My thoughts exactly. Tom Christiansen brings these things up as if they were a huge surprise, but they aren't, or at least shouldn't be. To put it another way,

My thoughts exactly. Tom Christiansen brings these things up as if they were a huge surprise, but they aren't, or at least shouldn't be. To put it another way, it isn't Apple doing something wrong. If your code depends on rounding errors in FP representations, your code is buggy. That's also why good numerical algorithms are hard to design. You need to worry about losing precision. As a simple example, if you calculate (a-b)c and a is almost equal to b, you're much better off doing ac-b*c. These tricks are important.