I would prefer to not change PyUnicodeWriter_WriteRepr() in a Python 3.14.x bugfix release. Can you leave it unchanged? Instead, modify list_repr() and tuple_repr() to write <NULL> string explicitly.

We need to modify not only list_repr() and tuple_repr(), but all code that uses PyUnicodeWriter_WriteRepr(). And what about the user code? We leave it with a mine which explodes in rare and most likely not tested case.

In the Python code base, if I read the code correctly, only tuple_repr() can call PyUnicodeWriter_WriteRepr() with NULL. For example, list_repr() cannot pass NULL, because item = Py_NewRef(v->ob_item[i]) crash if the item is NULL.

And what about the user code? We leave it with a mine which explodes in rare and most likely not tested case.

My concern is that if you develop on (the future) Python 3.14.4 with PyUnicodeWriter_WriteRepr() which accepts NULL, you can get crashes if an user uses Python 3.14.0 which doesn't accept NULL. I don't think that it's a good idea to change the API in a 3.14.x bugfix release.

Python 3.14.0 is broken. It crashes when you call repr() for uninitialized list or tuple, and it can crash in other cases. Does it mean that we should not backport the bugfix? I think that it is the nature of bugfix releases that they make the code which failed in earlier releases to pass.

The documentation says that PyUnicodeWriter_WriteRepr() calls PyObject_Repr() on obj and writes the output into writer. PyObject_Repr() works with NULL.

If you insist to change PyUnicodeWriter_WriteRepr() API, please add a versionchanged markup to document that the API changed.

If we don't fix it in 3.14, we will need a versionchanged for 3.15. But we rarely add versionchanged for bugfixes, especially if the feature was added in the same version.

I'm suggesting to add a versionchanged markup in the 3.14 change to say that the API changed in Python 3.14.x.

Done. Created also a PR for adding in in main.