| kislinsk | |
| Mar 20 2014, 5:44 PM |
| F1111: mylyn-context.zip | |
| Mar 26 2014, 2:51 PM |
http://docs.mitk.org/nightly-qt4/mitkVector_8h_source.html#l00467
The subtraction result in line 469 is converted implicitly to float by std::fabs() before it is compared to mitk::eps.
Use std::abs() instead to keep precision.
Attention! This might cause some tests to fail due to increased precision. *DO NOT* change mitk::eps again to fix this issue - fix the tests!
Which tests would fail after these changes? Is there a realistic chance that they will be fixed until Tuesday?
I did not fix it so far. It is just an assumption that a few tests might fail since we had many precision incidents regarding tests in the past. Fingers crossed, maybe all tests run successfully without any adaptions.
I'll give it a try next Tuesday if no one is going to do it the next days.
I do not think fabs converts to float. See e.g. mathcalls.h and math.h or
http://www.cplusplus.com/reference/cmath/fabs/
can it be that fabs was confused with fabsf, which is the explicly float variant of the function for some compilers?
I suggest no code alterations.
In that particular case I agree with Sebastian because fabs is overloaded for both float and double arguments.
In general I agree that as a rule of thumb "abs" should be preferred since in C++ by including the "cmath" header "abs" is overloaded for all primitive types and the compiler automatically selects the most appropriate one.