BUG: Fix StringArray use_inf_as_na bug

[dsaxton]

• closes BUG: pandas.options.mode.use_inf_as_na disables NA checking for StringArray #33655
• tests added / passed
• passes black pandas
• passes git diff upstream/master -u -- "*.py" | flake8 --diff
• whatsnew entry

[jreback]

you also have #33629 for this same issue?

[dsaxton]

you also have #33629 for this same issue?

We were thinking we may want to treat them separately since they're slightly different issues: #33629 (comment). Should both bugs be in the same PR?

[jorisvandenbossche]

Can you also add a test somewhere for an object dtype Series that has pd.NA in it? (that won't be covered by the string dtype tests)

This shouldn't be "draft" anymore?

[dsaxton]

This shouldn't be "draft" anymore?

I was hoping to finish up the PR linked above first because it overlaps somewhat with this one

[jbrockmendel]

Does this address #33209?

[dsaxton]

Does this address #33209?

Yes I think it should

[jorisvandenbossche]

Can you also add a base extension test for this? (there should already be isna() tests, so you can add a similar one there but which is using use_inf_as_na

[dsaxton]

Can you also add a base extension test for this? (there should already be isna() tests, so you can add a similar one there but which is using use_inf_as_na

Added some base extension use_inf_as_na tests, however I'm getting failures from test_sparse.py (dtype-related, I believe), test_json.py and test_decimal.py (the latter two I think because of improper handling of null values; e.g., {} and Decimal("NaN") under the use_inf_as_na context). Should I xfail for these for now and try to address in a follow-up?

[jbrockmendel]

can we short-circuit this (and avoid the object-cast of to_numpy) for e.g. PeriodArray?

[jorisvandenbossche]

Not in general, but we could short-circuit the known built-in dtypes that will never be able to contain inf (basically all our internal EA dtypes except for categorical ?)

[jorisvandenbossche]

@dsaxton can you merge master once more?

And what is left for this PR? I think only the comment of @jbrockmendel to not do the expensive libmissing.isnaobj_old(values.to_numpy()) check for certain dtypes where this will never be needed (like Periods) ?

[dsaxton]

And what is left for this PR?

@jorisvandenbossche There was also this issue that I still need to figure out (I think it could potentially be fixed with the short circuiting mentioned above): #33656 (comment)

[jreback]

lgtm. ping on green. I don't think its worth the trouble to deprecate this, possibly could do some more code cleaning, but so far so good (IIRC you de-duplicated things a bit already);

[dsaxton]

@jreback ping

[jreback]

thanks @dsaxton