refactor: Enforce LazyExpr._from_elementwise_horizontal_op

[dangotbanned]

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

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

Related issues

• See chore: use _expr_from_elementwise in IbisNamespace #2716 (comment)

Checklist

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

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

Having the method defined on the Namespace (current), is what I was trying to avoid since

• refactor: adds _compliant sub-package #2149

Prior to that, we had things like:

• Namespace._create_series_from_scalar
• Namespace._create_expr_from_callable
• Namespace._create_expr_from_series

I'd think it's much simpler to keep constructors defined on the class they produce, and they also benefit from shorter names 🙂

[dangotbanned]

@MarcoGorelli would you be open to adding a little doc here? 🙏

[dangotbanned]

Actually, another thought.

What about LazyExpr.from_horizontal_op?

That would mirror the same constructor these all use for ExprMetadata

narwhals/narwhals/functions.py

Lines 1369 to 1374 in f17acfc

	return Expr(
	lambda plx: apply_n_ary_operation(
	plx, plx.sum_horizontal, *flat_exprs, str_as_lit=False
	),
	ExprMetadata.from_horizontal_op(*flat_exprs),
	)

narwhals/narwhals/_expression_parsing.py

Lines 477 to 481 in f17acfc

	@classmethod
	def from_horizontal_op(cls, *exprs: IntoExpr) -> ExprMetadata:
	return combine_metadata(
	*exprs, str_as_lit=False, allow_multi_output=True, to_single_output=True
	)

I think the detail that we should use this for *_horizontal (and concat_str) is more relevant than the elementwise part

[dangotbanned]

I think these could just be:

return [func(chain.from_iterable(expr(df) for expr in exprs))]

[MarcoGorelli]

thanks @dangotbanned

i've renamed and prefixed with an underscore as I don't think this should be part of the compliant protocols #2713, but fine for now

[dangotbanned]

thanks @dangotbanned

i've renamed and prefixed with an underscore as I don't think this should be part of the compliant protocols #2713, but fine for now

Thanks @MarcoGorelli!

One thing though - the name in the PR title doesn't match the code 😉

I prefer the one without elementwise

[MarcoGorelli]

thanks! same, but it does need to be elementwise if we use window_function the way it's defined inside it