Forward all_equal() to dplyr

[krlmlr]

Slightly different implementation than in dplyr causes trouble (tidyverse/dplyr#2264). Throw an error if dplyr is not installed.

[krlmlr]

@hadley: Do you see a problem if tibble tries to forward to dplyr::all_equal() and fails if dplyr is not installed?

[hadley]

Yeah, I think that's ok.

[hadley]

Alternatively, maybe we tibble simply shouldn't define an all.equal() method? Although I guess that means you'd get different behaviour depending on whether or not dplyr was loaded, which we want to avoid.

[krlmlr]

Oh, tibble does have its own implementation of all_equal(), but the implementation is slightly different; behavior depends if dplyr is loaded or not. We'd have to move half of dplyr's C++ code here to use the same implementation.

[jennybc]

Alternatively, maybe we tibble simply shouldn't define an all.equal() method? Although I guess that means you'd get different behaviour depending on whether or not dplyr was loaded, which we want to avoid.

I just experienced exactly this and it was mystifying.

In my case, dplyr is not explicitly loaded -- a deliberate choice, on my part -- and yet because I have assign("%>%", dplyr::%>%, envir = .__Rprofile.env__.) in my .Rprofile, dplyr is "loaded via a namespace (and not attached)", in the sessionInfo() sense. Obviously I need to rethink the wisdom of this line of .Rprofile!

But the result is that all.equal(tibble_1, tibble_2) is calling tibble::all.equal.tbl_df(), which forwards to dplyr::all_equal(), which apparently has different default behaviour around attributes. I am accustomed to a default of check.attributes = TRUE and expected my comparison to behave that way. But dplyr::all_equal() seems to act as if check.attributes = FALSE. Or maybe it's just oblivious to attributes.

Has this ever been considered? The tidyverse/dplyr style of tibble comparison only happens with all_equal() (with the underscore) and never silently with all.equal().

[jennybc]

Related: I believe this line in my .Rprofile is why it took me so long to sympathize with people reporting that tibble printing or indexing changed during the course of a script or interactive session, as more packages gradually got loaded. I always got tibble-style printing and indexing and therefore never personally experienced this behaviour.

[hadley]

Probably the safest thing to do is explicitly forward to dplyr, with a clear error if dplyr isn't installed.

[krlmlr]

As of tibble 1.3.0, we always forward all_equal() to dplyr::all_equal() and abort if it's not installed. The implementation of all.equal.tbl_df() forwards to dplyr if it is available, otherwise it calls NextMethod() with a warning. This is to satisfy package checks for packages that suggest tibble but not dplyr.

Recent versions of import packages (readr, haven, readxl, ...) now always import tibble, so the incremental change of behavior should be rarer now. A package that returns tibbles and wants to ensure printing and [ needs to declare it in Imports: and actively import any object in its NAMESPACE.

[jennybc]

My main point is that I expected all.equal() to catch a difference in attributes, because it has always done so in the past (meaning "for most of my R life", not a comment on recent tidyverse developments). But all.equal.tbl_df() doesn't check attributes AFAICT. And so 2 tibbles with identical data but different attributes returned TRUE, which surprised me.

So all.equal() is arguably another function that should be messaged like intersect(), setdiff(), etc. Except in this case the default behaviour isn't even the same.

library(readxl)

tib1 <- read_excel(readxl_example("datasets.xlsx"), n_max = 2)
tib2 <- tib1

## new-to-me behavior re: attributes in all.equal()
attr(tib2, "foo") <- "bar"
all.equal(tib1, tib2) # huh?
#> [1] TRUE
all.equal(as.data.frame(tib1), as.data.frame(tib2))
#> [1] "Attributes: < Names: 1 string mismatch >"                            
#> [2] "Attributes: < Length mismatch: comparison on first 2 components >"   
#> [3] "Attributes: < Component 2: Modes: numeric, character >"              
#> [4] "Attributes: < Component 2: Lengths: 2, 1 >"                          
#> [5] "Attributes: < Component 2: target is numeric, current is character >"

Session info

devtools::session_info()
#> Session info -------------------------------------------------------------
#>  setting  value                       
#>  version  R version 3.3.2 (2016-10-31)
#>  system   x86_64, darwin13.4.0        
#>  ui       X11                         
#>  language (EN)                        
#>  collate  en_CA.UTF-8                 
#>  tz       America/Vancouver           
#>  date     2017-04-09
#> Packages -----------------------------------------------------------------
#>  package    * version     date       source                            
#>  assertthat   0.1         2013-12-06 CRAN (R 3.2.0)                    
#>  backports    1.0.5       2017-01-18 cran (@1.0.5)                     
#>  cellranger   1.1.0       2016-07-27 CRAN (R 3.3.0)                    
#>  devtools     1.12.0.9000 2017-02-09 Github (hadley/devtools@d8ab190)  
#>  digest       0.6.12      2017-01-27 cran (@0.6.12)                    
#>  dplyr        0.5.0.9001  2017-04-05 Github (hadley/dplyr@cf46c46)     
#>  evaluate     0.10        2016-10-11 cran (@0.10)                      
#>  glue         0.0.0.9000  2017-04-03 Github (tidyverse/glue@6e77c3b)   
#>  htmltools    0.3.5       2016-03-21 CRAN (R 3.2.4)                    
#>  knitr        1.15.1      2016-11-22 CRAN (R 3.3.2)                    
#>  magrittr     1.5         2014-11-22 CRAN (R 3.2.0)                    
#>  memoise      1.0.0.9001  2016-12-18 Github (hadley/memoise@884d565)   
#>  pkgbuild     0.0.0.9000  2017-03-13 Github (r-pkgs/pkgbuild@5ed87aa)  
#>  pkgload      0.0.0.9000  2017-04-03 Github (r-pkgs/pkgload@3a96cf2)   
#>  R6           2.2.0       2016-10-05 CRAN (R 3.3.0)                    
#>  Rcpp         0.12.10     2017-03-31 Github (RcppCore/Rcpp@886f5df)    
#>  readxl     * 0.1.1.9000  2017-04-09 local                             
#>  rlang        0.0.0.9006  2017-04-05 Github (hadley/rlang@0756eaf)     
#>  rmarkdown    1.4.0.9000  2017-03-31 Github (rstudio/rmarkdown@5f7cd3c)
#>  rprojroot    1.2         2017-01-16 CRAN (R 3.3.2)                    
#>  stringi      1.1.3       2017-03-21 cran (@1.1.3)                     
#>  stringr      1.2.0       2017-02-18 cran (@1.2.0)                     
#>  tibble       1.3.0       2017-04-01 cran (@1.3.0)                     
#>  withr        1.0.2       2016-06-20 cran (@1.0.2)                     
#>  yaml         2.1.14      2016-11-12 cran (@2.1.14)

[krlmlr]

Yes, all.equal.tbl_df() is different from all.equal.data.frame(), too. By default, row or column order doesn't matter either.

@hadley: Should we compare attributes in dplyr::all_equal(), and carry this over to tibble::all.equal.tbl_df()?

[hadley]

Yeah, we should attributes (probably with an option to turn off)