Consider adding `vec_expand_grid()`

[DavisVaughan]

It would be nice to have this in packages that can't use tidyr, and we could probably implement it in a slightly more performant way than tidyr.

> tidyr:::expand_grid
function (..., .name_repair = "check_unique") 
{
    dots <- dots_cols(...)
    ns <- map_int(dots, vec_size)
    n <- prod(ns)
    if (n == 0) {
        out <- map(dots, vec_slice, integer())
    }
    else {
        each <- n/cumprod(ns)
        times <- n/each/ns
        out <- pmap(list(x = dots, each = each, times = times), 
            vec_repeat)
    }
    out <- as_tibble(out, .name_repair = .name_repair)
    flatten_nested(out, attr(dots, "named"), .name_repair)
}

I'm not sure if we would need the flattening bit, which seems like it might be there for some backwards compat support

[DavisVaughan]

This seems pretty good - it is possible this should live in funs though

library(vctrs)
library(rlang)
library(purrr)

vec_expand_grid <- function(..., .name_repair = "check_unique") {
  args <- list2(...)
  
  names <- names2(args)
  names <- vec_as_names(names, repair = .name_repair, repair_arg = ".name_repair")
  args <- unname(args)
  
  sizes <- list_sizes(args)
  size <- as.integer(prod(sizes))
  
  if (size == 0L) {
    args <- map(args, vec_slice, NULL)
  } else {
    times <- size / cumprod(sizes)
    args <- map2(args, times, vec_rep_each)
    times <- size / times / sizes
    args <- map2(args, times, vec_rep)
  }
  
  names(args) <- names
  
  new_data_frame(args, n = size)
}

Examples

# require named input by default.
# i think this is a lot safer and more predictable,
# i don't particularly love auto-naming of quos for this
vec_expand_grid(1:2, 3:4)
#> Error: Names can't be empty.
#> x Empty names found at locations 1 and 2.

vec_expand_grid(x = 1:2, y = 3:5)
#>   x y
#> 1 1 3
#> 2 1 4
#> 3 1 5
#> 4 2 3
#> 5 2 4
#> 6 2 5

vec_expand_grid(1:2, 3:4, .name_repair = "unique")
#> New names:
#> * `` -> ...1
#> * `` -> ...2
#>   ...1 ...2
#> 1    1    3
#> 2    1    4
#> 3    2    3
#> 4    2    4

# with df-cols
df <- data_frame(a = 1:3, b = c("a", "b", "c"))
tibble::as_tibble(vec_expand_grid(x = 1:2, df = df))
#> # A tibble: 6 x 2
#>       x  df$a $b   
#>   <int> <int> <chr>
#> 1     1     1 a    
#> 2     1     2 b    
#> 3     1     3 c    
#> 4     2     1 a    
#> 5     2     2 b    
#> 6     2     3 c

Performance seems to be better than tidyr / base, at least in these simple examples.

x <- 1:1000
y <- 1:50
z <- letters

bench::mark(
  vctrs = vec_expand_grid(x = x, y = y, z = z),
  tidyr = as.data.frame(tidyr::expand_grid(x = x, y = y, z = z)),
  base = expand.grid(z = z, y = y, x = x, stringsAsFactors = FALSE)[c("x", "y", "z")],
  iterations = 50
)
#> # A tibble: 3 x 6
#>   expression      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 vctrs        9.38ms   9.38ms     107.     44.7MB   5330. 
#> 2 tidyr       61.32ms  64.98ms      14.9    43.7MB     27.3
#> 3 base        27.43ms  32.78ms      31.1    39.9MB     60.4