CoW: Ensure that iterrows does not allow mutating parent

[phofl]

• closes #xxxx (Replace xxxx with the GitHub issue number)
• Tests added and passed if fixing a bug or adding a new feature
• All code checks passed.
• Added type annotations to new arguments/methods/functions.
• Added an entry in the latest doc/source/whatsnew/vX.X.X.rst file if fixing a bug or adding a new feature.

cc @jorisvandenbossche

Any idea on how to do this nicer?

Do you think we need a way to hand over a ref object when constructing a Series from an array at some point?

[jorisvandenbossche]

Yeah, if this is a pattern that will be used more, then we should add a helper for it. In general I think we should still try to avoid this kind of pandas object -> values -> pandas object with views outside of the internals to avoid having to think about refs there.

Possible ideas to do this cleaner:

• Rely on indexing (self.iloc[i] to iterate over rows), and then refs will already be taken care of. But I suppose that will give a big overhead (because you repeat the dtype calculation for each row). And that's probably the same for self._mgr.fast_xs(i) which is special-case method in internals to get a row.
• Add a method to the BlockManager to have the core logic there (like a version of fast_xs but then not for a single row but all rows)

[phofl]

Yep I agree with all your points and would also
Like to avoid it as much as possible, but it’s tricky here without hurting Performance I think.

I‘ll explore adding something on the manager level, that sounds like a good compromise

[jorisvandenbossche]

Shall we short-term go for this "ugly" solution? (so at least to have correct behaviour)

[phofl]

Probably a good idea, I think I can get something done before the actual 2.0 is out