Add missing None return to RawConfigParser.get

[kkirsche]

Fix #7476

This adds support to RawConfigParser.get to return None as seen on these lines in cPython:
https://github.com/python/cpython/blob/main/Lib/configparser.py#L773-L774

[Akuli]

Presumably ConfigParser shouldn't inherit this behaviour. Usually you just want to parse a config file with ConfigParser, and in that case getting None seems to be impossible:

>>> p = configparser.ConfigParser()
>>> p.read_string("[foo]\nempty =")
>>> p.get("foo", "empty")
''
>>> p.get("foo", "asdasd")
Traceback (most recent call last):
   ...
configparser.NoOptionError: No option 'asdasd' in section: 'foo'

A couple ways to fix this:

• Add another get method to ConfigParser that doesn't have | None.
• Return str | Any, meaning that the user has to be prepared to get a string, but they can handle other types too if they want.

You will also have to adjust SectionProxy. Maybe make it generic, so that it can be a MutableMapping[str, str] for ConfigParser but MutableMapping[str, str | None] for RawConfigParser?

[kkirsche]

I've added the overloads for ConfigParser. Regarding SectionProxy, if I make it a generic that can accept either, SectionProxy no longer is compatible with `Mapping[str, str]

Regarding making section proxy generic, I can do that but it seems to really ripple throughout as we have to adjust the __getitem__ implementations, and related TypeAliases like _Parser to handle how it can change.

[Akuli]

Indeed, the inheritance is a bit unfortunate: RawConfigParser should behave like MutableMapping[str, SectionProxy[str | None]] but its subclass ConfigParser should behave like MutableMapping[str, SectionProxy[str]]. I don't think there's any way to express this. Maybe we should just return str | Any from SectionProxy's .get(), which makes it usable for both ConfigParser and RawConfigParser, although you won't get errors if you use RawConfigParser and you forget to handle None.

[Akuli]

Looking at zproject/config.py in zulip (from last mypy_primer output), mypy seems to think that _T is str, and complains when you do fallback=None. I will try to fix that.

Suggested change

	self, section: str, option: str, *, raw: bool = ..., vars: _Section | None = ..., fallback: _T
	self, section: str, option: str, *, raw: bool = ..., vars: _Section | None = ..., fallback: _T | None

[Akuli]

I reverted this. Returning a possibly None value is good if you pass fallback=None.

[Akuli]

This still seems a bit too disruptive to me: all mypy_primer results above (you need to click "1 similar comment") are usages of RawConfigParser objects that probably don't contain any Nones, because to get a None value in RawConfigParser, you would have to manually insert it instead of just reading a file.

We have a couple options to choose from:

• Just tell everyone to stop using RawConfigParser. This isn't great, because it's not necessarily the fault of whoever gets mypy errors. For example, flake8 provides a RawConfigParser object for plugin authors.
• Use | Any in RawConfigParser too

[github-actions]

According to mypy_primer, this change has no effect on the checked open source code. 🤖🎉

[Akuli]

Thanks!