bind_rows() using tibbles with attributes loses attributes

[DavisVaughan]

I assume this is very similar to #2457.

Using bind_rows() on two tibbles where either one has extra attributes removes all extra attributes. Perhaps an approach similar to #1692 can be taken where the attributes of the first are kept? Ideally I would like bind_rows() to be generic but I've read all the problems associated with that.

My use case is in tibbletime. My print.tbl_time() method relies on one of the attributes. Because the tbl_time class is kept after using bind_rows(), but the attributes are lost, this causes problems. It causes other problems too, but this shows up immediately.

library(dplyr)

iris_tbl <- as_tibble(iris)

attr(iris_tbl, "my_attr") <- "important thing"

# The classes of the first object are kept due to #1692
class(iris_tbl) <- c("my_new_class", class(iris_tbl))

attributes(iris_tbl)
#> $names
#> [1] "Sepal.Length" "Sepal.Width"  "Petal.Length" "Petal.Width" 
#> [5] "Species"     
#> 
#> $row.names
#>   ...truncated
#>
#> $class
#> [1] "my_new_class" "tbl_df"       "tbl"          "data.frame"  
#> 
#> $my_attr
#> [1] "important thing"

iris_doubled <- iris_tbl %>%
  bind_rows(iris_tbl)

# The attributes besides names / row.names / class are lost
iris_doubled %>%
  attributes()
#> $names
#> [1] "Sepal.Length" "Sepal.Width"  "Petal.Length" "Petal.Width" 
#> [5] "Species"     
#> 
#> $row.names
#>   ...truncated
#>
#> 
#> $class
#> [1] "my_new_class" "tbl_df"       "tbl"          "data.frame"

Session info

devtools::session_info()
#> Session info -------------------------------------------------------------
#>  setting  value                       
#>  version  R version 3.4.3 (2017-11-30)
#>  system   x86_64, darwin15.6.0        
#>  ui       X11                         
#>  language (EN)                        
#>  collate  en_US.UTF-8                 
#>  tz       America/New_York            
#>  date     2017-12-22
#> Packages -----------------------------------------------------------------
#>  package    * version    date      
#>  assertthat   0.2.0      2017-04-11
#>  backports    1.1.0      2017-05-22
#>  base       * 3.4.3      2017-12-07
#>  bindr        0.1        2016-11-13
#>  bindrcpp     0.2        2017-06-17
#>  compiler     3.4.3      2017-12-07
#>  datasets   * 3.4.3      2017-12-07
#>  devtools     1.13.3     2017-08-02
#>  digest       0.6.12     2017-01-27
#>  dplyr      * 0.7.4      2017-09-28
#>  evaluate     0.10.1     2017-06-24
#>  glue         1.2.0      2017-10-29
#>  graphics   * 3.4.3      2017-12-07
#>  grDevices  * 3.4.3      2017-12-07
#>  htmltools    0.3.6      2017-04-28
#>  knitr        1.17       2017-08-10
#>  magrittr     1.5        2014-11-22
#>  memoise      1.1.0      2017-10-26
#>  methods    * 3.4.3      2017-12-07
#>  pillar       0.0.0.9000 2017-11-29
#>  pkgconfig    2.0.1      2017-03-21
#>  R6           2.2.2      2017-06-17
#>  Rcpp         0.12.14    2017-11-23
#>  rlang        0.1.4      2017-11-05
#>  rmarkdown    1.6.0.9004 2017-09-23
#>  rprojroot    1.2        2017-01-16
#>  stats      * 3.4.3      2017-12-07
#>  stringi      1.1.5      2017-04-07
#>  stringr      1.2.0      2017-02-18
#>  tibble       1.3.4.9003 2017-11-29
#>  tools        3.4.3      2017-12-07
#>  utils      * 3.4.3      2017-12-07
#>  withr        2.1.0.9000 2017-11-29
#>  yaml         2.1.14     2016-11-12
#>  source                                 
#>  CRAN (R 3.4.0)                         
#>  CRAN (R 3.4.0)                         
#>  local                                  
#>  CRAN (R 3.4.0)                         
#>  CRAN (R 3.4.0)                         
#>  local                                  
#>  local                                  
#>  CRAN (R 3.4.1)                         
#>  CRAN (R 3.4.0)                         
#>  CRAN (R 3.4.2)                         
#>  CRAN (R 3.4.0)                         
#>  CRAN (R 3.4.2)                         
#>  local                                  
#>  local                                  
#>  CRAN (R 3.4.0)                         
#>  cran (@1.17)                           
#>  CRAN (R 3.4.0)                         
#>  Github (hadley/memoise@d63ae9c)        
#>  local                                  
#>  Github (r-lib/pillar@5a082e1)          
#>  CRAN (R 3.4.0)                         
#>  CRAN (R 3.4.0)                         
#>  CRAN (R 3.4.3)                         
#>  CRAN (R 3.4.2)                         
#>  Github (DavisVaughan/rmarkdown@f85ba35)
#>  CRAN (R 3.4.0)                         
#>  local                                  
#>  CRAN (R 3.4.0)                         
#>  CRAN (R 3.4.0)                         
#>  Github (tidyverse/tibble@60281b3)      
#>  local                                  
#>  local                                  
#>  Github (jimhester/withr@fe81c00)       
#>  CRAN (R 3.4.0)

