Thanks for confirming.

> Removing the check of len against 1024^3 would probably also stop this behavior from happening.

Yes, removing it or even checking against 1024 instead of 1024^3 is preventing the bug. Strange behaviour.

> it's usually a better idea to use unsigned types (such as size_t) for natural numbers like len, n, and i, and I would assume your code would compile fine with size_t instead of int.

When I don't need all the space of the unsigned type I like to use signed types to be able to return -1 as an indication of error. Well, should we report a bug?