memory not allocated when customColvar is used as a cvc with vector-valued sub-cvcs?

[zwpku]

Hi,
I just report a minor issue/bug on the definition of the variable cv_force through the line
colvarvalue cv_force(current_cv_value.type());
in the function void colvar::customColvar::apply_force(colvarvalue const &force) located in the file colvarcomp_combination.cpp.

It seems that this definition does not allocate memory when a sub-cvc is vector-valued. This may lead to segmentation fault, which can be recovered with a CV defined as follows:

colvar {
  name nn_0
  customFunction z
  lowerBoundary -2.5
  upperBoundary 3.2
  width 0.05
  extendedLagrangian on
  extendedFluctuation 0.1
  extendedTimeConstant 200

  customColvar {
    name z
    customFunction x1+y2
    cartesian {
      name x
      atoms { 
    atomNumbers {
      1 
    }
      }
    }
    cartesian {
      name y
      atoms { 
    atomNumbers {
      2  
    }
      }
    }
  }
}

This issue could be resolved by replacing the definition of cv_force with:

colvarvalue cv_force(current_cv_value);
cv_force.reset();

[HanatoK]

@zwpku Thanks for reporting the bug! I thought the two ways of construction, colvarvalue(Type const &vti) and colvarvalue(colvarvalue const &x); reset() should behave the same, but to my surprise they don't, as the type information of colvarvalue does not contain the size of the vector. Could you open a PR?

Update:
Your PR just reminds me another constraint of using nested CVCs in linearCombination, so I open a PR to fix both.

[giacomofiorin]

I agree that the behavior should be better documented, specifically to remind that the Type enum alone is not sufficient for variables that are vectors of arbitrary length. I added a commit to PR #571 to that sense.