tibble() behaves oddly with classed list columns

[moodymudskipper]

# I can't define this object
# https://stackoverflow.com/questions/54734771/sf-write-lat-long-from-geometry-into-separate-column-and-keep-id-column
fun <- function() {
  tibble::tibble(
  ID = c(1, 2, 4),
  geometry = list(
    c(117.2, 31.8) |>
      structure(class = c("XY", "POINT", "sfg")),
    c(116.4, 40.1) |>
      structure(class = c("XY", "POINT", "sfg")),
    c(117.9, 26) |>
      structure(class = c("XY", "POINT", "sfg"))
  ) |>
    structure(
      class = c("sfc_POINT", "sfc"),
      precision = 0,
      bbox = c(xmin = 116.4, ymin = 26, xmax = 117.9, ymax = 40.1) |>
        structure(class = "bbox"),
      crs = list(epsg = 4326L, proj4string = "+proj=longlat +datum=WGS84 +no_defs") |>
        structure(class = "crs"),
      n_empty = 0L
    ),
) |>
  structure(
    class = c("sf", "tbl_df", "tbl", "data.frame"),
    sf_column = "geometry",
    agr = factor(c(ID = NA_character_), levels = c("constant", "aggregate", "identity"))
  )
}

fun()
#> Error in `tibble::tibble()`:
#> ! All columns in a tibble must be vectors.
#> ✖ Column `geometry` is a `sfc_POINT/sfc` object.

# but it works if I load sf's namespace, despite it not changing the vector quality of the column
loadNamespace("sf")
#> <environment: namespace:sf>
fun()

#> Simple feature collection with 3 features and 1 field
#> Geometry type: POINT
#> Dimension:     XY
#> Bounding box:  xmin: 116.4 ymin: 26 xmax: 117.9 ymax: 40.1
#> old-style crs object detected; please recreate object with a recent sf::st_crs()
#> Geodetic CRS:  WGS 84
#> old-style crs object detected; please recreate object with a recent sf::st_crs()
#> # A tibble: 3 × 2
#>      ID     geometry
#> * <dbl>  <POINT [°]>
#> 1     1 (117.2 31.8)
#> 2     2 (116.4 40.1)
#> 3     4   (117.9 26)

# simpler example, tibble won't support list columns of any class
b <- structure(list(letters), class= "foo")

tibble::tibble(A = 1, B = b)
#> Error in `tibble::tibble()`:
#> ! All columns in a tibble must be vectors.
#> ✖ Column `B` is a `foo` object.

# That works if I use I() as in base R
tibble::tibble(A = 1, B = I(b))
#> # A tibble: 1 × 2
#>       A B         
#>   <dbl> <I<list>> 
#> 1     1 <chr [26]>

# I could define it with structure() but it won't print
df2 <- structure(list(A=1, B=b), class = c("tbl_df", "tbl", "data.frame"))
df2
#> Error in `[<-`:
#> ! Assigned data `map(.subset(x, unname), vectbl_set_names, NULL)` must
#>   be a vector.
#> Caused by error in `vec_size()`:
#> ! `x` must be a vector, not a <foo> object.

Is there a good reason to refuse classed list columns by default ? Is this something relating to subsetting or length methods ? My feeling is that {tibble} should use implementation wise subsetting and length (e.g. .subset2
rather than [[.foo), so that the object wouldn't depend on which packages are loaded, and the class shouldn't be a problem.