Merge pull request #2260 from RosettaCommons/roccomoretti/fix_ligand_database_io
Attempt to fix ligand_database_io test
I think I've tracked the instability in this test to the use of a std::set< ConstraintOP > in
ResidueTorsionRestraints. This was used in removing/re-adding constraints, such that the order of
the constraint addition was different run-to-run, as the address layout
is different. Different constraint orders means that there are slight
differences in energy summing (due to machine precision rounding),
which snowball during minimization. Changing the set to a vector1 appears to fix the instability issue.
Integration test changes in ligand docking tests expected, as they settle down to their (hopefully) stable trajectories.