's end arc on x = 60, the vertical edge x = 60, y = 40), then (60, 40, 8) picked: refused as a pick, SMLIB_ERR_PICK_NONE. Proof, SolidCraft's 016/018 expected.json: "the r3 rounds all three and the vertical edge is gone: its round is the cylinder r3 about (57, 37) ... The third pick (60, 40, 8) is 3 from each of those two lines and nearer no other edge: it names neither, and the right answer is a clean refusal asking for a point along one edge" (outcome refused, "give a point along one edge"). - 020 and 021 (the partial-chain copies: the R3 on its edge alone, then the vertical edge with its chain): r7 refused, SMLIB_ERR_NO_ROOM (020: "the r7's contact line on that face would be the arc offset 7 towards its centre, radius 5 - 7 < 0 ... refused (SMLib: no room for the rolling ball)"); r2 one body of 11 faces within the box, sound (021: body_count 1, faces 11, bbox (0, 0, 0) .. (60, 40, 20)). - The up-chain corner (the 020/021 body, radii 3.1, 4, 5, 7 refused naming 3; 1, 2, 2.5, 2.9 built) and rounded_corner (its r2.5 child for tangency-labs/solidcore#1173): the three points on the chain's edges become the vertical edge's chain as tangent_chain_points finds it (the same three edges here); expectations unchanged. The old ladder 016 copy (the vertical edge alone, built to 47418.926881 / 47546.786538 / 47591.811067 for r7 / r4 / r2) was a single-edge reading SolidCraft never sent (its 016 recipe rounds the vertical edge with the default chain): dropped for the four recipes. All pass today; they hold the smooth-vertex End to SolidCraft's answers. Refs #81 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01NvoxKMbMFSQU8tzRtoJhtT
…d the 851 middle edge at r0.25 is refused, no size building it (failing) Two expectations re-derived, each listed failing until the smooth-vertex End lands: - ladder 012 (SolidCraft 07-fillets/012: the variable round r0 -> 5 on the top back edge (30, 40, 20) from (0, 40, 20), r0 1 and 2, of the box whose vertical edge at (60, 40) is rounded R5; no tangent_chain, and a variable round is never chain-expanded): a partial selection, so the round ends at the smooth vertex x = 55, where the edge runs on along the top face into the R5 round's top arc, cut by the plane x = 55. It moves from main's body to the scenario block. Removed: Int_0^55 A(x) dx, A(x) the corner of the plane x cut off by the union of the law's balls (radius r(s) = r0 + (5 - r0) s / 55 centred at (s, 40 - r(s), 20 - r(s)), s in [0, 55]), measured by rays from the corner to their first entry into that union, A = Int_0^(pi/2) t^2 / 2 dphi (the previous expectation's own method, which ran s and x on past 55: 137.573156 and 168.865031; up to x = 55, 121.785383648 and 153.330530014); the end face at x = 55 of area A(55), 5.357125303 and 5.361201953. Graded to the round's certified bound (fillet tolerance times the ruled area for the volume, times the end's perimeter for the area). Today 137.573146201 and 168.864966948 removed, no end face. - 851's middle edge at r0.25 (middle_edge_of_an_arc_tangent_chain_r0_25): SMLIB_ERR_NO_ROOM, smlib_blend_fit_size 0 (no size builds). Its top end V, on the arc wall at y = -19, z = 10, is a smooth vertex: the round ends there on the plane through V square to the edge. Its rail on the R2 front round stays on that round only below the round's tangent edge with the top face (y = -19, z = 10), which passes through V. A ball of radius r touching the front round on that edge at x has its centre at (x, -19, 10 + r), 21 + r from the arc wall's axis (62 1/6, -21): (x - 62 1/6)^2 + 4 = (21 + r)^2. The plane meets that edge at V only, where (x_V - 62 1/6)^2 + 4 = 21^2: r = 0. For every r > 0 the rail crosses onto the top face before the plane (at r 0.25 by 0.024): it would roll over into the End, which is not built (#98), so no radius builds and the case cannot be checked at a named size. Today -1100 (not a solid). Refs #81 Refs #98 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01NvoxKMbMFSQU8tzRtoJhtT
