vec_match() fails on class "Date" and "IDate" with obscure error

[beansrowning]

I've noticed in several of my projects where I sometimes use data.table and dplyr together that an obscure error can arise if you attempt to perform a mutating join on a Date formated column with an inferred IDate/Date column as read in by data.table::fread()

Error: Can't join on `x$date` x `y$date` because of incompatible types.
i `x$date` is of type <date>>.
i `y$date` is of type <date>>.
Run `rlang::last_error()` to see where the error occurred.

This error was unintelligible to me and took a while to debug. I was able to browse into dplyr:::join_rows() and saw the offending line was the vec_match() call on line 7, yet the join columns of both were reported to be of type "Date" via the error and print.tbl_df:

I finally realized the column in the tibble originally sourced from fread() was of class "IDate", "Date". Which seems to trip vec_match()

I'm not sure what solution makes sense to make this more explicit to the end user, but as it stands it's hard to reason through.

Reprex

library(dplyr)
#> 
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#> 
#>     filter, lag
#> The following objects are masked from 'package:base':
#> 
#>     intersect, setdiff, setequal, union
library(data.table)
#> 
#> Attaching package: 'data.table'
#> The following objects are masked from 'package:dplyr':
#> 
#>     between, first, last
library(readr)

a <- tibble(
  a = letters,
  b = seq(Sys.Date(), Sys.Date() + 25, "days")
)

print(a)
#> # A tibble: 26 x 2
#>    a     b         
#>    <chr> <date>    
#>  1 a     2022-03-25
#>  2 b     2022-03-26
#>  3 c     2022-03-27
#>  4 d     2022-03-28
#>  5 e     2022-03-29
#>  6 f     2022-03-30
#>  7 g     2022-03-31
#>  8 h     2022-04-01
#>  9 i     2022-04-02
#> 10 j     2022-04-03
#> # ... with 16 more rows

write_csv(a, "test.csv")

b <- fread("test.csv") %>%
  mutate(d = LETTERS)

a %>%
  as_tibble() %>%
  print()
#> # A tibble: 26 x 2
#>    a     b         
#>    <chr> <date>    
#>  1 a     2022-03-25
#>  2 b     2022-03-26
#>  3 c     2022-03-27
#>  4 d     2022-03-28
#>  5 e     2022-03-29
#>  6 f     2022-03-30
#>  7 g     2022-03-31
#>  8 h     2022-04-01
#>  9 i     2022-04-02
#> 10 j     2022-04-03
#> # ... with 16 more rows

left_join(a, b)
#> Joining, by = c("a", "b")
#> Error: Can't join on `x$b` x `y$b` because of incompatible types.
#> i `x$b` is of type <date>>.
#> i `y$b` is of type <date>>.

sessionInfo()
#> R version 4.1.2 (2021-11-01)
#> Platform: x86_64-w64-mingw32/x64 (64-bit)
#> Running under: Windows 10 x64 (build 19044)
#> 
#> Matrix products: default
#> 
#> locale:
#> [1] LC_COLLATE=English_United States.1252 
#> [2] LC_CTYPE=English_United States.1252   
#> [3] LC_MONETARY=English_United States.1252
#> [4] LC_NUMERIC=C                          
#> [5] LC_TIME=English_United States.1252    
#> system code page: 65001
#> 
#> attached base packages:
#> [1] stats     graphics  grDevices utils     datasets  methods   base     
#> 
#> other attached packages:
#> [1] readr_2.1.0       data.table_1.14.2 dplyr_1.0.7      
#> 
#> loaded via a namespace (and not attached):
#>  [1] pillar_1.6.4      compiler_4.1.2    highr_0.9         R.methodsS3_1.8.1
#>  [5] R.utils_2.11.0    tools_4.1.2       bit_4.0.4         digest_0.6.28    
#>  [9] evaluate_0.14     lifecycle_1.0.1   tibble_3.1.6      R.cache_0.15.0   
#> [13] pkgconfig_2.0.3   rlang_0.4.12      reprex_2.0.1      cli_3.1.0        
#> [17] DBI_1.1.2         parallel_4.1.2    yaml_2.2.1        xfun_0.28        
#> [21] fastmap_1.1.0     withr_2.4.3       styler_1.6.2      stringr_1.4.0    
#> [25] knitr_1.37        hms_1.1.1         generics_0.1.1    fs_1.5.0         
#> [29] vctrs_0.3.8       bit64_4.0.5       tidyselect_1.1.1  glue_1.5.0       
#> [33] R6_2.5.1          fansi_0.5.0       vroom_1.5.6       rmarkdown_2.11   
#> [37] tzdb_0.2.0        purrr_0.3.4       magrittr_2.0.1    backports_1.3.0  
#> [41] ellipsis_0.3.2    htmltools_0.5.2   assertthat_0.2.1  utf8_1.2.2       
#> [45] stringi_1.7.5     crayon_1.4.2      R.oo_1.24.0

^{Created on 2022-03-25 by the reprex package (v2.0.1)}

[DavisVaughan]

It is somewhat confusing that vec_ptype_abbr() and vec_ptype_full() can utilize inheritance, meaning that the vec_ptype_abbr.Date method is hit here, which is why the results look so confusing.

library(vctrs)

x <- Sys.Date()
y <- data.table::as.IDate(x)

y
#> [1] "2022-03-25"
class(y)
#> [1] "IDate" "Date"

# Calls `vec_ptype_abbr.Date` through inheritance
vec_ptype_abbr(y)
#> [1] "date"

# Making this confusing
vec_ptype2(x, y)
#> Error in `stop_vctrs()`:
#> ! Can't combine <date> and <date>.



# Inject methods
vec_ptype_full.IDate <- function(x, ...) {
  "idate"
}
vec_ptype_abbr.IDate <- function(x, ...) {
  "idate"
}

vec_ptype_abbr(y)
#> [1] "idate"

vec_ptype2(x, y)
#> Error in `stop_vctrs()`:
#> ! Can't combine <date> and <idate>.

I'm not sure that we should try and provide ptype2 methods for Date and IDate. That doesn't seem like our place since we don't own IDate.

It is possible we should make vec_ptype_abbr() and vec_ptype_full() not utilize inheritance though.

[DavisVaughan]

See also https://github.com/tidyverse/dplyr/issues/6230#issuecomment-1100660750