I really liked that improvement, and didn't think it needed to be removed. That's why the PR reverts it partially. I think the os.stat improvements were in the other methods changed, and should not be changed in my PR - unless I changed it by accident.

So with the current PR for this issue, or with your suggested patch, both should have similar performance I think. (I hadn't seen that script to measure performance, thanks.)

I am happy if we revert partially, or if we build the fix on top of the current code consuming the iterator. Should I update the PR with your suggested fix then?