The idea behind the sample is that the heap usage is increased in such small doses that finally when there is an actual OOM it will propagate and just exit because there is not enough heap to construct the actual OOMError object.

I was not able to reproduce this with my JDK’s running on Linux with lots of ram. The ram plays a role because starting from certain versions of JDK the default heap is actually dynamic depending on the amount of ram.

Of course the arraylist resize might not be that small and depends on the current size a lot. My JDK 1.5 shows: int newCapacity = (oldCapacity * 3)/2 + 1;

So actually when all the planets are aligned your production system will just halt without a single warning. Bedazzled looks follow as anything could have happened :)