It seems as if the author of LKH has changed the behaviour of the algorithms, which resulted in not writing a found TSP path if the initial solution could not be improved during the local search phase.
Don't know how to proceed. Work around would be:
-If approximation failed due to any reason, switch to expensive exact solving.