I carefully read both the two PRs. The first one, easy to understand. The second one, I spend some time to figure out why the test doesn't crash, why we need to have reference count checks in two places and make some experiments to test in different cases, how the reference counts will be. I am afraid I have to repeat this procedure after some time when reading the code again. :-( And while in some cases the second approach increases performance. But in others cases it might hurt. Codes storing the array in a variable will go into the iterator branch.