dplyr functions change variables for no reason since dplyr_0.8.0

[mariodejung]

I couldn't thing about a better title, but I run into problems since the new dplyr release.

I narrowed it down to a single dplyr call, but it changes different variables in my script, even they are not "touched" by the dplyr call.

First of all, there is a group_by_at call, because 'if there is a column "species", I want to group by it'.
If the column does not exist, I get a warning, which was fine for me, but I don't understand the class changes for the other variables. This bringt problems downstream in my script because older functions can't handle the tibble yet.

library(dplyr)

set.seed(1)
# df <- data.frame(species=c('a','b'),
#                        Intensity=rnorm(1000, 25, 3))
df <- data.frame(Intensity=rnorm(1000, 25, 3))
class(df)
df_backup <- df

df_test <- 
  df %>% 
  dplyr::group_by_at(vars(matches('^species$'))) %>%
  dplyr::summarise(`5%`=stats::quantile(log10(Intensity),.05),
                   `50%`=stats::quantile(log10(Intensity),.50),
                   `95%`=stats::quantile(log10(Intensity),.95)) 
class(df)
class(df_test)
class(df_backup)

[batpigandme]

Hmm, that is strange. FFR, if you can run your example through reprex, it's helpful to see the input and the output right in the issue.

library(dplyr)

set.seed(1)
df <- data.frame(Intensity=rnorm(1000, 25, 3))
class(df)
#> [1] "data.frame"
df_backup <- df
class(df_backup)
#> [1] "data.frame"
df_test <- df %>% 
  dplyr::group_by_at(vars(matches('^species$'))) %>%
  dplyr::summarise(`5%`=stats::quantile(log10(Intensity),.05),
                   `50%`=stats::quantile(log10(Intensity),.50),
                   `95%`=stats::quantile(log10(Intensity),.95)) 
class(df)
#> [1] "tbl_df"     "tbl"        "data.frame"
class(df_test)
#> [1] "tbl_df"     "tbl"        "data.frame"
class(df_backup)
#> [1] "tbl_df"     "tbl"        "data.frame"

library(lobstr)
obj_addr(df)
#> [1] "0x7f86a3fb95d8"
obj_addr(df_backup)
#> [1] "0x7f86a3fb95d8"
obj_addr(df_test)
#> [1] "0x7f86a48584a8"

^{Created on 2019-02-25 by the reprex package (v0.2.1)}

I added the object address code from Binding basics in Advanced R. df and df_backup are just two names bound to the same value, but that is surprising that the copy on modify doesn't preserve df_backup as it was….

[mariodejung]

I even use this call within a function and it changes the objects outside the function!

library(dplyr)

set.seed(1)
df<- data.frame(Intensity=rnorm(1000, 25, 3))
class(df)
#> [1] "data.frame"
df_backup <- df
class(df_backup)
#> [1] "data.frame"
my_plotAbundanceRank <- function(data_set) {
    quantile_df <- 
        data_set %>% 
        dplyr::group_by_at(vars(matches('^species$'))) %>%
        dplyr::summarise(`5%`=stats::quantile(log10(Intensity),.05),
                         `50%`=stats::quantile(log10(Intensity),.50),
                         `95%`=stats::quantile(log10(Intensity),.95)) 
}
print(my_plotAbundanceRank(df))
#> # A tibble: 1 x 3
#>    `5%` `50%` `95%`
#>   <dbl> <dbl> <dbl>
#> 1  1.30  1.40  1.48
class(df)
#> [1] "tbl_df"     "tbl"        "data.frame"
class(df_backup)
#> [1] "tbl_df"     "tbl"        "data.frame"

[DavisVaughan]

I think we just need a shallow copy in the case when no groups are specified.

dplyr/src/group_indices.cpp

Line 645 in 16125d1

data.attr("class") = NaturalDataFrame::classes();

Basically it should do what ungroup_grouped_df() does:

dplyr/src/group_indices.cpp

Line 669 in 16125d1

DataFrame ungroup_grouped_df(DataFrame df) {

suppressPackageStartupMessages(library(dplyr))
#> Warning: package 'dplyr' was built under R version 3.5.2

x <- data.frame(y = 1)

x
#>   y
#> 1 1

dplyr::group_by(x)
#> # A tibble: 1 x 1
#>       y
#>   <dbl>
#> 1     1

x
#> # A tibble: 1 x 1
#>       y
#>   <dbl>
#> 1     1

suppressPackageStartupMessages(library(dplyr))
#> Warning: package 'dplyr' was built under R version 3.5.2

x <- data.frame(y = 1)

x
#>   y
#> 1 1

dplyr::group_by(x, y)
#> # A tibble: 1 x 1
#> # Groups:   y [1]
#>       y
#>   <dbl>
#> 1     1

x
#>   y
#> 1 1

[lock]

This old issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with reprex) and link to this issue. https://reprex.tidyverse.org/