Use a wall clock to determine the duration of the tests

[k15tfu]

Hi Raffi!

We run tests on build agents and I noticed that the real duration of test cases (build server calculates it using test output) is much smaller than that measured by Boost.Test. Currently it uses boost::timer which is based on std::clock(). Probably usage of something like high_resolution_clock will be more preferable.

Do you need help with PR?

[raffienficiaud]

Hi @k15tfu , would you please provide me with some hints? (measurement differences and platform). I do not believe the precision of std::clock is the issue there, and the time measurement reported are all wall time. I cannot really use high_resolution_clock as this part of the code should be compatible with C++03.

Thanks!

[k15tfu]

@raffienficiaud , VMs powered by VMware ESXi on MacPro6,1. According to the log file the test lasted 34m:36s, but Boost reported 4039503ms (~1h07m).

[raffienficiaud]

Thanks for the quick reply. Looking at the code I cannot see anything obviously wrong. I am just wondering: are your measurements reported by the Junit output?

[raffienficiaud]

BTW: one measurement seems to be the double of the other, my feeling is that we are counting the same thing twice.

[raffienficiaud]

Would it be possible for you to test the branch topic/GH-228-erroneous-tests-duration ? Thanks.

[raffienficiaud]

@k15tfu from my checks, I do not see the problem with the timing. The time for each tests is taken to be the wall clock time at the suite level, including some boost.test internal timing. This is why usually we have a test suite time that is greater than the accumulation of the its children test time.

The branch I mentioned topic/GH-228-erroneous-tests-duration does not fix any issue.

I need your input, thanks!