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
A boolean operand that touches itself along a line carried as two coincident edges fails (relate 1001 or not a closed solid) #13650
A boolean whose operand touches itself along a line fails when that line is carried as two coincident edges of two faces each, and works when it is one edge of four faces.
The body: a block with a hexagonal pocket through it and a 2.5 mm hole drilled from an end face that ends exactly at a corner of the pocket, so the hole's bottom meets the two pocket walls along a segment of the corner edge. The kernel's cut leaves that segment as one edge of four faces (15 faces, 37 edges, 24 vertices), and filling the pocket with the same hexagon as a union works (a valid solid, the exact volume).
After a record round trip (to_record, from_record: the body SolidCraft hands its next feature) the segment is two coincident edges of two faces each (split_touching: the record format writes one edge per face pair; 15 faces, 38 edges), the body still passes check(All), and the same union:
with the default fixes gives a body that is not a closed solid (the ops crate's NotSolid);
with SolidCraft's fix set fails in boolean::topology_intersector::p1::relate with 1001.
Gluing the two coincident edges back into one (GlueEdgesGeneral, same vertices, same curve, four distinct faces) before the union makes it work in both configurations. Through SolidCraft this is #12785 (Join filling a pocket refused on 0.30.1, relate 1001 on 0.40.0-rc.3).
Reproduction
crates/solidcore-ops/tests/boolean_join_fills_pocket_with_hole_at_its_corner.rs (the PR, committed first, failing): the block 100.29 x 32 x 10 at corner (-50.145, -16, 0), the regular hexagon of circumradius 6.3685 with a corner on +X cut through, the hole of radius 1.25 along -X from x = 50.145 to x = 6.3685, then the hexagon joined 10 mm; the join of the cut's own body passes, the join after the round trip fails, the pocket without the hole joins after the round trip.
Where
merge_boolean_by receives the operand with the coincident edges; the proto-topology merge relates the tool's corner edge to two coincident operand edges with the same vertices. The kernel's own form of a self-touch is the one edge of four faces (what its booleans produce), which the merge handles.
@bherbruck is draft #13576 still active? Its last commit was 2026-11-08, and #15621 (a duplicate) reports this as a regression since 0.30.1. If we don't hear back in 15 minutes, we'll open a PR that carries #13576's commits, credited to you. It will make the glue the default with no flag, and decide coincidence with a certified bound instead of the 3-sample check.
Written by Claude Code (Anthropic) on behalf of @brett19.
#16392 (draft) carries #13576's commits and fixes this issue. The glue of the two coincident edges is now always on, with no flag. Coincidence is decided by a bound over each edge's whole span, not by 3 samples. The issue's pocket-and-hole case now joins to a valid solid after the record round trip, with no fixes and with every fix.
#15621 is a different form: we reproduced its script with the ops crate. The kernel's own join gives a valid body, and the interference and cut with the cylinder both build (98174.770, 75000). After a record round trip, the touching line is one edge of four faces in which the cylinder's side face appears twice. It is not two coincident edges, so the glue in #16392 does not reach it. The interference and cut then fail with NotSolid / status 1001 on main and on #16392 alike. So #16392 does not claim #15621.
Written by Claude Code (Anthropic) on behalf of @brett19.
Activity
fix(boolean): an operand touching itself along two coincident edges i…
brett19 commentedon Nov 13, 2026
@bherbruck is draft #13576 still active? Its last commit was 2026-11-08, and #15621 (a duplicate) reports this as a regression since 0.30.1. If we don't hear back in 15 minutes, we'll open a PR that carries #13576's commits, credited to you. It will make the glue the default with no flag, and decide coincidence with a certified bound instead of the 3-sample check.
Written by Claude Code (Anthropic) on behalf of @brett19.
solidcore: the self-touch glue is always on and proves coincidence ov…
brett19 commentedon Nov 14, 2026
#16392 (draft) carries #13576's commits and fixes this issue. The glue of the two coincident edges is now always on, with no flag. Coincidence is decided by a bound over each edge's whole span, not by 3 samples. The issue's pocket-and-hole case now joins to a valid solid after the record round trip, with no fixes and with every fix.
#15621 is a different form: we reproduced its script with the ops crate. The kernel's own join gives a valid body, and the interference and cut with the cylinder both build (98174.770, 75000). After a record round trip, the touching line is one edge of four faces in which the cylinder's side face appears twice. It is not two coincident edges, so the glue in #16392 does not reach it. The interference and cut then fail with
NotSolid/ status 1001 on main and on #16392 alike. So #16392 does not claim #15621.Written by Claude Code (Anthropic) on behalf of @brett19.
solidcore: the self-touch glue is always on and proves coincidence ov…
solidcore: the self-touch glue never glues a crack edge or a copy sha…
solidcore: a self-touch gluing that fails leaves the operand as given (…
solidcore: a failed self-touch gluing also puts back the kernel's rng…
solidcore: a boolean operand touching itself along a line is glued in…