[Kst] [PATCH] skip off-by-one

[Barth Netterfield]

The new code is wrong.  It always rounds down, even if you start on a skip 
boundry:
new_f0 = 2, skip = 2:
old: ((2-1)/2 + 1) *2 = 2
new: (2-1)/2*2 = 0

new_f0 = 1, skip = 2
old: ((1-1)/2 + 1) * 2 = 2;
new: (1-1)/2*2 = 0;

new_f0 = 0 skip = 1 (eg, more than 1 spf, want only 1 spf)
old: (0-1)/1+1) * 1= 0
new: (0-1)/1 * 1 = -1

The new code rounds up to the next skip boundry.  The only trouble with this 
might come if new_f0 between the last skip boundry and the end of the file.

Check the nf calculations...

cbn

On September 22, 2005 09:07 am, George Staikos wrote:
> I haven't tested this enough to feel comfortable about it yet so I'd like
> some feedback.  I think we have an off-by-one bug in skip mode.  On planck
> data we were seeing occasional cases where the last value in the vector was
> not populated on an update, but there clearly was enough data for it.  This
> resulted in "going back in time" in the index/time vector, causing a line
> across the graph.  This patch fixes it.  The /Skip*Skip is necessary to get
> the integer rounding correct.  I will comment that before I commit.
>
> Index: kstrvector.cpp
> ===================================================================
> --- kstrvector.cpp      (revision 462954)
> +++ kstrvector.cpp      (working copy)
> @@ -468,7 +468,7 @@
>    if (DoSkip) {
>        // change new_f0 and new_nf so they both lie on skip boundaries
>        if (new_f0 != 0) {
> -        new_f0 = (((new_f0-1)/Skip)+1)*Skip;
> +        new_f0 = (new_f0-1)/Skip*Skip;
>        }
>        new_nf = (new_nf/Skip)*Skip;
>    }