feat(arithmetic): 🪜 floordiv #1452

cmp0xff · 2025-10-28T13:43:00Z

This PR adds floordiv for Index and Series.

Addresses Index arithmetic overloads show wrong types #1347 and Create tests for arithmetic for TimedeltaIndex and DatetimeIndex #1378
Tests added

Dr-Irv

A few things:

You're inconsistent when having timedelta divided by int where "divided" is either regular division or floor division, and timedelta is a single element or a sequence of some type. You have that as being not allowed in the stubs, but they work at runtime. Sometimes you do allow it, so you have to fix that.
There are a few places where you say you can't detect via static typing, and I'm surprised at that.

Also, you are making changes here that are independent of supporting floordiv, namely:

You changed the pyrefly override syntax
You changed some imports from from pandas import Series to from pandas.core.series import Series

In the future, if you are making changes where the changes don't have to do with the main intent of the PR, make a separate PR for those changes. E.g., in the above case, you could do one for pyrefly and another to change the imports.

tests/indexes/arithmetic/float/test_floordiv.py

Dr-Irv · 2025-11-04T16:48:52Z

tests/indexes/arithmetic/float/test_truediv.py

    check(assert_type(c / left, "pd.Index[complex]"), pd.Index, np.complexfloating)
+    if TYPE_CHECKING_INVALID_USAGE:
+        _14 = s / left  # type: ignore[operator] # pyright: ignore[reportOperatorIssue]
+        _15 = d / left  # type: ignore[operator] # pyright: ignore[reportOperatorIssue]


same comment as in indexes/test_floordiv.py . This does work:

>>> left = pd.Index([1.0, 2.0, 3.0]) >>> from datetime import timedelta >>> d = [timedelta(seconds=s) for s in (1, 2, 3)] >>> d / left Index([0:00:01, 0:00:01, 0:00:01], dtype='object')

This has been discussed in #1431 (review). The problem is that the result is Index[Timestamp], with element type Timestamp, but numpy dtype object. In #1431 we've decided to ban such usages.

In this case, can we just return Index without specifying what is inside? i.e., Index[Any] or just Index ?

Dr-Irv · 2025-11-04T16:57:22Z

tests/indexes/arithmetic/int/test_floordiv.py

+    if TYPE_CHECKING_INVALID_USAGE:
+        _13 = c // left  # type: ignore[operator] # pyright: ignore[reportOperatorIssue]
+        _14 = s // left  # type: ignore[operator] # pyright: ignore[reportOperatorIssue]
+        _15 = d // left  # type: ignore[operator] # pyright: ignore[reportOperatorIssue]


same comment - this should work

This has been discussed in #1431 (review). The problem is that the result is Index[Timestamp], with element type Timestamp, but numpy dtype object. In #1431 we've decided to ban such usages.

In this case, can we just return Index without specifying what is inside? i.e., Index[Any] or just Index ?

Dr-Irv · 2025-11-04T16:59:03Z

tests/indexes/arithmetic/int/test_floordiv.py

+    if TYPE_CHECKING_INVALID_USAGE:
+        assert_type(c // left, Any)
+        assert_type(s // left, Any)
+        assert_type(d // left, "np.typing.NDArray[np.int64]")


Dr-Irv · 2025-11-04T16:59:24Z

tests/indexes/arithmetic/int/test_truediv.py

    check(assert_type(c / left, "pd.Index[complex]"), pd.Index, np.complexfloating)
+    if TYPE_CHECKING_INVALID_USAGE:
+        _14 = s / left  # type: ignore[operator] # pyright: ignore[reportOperatorIssue]
+        _15 = d / left  # type: ignore[operator] # pyright: ignore[reportOperatorIssue]


This has been discussed in #1431 (review). The problem is that the result is Index[Timestamp], with element type Timestamp, but numpy dtype object. In #1431 we've decided to ban such usages.

In this case, can we just return Index without specifying what is inside? i.e., Index[Any] or just Index ?

Dr-Irv · 2025-11-04T17:17:36Z

tests/series/arithmetic/test_floordiv.py

+    if TYPE_CHECKING_INVALID_USAGE:
+        left_i.floordiv(c)  # type: ignore[arg-type] # pyright: ignore[reportArgumentType,reportCallIssue]
+        left_i.floordiv(s)  # type: ignore[arg-type] # pyright: ignore[reportArgumentType,reportCallIssue]
+        # left_i.floordiv(d)  # This invalid one cannot be detected by static type checking


This is very surprising. You were able to detect that left_i // d is invalid, so you should be able to do this one.

