https://issues.apache.org/ooo/show_bug.cgi?id=117841
--- Comment #8 from orcmid <orc...@apache.org> --- (In reply to comment #7) > This is not a defect. [ ... ] (The form of > arithmetic used can only represent fractional numbers as integers divided by > integer powers of 2. It's the same problem that there is in representing > 1/3 in decimal, only now 1/10, 2/10, and 3/10 also require approximations in > the computer arithmetic.) [ ... ] I just found this delightful article that will provide more than most ever want to know about floating-point arithmetic and its reliance on approximations: http://randomascii.wordpress.com/2013/02/07/float-precision-revisited-nine-digit-float-portability/ With current high-performance floating-point arithmetic, the closest approximation to 1/10 is 0.100000001490116119384765625. Sometimes these deviations cancel out, but it is a function of the order in which the computation is performed and whether some of those low-order digits are or not. When some intermediate results are large enough to lose some of the low order digits, subsequent results that would be 0 mathematically will fail to have cancelled everything out. -- You are receiving this mail because: You are on the CC list for the bug.