…dge of more than two faces bounds it TrimOriginalSurfaces collects the body's faces a fillet's rails enclose, without crossing the marked edges (the fillet's boundary imprinted on the body). An edge of the body with more than two faces (two lumps of the body touching along a line: tangency-labs/solidcore#453's bore wall touching the cube's bottom face) was crossed into the faces beyond it, so the collection reached the whole body, the trim was flagged as leaking and dropped, and the fillet's faces were merged into the body left whole: a blend that took nothing off, returned SMLIB_OK as built (sound, the volume unchanged). Now: - such an edge is marked before the trim: it bounds the material a blend takes off on its side, and the faces beyond it are not the blend's; - a leaking trim is an error (SM_FILERR_BAD_MERGE, "the trim leaks"), never dropped. On 5e8c9c1's corners #453's back rim still fails earlier (its arc ends at the bore's seam, a smooth vertex: Corner Edge Computation Failure); the smooth-vertex End that follows builds it. Refs #81 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01NvoxKMbMFSQU8tzRtoJhtT
…ge, and a rail past its side edge is refused naming the radius that reaches it A round selected on only part of a tangent chain (§25: SMLib blends exactly the edges it is given) ends where its edge runs on tangentially into an edge of its convexity that is not blended: the End at a smooth vertex (SM_BEND_AT_SMOOTH, SmBlendCornerEnd.cpp). The round is cut by the plane through the vertex square to its edge there (m_pEndCap, sized from the side faces' tight boxes; the fillet geometry's end and the rails' ends are found on it), and capped by the plane face between its end rib and the cap's edges on the side faces. Two edges leaving the vertex the same way (a knife-edge tip) are no continuation: the other edge must leave opposite. A rail that rolls over its face's tangent edge into such an End is refused, no size building it. A rail that runs past its side edge's far end P, outside a cliff overflow, is refused (SMLIB_ERR_NO_ROOM), naming the radius at which the rail reaches P: the root of dist(P + rho n_F, G) - rho, the ball touching the rail's face F at P that touches the other face G (positive at 0, negative at the round's radius when the rail lies beyond P). The miss is found on the side edge itself: the rail's end was found on a straight side edge's line continued past its end and taken as on the edge outside a cliff overflow (the up-chain corner at (60, 35, 20): (60, 35, 16) on the R3 cap's edge, z 17..20), and the global merge or an end rib off the edge then gave -1100. An End on more than one end face (no end rib) is checked the same way: the global merge builds it only where both rails reach their side edges. Where the ball's regularity check refuses first (r7 there: the R5 round, fit 5), each End's past-side-edge radius is a second cause and the refusal names the smaller (3). The Join's refusal of blends continuing each other across more than one edge on a side is SMLIB_ERR_NO_ROOM with no size building it (smlib_blend_fit_size 0), its reason saying so. The executive keeps a corner's NO_ROOM noted in CalcCornerEdgeGeom (it was overwritten by "Corner Edge Computation Failure"). Tests: - the up-chain corner (main): r3.1, r4, r5, r7 NO_ROOM, fit 3 and "a radius below 3" (as on 5e8c9c1; the reason is now the End's: the rail on x = 60, the R5 arc offset in by r, crosses y = 35 at z = 20 - r, below the cap edge's end z = 17 for r > 3); r1, r2, r2.5, r2.9 build. - tangency-labs/solidcore#453's back rim (main): refused on 5e8c9c1 by its corner's error (1001 at the seam, not a size); now chamfer 3 and blends round, chord and chamfer 3 build, each checked against its volume within 1e-7: 8000 - 500 pi less the section swept about the bore's axis over the arc's 270 degrees, clipped at the bottom face (sections at angle g from the touching point keep t < 5 / cos g - 5), by Simpson's rule (converged to 1e-9 mm^3); the chord blend of chord 3 between the square