Re: [Bug-gsl] Chebyshev series derivatives

[Brian Gough]

At Wed, 3 Mar 2010 10:03:27 +0100,
Jerry Gagelman wrote:
> I have found that the function gsl_cheb_calc_deriv() issues a segmentation
> fault when both the Chebyshev series and the series for its derivative are
> allocated for order N=1. No other value of N seems to effect the same error.
> There are two different installations of the GSL that I use at home versus
> at work: GSL 1.11 on Linux OpenSUSE 11.0, and GSL 1.12 on OS X 10.6
> (installed via macports). The test code that I have included below produces
> the same error on both systems.

Thanks for the bug report and example program.  I've confirmed the
problem and logged it at https://savannah.gnu.org/bugs/index.php?29139
There is a loop which does not handle the case n=1 properly.  A patch
is below and will be committed to the repository.

-- 
Brian Gough

GNU Scientific Library -
http://www.gnu.org/software/gsl/


diff --git a/cheb/deriv.c b/cheb/deriv.c
index 757d09f..8f7c223 100644
--- a/cheb/deriv.c
+++ b/cheb/deriv.c
@@ -49,10 +49,8 @@ int gsl_cheb_calc_deriv(gsl_cheb_series * deriv, const 
gsl_cheb_series * f)
   if(n > 1) {
     deriv->c[n-2] = 2.0 *(n-1.0) * f->c[n-1];
 
-    for(i = n-3; i>0; i--) 
-      deriv->c[i] = deriv->c[i+2] + 2.0 *(i+1.0) * f->c[i+1];
-
-    deriv->c[0] = deriv->c[2] + 2.0 * f->c[1];
+    for(i = n; i>=3; i--) 
+      deriv->c[i-3] = deriv->c[i-1] + 2.0 *(i-2.0) * f->c[i-2];
 
     for(i = 0  ; i<n ; i++) 
       deriv->c[i] *= con;