On Sun, Oct 31, 2010 at 03:57:33PM +0100, Luc Maisonobe wrote: > Le 31/10/2010 15:08, Gilles Sadowski a écrit : > > Hi. > > > > [Continued from the previous message.] > > > > Accepting that the interpolation abscissae are not sorted in strictly > > increasing order make it necessary to check (at every call to the "value" > > method) that no 2 interpolating points are the same. > > The code would be cleaner is we assume (as a pre-condition) that the "x" > > interval is sorted. [And it would be more similar to other interpolating > > classes that explicitely prohibit unsorted interpolation intervals.] > > So I guess you would put this check in the verifyInterpolationArray > method which is called from constructor and remove the d == 0 check in > computeCoefficient ?
Yes. > Beware that the static evaluate method can be called directly from > outside. We need to keep this public method, so we may perhaps split it > as a check followed by a call to a private internal method that does no > check, whereas the value instance method would directly call the > no-check version since the check would have been done beforehand in the > constructor. This is exactly what I've already implemented. What remains is the "problem" of allowing unsorted input arrays. > > > > If we nevertheless want to allow the user to pass an unsorted interval, I'd > > rather have a new utility method that would sort the "x" interval (and apply > > the corresponding permutations to "y") within the constructor. So, what do you think of just require that the array be sorted? [This would imply changing the test, where the array is not sorted.] Gilles --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org For additional commands, e-mail: dev-h...@commons.apache.org