faces is the round of 3 / sqrt 2. Got 6316.944420510 (chamfer, closed form 6316.944424728), 6382.568447992 (round, 6382.568731366), 6406.329342842 (chord, 6406.329313366). The residuals are each blend's end rib on the bottom face, the End's certified intersection (3.5e-7 off the cone, 8.6e-6 off the round's torus); the same rim closed (bore 1 higher, no ends) matches Pappus to 1e-9. The curvature-continuous blend stays refused (as on 5e8c9c1). - tangency-labs/solidcore#1173's child ran that chord blend first expecting it refused; it now builds, and the child runs it first as before (chord_blend_first): r2.5 is sound and the same body byte for byte in both orders. - scenarios no longer failing: round_ending_where_its_edge_runs_on, variable_round_ending_at_ the_r5_corner (ladder 012), middle_edge_of_an_arc_tangent_chain_r0_25 (-1103, fit 0). - constant_law_ending_at_the_r5_corner (r 1, 2, 3) passes, graded as ladder 012 to the law round's certified bound (a law round is the variable solver's certified B-spline): 100 times the part's tolerance (1e-3) times the ruled area 55 pi r / 2 for the volume (0.086 at r1), times the end's perimeter 2 r + pi r / 2 for its face (0.0036 at r1); it was graded to 1e-8. Got r1 11.803083866 removed (closed form 55 (1 - pi/4) = 11.803101013), end face 0.214602222546 (0.214601836603); r2 47.212387605 (47.212404053), 0.858408890200 (0.858407346410); r3 106.227893783 (106.227909118), 1.931420002975 (1.931416529423). - the end face at x = 55 is found within 1e-9 of the plane (its tight box is not exactly 55). Refs #81 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01NvoxKMbMFSQU8tzRtoJhtT
…gent chains, G1 along their edges and inside (tilt 20 and 40 failing) SolidCraft's junction rig (cov-junction-tee-* and cov-junction-boss-tilt-*-root-round): a pipe R8 along x with a branch joined on it (equal R8; R5; R5 2 off; R5 at 30 degrees; R4 at 45 degrees, 1 off), rounded 1; a cylinder R6 standing 2 into a plate 60 x 60 x 10, upright or tilted 20 or 40 degrees, its root edge rounded 1.5; each picked edge with its tangent chain (§27). The cylinders are made as solid.cylinder makes them (its frame, its seam 0.6131 rad from the frame's x), since where the seams fall decides the corners. Graded from the recipes' expected.json (one valid closed body whose round meets its faces G1; no closed form): SMLIB_OK, one solid, sound, and every blend face G1 along its edges with other faces (the tangent planes' angle at nine stations of each edge) and inside (the normals' limits on either side of each interior knot line, extrapolated from each side, so the surface's own turning cancels), both to 1e-6 rad. On b66c7ec (seams, inside; radians): - tee equal R8, 3 edges: 2.06e-7, 1.49e-8; R5: 4.71e-8, 2.11e-8; R5 2 off, 3 edges: 2.24e-7, 2.11e-8; R5 at 30 degrees: 4.21e-8, 2.11e-8; R4 at 45 degrees: 2.29e-7, 2.11e-8. - boss upright: 2.11e-8, 2.11e-8. - boss tilted 20 degrees (failing): 1001 at the End at (-5.848, +-2.409, 0), a mixed-convexity End (the cap's chord on the plate concave, its rim convex): the rail on the wall, about 1.5 up, passes over the whole rim, which rises only -2 cos 20 + 6 sin 20 = 0.173 above the plate. - boss tilted 40 degrees (failing): 1001 at the End at (-3.111, 5.506, 0): the cap's rim is split by the cylinder's seam vertex at (-3.93, 4.91, 0.69), and the rail on the wall crosses the rim past it, on the next edge of the same two faces. Refs #81 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01NvoxKMbMFSQU8tzRtoJhtT
