confirmed with GCC that these are actually enforced (a function with an enum type is not compatible with a function with an int type)
remove the functiondef reference which was necessary in several df analyses to compute the fixpoint. the reference is not necessary in the new infrastructure.
on travis-ci a few of the hundreds of external CPP calls typically fail for no apparent reason. it's unclear whether it's a problem on travis or with the scala process API or something else. it's now simply trying to preprocess a file up to three times if the output is empty to work around those problems
previously we implemented our own fixpoint computation because of performance issues in kiama. first performance tests show that the variant using kiama is faster than the old, self-written fixpoint implementation.
properly propagated using solely the size of the map to determine knowledge gain is not enough. we compare maps directly to track differences in feature expressions also.
this clutters the output of the test log, where errors are often expected and caught.