I am curious as to how the C code can use doubles to
determine that a value is less than DBL_MIN.

I am pretty sure you can have de-normalized numbers that are smaller than this. That is the smallest normalized number. De-normalized numbers would still have the minimum exponent but would have more leading zeros in the mantissa. With 53 bits of mantissa in a double you can go about 1E-15 smaller than that number. I just ran into this when I added the arbitrary precision math to Fracton. Here is a page that has the definition for all those terms for IEEE floating point.