[Scip] SCIPgetRelaxBranchCands

[Christian Puchert]

Hi all,

I am currently implementing some primal heuristics for Gerald Gamrath's
GCG which basically replaces the standard SCIP LP relaxation by a
special relaxator.

At the moment, I am trying to adjust the SCIP diving heuristics such
that they work with GCG. These heuristics need the fractional variables
of the current LP (relaxation) solution which they normally access via
SCIPgetLPBranchCands. Since the standard LP relaxation is not present
here, I use SCIPgetRelaxBranchCands instead.

While this function works fine on the root node, the returned number of
fractional variables on deeper nodes is always zero. However, accessing
the current relaxation solution via SCIPlinkRelaxSol (or
GCGrelaxGetCurrentOrigSol) works correctly (and having a look at this
solution reveals that it is actually not integer feasible).

What is the cause of that? Am I using SCIPgetRelaxBranchCands in a wrong
way, and can I use this function at all in my primal heuristics?

Thanks and regards,
Christian