[dzaima]

If you accept 2^32, then code using 32-bit ints is definitely broken on it and thus the OP question of the issue on half that is irrelevant. Which is my point - widening the acceptable input range from 2^31 to 2^32 (or in the case of signed integers, from 2^30 to 2^31; give or take 1 of course) just "fixes" one small case of the actual core issue of nearly any arithmetic anywhere being wrong if you don't explicitly constrain input sizes.

[wat10000]

I agree on there not being much difference between 2^30/31/32. But it’s not “nearly any arithmetic.” If your size is an actual data size, then 2^64 is fine.

[dzaima]

Right, with 64-bit ints things are a lot nicer. Though you can still run into some issues on "generate this much data" tasks as opposed to "operate over existing data of this size", though perhaps less exploitably so.