force-remaining-arguments-to-be-named.md

Force Remaining Arguments To Be Named

When the positional argument collector (*) is used in a function definition, all the arguments that follow it must now be explicitly named when invoking that function. This is regardless of whether you are collecting positional arguments into a named tuple (e.g. *rest) or are using the empty/dummy version (*).

A friend recently recommended I use this syntax to force some configuration values (often booleans) to be explicitly named at call sites.

Here is an example of this in my py-vmt project:

class CliContext:
    def __init__(self, *, verbose: bool, repo: SessionRepository | None = None) -> None:
        self.verbose: bool = verbose
        self.active_session: Session | None = None
        self.repo: SessionRepository = repo or JsonRepository()
        self.active_session = self.repo.active_session()

Notice that right after self is the * which means that both verbose and repo must be named arguments.

Here is a corresponding call site:

    ctx.obj = CliContext(verbose=True)

Now when I initialize CliContext, there is no choice for me to pass in an unnamed argument of True. It has to be paired with verbose. Above I've excluded repo because that falls back to a default value. If I were to include it, it would have to be named.

The same applies to this scenario where I'm collecting the positional arguments.

def build_identifier(first, *rest, delimiter="/"):
    if rest is None:
        return first

    return delimiter.join([first, *rest])

>>> print(build_identifier("taco"))
taco
>>> print(build_identifier("taco", "bell", "mas", delimiter="-"))
taco-bell-mas
>>> print(build_identifier("taco", "bell", "mas", "-"))
taco/bell/mas/-

Notice in the third example where I don't use a named argument for the delimiter. It instead gets swept up in the collection of positional arguments.

This language of positional argument collector I got from Python in a Nutshell, 4th Edition (see pg. 96).