atexit error

[ghost]

Originally reported by: agroszer (Bitbucket: agroszer, GitHub: agroszer)

---

http://winbot.zope.org/builders has now builders.
Note: some seem to pass, but fail on some atexit problem.
See http://winbot.zope.org/builders/setuptools_master%20py_265_win32/builds/0/steps/test/logs/stdio

---

• Bitbucket: https://bitbucket.org/pypa/setuptools/issue/38

[ghost]

Original comment by arfrever (Bitbucket: arfrever, GitHub: arfrever):

---

The atexit error is a bug in older versions of Python. See:

http://bugs.python.org/issue15881

https://bitbucket.org/tarek/distribute/issue/152

[ghost]

Original comment by agroszer (Bitbucket: agroszer, GitHub: agroszer):

---

Others:
http://winbot.zope.org/builders/setuptools_master%20py_323_win32/builds/0/steps/test/logs/stdio
http://winbot.zope.org/builders/setuptools_master%20py_330_win32/builds/0/steps/test/logs/stdio

the pain is that the test exists with exit code 0

[ghost]

Original comment by jaraco (Bitbucket: jaraco, GitHub: jaraco):

---

The tests exit with exit code 0 because the process is completing normally (and the tests are passing), but because the error is "atexit", it's only logged and doesn't affect the process exit code. We encounter these errors in our production systems all the time for other projects and simply disregard them.

Since the errors will go away in later versions of Python, is there any value in pursuing this issue? In Distribute 152, Nikolaus suggested that the problem was fixed in Setuptools, but I don't believe that to be the case. PJ Eby was pondering a less-than-optimal solution (remove the sandboxing of modules imported). To the best of my understanding, these are our two options:

• Wait until the problem goes away with regard to multiprocessing and hope it doesn't come back in another manifestation.
• Remove the save-and-restore on sys.modules.

Neither approach is optimal. I'm open to other suggestions. Failing other suggestions, I suspect this ticket will linger until the first option is elected by default.

[ghost]

Original comment by arfrever (Bitbucket: arfrever, GitHub: arfrever):

---

This bug will never be fixed by upstream in older versions of Python. We can just wait until Setuptools drops support for Python <2.7 and 3.1 in very distant future.

[ghost]

Original comment by jaraco (Bitbucket: jaraco, GitHub: jaraco):

---

These "error" messages will remain on older Python versions.

[ghost]

Original comment by jurko (Bitbucket: jurko, GitHub: jurko):

---

Just encountered this so wanted to add a note on a workaround that worked in my case -
importing the multiprocessing Python module before running the setuptools.setup() function.

Hope this helps someone.

Best regards,
Jurko Gospodnetić