API: Sort keys for DataFrame.assign

[TomAugspurger]

Closes #9777

Previously the order of new columns from .assign was arbitrary. For predictability, we'll sort before inserting.

We need to be comfortable with this change since we can't change behavior later with a keyword arg.

Technically we could allow referencing the a column defined within the same call to assign as long as they are sorted. e.g. df.assign(C=df.A + df.B, D=df.C**2) would work, but not df.assign(df.D=df.A +df.B, C=df.D**2). But I don't think we should.

cc @mrocklin

[shoyer]

👍 This looks great to me!

[jreback]

put the issue number here as a comment

[TomAugspurger]

✔️

[jreback]

agreed, you may want to add an example in the docs where you show using chained assignment (if for example you have an evaluation order requirement), .eg.

df = DataFrame({A = ..., B = ...})
df.assign(C=df.A+df.B).assign(D=df.C/df.A)

[TomAugspurger]

I've got one like that in the warning box

        (df.assign(C = lambda x: x['A'] + x['B'])
           .assign(D = lambda x: x['A'] + x['C']))

[jorisvandenbossche]

@TomAugspurger small not related note: you pushed your branch to upstream instead of master (no problem for now of course, but we should try to not do that for PRs I think)

[TomAugspurger]

@jorisvandenbossche are you talking about f00d6bb ? I reorganize the whatsnew entry when merging that commit. Or I could be confused about what I did. But I think I pushed this branch to origin.

[jorisvandenbossche]

no, I was talking about this pr, you apparantly pushed it to upstream (see at the top of the PR, I noticed it because did a fetch upstream). But as I said, no problem, it can always happen! Just be sure to delete it after the pr is merged

[TomAugspurger]

@jorisvandenbossche I see what you mean now. Sorry about that. I'll clean it up when this gets merged.

Speaking of which, any objections to merging?

[jreback]

this looks ok, go ahead and merge