subsetting an S3 vector fails with multiple arguments

[devinrkeane]

we recently created a custom S3 vector class to handle categorical data and are running into issues whenever this vector encounters base subsetting beyond a single dimension or argument.

Our class is similar to haven's labelled class and I provided an example where this gets us:

  # basic integer subsets fine with mutliple args
  y <- 1:2
  
  y[2, drop = FALSE]
  
# questionr contains this line of code, works fine here
  questionr::wtd.table(y, weights = c(1,1))
  
  # custom integer does not
  x <- haven::labelled(c(1:2), labels = c("apples" = 1L, "oranges" = 2L), label = "lab")
  
  x[2, drop = FALSE]
  
# questionr contains this line of code above and it's breaking things for us in a few areas
  questionr::wtd.table(x, weights = c(1,1))

Is this desired behavior and if so, should we be messaging to the user what the issue is or what we should be avoiding? This seems to be a flaw but i dont have a deep understanding of the vctrs theory, so any guidance here would be appreciated.

We're on 4.1.3, up to date with vctrs (0.6.3).

[DavisVaughan]

I think it is a bug on our end.

I think it is because

> vctrs:::`[.vctrs_vctr`
function (x, i, ...) 
{
    vec_index(x, i, ...)
}

doesn't have a named argument for drop in the signature so it ends up in the ... which gets passed on to vec_index()

> vctrs:::vec_index
function (x, i, ...) 
{
    i <- maybe_missing(i, TRUE)
    out <- vec_slice(x, i)
    if (!dots_n(...)) {
        return(out)
    }
    proxy <- vec_data(out)
    out <- proxy[, ..., drop = FALSE]
    vec_restore(out, x)
}

so we end up doing something like this:

> vctrs:::vec_index(1:2, 2, drop = FALSE)
Error in proxy[, ..., drop = FALSE] : incorrect number of dimensions

or, more simply:

> x <- 1:2
> x[, drop = FALSE, drop = FALSE]
Error in x[, drop = FALSE, drop = FALSE] : incorrect number of dimensions

[devinrkeane]

thanks! Makes sense, that's where i ended up too in my debugging,

 x <- haven::labelled(c(1:2), labels = c("apples" = 1L, "oranges" = 2L), label = "lab")
 vctrs:::vec_index(x, 2, drop=FALSE)

[hhhh5]

Is there a way to patch this on the user's side in the meantime? I used a combination of unlockBinding() and assign() to overwrite [.vctrs_vctr but there might be cleaner solutions.

[nalimilan]

FWIW, this also creates problem for me. In my lab we use a lot questionr's wtd.table with labelled arrays created by haven and it fails (juba/questionr#137).

[nalimilan]

@DavisVaughan Do you mean the fix is just that vec_index should take a drop argument, and ignore it (given that it doesn't make a difference for vectors)?