Davis Vaughan

I'm not seeing any difference after applying the patch above. It looks like it still shallow duplicates to me?

Oh I see it if I run in the R console and not in RStudio. I forgot that RStudio keeps track of `x` in a way that increments the namedness....

Recycling benchmarks for integers: Master: ``` r library(vctrs) x expression min median `itr/sec` mem_alloc `gc/sec` #> #> 1 vec_recycle(x, size) 22.2µs 47.7µs 17413. 402KB 119. size # A tibble: 1...

This PR now also includes `vctrs_compact_rep_dbl`. It was relatively straightforward with the integer infrastructure in place. Recycling benchmarks for doubles: Master: ``` r library(vctrs) x expression min median `itr/sec` mem_alloc...

It looks like the call to `vec_recycle()` itself is always faster or as fast. Recycling an integer of size 1 to size 2 is the fastest thing I can think...

I have pushed the special `vec_recycle()` ALTREP code further down to `vec_slice_impl()`. This is interesting because we now also get `vec_init()` generating ALTREP reps as well. Really anything that slices...

We now have support for the 4 core types: integer, double, logical, and character! Logicals are a bit tricky, as altlogical, altraw and altcomplex support was only added in R...

@hadley, no, right now `vec_slice_impl()` (and therefore `vec_init()` and `vec_recycle()`) is restricted to only generate a compact rep if the object is unclassed and has no attributes whatsoever. I haven't...

This is a useful draft, but we probably would never merge it as is and will want to build it from scratch some time in the future. We can come...

One reason for list-of in tidyr is consistent chop/unchop roundtripping https://github.com/tidyverse/tidyr/blob/a170e55890cbebe257acf83103003e9f4f123dff/R/chop.R#L20-L25