til/python/another-way-to-mark-keyword-only-dataclass-fields.md

Another Way To Mark Keyword-Only Dataclass Fields

In Configure Other Attributes Of Dataclass Field, I showed how the dataclasses.field constructor function can be used. One of the parameters I demonstrated was kw_only. Each field constructed with kw_only=True will be required to be passed as a keyword-only parameter when constructing an instance of that dataclass.

Another way to specify keyword-only parameters with dataclass fields is to segment them with KW_ONLY. This sentinel value can be included as a pseudo-field where all fields that come after it are treated as keyword-only.

Translating the example from that other post would look like this:

from dataclasses import dataclass, field, KW_ONLY
from datetime import datetime

@dataclass
class Session:
    start_time: datetime
    project_name: str
    _: KW_ONLY
    tags: list[str] = field(default_factory=list, kw_only=True)
    end_time: datetime | None = None

    # ...

sesh1 = Session(start1, "my-project", tags=["pytorch", "numpy"])
sesh2 = Session(start2, "other-project", end_time=datetime.now())

The field whose value is KW_ONLY is only used to signal that keyword-only boundary. It does not itself become a field of the dataclass.

On the one hand I like this approach because it feels closer to the way this is signaled in standard function definition syntax.

def build_session(start_time, project_name, *, tags, end_time=None)

On the other hand, it feels like magic dataclass syntax whereas the kw_only=True parameter is more explicit.