Note that comparisons between columns from different dataframes is outside the standard

[MarcoGorelli]

This has come up a few times

#249 (comment)

Are folks open to simply saying that the operation is undefined by the existing standard? Implementations can choose to raise

#287 (comment)

On way to do [Column.mean() whilst not having to materialise more times than necessary when filtering] is for the column to keep track of its originating DataFrame:

---

so let's document that it's outside the standard?

[rgommers]

Thanks Marco. This PR is useful in identifying all (?) the potentially problematic methods and operator support. It makes it more concrete where we have tension between current eager and lazy libraries.

Are folks open to simply saying that the operation is undefined by the existing standard? Implementations can choose to raise

I'm not sure this is actually a good way to go. I'd like to explore a little more if we really can't do better than this, because it's at best unsatisfactory - as commented on in #249.

[MarcoGorelli]

It's not just eager-vs-lazy, but also that eager libraries arguably shouldn't be doing more passes through their data then they need to #287 (comment)

[MarcoGorelli]

From today's call: Columns may have an "owner" DataFrame, need to define expectations around how they behave (e.g. same owner, different owners, no owner)

[MarcoGorelli]

From today's call: Columns may have an "owner" DataFrame, need to define expectations around how they behave (e.g. same owner, different owners, no owner)

taken forwards in #310

[MarcoGorelli]

reopening as maybe this is an alternative, see #310 (comment)