Code review comment for lp:~maddevelopers/mg5amcnlo/3.0.2

Revision history for this message
Olivier Mattelaer (olivier-mattelaer) wrote :

Hi,

Let's try to avoid any confusion here:
1) we do not have any issue with ninja.
   - we can still install it
   - when install it works

2) For some unknow reason, my version of 3.0.2 version was configured as if ninja was not installed in my machine, in that case collier is the next OLP provider and is used as default.

3) We have a bug related to collier that does not return the correct pole (I'm therefore worried about the other output like the finite part of the loop). This bug has been spotted by Eleni and Gauthier and independently by me (thanks to point 2 above)

>We must keep COLLIER as it is crucial for loop-induced efficiency.
I could not reproduce seg-faults in standalone (and neither did Richard I
believe).

Being fast is useless if we are wrong. Like this, this sounds like a bad bug.
If you are sure that this is a MadFKS related bug, then we can forbid the usage of COLLIER only in that case. But it would be nice that you take a real look at this. I prefer to be slow and correct than fast and wrong.

Cheers,

Olivier

« Back to merge proposal