The existing doctest fragments depend on fixture objects. Add three complete programs that create a window, rename a window without replacing it, and send literal input to a pane reader. Each creates a private socket, checks its advertised result, and explicitly cleans up its server.
Pytest executes the same files twice, checks their exit status and expected output, and verifies that another session and its shell remain alive. The docs show each whole file; source distributions include the programs, and mypy checks them. The programs use APIs already on master and require tmux 3.2a+.
The send-keys example uses ordinary channel acknowledgements, supported by tmux 3.2a's remembered-signal behavior. It asserts the captured response after the reader acknowledges input. It does not claim process completion or use the tmux 3.8 event-wait feature.
This is the initial executable-program slice of #815. Dedicated traversal, change watching, window movement, layouts, explicit argument lists, and run-to-exit examples remain open.
Validation:
- Focused wrapper passes on tmux 3.2a, 3.3a, 3.4, 3.5, 3.6, 3.7a, 3.7b, 3.8, and the installed 3.8a development build, using Python 3.15.0.
- Deliberately incorrect expectations in all three programs fail the wrapper. Incorrect stdout also fails it. Removing the reader's readiness signal produces the program's bounded error.
- Ruff format, Ruff lint, and full-tree mypy pass.
- Full configured pytest on tmux 3.2a: 1388 passed, 71 skipped, one warning from an existing display-message case; all 1456 prior collected cases retained plus the three examples checks.
- The documented Sphinx build passes; rendered code matches the three complete source files. The source archive contains the exact programs and wrapper.
The passing focused 3.8 examples do not imply a passing full suite on 3.8: master's independently recorded hook/popup/floating-pane failures are handled by #825.
The first full stable-3.8 branch run is FAIL/incomplete: it reached its 280-second limit during severe unrelated host contention. Its original failure/error markers and cleanup receipt are retained; no complete failure count or baseline-equivalence claim is available from that interrupted run. The focused examples remain separately qualified. After host recovery, a complete run at the same head finished in 106.34 seconds: 18 failed, 1,438 passed, 3 skipped. All failure IDs and messages match unchanged master after checkout-path normalization. The three example cases pass in that complete run; the full-suite gate remains FAIL. Both attempts are retained, with pytest reruns disabled.
Hosted checks at 4f589c4906f623e7aecb4ce4c4bd93f54c5cc582 remain FAIL: tmux 3.2a and 3.3a pass; 3.4, 3.5, 3.6, 3.7a and 3.7b fail. Both CodeQL analyses and the aggregate pass; codecov/project fails. The existing cached-binary ABI problem is tracked by #825. The development-tmux job reports success despite failing to start pytest because of that ABI problem; it is not a passing test run. Maintainer review remains pending.