The MacOS test is flaky. The specific test case that is failing is the comparison of the reprojection error for a default pet file. That is, what error des get written into the pet-file without having actually calibrated. This means that we do not have real values. On first glance it seems like we do not initialize the values either; so it could be that the code is just writing down whatever is at the current memory location.
I do not know why that would only be on Mac and only now, but since access to undefined variables is undefined behaviour, this still could be the case.
I changed the code such that the repojection error is zero-initalized by the default constructor. This should fix the problem.
To discuss is, whether we should just accept this change, once it is verified this does not break something or whether we want to run the MacOS regression test n times with this, before we trust in this being a fix.
Closes #397 (closed)