Only a month or two ago I had to debug a data roundtripping issue with our Hadoop cluster. Decimals being sent in came back as apparently completely random numbers.
More experimentation revealed it was only negative numbers that were being messed up (the original tests passed with positive numbers).
Further bitwise investigation and manual twos-complement calculation, and I discovered the problem to be a mismatch in fixed point precision, where negative numbers weren't being sign extended correctly.
It's not every day you rely on your binary calculation skills, but it does come up.
More experimentation revealed it was only negative numbers that were being messed up (the original tests passed with positive numbers).
Further bitwise investigation and manual twos-complement calculation, and I discovered the problem to be a mismatch in fixed point precision, where negative numbers weren't being sign extended correctly.
It's not every day you rely on your binary calculation skills, but it does come up.