Segmentation fault occurs when using rectangular lattices with large pitch values.

[MatteoZammataro]

Segmentation faults often occur when rectangular lattices with large pitch values are used. This is most likely due to a numerical cancellation issue in the boundary-crossing detection logic check in RectLattice::distance.

The original implementation computes the distance d to the nearest lattice boundary, then checks which boundaries were crossed using a floating-point proximity test:

if (u.x != 0.0 && std::abs(x + u.x * d - x0) < FP_PRECISION)
    lattice_trans[0] = copysign(1, u.x);

When pitch_[0] (and therefore x0) is large (eg. 100 cm), the expression x + u.x * d may suffer from cancellation: two large, nearly-equal floating-point numbers are subtracted, destroying significant digits. The tolerance threshold FP_PRECISION is a constant number, so it does not scale with the pitch. The result is that lattice_trans is sometimes computed as {0, 0, 0}, which causes the segmentation fault.