grouped_df throws errors in add_column

[t-kalinowski]

library(dplyr, warn.conflicts = FALSE)
library(tibble)

df <- tibble(x = 1:3, y = 1:3)

df %>% add_column(z = 4:6, .before = "y")
#> # A tibble: 3 x 3
#>       x     z     y
#>   <int> <int> <int>
#> 1     1     4     1
#> 2     2     5     2
#> 3     3     6     3
df %>% group_by(x) %>% add_column(z = 4:6, .before = "y")
#> Error: is.data.frame(df) is not TRUE

digging into this a little it seems to be caused by cbind() returning a matrix when called on a grouped_df

cbind(group_by(df, x), tibble(z = 1:3))
#>   [,1]      [,2]     
#> x Integer,3 Integer,3
#> y Integer,3 Integer,3
cbind(df, tibble(z = 1:3))
#>   x y z
#> 1 1 1 1
#> 2 2 2 2
#> 3 3 3 3

[krlmlr]

Does bind_cols() work?

[t-kalinowski]

Yes (with dev version of dplyr and tibble)

[krlmlr]

Is the behavior any different with the CRAN version?

[t-kalinowski]

No Sent from a mobile device.

…

On Sep 12, 2017, at 9:36 AM, Kirill Müller ***@***.***> wrote: Is the behavior any different with the CRAN version? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.

[krlmlr]

Closing, we don't currently support cbind().

[t-kalinowski]

The issue is aboutadd_column, not cbind.

[t-kalinowski]

The problem is caused by the call of cbind here:

tibble/R/add.R

Line 141 in 6320d77

out <- cbind(df, .data)

[krlmlr]

Bummer. This seems to mean that we need to make add_column() generic, with an implementation in dplyr.

[t-kalinowski]

Hmm, seems not ideal.
Would an approach based on [[<-to add the columns and then [ to reorder the columns be simpler?

[DavisVaughan]

@krlmlr I was just about to post an issue asking for add_column() to be generic.

For packages that extend tibble, like my new tibbletime package, I have to handle the dropping of extra classes and attributes that occurs with most tidyverse functions. Normally I create a tidyverse_fun.tbl_time generic version that drops the classes and attributes that I dont want to lose, calls the function and then adds them back.

Since add_column isn't generic, I can't do that there, and that call to cbind() removes the classes and attributes that I need to keep. At the end of add_column(), the classes are added back, but the attributes are not, which causes a lot of problems!

So for my problem, either add_column would need to be generic or it should also try and add the attributes of the original .data back on. Any help there is definitely welcome!

[t-kalinowski]

Most classes that build on top of tbl_df I think already need to define a [ method.... my preference would be to try to minimize the total number of methods that need to be defined in order to extend and build ontop of tbl_df. I think building add_column with [ is one such opportunity.

[DavisVaughan]

Solution

I think I have something that fixes everyone's problems using [.

https://github.com/DavisVaughan/tibble/blob/master/R/add.R#L141

Passes all tests for test-add.R.

Fixes t-kalinowski's problem

library(dplyr, warn.conflicts = FALSE)
library(tibble)

df <- tibble(x = 1:3, y = 1:3)

df %>% add_column(z = 4:6, .before = "y")
#> # A tibble: 3 x 3
#>       x     z     y
#>   <int> <int> <int>
#> 1     1     4     1
#> 2     2     5     2
#> 3     3     6     3

df %>% group_by(x) %>% add_column(z = 4:6, .before = "y")
#> # A tibble: 3 x 3
#> # Groups: x [3]
#>       x     z     y
#>   <int> <int> <int>
#> 1     1     4     1
#> 2     2     5     2
#> 3     3     6     3

Fixes my own problem

Classes and attrs are no longer lost.
Classes of the original .data are kept.

library(tibbletime)
#> 
#> Attaching package: 'tibbletime'
#> The following object is masked from 'package:stats':
#> 
#>     filter
library(tibble)

data(FB)
FB <- as_tbl_time(FB, date)

class(FB)
#> [1] "tbl_time"   "tbl_df"     "tbl"        "data.frame"
attr(FB, "index")
#> <quosure: frame>
#> ~date

FB <- FB %>% add_column(adj2 = FB$adjusted + 1)

class(FB)
#> [1] "tbl_time"   "tbl_df"     "tbl"        "data.frame"
attr(FB, "index")
#> <quosure: frame>
#> ~date

[krlmlr]

Thanks, happy to review a PR.

[krlmlr]

Closed by #306.

[github-actions]

This old thread has been automatically locked. If you think you have found something related to this, please open a new issue and link to this old issue if necessary.