NULL vs zero-length vectors

[hadley]

For vectorised arguments I think we should consistently treat NULL in the same way as if the argument had not been supplied. This is symmetric with our use of NULL for the default value of optional arguments that need complex calculations.

library(vctrs)
vec_c(TRUE, double())
#> [1] 1

vec_c(TRUE, NULL)
#> [1] TRUE
# Same as 
vec_c(TRUE, )
#> [1] TRUE
# Same as
vec_c(TRUE)
#> [1] TRUE

[hadley]

For glue this would imply that the following calls should be identical:

glue::glue("{a}: {b}", a = 1:2)
#> Error in eval(parse(text = text, keep.source = FALSE), envir): object 'b' not found
glue::glue("{a}: {b}", a = 1:2, b = )
#> Error in eval(parse(text = text, keep.source = FALSE), envir): argument is missing, with no default
glue::glue("{a}: {b}", a = 1:2, b = NULL)

And for purrr:

str(purrr::pmap(rlang::list2(1, 2), list))
#> List of 1
#>  $ :List of 2
#>   ..$ : num 1
#>   ..$ : num 2
str(purrr::pmap(rlang::list2(1, 2, ), list))
#> List of 1
#>  $ :List of 2
#>   ..$ : num 1
#>   ..$ : num 2
str(purrr::pmap(rlang::list2(1, 2, NULL), list))
#>  list()

For tibble:

tibble::tibble(x = 1:2)
#> # A tibble: 2 x 1
#>       x
#>   <int>
#> 1     1
#> 2     2
tibble::tibble(x = 1:2, y = NULL)
#> Error: Column `y` must be a 1d atomic vector or a list

# vs
tibble::tibble(x = 1:2, y = integer())
#> Error: Column `y` must be length 1 or 2, not 0

[hadley]

Also comes up in pluck()

[hadley]

Connected to #13

[krlmlr]

Is the current behavior of mutate(iris, Petal.Width = NULL) consistent with that description? (I admit I sometimes use it to say "I don't care if this column exists or not, just get rid of it", but it doesn't seem completely right.)

[hadley]

@krlmlr I think that's a special case where it's ok to have different behaviour.

And now implemented in tidyr::expand_grid(). This ensures that for vector inputs, vec_size(expand_grid(x1, x2, ..., xn)) = vec_size(x1) * vec_size(x2) * ... * vec_size(xn).