Test failure in: section on Trail' Line matches section on FixedSegment

[vcunat]

A test failed randomly for us. Tail of log:

    section on Trail' Line where a segment paramater is 0 or 1:                             OK
      +++ OK, passed 100 tests.
    section on Trail' Line matches section on FixedSegment:                                 FAIL
      *** Failed! Falsified (after 88 tests):
      lineFromSegments [bézier3  (V2 23.0 (-0.25)) (V2 (-0.7142857142857143) 4.24278998044767e-2) (V2 0.8608718340787933 (-67.95)),bézier3  (V2 (-2.6987951807228914) (-31.555555555555557)) (V2 0.7314740059995591 (-0.5)) (V2 19.0 79.183908045977),straight (V2 50.0 64.0),bézier3  (V2 8.545454545454545 0.2732184857924349) (V2 76.0 59.87755102040816) (V2 0.6363636363636364 (-19.714285714285715)),bézier3  (V2 (-56.298507462686565) 63.72727272727273) (V2 1.8510250624370617e-3 (-9.866666666666667)) (V2 11.36 0.9410574464856156),straight (V2 0.3254111873510118 43.0),bézier3  (V2 0.7152594007101306 0.7356060177472081) (V2 4.5711020478806486e-2 (-5.0)) (V2 0.1977369792893824 2.3333333333333335),straight (V2 1.6 (-2.3333333333333335)),bézier3  (V2 0.7057442034023289 (-3.5)) (V2 (-83.8) 0.2743888458907714) (V2 4.6922292333571036e-2 (-2.25)),bézier3  (V2 (-68.33333333333333) 44.375) (V2 (-68.18518518518519) 0.8593011719080682) (V2 0.5714285714285714 (-42.07692307692308)),bézier3  (V2 2.3333333333333335 73.11267605633803) (V2 (-0.3333333333333333) 0.5555555555555556) (V2 49.0 0.40414162544588983),straight (V2 5.9984667620534005e-3 0.40875436328888326),straight (V2 81.0 0.8252000613268063),bézier3  (V2 2.0 2.6666666666666665) (V2 (-60.8235294117647) (-54.195402298850574)) (V2 20.476190476190474 (-1.0)),straight (V2 68.62962962962963 3.1739130434782608),straight (V2 0.9597137107425706 (-70.97222222222223)),bézier3  (V2 0.647546743960823 (-47.88505747126437)) (V2 (-3.0) 52.888888888888886) (V2 1.5714285714285714 36.016666666666666),bézier3  (V2 21.844827586206897 (-57.0)) (V2 46.767857142857146 48.0) (V2 (-51.06666666666667) 2.5),bézier3  (V2 (-3.635135135135135) (-74.8695652173913)) (V2 0.5161720103112761 5.0) (V2 0.7870556348432733 59.79545454545455),straight (V2 (-2.6666666666666665) (-1.4)),straight (V2 0.48912524384396394 0.23938197883922974),bézier3  (V2 0.7712584983387674 (-1.0)) (V2 9.129515734180726e-2 (-85.54054054054055)) (V2 1.8 (-74.0)),straight (V2 3.268063163464541e-2 39.8125),straight (V2 70.59701492537313 8.204408303313271e-2),straight (V2 52.0 6.0),straight (V2 (-25.0) (-5.0)),straight (V2 61.857142857142854 (-64.27777777777777)),bézier3  (V2 0.7 7.061137322609556e-2) (V2 41.0 (-57.0)) (V2 (-64.0) 3.5),straight (V2 26.0 68.0),straight (V2 (-84.63333333333334) 0.4),bézier3  (V2 0.9739518529881794 69.35) (V2 (-78.02325581395348) (-39.0)) (V2 0.30014606155932955 0.33661775227641944),straight (V2 53.0 54.382716049382715),bézier3  (V2 0.8913931544296858 0.37071413386195573) (V2 (-6.133333333333334) (-18.352941176470587)) (V2 39.0 (-30.655172413793103)),straight (V2 0.16987629388600267 29.73076923076923),straight (V2 0.4444444444444444 (-2.75)),bézier3  (V2 13.333333333333334 32.95348837209303) (V2 0.4 0.20710160852121628) (V2 (-0.16666666666666666) 0.3694053750047174),bézier3  (V2 29.03448275862069 2.6348201757192657e-2) (V2 40.0 0.4408316023041149) (V2 0.5 78.85714285714286)] `at` P (V2 0.3891976780532279 0.8767513103107396)
      Param (-3.0798250574282093e-6)
      Param (-0.49523454069820794)
      Use --quickcheck-replay=87506 to reproduce.
      Use -p '/section on Trail'\'' Line matches section on FixedSegment/' to rerun this test only.

1 out of 64 tests failed (0.24s)
Test suite tests: FAIL
Test suite logged to: dist/test/diagrams-lib-1.4.6.2-tests.log
0 of 1 test suites (0 of 1 test cases) passed.

This is on aarch64-linux, NixOS and (as you can see) version 1.4.6.2 which isn't the latest but I haven't noticed anything remotely related in the changelog since then (or issues and PRs).

[byorgey]

Thanks for the report! I'm pretty sure that we can just chalk this up to floating-point error: the test in question checks that the results we get from sectioning a trail two different ways are equal up to some error bound, but I guess in this case the accumulated error is just a bit bigger than the allowed error. You can see that one of the parameters used to section the trail is about -3.1e-6 which definitely makes me suspicious.