Midpoint test failed

[Gustavo Gutierrez Sabogal]

Hello,

As i wrote in a previous personal mail, when i compile the library and do make check two tests are failing on my machine:
the midpoint test and the division test.

The test file i am considering is the midpoint created by the compilation of midpoint.cpp

After see that make check give me a two failed tests messages, i edit the midpoint.cpp file to get a trace of the inputs causing the error, so i modify it by adding only a cout line:

for (int i=0;i<nb_intervals;++i) {
    cout << "midpoint(" << test_array[i].desc << ") =";
    cout << test_array[i].I.midpoint();
    if (test_array[i].I.set_contains(test_array[i].I.midpoint())) {
      cout << "       included" << endl;
    } else {
      //if here, there is a test error (correct me if wrong!!)
      cout << "       NOT included!! " << test_array[i].I.midpoint() << " in "<< test_array[i].I << endl;
    }
  }

All tests prints an included message but there is one that doesn't:
midpoint([empty]) =0       NOT included!! 0 in [empty]

I am not so familiar with your library so i don't know if i am going in the right way.

BTW, congratulations, your library is very useful and it  has a good design. The best part is that get away the setting of the rounding modes, that's great.

Regards,
ggutierrez

[Frederic Goualard]

Hello Gustavo,

The statement is correct but the result is not!

Actually midpoint([empty]) should be NaN (which is, still, not included in [empty]) as stated in
the comment for that method. This is indeed what is
returned if the empty interval is interval::emptyset. However, for an interval which is empty but "symmetric", such as, say [oo, -oo], it returns 0.0 because the method lacks of an explicit check for emptyness.

I have corrected this bug in Version 0.2.6 that should
find its way on SourceForge in a matter of hours.

Thank you for the bug tracking!

Cheers,

Frederic.

[Gustavo Gutierrez Sabogal]

Hello Frederic,

Wiith the new 0.2.6 version the bug was corrected and the midpoint test pass on my machie, but the division test does not. Since the test code is a little different than the midpoint one, i can not give you a trace for the bug, i only can write off the output:

PASS: cos
[-0.6666666666666667, -0.3333333333333333]
FAIL: division
[-inf, -max] % [-inf, -max] = [0, inf]
[-inf, -max] % [-inf, -1] = [0, inf]
[-inf, -max] % [-inf, -min] = [0, inf]
[-inf, -max] % [-inf, -0] = [0, inf]
[-inf, -max] % [-inf, 0] = [0, inf]
[-inf, -max] % [-inf, min] = [-inf, inf]
[-inf, -max] % [-inf, 1] = [-inf, inf]

because of the sign '%' i think the relational division is the failing test, i don't know if all output is useful so i put only a few lines.

The main loop for the relational division test is the following, however it is  a little different and the result is not tested for inclusion with any other known (teoretical) interval.

for (int i=0;i<nb_intervals;++i) {
    for (int j=0;j<nb_intervals;++j) {
      cout << test_array[i].desc << " % " << test_array[j].desc << " = ";
      cout << test_array[i].I % test_array[j].I << endl;
                                                                                     }
Any help you can give me is highly appreciate,

Regards,
Gustavo.

[Frederic Goualard]

Hello Gustavo,

My mistake: division.cpp contains a call to exit(1)
that should be removed, together with the line just before it. These were added to make a specific test and I forgot to remove them afterwards.

Everything should go smoothly from that point.

Cheers,

Frederic.

[Gustavo Gutierrez Sabogal]

Hello Frederic,

I did what you told me and now all tests passed.

Thanks a lot,

Gustavo