Question

I’m writing GC friendly code to read and return to the user a series of byte[] messages. Internally I reuse the same ByteBuffer which means I’ll repeatedly return the same byte[] instance most of the time.

I’m considering writing cautionary javadoc and exposing this to the user as a Iterator<byte[]>. AFAIK it won’t violate the Iterator contract, but the user certainly could be surprised if they do Lists.newArrayList(myIterator) and get back a List populated with the same byte[] in each position!

The question: is it bad practice for a class that may mutate and return the same object to implement the Iterator interface?

• If so, what is the best alternative? “Don’t mutate/reuse your objects” is an easy answer. But it doesn’t address the cases when reuse is very desirable.

• If not, how do you justify violating the principle of least astonishment?

Two minor notes:

• I’m using Guava’s AbstractIterator so remove() isn’t really of concern.

• In my use case the user is me and the visibility of this class will be limited, but I’ve tried to ask this generally enough to apply more broadly.

Update: I’m accepting Louis’ answer because it has 3x more votes than Keith’s, but note that in my use case I’m planning to take the code that I left in a comment on Keith’s answer to production.

Answer 1

EnumMap did essentially exactly this in its entrySet() iterator, which causes confusing, crazy, depressing bugs to this day.

If I were you, I just wouldn’t use an Iterator — I’d write a different API (possibly quite dissimilar from Iterator, even) and implement that. For example, you might write a new API that takes as input the ByteBuffer to write the message into, so users of the API could control whether or not the buffer gets reused. That seems reasonably intuitive (the user can write code that obviously and cleanly reuses the ByteBuffer), without creating unnecessarily cluttered code.