 |
 |
Hi all, in the last days I noticed very strange behaviours of our analysis code and differences between the output of code versions with no actual difference. I discovered that if I simply add variables declared in the .hh file I got different numbers. I found a memory corruption for the variable tproTrk. It should have value in the range 0-1 (it is a probability and I verified this at root file level) but I got nan's, values<0. and so on. Its values depend on the number of variables declared in the .hh and on the position of the declaration of tproTrk itself. Since we use this variable to define the GTVL track criterion (tproTrk[i] >= 0.) and since we are using GTVL, our track selection could be screwed up. I am not able to find a configuration to get this variable correct at the moment. But this bad behaviour implies a more general comment. How can we know and discover whether the memory is overwritten or not? This problem could be present in many variables and it should be very subtle to be found. Moreover this effect can affect in a different way data and MC since we are handling more variables in the MC. Daniele