[fontanf]

compute_feasible_elements_and_points computes the intersections of the elements of the convex NFPs which are not strictly inside the union of the other convex NFPs. Many of these elements are on the boundary of the NFP, for example the collinear edges of the convex NFPs of all the parts along a straight side of a shape, and each of them intersects many others.

None of their points is strictly inside the NFP, so none of them is feasible. Their intersections with the other elements are not needed either:

• these intersections are not strictly inside the NFP, so they are not feasible points;
• the overlapping parts in opposite directions, which give the feasible elements, have convex NFPs on both sides, so they only reach the boundary of the NFP at their ends, which are candidates anyway.

These elements are now removed before computing the intersections: an element is removed if its start and its end are on an element of the same type of the boundary of the NFP containing its middle.

Measurements

On the 62 no-fit polygons computed by the periodic packings of packingsolver on alma_02 (alma2025 dataset), no_fit_polygon goes from 23.6 s to 20.9 s in total, with the same outputs (number of components and of elements of the NFP, number of feasible elements and points). On the self-NFP of a 64-vertex item type, 3,478 of the 10,197 remaining elements are removed, the number of intersecting pairs goes from 856,030 to 155,502, and no_fit_polygon goes from 1.59 s to 0.88 s.

All 714 tests pass.