…he largest radius whose rail still ends on the cap's rim (failing) Owner's ruling (2026-11-02): the root round of cov-junction-boss-tilt-20-root-round does not bridge the gap under the overhanging cap; it is a concave overflow, refused with the largest radius whose rail still ends on the rim (smlib_blend_fit_size, and in the reason). SolidCraft's recipe, which expected a body, is re-pinned with the ruling. Proof of the size. Tilted 20 degrees, the cylinder R6 standing 2 into the plate has its bottom cap out of the plate on the -x side, by 6 sin 20 - 2 cos 20 = 0.173 at its top: a wedge of air under the cap. The root round ends at the two vertices where the root ellipse meets the cap's chord on the plate, Ends of mixed convexity (the chord concave, the cap's rim convex), on the cap: the rail on the wall must end on the rim, so the ball's centre lies in the cap's plane. That centre, at axial height 0 above the base (-2 sin 20, 0, -2 cos 20), 6 + r from the axis at angle phi from the cap's frame x (cos 20, 0, -sin 20), is r above the plate: -2 cos 20 - (6 + r) sin 20 cos phi = r, cos phi = -(r + 2 cos 20) / ((6 + r) sin 20). It has a root only while r <= r* = (6 sin 20 - 2 cos 20) / (1 - sin 20) = 0.262524174; at r* both Ends' rails reach the rim at its top (phi = pi) and the Ends meet in a point, so the bound is open: r* less the fit margin, a thousand of the root edge's tolerance as SMLib works it (1.657e-5 there, though the body's edge reads 1e-5). Built today below r* (0.25, 0.2525 and 0.26 sound); at r 1.5 the rail on the wall, about 1.5 up, passes over the whole rim and the corner fails with SMLib's generic error (1001, Corner Edge Computation Failure, no size): listed failing until the End refuses it. Expectation: SMLIB_ERR_NO_ROOM at r 1.5; smlib_blend_fit_size below r* by at most 0.02 (two thousand tolerances: the margin is SMLib's working tolerance, not readable here); the reason's "a radius below" the same (to its printed digits); the round at the size named built and sound, and one 0.001 past r* refused. Refs #81 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01NvoxKMbMFSQU8tzRtoJhtT
… the other convexity is refused, naming the largest radius whose rail still ends on it A round's End of mixed convexity (MakeEndOnEndFace: one side edge of the round's convexity, one of the other) ends on the end face, its rail on the face F that the other convexity's side edge E bounds ending on E. Where the rail misses E (SM_FV_NO_INT_RAIL_EU) the round's rail passes over E: no ball touching F on E fits, and the round would run on past the end face. For a concave round (it adds material) that is a concave overflow, which the owner ruled refused (2026-11-02, cov-junction-boss-tilt-20-root-round): SMLIB_ERR_NO_ROOM, naming the largest radius whose rail still ends on E. It went on to the end rib, which failed with SMLib's generic "Corner Edge Computation Failure" (1001) and no size. The size, on the cause's own geometry (RefuseEndOverflowingSideEdge): the ball touching F at a point P of E and touching the round's other face G has radius rho(P), the root of dist(P + rho n_F, G) = rho (BallTouchingAt, by bisection, factored out of RadiusReachingSideEdgeEnd); over E rho runs from 0 where E meets G to its most, found by golden section over E's interval. At the most the two Ends' rails meet on E (the bound is open): the size named is it less the part's own tolerance (the input brep's; owner ruling, 2026-11-02: not a multiple of the round's working edge tolerance). The boss tilted 20 degrees (junction_boss_tilt_20_root_round, no longer listed failing): r 1.5 refused, naming r* - 1e-5 = 0.262514174, r* = (6 sin 20 - 2 cos 20) / (1 - sin 20) = 0.262524174 and 1e-5 the part's tolerance; the test pins it exactly (to rounding at the part's scale, 1e-12 x 60 = 6e-11) and checks that the round at that size builds, sound, and that r* + 0.001 is refused. Refs #81 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01NvoxKMbMFSQU8tzRtoJhtT
