Question

I’m seeing conflicting references in Oracles documentation. Is there any difference between how decimals are stored in a FLOAT and a NUMBER types in the database?

As I recall from C, et al, a float has accuracy limitations that an int doesn’t have. R.g., For ‘float’s, 0.1(Base 10) is approximated as 0.110011001100110011001101(Base 2) which equals roughtly something like 0.100000001490116119384765625 (Base 10). However, for ‘int’s, 5(Base 10) is exactly 101(Base 2).

Which is why the following won’t terminate as expected in C:

float i; i = 0; for (i=0; i != 10; ) {     i += 0.1 } 

However I see elsewhere in Oracle’s documentation that FLOAT has been defined as a NUMBER. And as I understand it, Oracle’s implementation of the NUMBER type does not run into the same problem as C’s float.

So, what’s the real story here? Has Oracle deviated from the norm of what I expect to happen with floats/FLOATs?

(I’m sure it’s a bee-fart-in-a-hurricane of difference for what I’ll be using them for, but I know I’m going to have questions if 0.1*10 comes out to 1.00000000000000001)

Answer 1

Oracle’s BINARY_FLOAT stores the data internally using IEEE 754 floating-point representation, like C and many other languages do. When you fetch them from the database, and typically store them in an IEEE 754 data type in the host language, it’s able to copy the value without transforming it.

Whereas Oracle’s FLOAT data type is a synonym for the ANSI SQL NUMERIC data type, called NUMBER in Oracle. This is an exact numeric, a scaled decimal data type that doesn’t have the rounding behavior of IEEE 754. But if you fetch these values from the database and put them into a C or Java float, you can lose precision during this step.