[ideas] Should tar_map complain if names does not exist in values?

[KaiAragaki]

Help

• I understand and agree to https://books.ropensci.org/targets/help.html.

Description

If a column given by values does not exist, random file names are given instead. Is this desired behavior, or would an error be appropriate?

library(targets)
tar_script({
    library(tarchetypes)
    values <- data.frame(names = c("A", "B"))
    list(tar_map(values = values, names = "i_dont_exist", tar_target(my_target, 
        "hi")))
})
tar_make()
#> ▶ start target my_target_6a0a0001
#> ● built target my_target_6a0a0001 [0.001 seconds]
#> ▶ start target my_target_df8a9eb9
#> ● built target my_target_df8a9eb9 [0 seconds]
#> ▶ end pipeline [0.035 seconds]
tar_manifest()
#> # A tibble: 2 × 2
#>   name               command 
#>   <chr>              <chr>   
#> 1 my_target_6a0a0001 "\"hi\""
#> 2 my_target_df8a9eb9 "\"hi\""

^{Created on 2023-08-08 with reprex v2.0.2}

Thanks for making a beautiful pkg

[wlandau]

Unfortunately, that is not feasible because tarchetypes uses tidyselect for names. It is designed for calls like tar_map(names = starts_with("my_target")), so names is automatically filtered to be a subset of names(values).

[KaiAragaki]

Sorry, I'm not sure if I explained myself properly - or perhaps I'm misunderstanding you. In this proposal, I'm saying it SHOULD error if a column does not exist/no columns that satisfy starts_with exists, but currently it does not. Sorry if I'm just missing something here.

[wlandau]

I think throwing an error could lead to edge cases and friction in the general case. For example, suppose a user sets names = -starts_with("bad_prefix") to exclude certain names. I think they would expect tar_map() to work even if the suffixes become random hashes.

[wlandau]

As always, tar_manifest() and tar_visnetwork() are good ways to check that you have the names you want.