Le 6 juin 2014 à 01:18, C. Scott Ananian <[email protected]> a écrit :
> On Thu, Jun 5, 2014 at 7:15 PM, Axel Rauschmayer <[email protected]> wrote: >> ```js >> // Compact ECMAScript 6 solution >> // Risk: number overflow >> [1, 5, 3, 12, 2].sort((a,b) => a-b) >> ``` > > Is this really an issue for IEEE floating point? > --scott Right. More precisely, if `a` and `b are both finite, but too far apart for `a - b` to be representable by a finite number, we get either `a - b == Infinity > 0`, or `a - b == -Infinitiy < 0`, so that `a` and `b` are still correctly ordered. There is a potential issue if some of the numbers you want to sort are infinite, because you get, in some occasions, `a - b is NaN` where you need `a - b == 0`. In that case, the behaviour of the sort function is explicitly left undefined by the current ES specification. This could be simply resolved spec-wise by requiring to interpret the result of the compare function as `0` whenever it produces `NaN`. For that case, I've filed: https://bugs.ecmascript.org/show_bug.cgi?id=2978 (Naturally, if some of the numbers are `NaN`, you are out of luck.) —Claude _______________________________________________ es-discuss mailing list [email protected] https://mail.mozilla.org/listinfo/es-discuss
