Floating point numbers are calculated internally by DGD with a precision of 36
bits. When converted to a string for display, however, only 8 digits of
precision will be seen.

Arithmetic operations on floats will contaminate other integer operands into
floats and return a floating point result. Note that operations on pure
integers will never return a floating point result.

The following builtin functions manipulate floats:

tofloat(), floor(), ceil(), frexp(), ldexp(), modf(), exp(), log(), log10(),
pow(), sin(), cos(), tan(), atan()

The sqrt() builtin will return a float if given a float argument. Also, min()
and max() will accept float arguments.

A new error, E_OVERFL, can result if a floating point operation causes a
floating point underflow or overflow.

A new builtin variable, FLOAT, can be used with the typeof() builtin function
to detect float values.

N.B. Floating point values are distinct from (integral) numbers; to convert
between types NUM and FLOAT, use tonum() and tofloat().