running alpha_s setup not properly working

[Juergen Reuter]

This is probably because of MD5 sums. The file alphas.sin from the test suite gives identical results for the second and third integration (which shouldn't be the case). Changing the order of the alpha_s evolution changes the result of the alpha_s_from_mz case, and accordingly the one from pdf_builtin. So this is broken.

[Juergen Reuter]

We need a QCD MD5 sum, going to do this. 3RD TRY TO TYPE THIS INTO THE FUCKING TRACKER!!!!

[Juergen Reuter]

Introduced MD5 sums for the QCD setup in r4897. The update of the phase-space works from fixed to non-fixed, but not from non-fixed to running from pdf_builtin. Still have to investigate this. Also, with 4.6.3/4 there is a double free corruption in the alpha_s case, which we have to look into.

[Juergen Reuter]

when using alphas.sin from the self tests, in the second case it says "qcd setup has changed" but not for the third integration, why?

[Juergen Reuter]

It actually does change, but not if it is three processes, only for the second one :(

[Juergen Reuter]

Maybe connected to #538.

[kilian]

This was actually an error in the test script: the flag ?alpha_s_from_mz takes precedence over ?alpha_s_from_pdf_builtin.

[Juergen Reuter]

Please recheck the phase space MD5 sum setup.

[Juergen Reuter]

As we talked about that offline... please set the defaults as agreed upon and close the ticket (if really phase space update happens in all cases in the test) (did you run it without the --r option outside the testsuite setup)

[kilian]

Gotcha ...

It turns out that the QCD setup is always checked for the phase space, but not for the grids. It should be the other way round -- there is no reason why the phase space should depend on the way of computing alpha_s. Will correct this.

[kilian]

Finally -

All over the place, the QCD setup was incorporated in the phase-space configuration, which it should not affect. Instead, it was missing from the process (i.e., integration object) configuration, which should depend on it. This is now corrected: r5017.

Also, we require now that exactly one flag for the QCD alpha calculation mode is set at a time.