Free42 possible accuracy flaw

[Albert Chan]

https://github.com/thomasokken/free42/bl...at.cc#L649

Code:

        // Integral power. bid128_pow doesn't handle these very well,
        // so I'm using repeated squaring instead. This way at least
        // we get exact results for integral powers of ten!
        if (x < -2147483647.0 || x > 2147483647.0)
            // For really huge exponents, the repeated-squaring
            // algorithm for integer exponents loses its accuracy
            // and speed advantage, and we switch to the
            // library's bid128_pow() instead.
            goto noninteger_exponent;

I would have reduce integer exponent range a bit, possibly remove it.
For exponent 1e8, that is already 27 squaring's, plus many multiplies !
No wonder the result is off.

Let's bypass the integer squaring part, and see what happen.

1 1e-8 + 32e8 Y^X       → 78962947548608.16088504664995794553
√ √ √ √ √                   → 2.718281814867636217652977243009177