…am vertex: tilted 40 degrees it builds, tilted 20 the refused round's named size builds (failing); not past a real edge The boss tilted 40 degrees into the plate, its root edge rounded 1.5: at each mixed End (the cap's chord concave, its rim convex) the rail on the wall ends where the ball's centre lies in the cap's plane, cos phi = -(r + 2 cos 40) / ((6 + r) sin 40) about the axis from the cap's frame x: phi = 128.97 degrees, (-4.17638, +-4.66470, 0.89358). The rim is the cap's circle, split where the wall's seam ends on it (125.13 degrees): at the +y End (113.39 degrees) the rim's edge stops at the seam's vertex, short of the rail's end, which lies on the rim's next edge between the same cap and wall. Today the rail's end is looked for on the first edge only, and the miss is read as the rail passing over the rim: SMLIB_ERR_NO_ROOM naming 1.09038, the radius whose rail ends at the seam's vertex (1.0903852), no bound; the rail passes over the rim only past (6 sin 40 - 2 cos 40) / (1 - sin 40) = 6.508. - junction_boss_tilt_40_root_round: builds, sound, G1 along its rails and inside (1e-6 rad), both rail ends vertices at the closed form within 1e-6. The edges with the cap are the End's ribs, where the round meets its end face at an angle: not graded G1 (blend_faces_g1 and junction_round take the end face; the angle printed). Listed failing with today's reason. - junction_boss_tilt_40_seam_past_the_rail_end: the same body, the seam turned to 140 degrees: builds today, rail ends at the closed form (9.9e-16, 7.7e-14), rails 2.11e-8 rad, ribs 1.48 rad (the edges with the cap). - junction_boss_tilt_40_wall_split_short_of_the_rail_end_refused: the wall cut by the plane through its generators at 120 and 240 degrees (three faces: the cut also makes the periodic wall's seam an edge): the rim's next edge past the 120 degree vertex lies between the cap and another face, past a real edge. Not continued: refused naming the radius whose rail ends at 120 degrees, (3 sin 40 - 2 cos 40) / (1 - sin 40 / 2) less the part's tolerance, 0.583942734 (to 6e-11). Passes today. - junction_boss_tilt_20_seam_over_the_rim_refused: the boss tilted 20 degrees with its seam turned to 170 degrees, inside the cap's overhang (156.3 to 203.7 degrees): each End's rail passes over both edges of the split arc, and the refusal names the most over the whole arc, at its top (180 degrees, on the -y End's first edge; the +y End's first edge reaches only 0.214 at 170 degrees): the same 0.262514174 (to 6e-11), and that size builds. Listed failing: today it names 0.262514174, but that size is refused too (-1103): at it the +y End's rail ends just short of the top, on the rim's edge past the seam's vertex. The rig's G1 along a blend face's edges now takes the blend face's normal at its own pcurve's point (on its contact line: the blend's pcurves lie on its surface to rounding) and the other face's at that point's foot on its surface, not the other face's normal at its pcurve's point: that pcurve lies off the contact line by its own error (the edge's tolerance), which read as an angle. (Taken at the 3D edge's point's feet on both surfaces instead, the edge curve's own offset from the contact line reads, times the faces' curvature difference: tees 1.5e-6 to 2e-6.) The round's tangency now reads 2.11e-8 rad on every junction case that builds; on b4ab8f1 the tees read 4.2e-8 to 2.29e-7 and the boss tilted 40 degrees with its seam at 140 degrees, whose wall pcurve of the rail lies up to 3.4e-5 off the rail (its edge's tolerance 1.46e-4), 4.7e-7. How far the pcurves lie off the edge is printed, not graded (a pcurve's fit, not the faces' tangency). cylinder_on takes the seam's angle (default solid.cylinder's 0.6131 rad). Refs #81 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01NvoxKMbMFSQU8tzRtoJhtT
… vertex, and a rail passing over the rim is refused on the whole of it A side edge is split where a periodic face's seam ends on it (a cylinder's seam on the rim of its cap) and nowhere else: that vertex is no corner of the body. An End's rail that runs past the side edge's far end there ends on the next edge, between the same two faces, which continues it G1. - sm_SideEdgeuseAcrossSeamVertex: the next edge across the side edge's end when that vertex has three edges, the side edge, one continuing it G1 between the same two faces, and a seam of one of them; its edgeuse on the side edgeuse's faceuse. Any other third edge (a real edge between two faces) is a corner: not continued. - SmFilletVertex::CalcRailIntEdgeuse: