`expand_grid()` evaluation environment

[moodymudskipper]

library(tidyr)
data <- tibble(foo=1)
foo <- bar <- 1:2

expand_grid(data, new = bar)
#> # A tibble: 2 × 2
#>     foo   new
#>   <dbl> <int>
#> 1     1     1
#> 2     1     2

expand_grid(data, new = foo)
#> # A tibble: 1 × 2
#>     foo   new
#>   <dbl> <dbl>
#> 1     1     1

This puzzled me then hit me that it behaves like a mutating function and this works :

expand_grid(data, new = .env$foo)
#> # A tibble: 2 × 2
#>     foo   new
#>   <dbl> <int>
#> 1     1     1
#> 2     1     2

expand_grid(data, new = !!foo)
#> # A tibble: 2 × 2
#>     foo   new
#>   <dbl> <int>
#> 1     1     1
#> 2     1     2

I'm not sure if the current behaviour is intended but it's undocumented. As far as I'm concerned I wouldn't mind have the dots evaluated outside of the data.

[DavisVaughan]

expand_grid() takes ..., the signature isn't .data, ... like mutate(). Just want to clarify that up front.

Beyond that, this is just the fact that expand_grid() evaluates its arguments sequentially like tibble() and mutate() (for better or worse), which makes your example a special case of this behavior

# second `foo` references first `foo` result
tidyr::expand_grid(foo = 1, new = foo)
#> # A tibble: 1 × 2
#>     foo   new
#>   <dbl> <dbl>
#> 1     1     1

[hadley]

expand_grid() seems overly complicated for it's purpose. Maybe we can simplify it.

[DavisVaughan]

In my brain it should just capture ... with list2(), so yea, agreed

Like, when do you really use the "sequential evaluation" feature with expand_grid()? I don't think I ever have

[hadley]

I assume at the time we wanted to make it like tibble(), but I don't see why that matters. As long as it doesn't break a bunch of packages on CRAN, I think we can simplify.