Breakage due to dplyr 1.10.0

[bwiernik]

dplyr 1.10.0 made many changes under the hood, and this is breaking several different different areas of functionality with posterior:

library(dplyr)
#> 
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#> 
#>     filter, lag
#> The following objects are masked from 'package:base':
#> 
#>     intersect, setdiff, setequal, union
library(posterior)
#> This is posterior version 1.3.1.9000
#> 
#> Attaching package: 'posterior'
#> The following objects are masked from 'package:stats':
#> 
#>     mad, sd, var
#> The following objects are masked from 'package:base':
#> 
#>     %in%, match

n_param <- 4
dat <- tibble(
  a = rvar(array(rnorm(4000 * n_param, mean = rep(1:n_param, each = 4000), sd = 1), dim = c(4000, n_param)))
)

mutate(dat, b = first(a))
#> Error in `mutate()`:
#> ℹ In argument: `b = first(a)`.
#> Caused by error in `.subset2()`:
#> ! subscript out of bounds

#> Backtrace:
#>      ▆
#>   1. ├─dplyr::mutate(dat, b = first(a))
#>   2. ├─dplyr:::mutate.data.frame(dat, b = first(a))
#>   3. │ └─dplyr:::mutate_cols(.data, dplyr_quosures(...), by)
#>   4. │   ├─base::withCallingHandlers(...)
#>   5. │   └─dplyr:::mutate_col(dots[[i]], data, mask, new_columns)
#>   6. │     └─mask$eval_all_mutate(quo)
#>   7. │       └─dplyr (local) eval()
#>   8. ├─dplyr::first(a)
#>   9. │ └─dplyr::nth(x, 1L, order_by = order_by, default = default, na_rm = na_rm)
#>  10. │   └─dplyr:::vec_slice2(x, n)
#>  11. └─dplyr (local) `<fn>`(`<sbscOOBE>`)
#>  12.   └─rlang::abort(message, class = error_class, parent = parent, call = error_call)


n_param <- 4
n_col <- 3
dat <- tibble(
  a = rvar(array(rnorm(4000 * n_param * n_col, mean = rep(1:n_param, each = 4000), sd = 1), dim = c(4000, n_param, n_col)))
)

dat |> 
  rowwise() |> 
  mutate(b = a[,3])
#> Error in `mutate()`:
#> ℹ In argument: `b = a[, 3]`.
#> ℹ In row 1.
#> Caused by error in `a[, 3]`:
#> ! incorrect number of dimensions

#> Backtrace:
#>      ▆
#>   1. ├─dplyr::mutate(rowwise(dat), b = a[, 3])
#>   2. ├─dplyr:::mutate.data.frame(rowwise(dat), b = a[, 3])
#>   3. │ └─dplyr:::mutate_cols(.data, dplyr_quosures(...), by)
#>   4. │   ├─base::withCallingHandlers(...)
#>   5. │   └─dplyr:::mutate_col(dots[[i]], data, mask, new_columns)
#>   6. │     └─mask$eval_all_mutate(quo)
#>   7. │       └─dplyr (local) eval()
#>   8. └─base::.handleSimpleError(...)
#>   9.   └─dplyr (local) h(simpleError(msg, call))
#>  10.     └─rlang::abort(message, class = error_class, parent = parent, call = error_call)

^{Created on 2023-01-31 with reprex v2.0.2}

Both of these worked correctly with dplyr 1.9

[mjskay]

Thanks @bwiernik for spotting this. Fortunately I think this might be an easy fix (and one that should incidentally also close #267 ). Basically vctrs::vec_is_list() should be FALSE for rvars, and isn't, because we include "list" in the class definition. I had thought it was necessary, but removing it does not seem to break anything (might need to double check on that...).

@bwiernik if you have a chance, can you check and see if you have any other breakages if you install this branch: remotes::install_github("stan-dev/posterior@new_dplyr")? Thanks!

[bwiernik]

yep, all of my functions are working fine with that branch

[paul-buerkner]

Feel free to merge @mjskay

[mjskay]

Thanks @bwiernik and @paul-buerkner! I'll merge this later today or tomorrow once I do a few more checks.