TST: Ensure dtypes are set correctly for empty integer columns #24386

[avinashpancham]

• [x ] closes DataFrame constructor ignores integer dtype when dict-data and non-overlapping columns #24386
• [x ] tests added / passed
• [x ] passes black pandas
• [x ] passes git diff upstream/master -u -- "*.py" | flake8 --diff
• [x ] added tests to verify whether empty integer columns are loaded in as integer columns

[charlesdong1991]

thanks! one nitpick

[jorisvandenbossche]

Thanks for your contribution!

I would maybe move the test to tests/frame/test_constructors.py

[jorisvandenbossche]

Maybe add to the comment "dictionary data with non-overlapping columns, resulting in an empty DataFrame", as it is specifically this case (not just an empty dataframe)

[avinashpancham]

Thanks. Can I add the test at the bottom of the class TestDataFrameConstructors in tests/frame/test_constructors.py or should I add it in a specific section of the class?

[jreback]

also pls parameterize over the dtypes mentioned in the OP

[jreback]

in test_constructors.py there are already similar tests, please move near them

[jreback]

great as long as you are adding a list, can you add the rest of the types. you ca easily do this by using the types defined in pandas._testing (we also have fixtures for this, but maybe not one that covers all types); would be ok with one for that as well. Note that I don't think we have Boolean, String defined pandas._testing (could add them there)

[avinashpancham]

Added more dtypes check, but now the list is very long. Any ideas on how to makes more pretty? I did not want to add it to the lists in pandas._testing, since I was not sure it would be compatible with other tests that rely on them.

[jreback]

you don't need to do it like this, rather use the already defined ones that are ther
e.g. ALL_EA_INT_DTYPES

[avinashpancham]

Do you have any tips on how to refactor the parametrization if I use ALL_EA_INT_DTYPES? The parametrization now has the form [(input_value, expected_value)], but if I loop over ALL_EA_INT_DTYPES I think I still need to manually provide the expected value every time.

[jreback]

how's that? the expected IS the same as the input

[jreback]

of course, you can easily tranform these:

In [1]: pd.Int64Dtype().name                                                                                                                                                                                        
Out[1]: 'Int64'

[jreback]

so my points is that you can simply pass in the ALL_* (just concatenate them all), and check the dtypes of the result columns

[avinashpancham]

Great, that already helps me a lot. But I for int this would not work since the expected is 'Int64'. Similarly for float, str.

Sorry for the large number of questions, this is my first PR :)

[jreback]

sure so what you can do is break this into 2 tests, one for the easy cases where this is true and one for the special cases (where you can input 2 args for input and expected)

[avinashpancham]

Thanks for the feedback, I think I came to a much better test now.

[jreback]

thanks @avinashpancham