Re: make check error

[Matt Newville <newville at cars6 dot cars dot aps dot anl dot gov>]

> Can you try the program below to see if the wrong behavior of fabs()
> is reproducible. If so, that is the problem.
> 
> #include <math.h>
> #include <stdio.h>
> 
> int main()
> {
>   double x, y;
>   x = DBL_MIN;
>   y = fabs(x);
>   printf("x= %e fabs(x)= %e\n", x, y);
> }
> 
> $ ./a.out 
> x= 2.225074e-308 fabs(x)= 2.225074e-308

This works fine with RH7.0 and the doomed gcc2.96.  In fact,
adding a simple printf(" "); statement to the "volatile section"
of gsl_sf_multiply_impl() will cause the test to work correctly
(ie, return Underflow).  Using CFLAGS =-O0 when compiling
elementary.c will also successfully evaluate an Underflow.

Sorry I can't be of more help right now.

--Matt Newville

PS: And for the record, I'm usually for erring on the side of
    pedantic, and definitely in favor of holding on to that
    illusion of robustness and correctness.  Keep it up!