how to combine `quietly_mapped` and `safely_mapped` outputs? about collateral HOT 3 CLOSED

The way collateral currently works is that map_safely() and map_quietly() just use purrr::map(), but they add S3 classes (safely_mapped and quietly_mapped, respectively—not collat) to the list that comes back. Then S3 methods are registered to those classes to do the fancy output. S3 classes generally dispatch by checking an object's classes from first to last and dispatching to the method of the first registered class.

So what I'm thinking is that when you rbind two data frames where one has one class and the other has the other class, the list columns are probably concatenated successfully, but the classes just get thrown together and the method dispatch goes to whichever class ended up in front. I didn't really consider a scenario where someone might concatenate a safe column and a quiet column, and since the methods for printing just check whether components like error or warning are present, they don't crash if fed the wrong class—they just don't check for one or the other!

from collateral.

One misleading thing here is that the type shown at the top of the tibble column, collat, is not actually one of the classes of the list. That might make it falsely seem like a safe column and a quiet column could be concatenated. Anticipating your use case here, I think this is probably less of an issue if we fix #8 and provide a way to check for both errors and warnings at once 😄

from collateral.

I'm looking back into this, as some workflows I'm testing that involve using dplyr::bind_rows() on data frames that have collateral columns cause them to lose their formatting:

library(tidyverse)
library(collateral)

test =
  # tidy up and trim down for the example
  mtcars %>%
  rownames_to_column(var = "car") %>%
  as_tibble() %>%
  select(car, cyl, disp, wt) %>%
  # spike some rows in cyl == 4 to make them fail
  mutate(wt = dplyr::case_when(
    wt < 2 ~ -wt,
    TRUE ~ wt)) %>%
  # nest and do some operations quietly()
  nest(-cyl) %>%
  mutate(qlog = map_quietly(data, ~ log(.$wt)))

# now slice them up...
test1 = test %>% slice(1)
test2 = test %>% slice(2)
test3 = test %>% slice(3)

# ... and combine them. problem!
test_rejoined = bind_rows(test1, test2, test3)
#> Warning in bind_rows_(x, .id): Vectorizing 'quietly_mapped' elements may
#> not preserve their attributes

#> Warning in bind_rows_(x, .id): Vectorizing 'quietly_mapped' elements may
#> not preserve their attributes

#> Warning in bind_rows_(x, .id): Vectorizing 'quietly_mapped' elements may
#> not preserve their attributes
test_rejoined
#> # A tibble: 3 x 3
#>     cyl data              qlog      
#>   <dbl> <list>            <list>    
#> 1     6 <tibble [7 x 3]>  <list [4]>
#> 2     4 <tibble [11 x 3]> <list [4]>
#> 3     8 <tibble [14 x 3]> <list [4]>
class(test_rejoined$qlog)
#> [1] "list"

^{Created on 2019-07-09 by the reprex package (v0.3.0)}

This took me by surprise a little, as my understanding is that classes are supposed to be exempt from attributes getting wiped when a vector is modified. So I'm not 100% sure what's happening, but it may be that I need to handle collateral lists getting concatenated or subsetted a little more explicitly.

R/format.r has some extra S3 methods that I mostly added because I was trying to follow along with the pillar documentation, but I wonder if—for example—these two might deserve some more attention:

#' @rdname collateral_extras
#' @export
c.safely_mapped <- function(x, ...) {
  as_safely_mapped(NextMethod())
}

#' @rdname collateral_extras
#' @export
`[.safely_mapped` <- function(x, i) {
  as_safely_mapped(NextMethod())
}

from collateral.