Use repr of fill_value in SparseDtype repr

[topper-123]

This proposes changing the StringDtype repr to show the fill_value's repr rather than its string representation:

>>> arr = pd.arrays.SparseArray([0, 1])
>>> arr.dtype
Sparse[int64, 0]  # master & this PR, unchanged
>>> arr.astype(str).dtype
Sparse[object, 0]  # master
Sparse[object, '0']  # this PR, notice quotes

[jorisvandenbossche]

Good idea!

[jorisvandenbossche]

Can you put this test in one of the files in tests/arrays/sparse/.. (eg in test_dtype.py)?

(trying to keep those extension tests only the shared common ones inherted from the base tests, any custom test goes into tests/arrays instead of tests/extension/)

[topper-123]

Ok, I've moved the tests.

[jreback]

looks good can you add a release note (sparse bug fix section is ok). ping on green.

[jorisvandenbossche]

Thanks!

[jreback]

btw if would be helpful to survey other dtypes (mainly category) to see if we are consistent for various types

[jorisvandenbossche]

Non of the other dtypes have the concept of a "fill_value" (or print another value in its dtype repr), so I am not sure what should be checked?

[jreback]

of course; i am speaking generally

we expose other things in the repr eg categories

[jorisvandenbossche]

The CategoricalDtype is fine:

In [9]: pd.Categorical([1, 2, '3']).dtype 
Out[9]: CategoricalDtype(categories=[1, 2, '3'], ordered=False)

Categorical itself is not:

In [11]: pd.Categorical([1, 2, '3']) 
Out[11]: 
[1, 2, 3]
Categories (3, object): [1, 2, 3]

but that is actually already being handled in #34222