rename_with() errors when no match/clarity of error message

[asadow]

I thought I could use map over a list of data frames to rename them using rename_with() and a tidyselector like matches(), whether or not each data frame has a matching column name. Now it looks as though that is not longer possible (#6688).

That's all fine and good. Now I see I can just use rename(), which does not error. But I think the error for rename_with() could be more clear by saying that there is no match.

Lastly, I'm curious: why the difference in erroring vs. not between rename() and rename_with()?

library(tidyverse)

# No error
mtcars |> rename("joe" = matches("qwerty")) |> head(1)
#>           mpg cyl disp  hp drat   wt  qsec vs am gear carb
#> Mazda RX4  21   6  160 110  3.9 2.62 16.46  0  1    4    4

# Hinting error
mtcars |> rename_with(~ "joe", qwerty)
#> Error in `rename_with()`:
#> ! Can't subset columns that don't exist.
#> ✖ Column `qwerty` doesn't exist.
#> Backtrace:
#>      ▆
#>   1. ├─dplyr::rename_with(mtcars, ~"joe", qwerty)
#>   2. ├─dplyr:::rename_with.data.frame(mtcars, ~"joe", qwerty)
#>   3. │ └─tidyselect::eval_select(enquo(.cols), .data, allow_rename = FALSE)
#>   4. │   └─tidyselect:::eval_select_impl(...)
#>   5. │     ├─tidyselect:::with_subscript_errors(...)
#>   6. │     │ └─rlang::try_fetch(...)
#>   7. │     │   └─base::withCallingHandlers(...)
#>   8. │     └─tidyselect:::vars_select_eval(...)
#>   9. │       └─tidyselect:::walk_data_tree(expr, data_mask, context_mask)
#>  10. │         └─tidyselect:::as_indices_sel_impl(...)
#>  11. │           └─tidyselect:::as_indices_impl(...)
#>  12. │             └─tidyselect:::chr_as_locations(x, vars, call = call, arg = arg)
#>  13. │               └─vctrs::vec_as_location(...)
#>  14. └─vctrs (local) `<fn>`()
#>  15.   └─vctrs:::stop_subscript_oob(...)
#>  16.     └─vctrs:::stop_subscript(...)
#>  17.       └─rlang::abort(...)

# Less helpful error
mtcars |> rename_with(~ "joe", matches("qwerty"))
#> Error in `rename_with()`:
#> ! `.fn` must return a vector of length 0, not 1.
#> Backtrace:
#>     ▆
#>  1. ├─dplyr::rename_with(mtcars, ~"joe", matches("qwerty"))
#>  2. └─dplyr:::rename_with.data.frame(mtcars, ~"joe", matches("qwerty"))
#>  3.   └─cli::cli_abort("{.arg .fn} must return a vector of length {length(sel)}, not {length(new)}.")
#>  4.     └─rlang::abort(...)

^{Created on 2023-10-26 with reprex v2.0.2}

[DavisVaughan]

It isn't really the fact that there is no match in matches("qwerty") that is the problem. That's totally fine. It's the fact that you matched 0 column names (i.e. the input to your .fn) but the output has length 1 ("joe").

rename_with() isn't really meant to be used with a static result like that, it is more often used where you take the input, apply some transformation, and return some output of the same size, like .fn = ~ stringr::str_c("blah.", .x).

So I think it is probably just the wrong function for what you were trying to do

[asadow]

I agree; the doc's say rename() should be used for individual columns.

I think my 2 points are still worth considering:

1. The rename_with() error is not very user-friendly.
2. rename_with() and rename() behave inconsistently: mtcars |> rename("joe" = matches("qwerty")) returns the data, but mtcars |> rename_with(~ "joe", matches("qwerty")) does not.