refactor: Simplify calculate_timestamp_datetime and calculate_timestamp_date

[FBruzzesi]

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

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

Related issues

• Related issue Simplify misc #2492

Checklist

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

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

Heavily inspired by @dangotbanned code for arrow

[dangotbanned]

#2838 (comment)

I think the issue is that BinOpBroadcast is generic, but using a TypeVar with a default.

So you don't get a warning when using BinOpBroadcast without specialising - but it causes an issue later when the default conflicts with the constraints of the function signature.

Hopefully this works, I'm replying from my phone 😅

Suggested change

	tuple[UnitCurrent, UnitTarget], tuple[BinOpBroadcast, IntoRhs]
	tuple[UnitCurrent, UnitTarget], tuple[BinOpBroadcast[NativeSeriesT], IntoRhs]

[dangotbanned]

narwhals/_pandas_like/utils.py:583: error: Argument 1 has incompatible type "_CuDFSeries"; expected "Series[Any]"  [arg-type]
narwhals/_pandas_like/utils.py:583: error: Argument 1 has incompatible type "_ModinSeries"; expected "Series[Any]"  [arg-type]

narwhals/narwhals/_pandas_like/typing.py

Lines 25 to 31 in bb03703

	NativeSeriesT = TypeVar(
	"NativeSeriesT",
	"pd.Series[Any]",
	"_CuDFSeries",
	"_ModinSeries",
	default="pd.Series[Any]",
	)

NativeSeriesT means exactly one of pd.Series[Any], _CuDFSeries, _ModinSeries, but default to pd.Series[Any] when not provided.

So the conflict is that

• NativeSeriesT is still used for the function parameter and return type
• But _TIMESTAMP_DATETIME_OP_FACTOR is currently saying it only accepts pd.Series[Any]

narwhals/narwhals/_pandas_like/utils.py

Lines 576 to 578 in 0e6def9

	def calculate_timestamp_datetime(
	s: NativeSeriesT, current: TimeUnit, time_unit: TimeUnit
	) -> NativeSeriesT:

[FBruzzesi]

Hey @dangotbanned thanks for investigating this.

But _TIMESTAMP_DATETIME_OP_FACTOR is currently saying it only accepts pd.Series[Any]

Why would that be the case? The definition is:

BinOpBroadcast: TypeAlias = Callable[[NativeSeriesT, int], NativeSeriesT]

Regarding #2838 (comment), BinOpBroadcast is not a generic, and it would result in:

Type variable "NativeSeriesT" has no meaning in this contextPylance[reportGeneralTypeIssues]

🧐

[dangotbanned]

Hey @dangotbanned thanks for investigating this.

But _TIMESTAMP_DATETIME_OP_FACTOR is currently saying it only accepts pd.Series[Any]

Why would that be the case? The definition is:

BinOpBroadcast: TypeAlias = Callable[[NativeSeriesT, int], NativeSeriesT]

Regarding #2838 (comment), BinOpBroadcast is not a generic, and it would result in:

BinOpBroadcast is a generic TypeAlias because it uses a TypeVar (NativeSeriesT) in it's definition.

If we compare that to when you hover over the current use, it uses the default type - because it wasn't subscribed

---

Type variable "NativeSeriesT" has no meaning in this contextPylance[reportGeneralTypeIssues]

🧐

Ahh, so that message it about scoping - not that the TypeVar can't be used at all

So this is because _TIMESTAMP_DATETIME_OP_FACTOR is in a global scope, but we're trying to solve the TypeVar in a function scope

[dangotbanned]

I've just used Any instead in (bdb46aa)

The main reason I had that typing in _arrow was because the stubs caused too many false positives when I used the functions directly.

Here, we're just using operator functions, which are mostly typed with Any anyway 😉

https://github.com/python/typeshed/blob/8e25bda401332ac56d4750cffda2127a4302cd2f/stdlib/_operator.pyi#L42-L71

[dangotbanned]

We still have NativeSeriesT preserved on the outside, so this is simpler than refactoring everything to provide a scope for the TypeVar 😅

[dangotbanned]

Thanks @FBruzzesi

[FBruzzesi]

Thank you @dangotbanned 🎉