> > > > counter less than total_count/N (they are not useful anyway). > > > > However it seems that I missed a problem here. With step 2 and 4 the > > merge is not distributive. > > > > I added 2 and 4 trying to work around the fact that we have no > > convenient place to do pruning if merging is not performed (since merge > > functions are not called at all). I think we need to invent such place > > - just do that on the begginign of dump_one_gcov. > > Hm, anyway, stage3 closes today and it's definitely next stage1 material.
It is a regression, so we can still fix it in stage4 :) Honza
