type_sum methods ignored for atomic vector based objects in a list column

[rundel]

In a recent change to pillar atomic vector based objects are no longer using type_sum methods when printing in a column, see the reprex below for an example:

---
author: rundel
date: 2021-05-17
output: "reprex::reprex\\_document"
title: nerdy-ram\_reprex.R
---

``` r
---
author: rundel
date: 2021-05-17
output: "reprex::reprex\\_document"
title: choky-barb\_reprex.R
---

``` r
library(pillar)

type_sum.list_class = function(x) {
  "list obj"
}

type_sum.atomic_class = function(x) {
  "atomic obj"
}

a = structure(list(1:10), class = "list_class")
b = structure(1:10, class = "atomic_class")
x = structure(
  list(a, b),
  class = c("container", "list")
)

tibble::tibble(x = x)
#> # A tibble: 2 x 1
#>   x              
#>   <containr>     
#> 1 <list obj>     
#> 2 <atmc_cls [10]>

Previously the behavior was that the 2nd row would have contained <atomic obj>.

[krlmlr]

Thanks. It works if we override vec_ptype_abbr(), which is called by the default method of type_sum() . I don't know yet when this was introduced.

What is your use case? Do you think we need to fix this here?

library(pillar)
library(vctrs)

vec_ptype_abbr.list_class <- function(x) {
  "list obj"
}

vec_ptype_abbr.atomic_class <- function(x) {
  "atomic obj"
}

a <- structure(list(1:10), class = "list_class")
b <- structure(1:10, class = "atomic_class")
x <- structure(
  list(a, b),
  class = c("container", "list")
)

tibble::tibble(x = x)
#> # A tibble: 2 x 1
#>   x                
#>   <containr>       
#> 1 <list obj>       
#> 2 <atomic obj [10]>

^{Created on 2021-05-21 by the reprex package (v2.0.0)}

[rundel]

I am using it in parsermd when displaying the ast in a tibble format (as a column). The overall ast has a custom s3 class + list and then each node has their own S3 classes, all of these have their own specialization of type_sum. One of the node types was just a character vector of text lines to keep things simple, this is the failing case for me currently.

This doesn't break anything critical for me, but it introduces an undesirable display glitch (and was causing snapshot tests to fail).

[krlmlr]

Introduced in cd3b4e7, pillar 1.6.1 .

[krlmlr]

Making it work with type_sum() again requires #250, which will take quite some time because it affects downstream CRAN packages. Please implement vec_ptype_abbr().

[github-actions]

This old thread has been automatically locked. If you think you have found something related to this, please open a new issue and link to this old issue if necessary.