mirror of
https://github.com/bherbruck/solidcraft.git
synced 2026-11-08 13:01:52 +00:00
Kai Evans
a74e78fba8
A chamfer or fillet without `body` on a moved part failed with "[20, 0, 0] is where 2 edges meet". This happened when another part had an edge at that point in its own frame. Take 2 parts that meet in the world, but not in their own frames. A pick on the shared edge cut the first part. edge_points looked for edges on every body in that body's own frame. It ignored the world point of the pick. - frames.rs remembers the world point of every pick, face or not. picked_world(local, nth) returns it. nth keeps 2 picks at one local point apart. - edge_points maps the world point into each placed body's frame, hidden placements too. The world rows are built on first use. A body's box, grown by the reach, culls it. The polylines short-list edges. The exact curves decide (edge_foot: 33 samples, then a golden-section search). Bodies within 2 d + 1e-3 mm of the nearest refuse the pick. The reply names both: "[20, 0, 0] lies on edges of 2 bodies, Base block and Cap block: name the one to chamfer with `body`". A hidden body adds "(X is hidden)". - A seam at the pick, with no edge there, adds no second body. - 2 placements of one body at one edge, or one local point, are 1 edge. - With no placement to look through, the pick uses the bodies' own frames. That happens when every frame is singular. The pick is not passed unchecked. - A pick's own `body` must exist and agree with `body`. Else the command refuses: "an edge names `Cap block`, but `body` is `Base block`: give one". Main finds the pick on Cap and cuts Base. - The corner check runs on the body found. Corner and no-edge refusals print the world point. - With 2 or more bodies, the feature keeps the body found as `body`. Picks on 2 or more bodies get 1 more check. If another body's frame has an edge at a pick's point, the command refuses. The reply asks for "each body in a command of its own". The ghost check skips edges whose polyline is too far before it runs edge_foot. - solid.press_pull forwards `body` to the fillet. Its frames entry maps `body`, and its params name it. - The fillet schema and the fillet and chamfer params describe the lookup. Changes a default: no. Calls that cut the wrong part now refuse, naming both parts. So do picks on the cut of a split body. A pick whose own `body` disagrees with `body` refuses too. Dialogs already send `body`. Synthetic scripts ran against6a1fe7468(solidcraft-cli run, --profile ci). 4 picks now cut the picked part, where main refused: - Base's edge, with Cap moved 50 mm up. - The same, with the root active and Cap moved 500 mm along x. - Cap's edge at its world point. - Base's edge with tangent_chain false. Parts meeting in one frame or in the world are refused, naming both. Press Pull there is refused too; main rounds Base block. A box edge on a cylinder's seam is cut; main tries the seam and fails. Tests: 19 in frames_tests.rs (16) and tests.rs (3). 16 fail on6a1fe7468. 3 pass there: hostile picks, a singular placement, and 2 placements of one part. action_matrix edge_round_actions_keep_the_picked_body expects the new refusal. cargo xtask ci: 13948 passed, 32 failed, 67 ignored, 89 binaries. - 22 failures are in tests/KNOWN_FAILING_TESTS. - 7 fail alone on6a1fe7468too: 3 pleat tests (FilletRearOuter, SMLib status 1001), smlib a_cylinder_side_section_is_its_circle, gpu gpu_thin_lens_focus_and_defocus_match_geometric_oracle, ui-egui the_extrude_arrow_passes_faces_beside_the_profile_and_in_a_free_drag, kernel boolean_props translation_invariant (a random 2-sphere case). - 3 ui-egui agent_chat tests pass alone on both trees.