GH-73991: Prune pathlib.Path.delete() arguments

[barneygale]

Remove the ignore_errors and on_error arguments from Path.delete(). This functionality was carried over from shutil.rmtree(), but its design needs to be re-evaluated in its new context. Some possible alternate designs:

• We may wish to support a missing_ok argument (like Path.unlink())
• Or automatically chmod() and retry operations when we hit a permission error (like tempfile.TemporaryDirectory)
• Or retry operations with a backoff (like test.support.os_helper.rmtree()),
• Or utilise exception groups

It's best to leave our options open for now. I don't want to risk these arguments landing in 3.14 final because I got hit by a bus or something.

No news because Path.delete() remains unreleased.

• Issue: Support moving across filesystems in pathlib.Path, as shutil.move() does #73991

---

📚 Documentation preview 📚: https://cpython-previews--123158.org.readthedocs.build/

[pfmoore]

I can understand the logic that we omit the arguments until we're sure we have the right design, but in my experience, a directory tree delete very often needs these arguments (specifically on_error)¹. Finding a read-only file partway through a directory deletion can cause the deletion to fail half-done, leaving the filesystem in a mess. And having to traverse the directory tree, either in advance or after the fact, to locate the offending file and fix the issue, can be extremely costly in a large directory structure.

To put this another way, the key benefit of on_error is to allow problems to be handled in context. Without that capability, I think I'd always advise using shutil.rmtree over Path.delete.

I'd rather defer making Path.delete public until we have the error handling sorted out, over releasing it in a state that makes it strictly worse than shutil.rmtree.

Footnotes

1. For example, I thought that the way .git directories were created on Windows used to mean that they would always cause failures without corrective action in on_error, but when I tested just now, it looks like things might have changed there. Nevertheless, the point about fixing errors in context remains. ↩

[barneygale]

Thanks very much, that's most helpful. I'll close this PR for now, and open a new topic on the forum (in a few weeks time) about what to do with these arguments. I'll try to make the case that the number of practical on_error implementations is limited - in general you retry the unlink() or rmdir(), possibly after a chmod() and/or a time delay - and that there might be a more pathlib-y way to expose these options.

[tjg-global]

A tiny point, somewhat after the event. In the current shutil.rmtree the onerror callback has been deprecated in favour of a similar but different onexc. Same arguments apply as above, and I rather agree with @pfmoore ("the key benefit of on_error is to allow problems to be handled in context").