Handle construction of string ExtensionArray from lists

[xhochy]

I had to add a string-based Arrow Extension array to trigger the bug but did not run the same test suite as we do on the boolean array as I don't see it adding value but just runtime at the moment.

• closes String-based ExtensionArrays force cast to numpy in list based construction #27673
• tests added / passed
• passes black pandas
• passes git diff upstream/master -u -- "*.py" | flake8 --diff
• whatsnew entry

[TomAugspurger]

Thanks.

I'm wondering a bit more general fix. What about something like

# in sanitize_array. Somewhere before `data = extract_array(...)`

if not is_array_like(data) and is_extension_array_dtype(dtype):
    data = dtype.construct_array_type()._from_sequence(data, dtype=dtype, copy=copy)

Would that work? Then we get to the EA handling earlier on, and don't require special casing later on.

[xhochy]

I moved the check slightly up but didn't move it significantly up as the above code could actually be taken by ExtensionArray data. If I run into other issues, I would revisit this if I run into other problems while workin on fletcher but moving it to a more broad scope at the moment seems to me a bit risky as sanitize_array is not directly tested but only indirectly via Series constructions.

[TomAugspurger]

OK, thanks.

@jbrockmendel can you take a look at this?

[jbrockmendel]

might as well share this method by using "{cls}({data})".format(cls=type(self).__name__, data=repr(self._data))

[jbrockmendel]

small comment, otherwise looks good

[xhochy]

@jbrockmendel Adapted that but Travis keeps faling for unrelated reasons.

[xhochy]

Rebased on master and everything is green now.

[jorisvandenbossche]

Thanks @xhochy !