I agree that doubling the range may be a good solution depending on the input. My experience has shown me that increasing a valid range by a factor of two is generally stop-gap, but without more knowledge of the problem this is pure conjecture.

I would certainly not promote arbitrary precision in the general case, but in this case the OP has (probably) encountered the native limit, and appears to expect an unreasonable 20-digits of precision based on his format string.