breaking compatibility is like nicotin

Yes, I've read that, and–among other things–sure indicates that this change wasn't quite done well. From what I've seen, the setuptools maintainers sometimes seem to have... peculiar opinions.

But we should still consider the number in perspective.

The setuptools maintainers really don't have chance at all to assess the impact of their changes upfront. The whole Python ecosystem is neither prepared nor equipped for large-scale testing of its foundations. It desperately lacks the equivalent of Rust's crater, all the more for its fragmented tooling.

Their only way to see what they'd break is to rollout the change and see. Consider this, it's quite surprising that this doesn't break more frequently, and in worse ways.