DEPR: making copies when indexing along columns

[jbrockmendel]

• closes API: Should indexing on columns give views or copies? #33780
• tests added / passed
• passes black pandas
• passes git diff upstream/master -u -- "*.py" | flake8 --diff
• whatsnew entry

xref #33597 for what this might look like once the deprecation is enforced (nice perf bump)

cc @jorisvandenbossche my understanding is that we will need this deprecation for the all-1D world too. Is that correct? Any thoughts on nice ways to test this and avoid putting tm.assert_produces_warning in a zillion places across the code?

[jorisvandenbossche]

my understanding is that we will need this deprecation for the all-1D world too. Is that correct?

We don't have to (I mean it's not tied to being 1D), but yes, I suppose we will want that to return views in the long term (although there wasn't much discussion around it in the issue).

Any thoughts on nice ways to test this and avoid putting tm.assert_produces_warning in a zillion places across the code?

This also indicates that users will see a zillion of such warnings .. Also, what is the way for users to silence the warning?
So therefore, I am not really sure we should already start doing that now. Certainly because in many cases it won't actually matter for their code. The other option is to "simply" make a breaking change for pandas 2.0.

[jbrockmendel]

The other option is to "simply" make a breaking change for pandas 2.0

Is there precedent for this? Per #33597, I'm OK with just making the switch immediately.

This also indicates that users will see a zillion of such warnings .. Also, what is the way for users to silence the warning?

Good question. I guess we'd have to give a way to opt in to the always-view behavior, or maybe a global flag to silence the warning. Neither is particularly pretty.

[jorisvandenbossche]

What do you mean with precedent? (what kind of precedent are you looking for in this case?)

[jbrockmendel]

What do you mean with precedent? (what kind of precedent are you looking for in this case?)

Changes that would require a deprecation cycle being made without one

[jorisvandenbossche]

Ah, we certainly have that ;) Check the sections "Backwards incompatible API changes" in all previous releases?

[jbrockmendel]

makes sense, thanks