Merge pull request #1701 from brett19/docs/fillet-geom-p3-xsect-neighbour

docs: blend.fillet_geom_p3 — CalcCrossSection neighbour-solver stub re-checked unreachable

plan/findings/blend.fillet_geom_p3--calc-cross-section-neighbour-unreachable--addendum-2026-10-30-recheck.md

@@ -0,0 +1,72 @@
+# blend.fillet_geom_p3: neighbour-solver CalcCrossSection stub re-checked, still unreachable (2026-10-30)
+
+Updates: plan/findings/blend.fillet_geom_p3--calc-cross-section-neighbour-unreachable.md
+
+This re-check covers the `UNPORTED(blend.fillet_geom_p5)` lookup in `re_calc_fillet_geom`
+(`crates/solidcore/src/blend/fillet_geom/p3.rs`) on main d4cb45e1f, after the fillet
+fixtures added since the original check (84d080fbd). The tag stays.
+
+## Additional static check: writers outside the corner
+
+The original record covered the `m_vEdges` writers and the corner's `SetFilletGeom`
+calls. A neighbour solver is needed only if a use of an edge in some geometry's
+`m_vEdges` belongs to a geometry of a *different* solver. Every remaining
+`SetFilletGeom` site under `source/SMLib/src` either keeps a single solver or acts on an
+edge that is never in a geometry's `m_vEdges`:
+
+- `SmFilletGeom.cpp:10053`, `:12714`: rail edge uses (`MakeRailEdgeuse`, the split rail
+  end). Rails are never in `m_vEdges`.
+- `SmFilletGeom.cpp:12824` (`TopologySplit`): moves `m_vRefEUs` side uses, which can
+  include a divide edge's mate, to `rpNewFilletGeom`. That geometry is created by the
+  same solver (`new SmFilletGeom(m_pFilletSolver, ..)`, `:12603`), so the mate's geometry shares
+  the recalculated geometry's solver.
+- `SmFilletSolver.cpp:6049`, `:6819`: rail uses. `:6860`, `:6884`: `SM_FE_ON_EXTEND_EDGE`
+  and `SM_FE_SPLIT_FACE` support edges, which are not cross-section edges and are not
+  in `m_vEdges`. `rpSideFG` is created by the same solver (`new SmFilletGeom(this, ..)`,
+  `:6813`).
+- `SmFilletExecutive.cpp:4286-4292` (`SM_MODIFY_CORNEREDGE_FILLET_RELATION`) acts on a
+  corner's `m_vEdges[0]` (`:4450`, `:4463`) and only moves the primary's geometry to the
+  mate's or clears it. `:4427-4428` turns a squeezed rail into a cross-section edge and
+  clears both uses' geometries. In neither case does a use gain a geometry of another
+  solver.
+
+`SmFilletGeom::m_pFilletSolver` is assigned only in the constructor
+(`SmFilletGeom.cpp:9018`). Upstream has no `SetFilletSolver`.
+
+## Runtime check
+
+Temporary `eprintln!` probes went at the cross-section loop body, at the neighbour
+lookup (the stub) and at the `re_calc_fillet_geom` entry, followed by a serial
+`--unit` difftest of each unit below. All passed with the probes in place, and the
+probes were then removed.
+
+| unit | loop hits (own solver) | stub hits |
+|------|-----------------------:|----------:|
+| blend.fillet_geom_p3 | 0 | 0 |
+| blend.fillet_executive_p1 | 0 | 0 |
+| blend.fillet_executive_p2 | 0 | 0 |
+| blend.fillet_executive_p3 | 3 | 0 |
+| blend.fillet_corner_p1 | 0 | 0 |
+| blend.fillet_corner_p2 | 0 | 0 |
+| blend.fillet_corner_p3 | 0 | 0 |
+| blend.fillet_solver_p4 | 0 | 0 |
+| blend.fillet_solver_p5 | 0 | 0 |
+| blend.fillet_standard_solver_p2 | 0 | 0 |
+| topology.api_files | 0 | 0 |
+| topology.api_projection | 0 | 0 |
+| topology.brep_cutting | 0 | 0 |
+| topology.brep_planar_section | 0 | 0 |
+| primitive.api_trim_ops | 0 | 0 |
+
+The loop is now reached: 3 cross-section edges in blend.fillet_executive_p3, compared
+against C++ and passing. Each one resolved to the recalculated geometry's own solver,
+which the existing lookup returns. No case asked for another solver.
+
+## Consequence
+
+The conclusion is unchanged. Wiring a real neighbour lookup would mean threading the
+executive's solver store through all five `re_calc_fillet_geom` callers, while the
+caller's own solver is mutably borrowed, for a branch that no upstream topology can
+produce. Only a synthetic model could exercise it, so no difftest case or regression
+test can reach it with an upstream-valid input. The `UNPORTED` tag and its reason
+comment stay as they are.