Possible cause for object '.x' not found error with tar_combine

[nicolaspayette]

I had a problem, for which I eventually found the cause, but it took me quite some time to figure it out, so I'm posting it here in case some else runs into the same issue.

Take the following example:

library(targets)
tar_script({
  library(tibble)
  library(tarchetypes)
  targets <- tar_map(
    values = lst(letter = c("A", "B", "C")),
    tar_target(x, tibble(value = letter))
  )
  lst(
    targets,
    tar_combine(y, targets, command = dplyr::bind_rows(!!!.x, .id = "name"))
  )
})

tar_manifest(fields = command)
#> Error:
#> ! Error running targets::tar_manifest()
#>   Target errors: targets::tar_meta(fields = error, complete_only = TRUE)
#>   Tips: https://books.ropensci.org/targets/debugging.html
#>   Last error: object '.x' not found

#> Backtrace:
#>     ▆
#>  1. └─targets::tar_manifest(fields = command)
#>  2.   └─targets:::callr_outer(...)
#>  3.     ├─targets:::if_any(...)
#>  4.     └─targets:::callr_error(condition = out, fun = fun)
#>  5.       └─targets::tar_throw_run(message)
#>  6.         └─targets::tar_error(...)
#>  7.           └─rlang::abort(message = message, class = class, call = tar_empty_envir)

^{Created on 2022-11-30 with reprex v2.0.2}

It fails with object '.x' not found. Can you see why?

The culprit is the tibble::lst function, which I have used here to build the target list. It doesn't work with the slicing operator !!!.x, presumably because it's doing something fancy under the hood.

Using plain old base list instead fixes the issue:

library(targets)
tar_script({
  library(tibble)
  library(tarchetypes)
  targets <- tar_map(
    values = lst(letter = c("A", "B", "C")),
    tar_target(x, tibble(value = letter))
  )
  list(
    targets,
    tar_combine(y, targets, command = dplyr::bind_rows(!!!.x, .id = "name"))
  )
})

tar_manifest(fields = command)
#> # A tibble: 4 × 2
#>   name  command                                                                 
#>   <chr> <chr>                                                                   
#> 1 x_B   "tibble(value = \"B\")"                                                 
#> 2 x_C   "tibble(value = \"C\")"                                                 
#> 3 x_A   "tibble(value = \"A\")"                                                 
#> 4 y     "dplyr::bind_rows(x_A = x_A, x_B = x_B, x_C = x_C, \n     .id = \"name\…

^{Created on 2022-11-30 with reprex v2.0.2}

I'm happy to carry on from here, but if someone could explain why lst doesn't play well with !!!.x, I'd be very curious.

[wlandau]

The problem is that lst() tries to tidy-evaluate .x. We instead want lst() to ignore .x and leave it for tar_combine(). This works:

targets <- tar_map(
  values = lst(letter = c("A", "B", "C")),
  tar_target(x, tibble(value = letter))
)
combined <- tar_combine(y, targets, command = dplyr::bind_rows(!!!.x, .id = "name"))
lst(targets, combined)

[motorlearner]

Is there no way to solve this while having the tar_combine call inside lst?

When not specifying the command argument inside tar_combine, there is no object .x not found error (but a different one that asks to supply a .name_spec specification to vctrs::vec_c).

When specifying the command argument, e.g. just specifying the default, the object .x not found error appears.

[wlandau]

Is there no way to solve this while having the tar_combine call inside lst?

Is there a need to call tibble::lst() instead of base::list()?

[motorlearner]

No -- just I had already written it that way and don't want to re-run targets that take very long.

For the record, it works fine when combining lists of tibbles (without specifying a command). So I just converted the relevant targets (nested lists to lists of tibbles).