NaN currency
The Daily WTF, an excellent website that I recommend everyone remotely involved in computers has a look at every now and then, has one series of posts where it shows 'strange' or 'funny' error messages from not just desktop computers, but also ATMs and other embedded devices.
One of the more recent entries of these Error'd series contained a picture that made me cringe; it showed 'NaN' for a currency value. Obviously this is an unhandled floating point error, but that's not what made me cringe.
What did make me cringe is that people who are supposed to work with shitloads of money are using floating point numbers to store monetary values. This is stupid. By their nature, floating point numbers are imprecise and inaccurate. The representation of a floating point number is an approximation. This is not the kind of thing one would like for bank transfers. "Mr Foo would like to transfer approximately €100 from his account to that of his neighbour." Err, no; it doesn't work that way.
Instead, whenever you use currency, you should use integer numbers. If your local currency is the US Dollar or the Euro or another similar currency where zero-point-something amounts of money are regularly used, work on the convention that "1" in your variable actually means ".01", or ".1", or whatever is most convenient, and make the right conversion whenever you need to do some output. Do not think you can deal with rounding errors in floating point, because you cannot.
COBOL (I can't believe I'm saying this) actually makes the latter easy, because you can create a statement that says "to output this variable, show the first three digits, then a point, then the last two digits". I believe it doesn't even have support for floating point numbers. Unfortunately this kind of thing is harder in C (and similar languages), but that doesn't mean you should just go ahead and assume that floating points are a good idea...
This is the kind of error I could accept from a first-year graduate student, but not from someone who's supposed to work with money on a daily basis.
Oh well. At least the ATM is about US Dollars, so it's not likely that it's going to be my problem any time soon.
They call that "fixed point".
You can implement fixed point easily with something as "multiply by 100". But there are also libraries with arbitrary precision like gmp.
In "Embedded-C" there is support for fixed point types. gcc has some support for that.
... that would work too, though arbitrary precision is a bit overkill for monetary transactions.
The main point being that floating point math is seriously wrong for that kind of stuff.