mirror of
https://github.com/bherbruck/solidcraft.git
synced 2026-11-20 23:52:23 +00:00
916b91e8d0
solidcore's main at c0d3cc0124, rewritten to one commit per same-author run per PR (2,800 commits, author credit kept; tree identical). Full original history: tangency-labs/solidcore archive/full-history. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> git-subtree-dir: solidcore git-subtree-mainline:44525da669git-subtree-split:c440e5f892
13 lines
5.3 KiB
TOML
Vendored
13 lines
5.3 KiB
TOML
Vendored
id = "surface.trim_srf_cache"
|
|
upstream = ["source/SMLib/inc/SmTrimSrfCache.h", "source/SMLib/src/SmTrimSrfCache.cpp"]
|
|
rust_module = "surface::trim_srf_cache"
|
|
depends_on = ["curve.curve_cache", "geom.extent1d", "geom.extent2d", "geom.extent3d", "geom.object", "geom.tol", "geom.vector2d", "geom.vector3d", "surface.surface_cache_p1"]
|
|
soft_depends_on = ["geom.tree", "nurbs.b_spline_curve_p1", "nurbs.b_spline_curve_p2", "surface.surface_p2"]
|
|
est_lines = 966
|
|
layer = 23
|
|
tests = ["prog_test:my_test_extent3d", "prog_test:my_test_vector2d", "prog_test:my_test_surface_closest_point", "prog_test:my_test_drop_point", "prog_test:my_test_matrix", "prog_test:my_test_reversed_line_join"]
|
|
status = "green"
|
|
claimed_by = ""
|
|
claimed_at = ""
|
|
notes = "2026-10-31 (nightly fuzz00295, run 41040730716: a surface with one rational control point timed out on the Rust side): the runner's Rust side applies solidcore_ffi surface_trim_srf_cache::validate to the same input as the C++ side, so an input the FFI refuses is a harness error on both sides; regression crates/difftest/tests/fuzz_corpus/surface.trim_srf_cache.json, plan/findings/surface.trim_srf_cache--mixed-rationality-rust-side-runs-past-fuzz-limit-fuzz00295.md. FACE-OWNED GetTree (wire: face-owned tree build, 2026-10-28): TrimSrfCache::get_tree_with_face is SmSurfaceCache::GetTree with the virtual BuildTree dispatched to build_tree_with_face; blend.hcr IsPointVisible uses it (its three deferrals compare again). 7 new direct native-face cases (build_mode 2..5 in the faces shim: GetTree after an editing-on build rebuilds standalone; editing turned off before GetTree implants the trims into the existing tree; trimmed tree is not rebuilt; GetTree builds a missing tree), all pass with values, max_rel_err=0. SurfaceCache::get_tree without a face keeps its Unported stub for a face-owned trimmed cache. Earlier: OWNED UV CURVE TOLERANCE (PR #1307, 2026-10-27): ARM64 native gate 772 cases, 772 value-bearing passes, 0 failed/deferred/rejected/vacuous, max_rel_err=0; all 762 baseline cases retained. Ten new nonlinear/control/reimplantation cases exercise the real owning-edge tolerance. EdgeuseInput transfers exact geometry and resolved optional tolerance into retention; BuildTree resolves edgeuse ownership, child subdivision reuses retained geometry without cloning, and handle mapping preserves its tolerance. Ownerless curves retain system tolerance. Native guards keep unrelated coordinate/scalar/z limits unchanged; the larger 1e11 coordinate envelope is isolated to ImplantEu. No core CurveCache or numerical tolerance change. Prior history: POLY-EDGE SUBDIVISION (PR #1016, 2026-10-25): 762 cases, 762 passed with values, 0 failed/deferred/rejected/vacuous, max_rel_err=0, rel_tol=1e-9, no absolute floors. All 630 prior cases retain their value coverage; 128 separate fixed-seed cases and 4 tolerance-boundary fixtures now exercise polygon implantation followed by repeated subdivision, including mixed edge-use lists, zero/negative native tolerances, different original/cache surfaces and handle mapping. SubdivideNode reimplants parent polygon edges head first into child 1 then child 2, preserving the parent lists. Endpoints, tolerance and an exact bounded evaluation copy of the original face surface are retained per handle; reimplant after edits to refresh the snapshot. GREEN (PR #594, 2026-10-23): canonical Linux CI run 39936710786 on 3eefd5fc8 (includes upstream/main 38b0745b3) compares 630 cases, 630 passed with values, 0 failed, 0 deferred, max_rel_err=0, rel_tol=1e-9, no absolute floors. The full differential gate passes all 260 units (177 passed, 83 partial, 0 failed); native runner tests pass (253 passed, 0 failed, 32 ignored), including direct C++/Rust equality for built/unbuilt nonplanar face interior/exterior/edge/vertex queries. That run's only failure was a test-only Clippy chunks_exact suggestion, corrected with this promotion; final-head all-seven CI remains the review gate. All 628 original cases and 568 original passes are retained: 60 former runner-only SmBrep::CreateFaceFromSurface deferrals (4 file fixtures, 56 generated) now compare, plus 2 new saddle boundary fixtures. OpIn::Face independently builds real topology through landed brep p8 and uses the existing BrepCacheFace; build=false stays lazy, statuses/object types/owner tolerances are preserved. No production kernel, shared native_face helper, C++/FFI, seeds, guards or tolerances changed; no primitive/sweep parity improvement is claimed. No local native tests, baseline repeats or C++ builds were run. Prior #533/#537 already supplied the shared trimmed-cache machinery and 568 baseline passes, including seven real holed/nonplanar face probes; those gains are not attributed to this follow-up. Earlier notes' construction/classifier gaps are superseded by the landed adapters and this measured evidence. 11 fns; largest: SmTrimSrfCache::ImplantEdgeuse, SmTrimSrfCache::BuildTree, SmTrimSrfCache::ImplantVertexuse, SmTrimSrfCache::ImplantPolyEdge; cycle back-edges: intersect.point_class(7), query.poly(5), surface.surface_cache_p3(5), query.face_p2(2), topology.face_p2(2), topology.face_p4(2), topology.loopuse(2), topology.vertexuse(2), query.tess(1), surface.surface_p1(1), surface.surface_p3(1), topology.edgeuse_p1(1), topology.face_p3(1), topology.vertex(1); unit-SCC #22 (220 units); 319 upstream test entry points reach this unit."
|