a rail end not found on the side edge is looked for on the next edge across such a vertex, from that vertex on, and the vertex classified to it. - SmFilletEdge::CalcEdgeOnEdge: the End's edge on the side edge runs from the corner's vertex along the chain to the rail's end, the chain's edges joined end to end. The End takes the side edge's first edge and the seam's vertex into the cut-away part (the merge removes them; the rail crosses the seam's edge as anywhere on the face). - RefuseEndOverflowingSideEdge (SideEdgeChainTo): a concave round's End whose rail passes over the rim is refused only once its end was looked for along the whole chain, naming the most over all of the chain's edges, less the part's tolerance. The boss tilted 40 degrees (junction_boss_tilt_40_root_round): its +y End's rim edge stops at the seam's vertex (125.13 degrees) short of the rail's end (128.97 degrees). It was refused naming 1.09038, the radius whose rail ends at the seam's vertex, no bound (the rail passes over the rim only past (6 sin 40 - 2 cos 40) / (1 - sin 40) = 6.508). Now it builds: one solid, sound, G1 along its rails to 2.11e-8 rad and inside to 1.49e-8, both rail ends vertices at the closed form, (-4.176377, +-4.664695, 0.893582) at 2.0e-15 and 9.9e-16. Unlisted. The boss tilted 20 degrees with its seam at 170 degrees in the cap's overhang (junction_boss_tilt_20_seam_over_the_rim_refused): still refused naming 0.262514174 (r* less the part's tolerance), the most over both edges of the split arc, and that size now builds (its +y End's rail ends past the seam's vertex). Unlisted. Over the last edge only it would name 0.213446624 (checked with the loop cut to the last edge, not committed). Unchanged: the boss tilted 20 degrees (junction_boss_tilt_20_root_round) still names exactly 0.262514174; the seam turned past the rail's end builds as before; the wall split by a real edge at 120 degrees is not continued and is refused as before, naming 0.583942734. Found, not changed: on the tilted bosses the wall's pcurve of the rail lies up to 4.4e-5 off the rail here (3.4e-5 with the seam past the rail's end, on the base too), its edge's tolerance raised to match; printed by the rig, not graded. Unlisted from the scenarios: junction_boss_tilt_40_root_round and junction_boss_tilt_20_seam_over_the_rim_refused. (The first's list line is reindented.) Refs #81 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01NvoxKMbMFSQU8tzRtoJhtT
… its size, the side edge continued across a seam's vertex capi/README.md "Changes from upstream": three entries. CHANGELOG.md [Unreleased] ### Changed: the three user-visible changes to smlib_fillet and smlib_blend_fit_size. docs/generic-blend.md §6: Refs #81 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01NvoxKMbMFSQU8tzRtoJhtT
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #81.
The End at a smooth vertex. A round whose edge runs on into an unselected round now ends at the smooth vertex, on the plane square to its edge. Before, it folded or failed.
Refusals that name a size. Where an End's rail would pass a side edge of the other convexity, the round would add material past the end face. It is refused, naming the largest radius whose rail still ends on that edge, which builds. That radius is the ball's ρ with dist(P + ρ n_F, G) = ρ on (0, r], P being where the rail meets the side edge. It is solved by bracketed bisection to rounding (
SmBlendCorner::BallTouchingAt), and only names the size, so it decides no geometry.Seams. The side edge is continued across a seam's vertex, so a rail ending past the wall's seam still finds its edge.
Guards.
Tests (capi
ops_kernel.cpp):Each was committed failing first.
Gate: based on b785227. This PR's C API CI covers all five targets, Linux x86_64 with
-DSMLIB_STRICT_WARNINGS=ON. The worker's local gate report is not attached here. Docs: CHANGELOG,capi/README.md,docs/generic-blend.md.🤖 Generated with Claude Code
https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