The BigNumber intrinsic class lets you perform floating-point and integer arithmetic with (almost) any desired precision. BigNumber uses a decimal representation, which means that decimal values can be represented exactly (i.e., with no rounding errors, as can happen with IEEE 'double' and 'float' values that languages like C typically support). BigNumber combines a varying-length mantissa with an exponent; the length of the mantissa determines how many digits of precision a given BigNumber can store, and the exponent lets you represent very large or very small values with minimal storage. You can specify the desired precision when you create a BigNumber explicitly; when BigNumber values are created implicitly by computations, the system chooses a precision based on the inputs to the calculations, typically equal to the largest of the precisions of the input values.

The maximum precision for a BigNumber is about 64,000 digits, and the exponent can range from -32768 to +32767. Since this is a decimal exponent, this implies an absolute value range from 1.0e-32768 to 1.0e+32767. The more digits of precision stored in a given BigNumber value, the more memory the object consumes, and the more time it takes to perform calculations using the value.

Modified in reflect.t[356]:
If desired, modify the BigNumber intrinsic class to provide a to-symbol mapping. We only include this modification if the program is compiled with REFLECT_BIGNUM defined.

intrinsic class BigNumber :   Object