TST: [ArrowStringArray] more parameterised testing - part 1

[simonjayhawkins]

aim to reduce diff to address #39908 (comment), but good to do as follow-up to #35259 anyway.

This also includes a change to inference, which could be broken off if not wanted as part of a TST PR.

more tests to change once fixture in play and also some parameterisation to add for the str accessor testing (which could be a precursor with xfails for ArrowStringArray or combined with PR that adds the accessor.. will open a draft shortly for discussion)

[simonjayhawkins]

could add this instead of changing.

both StringDtype and ArrowStringDtype have type attrribute as str

at the moment the inference matches on name so this would need to change for 'string[python]' if we wanted to match on name

edit: kind -> type

[jreback]

hmm, this doesn't break anything? cause any perf issues?

[simonjayhawkins]

this will change in #39908, but getting the fixture in play as a pre-cursor should still be an advantage.

[simonjayhawkins]

ArrowStringDtype is removed in #39908 and therefore this change will be reverted.

[jreback]

hmm, this doesn't break anything? cause any perf issues?

[simonjayhawkins]

only the tests that test inference seem to break. so we currently have ArrowStringArray not infered as string in master. maybe not used but I'll look in detail and test perf tomorrow.

This will slow down this PR and I want to get the fixture in play so that I can parameterise many other tests that test for only dtype="string" so I will remove the inference change and pararmeterisation of the inference test in this PR and do that as a separate PR. (after this PR with the fixture in conftest is merged)

[simonjayhawkins]

on master

>>> pd.__version__
'1.3.0.dev0+1180.gdd697e1cca'
>>> 
>>> from pandas.core.arrays.string_arrow import ArrowStringDtype
>>> 
>>> arr = ArrowStringDtype.construct_array_type()._from_sequence(["B", pd.NA, "A"])
>>> 
>>> arr
<ArrowStringArray>
['B', <NA>, 'A']
Length: 3, dtype: arrow_string
>>> 
>>> pd._libs.lib.infer_dtype(arr)
'unknown-array'
>>> ```

[simonjayhawkins]

OK have reverted the inference change... so can open a PR on the str accessor and parameterise existing str accessor tests for that, but nbd if this is not ready as can just dup the fixture and resolve conflicts if necessary.

[jorisvandenbossche]

Looks good to me!

[jorisvandenbossche]

Failures are the matplotlib ones