[ArrowStringArray] TST: parametrize str.extractall tests

[simonjayhawkins]

No description provided.

[jbrockmendel]

thanks @simonjayhawkins

[jbrockmendel]

for future PRs of this sort, it would be somewhat easier to review if you did variable renaming (which I heartily approve of, and in fact would ask you to go one step further and rename s to ser to avoid 1-letter variable names) in a pre-cursor PR

[simonjayhawkins]

I think changes I may need to be made to _wrap_result for the expand path in the string accessor so this affects split/rsplit, extract/extractall, partition/rpartition and get_dummies. I want to make sure that all dtypes are tested before that change is made.

so these test parameterisation PRs are 'blockers' for the next step of the pyarrow implementation. So not keen to do precursor to the precursors, so can maybe leave out the other cleanups and just parameterise on dtype?

[simonjayhawkins]

we should maybe discuss in today's dev meeting, but I'm getting concerned that I won't get this done for 1.3.

the rc is only a few weeks away and if it doesn't make the rc it'll have to wait for 1.4

In the original PR adding the dtype #27949 there was some discussion on why the StringDtype was even needed and the gist, as I understand it, is that it was added so that the api could be established/tested ready to swap out the memory data model in the future.

#35169 has an updated discussion where having a parameterised dtype has been proposed (but AFAICT not agreed), leading to breaking api change for users.

so far all the changes merged for ArrowStringArray have the same behaviour as StringArray (I hope). There is draft PRs, that I don't want to merge yet if a) performance using native pyarrow is slower than object fallback, or b) the behaviour of ArrowStringArray is different from StringArray. Doing this may preclude the simple swap out memory model option.

[jbrockmendel]

So not keen to do precursor to the precursors, so can maybe leave out the other cleanups and just parameterise on dtype?

that would have the same effect vis-a-vis making review easier