homepage

I don't recall the specifics of how we'd talked about making use of module specs in pickle. I vaguely remember (or misremember <wink>) something related to saving __main__.__spec__.name in the pickle rather than __main__.__name__. Anyone have anything more concrete than that? I'm willing to work this out but only with a more specific goal relative to the pickle module.

The specific proposal was to use __spec__.name when __name__ == "__main__"
to avoid the pickle compatibility issues described in PEP 395 when using
the -m switch.
runpy has to be updated first, though.

Issue 19700 means that runpy now ensures that __main__.__spec__ is set appropriately when __main__ is executed via the import system.
Issue 19946 means that multiprocessing now ensures that __main__ is configured correctly in child processes to reference a properly initialised "fake main" to allow pickle compatibility with classes and functions defined in __main__ outside "if __name__ == '__main__'" guards.
The proposal here is that we make the following changes:
- runpy will ensure that when __main__ is executed via the import system, it will also be aliased in sys.modules as __spec__.name
- if __main__.__spec__ is set, pickle will use __spec__.name rather than __name__ to pickle classes, functions and methods defined in __main__
- multiprocessing is updated appropriately to skip creating __mp_main__ in child processes when __main__.__spec__ is set in the parent process
While I still think this is a reasonable idea, I think it qualifies as a new feature, and hence is better postponed to Python 3.5

So far I agree that this should be postponed to 3.5.

Just noting that PEP 499 covers adding modules executed with `-m` to `sys.modules` under their real name in addition to `__main__`.