Dr-Irv · 2025-11-04T17:21:41Z

tests/series/arithmetic/test_floordiv.py

+    if TYPE_CHECKING_INVALID_USAGE:
+        _03 = left_i // c  # type: ignore[operator] # pyright: ignore[reportOperatorIssue]
+        _04 = left_i // s  # type: ignore[operator] # pyright: ignore[reportOperatorIssue]
+        # _05 = left_i // d  # This invalid one cannot be detected by static type checking


A bit surprising, but then I see that you didn't add s and d in the tests in tests/series/arithmetic/test_truediv.py, so you should add that and things should be the same between floordiv and truediv

Dr-Irv · 2025-11-04T17:21:57Z

tests/series/arithmetic/test_floordiv.py

+    if TYPE_CHECKING_INVALID_USAGE:
+        left_i.floordiv(c)  # type: ignore[arg-type] # pyright: ignore[reportArgumentType,reportCallIssue]
+        left_i.floordiv(s)  # type: ignore[arg-type] # pyright: ignore[reportArgumentType,reportCallIssue]
+        # left_i.floordiv(d)  # This invalid one cannot be detected by static type checking


surprised at this

Dr-Irv · 2025-11-04T17:22:11Z

tests/series/arithmetic/test_floordiv.py

+    if TYPE_CHECKING_INVALID_USAGE:
+        _03 = left_i // c  # type: ignore[operator] # pyright: ignore[reportOperatorIssue]
+        _04 = left_i // s  # type: ignore[operator] # pyright: ignore[reportOperatorIssue]
+        # _05 = left_i // d  # This invalid one cannot be detected by static type checking


same comment

Dr-Irv · 2025-11-04T17:22:26Z

tests/series/arithmetic/test_floordiv.py

+    if TYPE_CHECKING_INVALID_USAGE:
+        left_i.floordiv(c)  # type: ignore[arg-type] # pyright: ignore[reportArgumentType,reportCallIssue]
+        left_i.floordiv(s)  # type: ignore[arg-type] # pyright: ignore[reportArgumentType,reportCallIssue]
+        # left_i.floordiv(d)  # This invalid one cannot be detected by static type checking


same comment

cmp0xff

1. You're inconsistent when having `timedelta` divided by `int` where "divided" is either regular division or floor division, and `timedelta` is a single element or a sequence of some type.  You have that as being not allowed in the stubs, but they work at runtime.  Sometimes you do allow it, so you have to fix that.

When the runtime result has the correct dtype, this PR should give type hints. When the operation crashes at runtime or gives object as dtype, this PR should show error. This was discussed in e.g. #1431 (comment), where Index[Timedelta] (which has dtype object) is banned.

2. There are a few places where you say you can't detect via static typing, and I'm surprised at that.

Still working on them. Will ping when finished.

1. You changed the `pyrefly` override syntax

Will do separately

2. You changed some imports from `from pandas import Series` to `from pandas.core.series import Series`

Will do separately.

tests/indexes/arithmetic/float/test_floordiv.py

cmp0xff · 2025-11-05T12:19:48Z

tests/indexes/arithmetic/float/test_truediv.py

    check(assert_type(c / left, "pd.Index[complex]"), pd.Index, np.complexfloating)
+    if TYPE_CHECKING_INVALID_USAGE:
+        _14 = s / left  # type: ignore[operator] # pyright: ignore[reportOperatorIssue]
+        _15 = d / left  # type: ignore[operator] # pyright: ignore[reportOperatorIssue]


This has been discussed in #1431 (review). The problem is that the result is Index[Timestamp], with element type Timestamp, but numpy dtype object. In #1431 we've decided to ban such usages.

cmp0xff · 2025-11-05T12:19:56Z

tests/indexes/arithmetic/int/test_floordiv.py

+    if TYPE_CHECKING_INVALID_USAGE:
+        _13 = c // left  # type: ignore[operator] # pyright: ignore[reportOperatorIssue]
+        _14 = s // left  # type: ignore[operator] # pyright: ignore[reportOperatorIssue]
+        _15 = d // left  # type: ignore[operator] # pyright: ignore[reportOperatorIssue]


This has been discussed in #1431 (review). The problem is that the result is Index[Timestamp], with element type Timestamp, but numpy dtype object. In #1431 we've decided to ban such usages.

