We focus on the Golomb ruler problem, a hard constrained
combinatorial optimization problem. Two alternative encodings are
considered, one based on the direct representation of solutions,
and one based on the use of an auxiliary decoder. The properties
of the corresponding fitness landscapes are analyzed. It turns out
that the landscape for the direct encoding is highly irregular,
causing drift to low-fitness regions. On the contrary, the
landscape for the indirect representation is regular, and exhibits
comparable fitness-distance correlation to that of the former
landscape. These findings are validated in the context of variable
neighborhood search.