I realized that the poster had cut-paste from the link he has included in the answer. This could have been avoided by careful proof-reading; It might have been a case of oversight.

However I am not clear what exactly the question is testing here? How is it that the rounding behaves differently for two very similar numbers?

When I use float(24) for I get 165.76 in the first select ---> implying rounding behaves as expected. When I use float(25) I get 165.75 --> implying there is no rouding; rounding does not behave as expected.

In the second select statement - the situation is reversed: Now when I use float(24), I get the result - 165.55 implying no rounding. When I use float(25) - I get 165.56 - implying rounding happens as expected.

So when a double byte precision is used for 165.755 - no rounding happens; When double byte precision is used for 165.555 rounding does happen.

What is so drastically different between these two numbers that's causing the difference in behavior?

Does any one know?