[DavisVaughan]

Thinking about what might be coming in the future, what if bind_rows() called sloop::reconstruct() using the new object and the first object before returning? This could retain class and attributes of the first object, and is generic, but there may be problems I'm not thinking of.

[krlmlr]

Thanks. Calling reconstruct() feels right, but we might want a "symmetric" version of reconstruct() to support use cases where we keep the attributes of the second (third, ...) argument.

[DavisVaughan]

By "symmetric" is that meaning keeping all custom attributes for each argument? Or just specifically attributes of the second argument, and not the first?

If it's the former, I can think of edge cases where the first and second arguments have attributes of the same name but with different data. Perhaps in this case the first argument should have precedence.

I assume this should propagate to bind_cols() too. It currently keeps all attributes of just the first object (different than bind_rows()).

library(dplyr)

iris_tbl1 <- as_tibble(iris)
iris_tbl2 <- as_tibble(iris)

attr(iris_tbl1, "my_attr")  <- "important thing 1"
attr(iris_tbl2, "my_attr")  <- "important thing 2"
attr(iris_tbl2, "my_attr2") <- "attr unique to 2"

# bind_cols() keeps attributes of the first object ONLY
bind_cols(iris_tbl1, iris_tbl2) %>%
  attributes()
#> $row.names
#> ...truncated
#> 
#> $class
#> [1] "tbl_df"     "tbl"        "data.frame"
#> 
#> $my_attr
#> [1] "important thing 1"
#> 
#> $names
#>  [1] "Sepal.Length"  "Sepal.Width"   "Petal.Length"  "Petal.Width"  
#>  [5] "Species"       "Sepal.Length1" "Sepal.Width1"  "Petal.Length1"
#>  [9] "Petal.Width1"  "Species1"

# bind_rows() of course loses all custom attributes
bind_rows(iris_tbl1, iris_tbl2) %>%
  attributes()
#> $names
#> [1] "Sepal.Length" "Sepal.Width"  "Petal.Length" "Petal.Width" 
#> [5] "Species"     
#> 
#> $row.names
#> ...truncated
#> 
#> $class
#> [1] "tbl_df"     "tbl"        "data.frame"

[krlmlr]

"Symmetric" means that the logic decides which attributes are inherited from which arguments, and that the attributes of bind_rows(a, b) and bind_rows(b, a) should be the same. Associativity would be great, too.

[krlmlr]

Related to #3429?

[hadley]

Duplicate of #2457 and will be resolved by vctrs.

[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/