`DynamicClassAttribute` drops explicitly provided empty docstrings

[T90REAL]

Bug report

Bug description:

types.DynamicClassAttribute loses an explicitly provided empty docstring. Its constructor assigns self.__doc__ = doc or fget.__doc__, so when you pass doc='' and fget is None, the descriptor ends up with None.__doc__ (“The type of the None singleton.”) instead of retaining the empty string. This breaks parity with property, which preserves an empty doc, and cascades through setter()/deleter().

from types import DynamicClassAttribute

attr = DynamicClassAttribute(fget=None, fset=None, fdel=None, doc='')
print(repr(attr.__doc__))  # Expected '', but prints 'The type of the None singleton.'

None

Easy Fix:

 class DynamicClassAttribute:
     def __init__(self, fget=None, fset=None, fdel=None, doc=None):
         self.fget = fget
         self.fset = fset
         self.fdel = fdel
-        self.__doc__ = doc or fget.__doc__
-        self.overwrite_doc = doc is None
+        if doc is None and fget is not None:
+            doc = fget.__doc__
+        self.__doc__ = doc
+        self.overwrite_doc = doc is None

CPython versions tested on:

3.12

Operating systems tested on:

Linux

Linked PRs

• gh-140972: Fix: DynamicClassAttribute drops explicitly provided empty docstrings #140975

[StanFromIreland]

cc @JelleZijlstra

[JelleZijlstra]

This isn't related to typing, types.py has runtime helpers.

[JelleZijlstra]

As for the bug itself, I think there are really two bugs: (1) empty strings get replaced and (2) we use None.__doc__ as the docstring. I agree the proposed patch makes for more sensible behavior. Feel free to create a PR with the change, plus tests and a NEWS entry.

We'll have to think about whether to backport it. Given that the current behavior seems clearly wrong to me and the impact is only on docstrings, I think it should be safe to backport to the bugfix releases (currently 3.13 and 3.14).