Nick Crews

I'm gonna be out of office for this week. So I hope when I get back I see that y'all will have figured it all out and everything is merged...

@cpcloud whenever you get the chance, I responded to each of the individual comments, curious what you think. I agree it's sorta mungy, but I can't think of a cleaner...

Perhaps the larger question here is the tension between - provide a high level API that is very flexible in what it takes in (None, python lists of python vals,...

> Can we please leave all the additional complexity and decisions out here beyond addressing the empty array case and allow the user to specify a type? Yeah I think...

I currently have lots of code that looks like `t.mutate(emails=ibis.array([t.email0, t.email1, t.email2])`. Am I missing something here, how can I accomplish this goal?

Yes, I agree it is really hairy, and I think that would help. Maybe I start with a PR that just adds or changes tests to precisely show what the...

> I don't think this is supported in many of our backends. How is this spelled in DuckDB for example? Indeed, it is only supported on bigquery, dask, and pandas...

@jcrist @kszucs @gforsyth curious what yall think about the costs/benefits of allowing/disallowing empty structs

hmm, our pyarrow type roundtrip fuzzing tests are failing. Can I get some help with hypothesis to limit it to only give us fields of length 1+? I don't understand...

Just to be clear, I swapped this PR to now be checking for disallowment