This feels too magical (overloaded) to me. I think I’d prefer to have dedicated methods for converting from array-of-objects to object-of-arrays and vice versa.

I'd tend to disagree: for me iterating over Object.keys(rows) and Object.keys(row) is unifying both structures and feels "natural" (maybe because that's how I represent these things in my head).

The weird part might be length ? [] : {}; because I don't know how to distinguish [] and {} otherwise?

The weird part might be length ? [] : {}; because I don't know how to distinguish [] and {} otherwise?

That’s exactly it. There are a variety of other ways to test for arrays (or iterables), and it’s not clear to me that the truthy length property is what we want. My thought was that if we don’t try to handle all of these cases in one method then we can more precisely define what the behavior should be. The current documentation doesn’t really say what will happen, and I think it would be cumbersome to define it.

d3.permute is agnostic to [] {}, but its keys are passed explicitly
https://github.com/d3/d3-array/blob/master/src/permute.js

Would it be more acceptable to test for x instanceof Array instead?

In any case, whether we instantiate the row or column as a [] or a {} is a bit cosmetic, since both arrays and objects can have any type of keys (integers or strings).

Also, if these are iterables we should create an Array.from them?

closing due to inactivity