#4027 core test failures due to numerical tolerances exceeded

[Michael Orlitzky]

In bug 3974 I reported some test failures, most of which are a result of passing share_tests=true to the test suite. Without the share_tests we still get a few failures though.

With ecl-21.2.1:

Running tests in rtest_gamma: 
********************** Problem 384 (line 1846) ***************
Input:
relerror(gamma_incomplete(0.5, (- 2.5) %i), 
                    (- 0.5969141790423885) + 0.009214957317429537 %i, 2.81e-15)


Result:
3.078728538817989e-15

This differed from the expected result:
true

********************** Problem 390 (line 1882) ***************
Input:
relerror(gamma_incomplete(0.5, 2.5 %i), 
                    (- 0.5969141790423885) - 0.009214957317429537 %i, 2.81e-15)


Result:
3.078728538817989e-15

This differed from the expected result:
true

825/827 tests passed

With clisp-2.49.92:

********************** Problem 209 (line 990) ***************
Input:
                                                        3
                                      3 sqrt(%pi) gamma(-)
                                                        4
closeto(carlson_rd(0, 2, 1.0) - float(--------------------), 6.66134E-16)
                                                  1
                                            gamma(-)
                                                  4


Result:
8.881784197001252E-16

This differed from the expected result:
true

[Robert Dodier]

Thanks for taking a look at it. Looks like these are due to the tolerance for floating point comparison being exceeded by a little bit.