Sorry, can you please calrify why do you want to change this?
Right now this reads perfectly fine to me: 'sum' and 'max' are strings, defined as choices=['sum', 'max']

I think we have a small discrepency here:

Technically, we shouldn't have quotes because we do map(str, ...) and not map(repr, ...). However!!! on 3.12.6 (not latest one):

Python 3.12.6 (main, Sep 13 2024, 13:53:54) [GCC 7.5.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import argparse
>>> p = argparse.ArgumentParser()
>>> p.add_argument('a', choices=["a", "b"])
_StoreAction(option_strings=[], dest='a', nargs=None, const=None, default=None, type=None, choices=['a', 'b'], required=True, help=None, metavar=None)
>>> p.parse_args(['1234'])
usage: [-h] {a,b}
: error: argument a: invalid choice: '1234' (choose from 'a', 'b')

Whereas on main or latest 3.12:

Python 3.14.0a5+ (heads/fix/repl/show-source-129098-dirty:8eb3cca8f76, Mar  2 2025, 12:02:03) [GCC 7.5.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import argparse
>>> p = argparse.ArgumentParser()
>>> p.add_argument('a', choices=["a", "b"])
_StoreAction(option_strings=[], dest='a', nargs=None, const=None, default=None, type=None, choices=['a', 'b'], required=True, help=None, metavar=None, deprecated=False)
>>> p.parse_args(['1234'])
usage: python -m _pyrepl [-h] {a,b}
python -m _pyrepl: error: argument a: invalid choice: '1234' (choose from a, b)

We have indeed a discrepency. We actually changed this in 21524ee, so it was an omission during the backport. But I'm not sure whether str is actually better.

Oh, interesting. Tbh, I prefer the readability of the explicit string, even though it's obvious in the example in question. It reads strangely to me that we'd have the invalid choice with quotes and then the supported choices without quotes. Consistency is good :)

I think the issue with the docs is that we're using 1-char choices which make the output slightly disturbing. I would suggest using words such as foo and bar to make it more readable.

However, I personally find it better to use repr() because it also highlights what's being expected. It's quite common to have error messages that contain the repr of the invalid argument instead of its str() so I wouldn't change that part.

Thus, my personal choice would be to use the repr() for the choices as well, though for enumeration choices, I don't know how it renders (for enum members the str() is usually better I think?)

I don't have any preference for this. I was building a package from source that depended on cpython and had a test failing due to the expected stderr being different than the actual output because of the quotation marks. Upon looking into your code and documentation, I noticed the discrepancy and decided to create this PR to make the docs match the actual output. If you'd prefer to go back to the quoted version, then please close this PR and change the code to include the quotes again. Changing the output is a breaking change because tests that are expected to fail usually rely on the error message.

Also, as I mentioned in the issue for this PR, even though argparse doesn't quote anymore, optparse still does, so it's inconsistent.