[KennyMcSimpson]

set_window_title() now flushes terminal title updates immediately and keeps them on the output owned by the active prompt-toolkit application.

What changed

• A title update is written and flushed in one callback. When an application loop is running, the callback is scheduled on that loop; an inactive or closed loop keeps the existing direct-output fallback and unrelated scheduling errors still propagate.
• Cmd.select() now records its temporary child AppSession while choice() is active. set_window_title() follows the parent-to-child route so an ordinary background thread reaches the menu session even though it does not inherit the child context.
• The registry is private, lock-protected, cleaned by identity in finally, supports nested menus and out-of-order completion, and never holds its lock while writing output or scheduling callbacks.
• Running applications use app.output; before an application starts, the routed session output is used. Unrelated custom sessions remain isolated.
• The prompt guide and the 4.3.2 changelog entry document the behavior.

Validation

• The branch includes upstream main at e9c64db568404a6ec384f156656768d77a026c04; the 4.3.1 release section is preserved and this fix is listed under 4.3.2.
• Against the exact previous published implementation (2997f51), an independent reference using the public Cmd.select() and Cmd.set_window_title() APIs reports 24 failures in 55 checks. Commit e49ec13b reports 0 failures in all 55 checks.
• The independent reference covers ordinary no-context workers, distinct ambient and application outputs, Ctrl-C cleanup, foreign session isolation, and active applications whose app.output differs from the ambient session output. Repository regressions also cover nested menus and out-of-order menu completion.
• make check: all hooks, Ruff, ty, and mypy pass. make test on Python 3.13: 1,879 passed / 159 skipped. Python 3.11: 1,878 passed / 160 skipped. make docs-test: no issues. Focused title/select tests: 27 passed; the complete tests/test_cmd2.py module: 351 passed / 34 skipped.

The current hosted head c78994f8 has 29 successful checks, including quality, type checks, docs, CodeQL, coverage, and 17 of 18 OS/Python test jobs. The macOS/Python 3.15t job again fails in the unchanged test_pipeline_stops_with_cmd2_and_returns_terminal[wrapper-stop-and-interrupt-busy-shell-producer]: it times out at tests/test_pipeline_job_control.py:335 after a job-control signal, with 2,035 tests passed, 1 failed, and 2 skipped. The aggregate test check consequently fails as well. That case exercises a shell pipeline and does not call select() or set_window_title(), and the test file matches upstream main. The precise cause has not been established in a controlled macOS reproduction. GitHub denied the earlier attempted failed-job rerun with 403 because this account lacks the required upstream repository permissions.

Commit c78994f8 adds the requested traversal comment: workers may still see the parent session, so registered children are followed to the innermost active select() session, with the latest active child selected when menus share a parent. This follow-up changes comments only. make check, make test (Python 3.13: 1,879 passed / 159 skipped), and make docs-test pass locally.

AI assistance was used to investigate, implement, and validate this pull request.

Closes #1782.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.75%. Comparing base (e9c64db) to head (c78994f).

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1783   +/-   ##
=======================================
  Coverage   99.75%   99.75%           
=======================================
  Files          23       23           
  Lines        6474     6506   +32     
=======================================
+ Hits         6458     6490   +32     
  Misses         16       16           

Flag	Coverage Δ
unittests	99.75% <100.00%> (+<0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

[tleonhardt]

select() still has the race the PR is meant to fix.

A plain background thread gets prompt-toolkit's shared default session, so it can see the main prompt while it's running. But select() runs choice() inside a new session created at cmd2/cmd2.py:4793, and only the main thread can see that one. So during a menu, a thread like the alerter in examples/async_printing.py gets no app back and writes directly to the same output object the menu is drawing with, from the wrong thread. choice() takes no input/output arguments, so that wrapper can't simply be removed.

One fix is for cmd2 to keep track of the app or loop it is running, instead of relying on the calling thread's context. A different fix would be for select() to handle the title itself.

[tleonhardt]

The title is written to a different output from the app it's scheduled on.

But this isn't new: the old set_title() also wrote to the default session's output.

Now that the code has the running app, writing to app.output when one exists, and to get_app_session().output otherwise, is a one-line fix. It sends the title to the right terminal when stdout isn't sys.stdout.

[tleonhardt]

@KennyMcSimpson Thanks for the PR! I left a couple comments for things I think likely need to be tweaked.

I also just put out a 4.3.1 patch release, so you will need to pull latest and merge main into this branch and update the CHANGELOG entry to target 4.3.2.

[KennyMcSimpson]

Thanks for the review. Commit e49ec13b addresses both inline comments and includes the requested upstream release update.

• select() now registers each temporary child AppSession for the full choice() lifetime. set_window_title() follows that route for ordinary workers, cleans the exact child in finally, and supports nested and out-of-order menu completion without stale mappings.
• When a prompt-toolkit application is active, the scheduled callback captures app.output; when no application is active, it uses the routed session output. Unrelated sessions remain isolated.
• I merged upstream main at e9c64db5 and moved the changelog entry to 4.3.2 while preserving the 4.3.1 release section.

Validation after the follow-up: independent public-API reference 55/55 (previous implementation 31/55), Python 3.13 suite 1,879 passed / 159 skipped, Python 3.11 suite 1,878 passed / 160 skipped, make check all green, and strict docs build successful.

Hosted follow-up: 29 checks pass, including quality, type checks, docs, CodeQL, coverage, and 17 of 18 OS/Python test jobs. The macOS/Python 3.15t job times out in the unchanged test_pipeline_stops_with_cmd2_and_returns_terminal[wrapper-stop-and-interrupt-busy-shell-producer] after its job-control signal; this also fails the aggregate test check. That case runs a shell pipeline without calling select() or set_window_title(), and the test file matches upstream main. I have not established its cause in a controlled macOS reproduction. GitHub returned 403 when I attempted to rerun the failed jobs, so I cannot rerun that upstream job with this account.

[kmvanbrunt]

Please comment this while loop to explain why you're doing this traversal.

[KennyMcSimpson]

Added in c78994f8. The comment explains that a worker may still see the parent session, so the loop follows registered children to the innermost active select() session; [-1] selects the most recently registered active child when menus share a parent. make check, make test (1,879 passed / 159 skipped), and make docs-test pass locally.