Floating-point exceptions in the GSL

[Thanh Vo <thanhvv@gmail.com>]

Hi,

My name's Thanh Vo. My research group is developing a tool to detect
floating point exceptions in numerical code. On Aug 11th 2010, we contacted
the list to report exception triggering inputs we had found in the GSL.
Brian Gough confirmed the exceptions were valid warnings.  We have continued
our work since then.  In this email, we report more, potentially problematic
exception-triggering inputs.

For each exception, we describe the name of function, the type of exception
thrown, the inputs that trigger the exception, and the source line that
throws the exception.

1.  gsl_sf_bessel_Inu_scaled_asymp_unif_e
Invalid: nu = 0.000000e+00 and  x = 0.000000e+00
Location: line 359 at bessel.c

2.  fg_asymp
Divide ByZero: x = 0.000000e+00
Location:  line 228 at sinint.c

3.  gsl_sf_legendre_P3
Overflow: x = -5.996154e+153
Location: line 95 at legendre_poly.c

4.  gsl_sf_gammastar
Underflow: x = 8.988466e+307
Location: line 1342 at gamma.c

You can see the full list of potential bugs at

barr.softwareprocess.es

If anyone could confirm whether any of these exceptions are useful warnings
or actual bugs, we would appreciate it.

We are finishing the tool and will release it as open source after
publishing the results.

Looking forward to hearing from you,

Best regards,

Thanh Vo.