test_timer failure in v4.0.x

I noticed the following failure in the v4.0.x CI (https://github.com/astropy/astropy/pull/11398/checks?check_run_id=2147521346)

    @pytest.mark.filterwarnings("ignore")
1081
    def test_timer():
1082
        """Test function timer."""
1083
        from astropy.utils.timer import RunTimePredictor
1084
        p = RunTimePredictor(func_to_time)
1085
    
1086
        # --- These must run before data points are introduced. ---
1087
    
1088
        with pytest.raises(ValueError):
1089
            p.do_fit()
1090
    
1091
        with pytest.raises(RuntimeError):
1092
            p.predict_time(100)
1093
    
1094
        # --- These must run next to set up data points. ---
1095
    
1096
        p.time_func([2.02, 2.04, 2.1, 'a', 2.3])
1097
        p.time_func(2.2)  # Test OrderedDict
1098
    
1099
        assert p._funcname == 'func_to_time'
1100
        assert p._cache_bad == ['a']
1101
    
1102
        k = list(p.results.keys())
1103
        v = list(p.results.values())
1104
        np.testing.assert_array_equal(k, [2.02, 2.04, 2.1, 2.3, 2.2])
1105
        np.testing.assert_allclose(v, [0.1, 0.2, 0.5, 1.5, 1.0])
1106
    
1107
        # --- These should only run once baseline is established. ---
1108
    
1109
        with pytest.raises(ModelsError):
1110
            a = p.do_fit(model='foo')
1111
    
1112
        with pytest.raises(ModelsError):
1113
            a = p.do_fit(fitter='foo')
1114
    
1115
        a = p.do_fit()
1116
    
1117
        assert p._power == 1
1118
    
1119
        # Perfect slope is 5, with 10% uncertainty
1120
        assert 4.5 <= a[1] <= 5.5
1121
    
1122
        # Perfect intercept is -10, with 1-sec uncertainty
1123
>       assert -11 <= a[0] <= -9
1124
E       assert -8.99263929034811 <= -9
1125

This might be a non-deterministic numerical accuracy failure? It’s curious that the value is so close to -9 when it should be -10?

[github-actions[bot]]

This issue was labeled as Close?. Remove Close? label or this will be closed after 7 days. (This is currently a dry-run.)

[pllim]

Re: bot – Actions is only in dry run mode because upstream needs more fixes. astropy-bot is, uh, on vacation.