dplyr bind_rows coerces inputs where data.table rbindlist does not

[tchakravarty]

@hadley I am trying to call bind_rows on a list of data.frames which have the identical set of columns with the identical attributes (or at least to the extent that class can tell). The columns that are getting coerced to numeric have the class attributes c("chron", "date", "times"). data.table::rbindlist is correctly handling these types. Is there a reason for this difference and is dplyr doing the right thing and data.table is not? The latter certainly seems to have the more desirable outcome.

Here is a reproducible example (apologies for the formatting of the dput output):

li_dplyr = structure(
  list(
    df1 = structure(
      list(
        start_datetime = structure(
          c(16802.0833333333, 16802.0833333333, 16802.092025463, 16802.0922337963, 
            16802.1046643519, 16802.1048726852, 16802.1153819444, 16802.1155902778, 
            16802.1261574074), 
          format = structure(
            c("m/d/y", "h:m:s"), 
            .Names = c("dates", "times")
          ), 
          origin = structure(
            c(1, 1, 1970), 
            .Names = c(
              "month", "day", "year"
            )
          ), 
          class = c("chron", "dates", "times")
        )
      ), 
      class = "data.frame", 
      row.names = c(NA,  -9L), 
      .Names = c("start_datetime")
    ), 
    df2 = structure(
      list(
        start_datetime = structure(
          c(16809.0833333333, 16809.0833333333, 
            16809.084849537, 16809.0851041667, 16809.0921180556, 16809.0922569444, 
            16809.1040046296, 16809.1041435185, 16809.115787037), 
          format = structure(
            c("m/d/y", "h:m:s"), 
            .Names = c("dates", "times")
          ), 
          origin = structure(
            c(1, 1, 1970), 
            .Names = c(
              "month", "day", "year"
            )
          ), 
          class = c("chron", "dates", "times")
        )
      ), 
      class = "data.frame", 
      row.names = c(NA,  -9L), 
      .Names = c("start_datetime")
    )
  ), 
  .Names = c("df1", "df2")
)

# original column classes                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                      
lapply(li_dplyr, function(x) lapply(x, class))

# dplyr
lapply(dplyr::bind_rows(li_dplyr), class)

# data.table
lapply(data.table::rbindlist(li_dplyr), class)

[hadley]

Please provide a minimal reproducible example.

[tchakravarty]

@hadley Does the example included work for you?

[hadley]

It doesn't even fit all on one screen - please make it MINIMAL.

[tchakravarty]

@hadley Haha. Sure thing. Now painstakingly hand-formatted for your perusal.

[hadley]

It's still not exactly minimal - the shorter it is the more likely that I can quickly locate the bug...

[tchakravarty]

@hadley Updated to what I would say is minimal.

[hadley]

Here is a minimal reprex:

l <- list(data_frame(x = chron::times(1)), data_frame(x = chron::times(2)))
bind_rows(l)
#> Source: local data frame [2 x 1]
#> 
#>       x
#>   <dbl>
#> 1     1
#> 2     2

[tchakravarty]

@hadley Thanks for taking this on as an enhancement request. Just wondering what the issue is?

[gavinsimpson]

The problem seems more pervasive that just chron objects. For example the following coercion to "yearmon" "numeric" upon a bind_rows():

l <- list(data_frame(x = zoo::as.yearmon(2016 + (c(0:3) / 12))),
          data_frame(x = zoo::as.yearmon(2016 + (c(0:3) / 12))))
bind_rows(l)

# A tibble: 8 × 1
         x
     <dbl>
1 2016.000
2 2016.083
3 2016.167
4 2016.250
5 2016.000
6 2016.083
7 2016.167
8 2016.250

With the problem occurring whether the data frames are supplied as a list or as individual data frames.

It's quite an impediment to have to check/re-coerce variables types after binding.

[hadley]

Now part of #2432