[BUG]: The results of pyclipper calculations are inconsistent on x86 and arm architectures

[GenerallyCovetous]

trafficstars

As the title suggests, the single-case test is reproduced below: import numpy as np import pyclipper poly = np.array([[632, 970], [631, 971], [630, 971], [630, 973], [631, 974], [631, 975], [632, 975], [633, 974], [633, 973], [634, 972], [634, 971], [633, 971]]) distance = 0.9118854184679032 pco = pyclipper.PyclipperOffset() pco.AddPath(poly, pyclipper.JT_ROUND, pyclipper.ET_CLOSEDPOLYGON) result = np.array(pco.Execute(distance), dtype=object) print(result)

numpy == 1.25.2 pyclipper == 1.2.1 the results on arm machine is [[[634 970] [635 970] [635 971] [634 972] [634 973] [633 974] [632 976] [631 976] [630 975] [630 974] [629 973] [629 971] [630 970] [631 970] [632 969]]], while on x86_64 machine is [[[634 970] [635 970] [635 973] [634 974] [634 975] [633 976] [632 976] [630 975] [630 974] [629 973] [629 971] [630 970] [631 970] [632 969]]]