Unlock a world of possibilities! Login now and discover the exclusive benefits awaiting you.
Accepted Solutions
.png "Former Employee"){kind=icon}
There is a design difference. Or rather, it is a bug-fix. There was a bug that caused a small unnecessary internal rounding, and that resulted in some corner cases being incorrectly rounded. But it was really only visible in cases where you wanted a tie-breaking, e.g.: Should 0.65 be rounded upwards to 0.7 or downwards to 0.6? So the version 12 behaviour should be more correct.
Note also that 102.065 cannot be exactly represented internally - it has no exact binary representation. Hence, it is really not a tie-breaking issue. The number is always below or above the cutoff limit. See Rounding Errors.
If you want to test how the QIX engine handles real tie-breaks, you need to use exact binary numbers. Then you will find that it always rounds upwards, and that it hasn't changed between v11.2 and v12.
There is a design difference. Or rather, it is a bug-fix. There was a bug that caused a small unnecessary internal rounding, and that resulted in some corner cases being incorrectly rounded. But it was really only visible in cases where you wanted a tie-breaking, e.g.: Should 0.65 be rounded upwards to 0.7 or downwards to 0.6? So the version 12 behaviour should be more correct.
Note also that 102.065 cannot be exactly represented internally - it has no exact binary representation. Hence, it is really not a tie-breaking issue. The number is always below or above the cutoff limit. See Rounding Errors.
If you want to test how the QIX engine handles real tie-breaks, you need to use exact binary numbers. Then you will find that it always rounds upwards, and that it hasn't changed between v11.2 and v12.
