Should complete() and friends detect data masks in caller environment?

[lionel-]

And should lst() take an optional mask as argument? cc @krlmlr

https://community.rstudio.com/t/how-to-make-complete-nesting-work-with-quosures-and-tidyeval/16473/4

[lionel-]

Related to dplyr::case_when() issues.

[krlmlr]

Are you suggesting that complete() and nesting() behave differently depending on the caller? This feels like opening a new can of worms to me, but I don't fully understand the implications yet.

How is this related to case_when()?

As for lst(), perhaps we could use sth like

eval_tidy(tibble::lst(!!!ensyms(...)), .data)

? Do you see other advantages of having an explicit data mask in lst() ?

[lionel-]

Behaving differently based on the caller is often the way to implement lexically scoped behaviour, i.e. the right thing to do.

Passing the data mask is a way to pass the user's lexical context to lst() where the evaluation happens. But good suggestion about quoting lst(), I think this might just work!

eval_tidy(expr(tibble::lst(!!!quos(...))), mask)

[lionel-]

It doesn't work because:

• lst() does not forward the caller env to lst_quos(), which does not forward it to eval_tidy(). So the lexical context passed to eval_tidy() (default value of the env argument) is the execution env of tibble:::lst_quos().

• Even if the caller env is correctly forwarded, lst_quos() also uses a data mask. So rlang would need to be able to combine data masks passed as data with data masks found in env. This is possible but would take some work.

In any case it'd be cleaner if tibble did forward the user context to eval_tidy().

[hadley]

Could you please inline the discussion from community, including a small reprex?

[batpigandme]

Question was basically why nesting() doesn't work with quosures.

Now there is a better error message thrown that tells you what to do so, FWIW, I don't think it's really problematic at this point.

library(tidyverse)

d <- mtcars %>%
  mutate_at(vars(carb, am), as.factor)

# desired behaviour works with this syntax
d %>%
  group_by(carb, am) %>%
  tally() %>%
  complete(am, nesting(carb), fill = list(n = 0))
#> # A tibble: 12 x 3
#> # Groups:   carb [6]
#>    am    carb      n
#>    <fct> <fct> <dbl>
#>  1 0     1         3
#>  2 1     1         4
#>  3 0     2         6
#>  4 1     2         4
#>  5 0     3         3
#>  6 1     3         0

# in a function, `nest` does not work with a quosure
fnc <- function(group, nest, data) {
  group <- enquo(group)
  nest <- enquo(nest)

  data %>%
    group_by(!!group, !!nest) %>%
    tally() %>%
    complete(!!group, nesting(!!nest), fill = list(n = 0))
}

fnc(am, carb, d)
#> Error in eval_tidy(xs[[i]], unique_output): object 'carb' not found

# uses `ensym()` for nest rather than `enquo()`

fnc2 <- function(group, nest, data) {
  group <- enquo(group)
  nest <- ensym(nest)
  
  data %>%
    group_by(!!nest, !!group) %>%
    tally() %>%
    complete(!!group, nesting(!!nest), fill = list(n = 0))
}

fnc2(am, carb, d)
#> # A tibble: 12 x 3
#> # Groups:   carb [6]
#>    am    carb      n
#>    <fct> <fct> <dbl>
#>  1 0     1         3
#>  2 1     1         4
#>  3 0     2         6
#>  4 1     2         4
#>  5 0     3         3
#>  6 1     3         0

^{Created on 2019-01-04 by the reprex package (v0.2.1.9000)}