[Neallin-917]

Summary

• treat an empty platform.mac_ver() result as an unknown macOS version and try the modern backend first
• preserve the legacy backend for known pre-10.9 systems and as the existing fallback when the modern backend cannot be imported
• add selector-level regression coverage without calling either trash implementation

Fixes #110.

Validation

• python -m pytest tests/test_plat_mac.py -q — 6 passed
• python -m flake8 . — passed
• python -m pytest -q with an isolated HOME — 17 passed, 1 Windows-only module skipped
• python -m build — sdist and wheel built successfully

The regression executes the real backend selector with platform.mac_ver and the native backend import boundaries controlled. It does not claim to reproduce the reporter's complete Homebrew/Jupyter environment. The full macOS suite only operated on generated test fixtures; no existing user files were read or trashed.

AI assistance

Codex assisted with implementation and verification. A separate Codex agent independently reviewed the backend-selection semantics and regression-test path. No human code review is claimed.

[Neallin-917]

The CI workflows on the current head are still awaiting approval. When convenient, could a maintainer approve the runs so the upstream platform checks can execute? The local macOS results and their limitations are recorded above. Happy to address any findings.