Davis Vaughan

So I gave an internal presentation of the current state of `rray` to the tidyverse team, they seemed fairly interested! As I was describing how broadcasting works for those who...

I just realized that Julia uses column-major arrays, and they have broadcasting baked in, so I was curious about whether they append or prepend new dimensions when doing broadcasting operations....

That's tough because the first thing we do in `vec_slice()` is call `vec_as_location()`, which forces us to lose the context of where the `NA` originated from ``` r vctrs::vec_as_location(c(TRUE, NA,...

I'm fairly certain the `vec_ptype2()` vs `vec_ptype_common()` behavior was intentional, with the intention being that `vec_ptype2()` is used when computing the common type by doing something like `reduce(xs, vec_ptype2)`, so...

Oh good point about `vec_ptype2()` using `vec_ptype()`, that logic seems correct to me

~Not quite right because extra attributes should not be kept~ ``` > vctrs::vec_get(structure(1, extra = "x", names = "x"), 1) [1] 1 attr(,"extra") [1] "x" ``` I take it back,...

With `vec_get()` and the extension, `vec_chop2()`, it looks like this dplyr section where columns are sliced could be reduced to: https://github.com/tidyverse/dplyr/blob/057b453428f0139dd013ee4db247d1a1ca51032f/R/tbl-df.r#L38-L49 ```r resolve_chunks

Should rename to `vec_slice2()`, and add `vec_chop2()`

Idea from @lionel-: `vec_chop()` always keeps names on the inner vectors, `vec_chop2()` moves the names from the inner vectors to the outer list. This is consistent with `vec_slice()`, which keeps...

Will be `vec_slice2()`, like in #1228