From koffice-devel Sun Feb 11 06:56:25 2007 From: "Ariya Hidayat" Date: Sun, 11 Feb 2007 06:56:25 +0000 To: koffice-devel Subject: Re: Notes on koffice unit tests Message-Id: X-MARC-Message: https://marc.info/?l=koffice-devel&m=117117699008302 > 10, 36 and 38 may have something wrong with the test setup - dies in > QCoreApplication. Or maybe it is in my setup? Does this happen if you change the test (temporarily) to use KApplication directly? This is what I have to do for KSpread, at least until the issue is finally solved in kdelibs. > 23 is failing for some reason I don't understand - kspread formula parser > looks to be at fault. I'll have a look at this, if Stefan is not faster than me.... > 28 and 29 appear to be failing because of excess precision in the comparison > of floating point numbers. Will have a look as well. > Are we comfortable with using less than 15 digits in the floating point > comparisons? Note than the openformula spec uses epsilon of 1e-6 for some > tests, but I couldn't find a definition of what the "=" operator should treat > as "close enough". I guess, yes. I believe 1e-6 is too large for the tolerance. Ideally of course, this should be proportional to the result. In addition, the problem is not only the machine epsilon, but the different algorithms as well. So if a function has two different implementations, it is not always easy to find out which one is the "most correct" one. Regards, Ariya _______________________________________________ koffice-devel mailing list koffice-devel@kde.org https://mail.kde.org/mailman/listinfo/koffice-devel