cmp0xff · 2025-11-05T12:31:26Z

tests/indexes/arithmetic/int/test_floordiv.py

+    if TYPE_CHECKING_INVALID_USAGE:
+        assert_type(c // left, Any)
+        assert_type(s // left, Any)
+        assert_type(d // left, "np.typing.NDArray[np.int64]")


cmp0xff · 2025-11-05T12:32:20Z

tests/indexes/arithmetic/int/test_truediv.py

    check(assert_type(c / left, "pd.Index[complex]"), pd.Index, np.complexfloating)
+    if TYPE_CHECKING_INVALID_USAGE:
+        _14 = s / left  # type: ignore[operator] # pyright: ignore[reportOperatorIssue]
+        _15 = d / left  # type: ignore[operator] # pyright: ignore[reportOperatorIssue]


This has been discussed in #1431 (review). The problem is that the result is Index[Timestamp], with element type Timestamp, but numpy dtype object. In #1431 we've decided to ban such usages.

cmp0xff · 2025-11-05T12:36:55Z

tests/series/arithmetic/test_floordiv.py

+    if TYPE_CHECKING_INVALID_USAGE:
+        _13 = c // left_i  # type: ignore[operator] # pyright: ignore[reportOperatorIssue]
+        _14 = s // left_i  # type: ignore[operator] # pyright: ignore[reportOperatorIssue]
+        _15 = d // left_i  # type: ignore[operator] # pyright: ignore[reportOperatorIssue]


The result is Series with dtype object, not timedelta64 as expected. From #1431 I guess we also ban it.

tests/series/arithmetic/float/test_floordiv.py

cmp0xff · 2025-11-05T12:42:43Z

tests/series/arithmetic/test_floordiv.py

+    if TYPE_CHECKING_INVALID_USAGE:
+        assert_type(c // left_i, Any)
+        assert_type(s // left_i, Any)
+        assert_type(d // left_i, Any)  # pyright: ignore[reportAssertTypeFailure]


The result is Series with dtype object, not timedelta64 as expected. From #1431 I guess we also ban it.

Dr-Irv · 2025-11-05T19:22:00Z

When the runtime result has the correct dtype, this PR should give type hints. When the operation crashes at runtime or gives object as dtype, this PR should show error. This was discussed in e.g. #1431 (comment), where Index[Timedelta] (which has dtype object) is banned.

I think this situation is different. When we banned Index[Timedelta] when the dtype was object, my concern was that you were doing tests using that type, and we don't want to assume it could be created.

Here, this is about the results of the computation. If we know the result has object dtype, I think the stubs can return a result without a specified value of S1 (i.e., either Index or Index[Any]; Series or Series[Any])

cmp0xff added Index Related to the Index class or subclasses Numeric Operations Arithmetic, Comparison, and Logical operations Series Series data structure labels Oct 28, 2025

cmp0xff mentioned this pull request Oct 28, 2025

Type RangeIndex #1453

Merged

2 tasks

cmp0xff force-pushed the feature/floordiv branch 4 times, most recently from 3df344a to 674b994 Compare October 31, 2025 11:02

cmp0xff added 9 commits October 31, 2025 20:53

floordiv

a074123

remove irrelevant changes

7a32e45

floordiv

cb1506a

remove irrelevant change

0f86bc8

floordiv

e1090cf

pyright

32bf36c

mypy

7f7cab4

pytest

5b997b6

pyrefly

dca5fa2

cmp0xff force-pushed the feature/floordiv branch from 674b994 to dca5fa2 Compare October 31, 2025 19:53

cmp0xff requested review from Dr-Irv and loicdiridollou November 4, 2025 12:42

Dr-Irv requested changes Nov 4, 2025

View reviewed changes

cmp0xff marked this pull request as draft November 4, 2025 22:33

cmp0xff added 3 commits November 5, 2025 13:31

fix: np.ndarray // pd.IndexOpsMixin

c95aa39

mypy

eb648fb

series

c70a3ce

cmp0xff commented Nov 5, 2025

View reviewed changes

mypy

e2482cb

cmp0xff added 2 commits November 5, 2025 23:17

series

ea5d6bf

series

5d1c228

cmp0xff mentioned this pull request Nov 6, 2025

BUG: Inconsistency between Series[float].__rfloordiv__([timedelta]) and rfloordiv and further pandas-dev/pandas#63007

Open

3 tasks

cmp0xff added 3 commits November 6, 2025 11:42

fix: pandas-dev/pandas#63007 pytest

adc6118

py310

cf7e941

index

24deb37

Uh oh!

feat(arithmetic): 🪜 floordiv #1452

Are you sure you want to change the base?

feat(arithmetic): 🪜 floordiv #1452

Conversation

cmp0xff commented Oct 28, 2025

Uh oh!

Dr-Irv left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

cmp0xff left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Dr-Irv commented Nov 5, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants