This issue tracker has been migrated to GitHub, and is currently read-only.
For more information, see the GitHub FAQs in the Python's Developer Guide.

classification
Title: Possibly misleading/wrong documentation about PyModuleDef.m_free
Type: enhancement Stage:
Components: Documentation Versions: Python 3.8, Python 3.7
process
Status: open Resolution:
Dependencies: Superseder:
Assigned To: docs@python Nosy List: Segev Finer, docs@python
Priority: normal Keywords:

Created on 2018-08-23 15:07 by Segev Finer, last changed 2022-04-11 14:59 by admin.

Messages (1)
msg323955 - (view) Author: Segev Finer (Segev Finer) * Date: 2018-08-23 15:07
This section in the documentation https://docs.python.org/3.7/howto/cporting.html#module-initialization-and-state shows how to implement an extension module that works for both Python 2 and 3. It also shows how to use m_size to manage module state instead of using static globals. But it omits implementing m_free. This is also done in https://www.python.org/dev/peps/pep-3121/#example

To my understanding, m_clear is *only* called by the Python GC when the object is a part of a reference cycle. It won't be called for a normal free when the object wasn't a part of a reference cycle. This means that for such a case the objects referenced in the state structure will leak!

Looking around the standard library, it seems the extension modules there point m_free to calling the function used to implement m_clear.

The documentation of the structure is also quite unclear on the subject https://docs.python.org/3/c-api/module.html?highlight=pymoduledef#c.PyModuleDef.

P.S. There is also no freeing/DECREF-ing stuff in the state on error in the Python 2 case, but I think even the standard library is like this.
History
Date User Action Args
2022-04-11 14:59:05adminsetgithub: 78659
2018-08-23 15:07:34Segev Finercreate