------- Comment #3 from jkherciueh at gmx dot net 2007-10-19 10:13 ------- > Actually, the problem happens only for some specific values of the range, like > eng.max() / 5.5, dividing by 2 or 10 is ok. Indeed, seems a binary arithmetic > problem.
I conjecture that the problem happens if and only if the range size does not divide evenly into eng.max(). In fact, the fractional part determines which buckets get hit more often. If you do 4.3 instead of 5.5, the first 3 out of 10 or the first 6 ot out 20 buckets get the load. If this conjecture is correct, it would be more accurate to say that the algorithm performs as promised only for a small set of range sizes (the divisors of eng.max()) and is biased for all other values. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33815
