Tibble allows duplicate column names after rename().

move[bot] commented on Jul 21, 2018, 9:58 PM UTC:

@kovla commented on Jul 12, 2018, 9:35 AM UTC:

I imagine that rename() of a column to a name that already exists in the tibble (in another column) should warrant at least a warning. Otherwise it can result in two indistinguishable variables, with same name and type:

x <- tibble(a = 1:3, b = 4:6, c = 7:9)
x %>% rename(b = c)

# A tibble: 3 x 3
      a     b     b
  <int> <int> <int>
1     1     4     7
2     2     5     8
3     3     6     9

This issue was moved by krlmlr from tidyverse/tibble#431.

This issue was moved by romainfrancois from tidyverse/dplyr#3723.

@krlmlr commented on Jul 21, 2018, 9:59 PM UTC:

Thanks, confirmed. We shouldn't create duplicates here.

@romainfrancois commented on Aug 13, 2018, 8:50 AM UTC:

Is this a tidyselect issue @lionel- ?

> dplyr:::rename.data.frame
function(.data, ...) {
  vars <- tidyselect::vars_rename(names(.data), !!!quos(...))
  select_impl(.data, vars)
}
<bytecode: 0x10f1e6ed0>
<environment: namespace:dplyr>
> x <- tibble(a = 1:3, b = 4:6, c = 7:9)
> tidyselect::vars_rename(names(x), b = c)
  a   b   b 
"a" "b" "c" 

We can be defensive in dplyr otherwise, but I feel vars_rename should be.

@lionel- commented on Aug 13, 2018, 9:03 AM UTC:

It looks like it.

[krlmlr]

What do you suggest that tibble should do about it?

[romainfrancois]

I'm suggesting that tidyselect::vars_rename result should not have the same name more than once.

[krlmlr]

Thought this is a tibble issue now... 🤦‍♂️ 🤦‍♂️

[profandyfield]

There are times you might want duplicate names. For example, creating a tibble to pass to kable() or apa_table() in papaja. Let's say your table will show mean and CI of some outcome measures for treatment and control groups. Different measures are in rows of the table, named in column 1. Column 2 is mean for treatment, column 3 the CI for treatment, column 4 the mean for control, column 5 the CI for control. You're going to put column spans above the header row to denote treatment (cols 2 & 3) and control (cols 4 and 5). You'd want column 2 and 4 in the tibble to both be labelled 'Mean' and columns 3 and 5 to both be labelled '95% CI' because (and I could be wrong) in apa_table you can't override the variable names from the tibble you pass into it. Even if I'm wrong about that, it makes senseto have an option to override a default of no duplicate names IMO.

[lionel-]

@profandyfield You'd probably want to use nested tibbles in that case (i.e. a whole data frame in a column):

tibble(
  measure = c("foo", "bar"),
  treatment = tibble(mean = 1:2, se = rnorm(2)),
  control = tibble(mean = 3:4, se = rnorm(2))
)
#> # A tibble: 2 x 3
#>   measure treatment$mean    $se control$mean    $se
#>   <chr>            <int>  <dbl>        <int>  <dbl>
#> 1 foo                  1 -0.453            3 -0.577
#> 2 bar                  2  0.239            4 -1.18