Debug launch kills Tkinter GUI process after recent update; plain terminal run is stable

Summary
After recent updates, debugging a Tkinter GUI with F5 becomes unstable. The GUI process exits suddenly without Python traceback. Running the same app with plain Python is stable.

Environment

VS Code: 1.124.0 (commit 1b50d58d73426c9171299ec4037d01365d995b78, x64)
Python extension: ms-python.python-2026.4.0-win32-x64
Debugpy extension: ms-python.debugpy-2026.6.0-win32-x64
Pylance: ms-python.vscode-pylance-2026.2.1
OS: Windows
Interpreter: workspace venv [python.exe]
Repro steps

Open a Tkinter GUI app in VS Code.
Start debugging with F5 using a launch config of type debugpy, request launch, program app_gui.py, console integratedTerminal or externalTerminal.
Let the app run without interaction for about 30 to 90 seconds.
GUI exits unexpectedly.
Expected behavior
GUI should keep running until user closes it or app throws a visible Python exception.

Actual behavior

Process exits abruptly.
No Python traceback.
No OS crash event for python.
Internal app heartbeat logs stop suddenly.
Important control test
Running the same GUI via plain terminal command is stable:
[python.exe] [app_gui.py]

Additional evidence

Python extension log repeatedly shows debug launch commands through debugpy launcher:
Send text to terminal ... debugpy\launcher ... -- [app_gui.py]
App-side diagnostics show debugger_attached=False in failing runs, and process chain python -> python -> app_gui before sudden exit.
No fault traceback captured.
Workaround
Attach mode is stable:

Start app manually with debugpy listen.
Use debugpy attach config from VS Code.
This avoids VS Code owning the app process lifecycle.
Question
Was there a recent regression/change in debug launch process lifecycle handling on Windows in ms-python.python 2026.4.0 or debugpy 2026.6.0?

Aucun fichier source ni aucun test n’est indiqué. Commencez par reproduire l’application Tkinter avec la configuration de lancement F5 debugpy sous Windows, puis comparez le cycle de vie de son processus avec celui des exécutions stables dans un terminal simple et en mode attach. Suivez les logs du launcher debugpy et de Python extension autour de la sortie inattendue ; le travail est terminé lorsque la régression ou un test ciblé qui échoue pour le comportement du lancement est identifié.