The last commit fixes a long-standing problem of t/chkjac.t failing for float. The fix was to just cast the return value of exp in the lpp function to (FLOAT). Otherwise a floating-point difference happens, which only for the first value of the function gets amplified into a difference of about 1e-1, far outside of the epsilon which is 1e-4.
The last commit fixes a long-standing problem of
t/chkjac.tfailing forfloat. The fix was to just cast the return value ofexpin thelppfunction to(FLOAT). Otherwise a floating-point difference happens, which only for the first value of the function gets amplified into a difference of about 1e-1, far outside of the epsilon which is 1e-4.