GroupBy[Series].count() return type should be Series[int]

[chrisyeh96]

• Closes GroupBy[Series].count() should have return type Series[int] #965
• Tests added: Please use assert_type() to assert the type of any return value

[Dr-Irv]

I'd prefer removing def count() from here, and having declarations in DataFrameGroupBy and SeriesGroupBy in core/groupby/generic.pyi

[chrisyeh96]

I'm a new contributor to this repo and eager to learn more about these choices. Is there a specific reason for moving the declarations to core/groupby/generic.pyi?

In the actual pandas repo, the count() method is implemented in core/groupby/groupby.py, not core/groupby/generic.py.

Thanks in advance for taking the time to explain.

[Dr-Irv]

I'm a new contributor to this repo and eager to learn more about these choices. Is there a specific reason for moving the declarations to core/groupby/generic.pyi?

In the actual pandas repo, the count() method is implemented in core/groupby/groupby.py, not core/groupby/generic.py.

Thanks in advance for taking the time to explain.

It probably is due to that we had issues in the past with doing overloads where self was declared as Series vs. DataFrame, so putting the stubs in the lower classes was a way to get around that problem.

The type checkers have improved, so it's possible your approach could be used going forward.

Having said that, I see that for any() and all() in the file you changed, that was the pattern used, so maybe your change was fine the way it was!

As an example, there are a number of methods shared between Series and DataFrame that are implemented in NDFrame, but instead of putting the stub in NDFrame, we put them in Series and DataFrame to clarify that the results are different in the two subclasses.

[Dr-Irv]

We just uncovered an issue with our CI that was fixed earlier today. Can you merge with current main and push? Then I will start the CI runs.

[chrisyeh96]

I have merged pandas-dev/main into my fork

[Dr-Irv]

Ugh, so now CI is failing because when you first tested, it used pyright 1.1.373 and that got locked for further tests.

Can you include this in your PR? In pyproject.toml, change:

pyright = ">=1.1.369"

to

pyright = ">= 1.1.374"

[chrisyeh96]

I have updated the pyright dependency version in pyproject.toml as you suggested.

[Dr-Irv]

thanks @chrisyeh96