configparser: allow default_section=None

[ekanshul]

Closes #12700

Widen RawConfigParser.default_section and the default_section parameter of RawConfigParser.__init__ to str | None, since passing None works at runtime to disable the default section and is relied on in the wild (e.g. setuptools).

Agent used: Claude Code

[github-actions]

Diff from mypy_primer, showing the effect of this PR on open source code:

setuptools (https://github.com/pypa/setuptools)
+ setuptools/config/expand.py:371: error: Unused "type: ignore" comment  [unused-ignore]
+ setuptools/config/expand.py:372: error: Cannot assign to a method  [method-assign]
+ setuptools/config/expand.py:372: error: Incompatible types in assignment (expression has type "type[str]", variable has type "Callable[[str], str]")  [assignment]
+ setuptools/config/expand.py:375: error: Argument 1 to "pop" of "dict" has incompatible type "str | None"; expected "str"  [arg-type]

[JelleZijlstra]

A problem with this is that if default_section is given as None, you will get None in __iter__.

[ekanshul]

You're right, and it goes further than __iter__. On 3.12 with default_section=None:

>>> p = ConfigParser(default_section=None)
>>> p.read_string("[DEFAULT]\na = 1\n[sec]\nb = 2\n")
>>> list(p)
[None, 'DEFAULT', 'sec']
>>> len(p), None in p, p[None]
(3, True, <Section: None>)

__iter__ chains self.default_section in front of the sections, so __iter__, __len__, __contains__ and the keys()/items()/values() mixins all surface None as a key. setuptools actually relies on that: it iterates parser.items() and then does groups.pop(parser.default_section, None) to drop the None entry, which is the pop() error in the mypy_primer output. (The two new errors on the parser.optionxform = str line are pre-existing problems that were hidden while the constructor call failed to match and parser was Any.)

Modelling this properly would mean Iterator[str | None] and friends, or making the parser generic in the section-name type, and both are a heavy price for undocumented behaviour with one known user. Even this narrower version has a cost for everybody: with default_section: str | None, parser[parser.default_section] and parser.get(parser.default_section, ...) no longer type check.

So I don't think this is worth doing. I'd suggest closing this PR and #12700 as "not modelled"; setuptools already carries a # type: ignore with a comment pointing at #12700, which seems like the right place for it. If you'd rather accept the imprecision in __iter__ with a comment, I can do that instead, but I lean towards closing.