On 18 January 2014 18:16, Brendan Eich <[email protected]> wrote: > Nick Krempel wrote: > >> Whether log(0) is -Infinity or NaN should depend in some sense on what >> side you approach 0 from (I arbitrarily claim to be approaching it from the >> left in my formula, to give a NaN result there too). >> >> I feel Math.log(-0) should be NaN in js for that reason, but it is >> defined to be -Infinity in the standard. Maybe there are industry standards >> pertaining to floating point transcendental functions that mandate this? >> Similarly, Math.sqrt(-0) is -0 rather than NaN. Perhaps using the more >> correct NaN values in these cases have caused more problems than they have >> solved in practice? >> > > These cases (log(-0), sqrt(-0)) conform to IEEE 754. See >
Thanks. Didn't know it covered log. > The "Signed Zero" section of > > http://docs.oracle.com/cd/E19957-01/806-3568/ncg_goldberg.html > > is worth reading. > Yes, and it supports my case: "Another example of the use of signed zero concerns underflow and functions that have a discontinuity at 0, such as log. In IEEE arithmetic, it is natural to define log 0 = -∞and log x to be a NaN when x < 0. Suppose that x represents a small negative number that has underflowed to zero. Thanks to signed zero, x will be negative, so log can return a NaN. However, if there were no signed zero, the log function could not distinguish an underflowed negative number from 0, and would therefore have to return -∞." Further reading from Prof. Wm. Kahan: > > http://www.cs.berkeley.edu/~wkahan/ > http://www.cs.berkeley.edu/~wkahan/ieee754status/ > > Also good links. Think I've read most of it already. Kahan's paper on the correct definitions of transcendental functions along branch cuts mostly relates to the complex-valued versions, but IIRC has sqrt(-0+0i) = +0, not -0 in any case. Nick
_______________________________________________ es-discuss mailing list [email protected] https://mail.mozilla.org/listinfo/es-discuss
