chore: remove more kwargs from dask, dont treat ddof as expressifiable arg

[MarcoGorelli]

I don't like how currently we're muddying the waters between "expressifiable arguments" and "kwargs", which leads us to having to pass ddof as an expressifiable argument in std and var

I'd like to start by separating the concepts, and not passing ddof as an expressifiable argument

What type of PR is this? (check all applicable)

• 💾 Refactor
• ✨ Feature
• 🐛 Bug Fix
• 🔧 Optimization
• 📝 Documentation
• ✅ Test
• 🐳 Other

Related issues

• Related issue #<issue number>
• Closes #<issue number>

Checklist

• Code follows style guide (ruff)
• Tests added
• Documented the changes

If you have comments or can explain your changes, please do so below

[EdAbati]

What if we call this call_kwargs i.e. kwargs for the call function ?

(Just to clarify these kwargs are not passed to _from_call itself)

[dangotbanned]

@EdAbati Couldn't you just pass call as a partial in the cases where this is needed?

• https://docs.python.org/3/library/functools.html#partial-objects
• https://docs.python.org/3/library/functools.html#functools.partial

---

• Avoids needing this parameter entirely
• "binding" of arguments is done once, instead of on each use
• Also supports positional arguments

So instead of call, call_kwargs - you'd have call.func, call.keywords.
But the keywords themselves aren't exposed here

[MarcoGorelli]

I did not know that functools.partial stored keywords like that, thanks!

[dangotbanned]

I did not know that functools.partial stored keywords like that, thanks!

You might need to use operator.methodcaller instead, if you don't know the class that would be passed.

E.g. partial would work if you could do partial(dx.Series.std, ddof=ddof).
But where it might be dx._groupby.GroupBy.std, using the method name would be more widely accepted

narwhals/narwhals/_dask/group_by.py

Lines 54 to 55 in 4dda548

	def std(ddof: int = 1) -> _AggFn:
	return partial(_DaskGroupBy.std, ddof=ddof)

[dangotbanned]

if you don't know the class that would be passed.

Is that why you're storing function_name, in addition to the lambda?

[MarcoGorelli]

yup, that's right

[dangotbanned]

Related to #2059 (comment)

and

https://github.com/MarcoGorelli/narwhals/blob/52b80b240f0fc0e7eb069a84f0ee001f218bedaf/narwhals/_dask/expr.py#L51-L53

        # Kwargs with metadata which we may need in group-by agg
        # (e.g. `ddof` for `std` and `var`).
        kwargs: dict[str, Any] | None = None,

I'm not following why kwargs is used at all here:

narwhals/narwhals/_dask/group_by.py

Lines 54 to 55 in 4dda548

	def std(ddof: int = 1) -> _AggFn:
	return partial(_DaskGroupBy.std, ddof=ddof)

narwhals/narwhals/_dask/group_by.py

Lines 148 to 150 in 4dda548

	kwargs: dict[str, Any] = (
	{"ddof": expr._kwargs["ddof"]} if function_name in {"std", "var"} else {} # type: ignore[attr-defined]
	)

Am I reading this wrong, or are the same kwargs applied twice?

[MarcoGorelli]

thanks both for comments!

i'd like to refactor this mechanism more broadly, so i'll hold off functools gymnastics for the time being (even though it does look really nice!), but definitely interested in revisiting in the future

[dangotbanned]

Sounds good @MarcoGorelli

[MarcoGorelli]

thanks all!