[gang-zh]

Since #73503, when [core] execute_tasks_new_python_interpreter is on (and always on macOS), the task supervisor starts a fresh interpreter with os.posix_spawn instead of fork() followed by execv(). The option's description, its 3.4.0 newsfragment and the two security pages still say the task process is forked and then execs. That describes a mechanism the code no longer uses, and it leaves out the property that motivated #73503: at-fork handlers registered in the supervisor no longer run in the task.

This is a docs-only change with no change in behaviour.

On v3-4-test the newsfragment has already been compiled into RELEASE_NOTES.rst, so this will not cherry-pick cleanly. A manual backport that edits RELEASE_NOTES.rst instead will follow.

Checks run:

• prek run --files on the four files: every hook passed except check-default-configuration, which needs a Breeze CI image that I couldn't build locally. I ran its two commands, airflow config list --default and then airflow config lint, in apache/airflow:3.4.0rc1-python3.12 with this branch's config.yml mounted. The new description renders, and lint reports no issues.
• In the same image, os.posix_spawn did not run os.register_at_fork handlers, while a bare os.fork() ran both before and after_in_child.
• On 3.4.0rc1 with the option on, task processes ran in the spawned bootstrap interpreter (sys.argv == ['-c']) with PR_GET_DUMPABLE at 0, which matches the updated security docs.

related: #72164, #73503

---

Was generative AI tooling used to co-author this PR?

• Yes — Claude Code (Opus 5.5)

Generated-by: Claude Code (Opus 5.5) following the guidelines

🤖 Generated with Claude Code