Is using vec_is() in check_valid_cols() too strict?

[DavisVaughan]

Immediately breaks bench. I faintly remember we mentioned bench using expressions where it didn't have to?

packageVersion("tibble")
#> [1] '2.99.99.9006'

bench::mark(x = 1)
#> All columns in a tibble must be 1d or 2d objects:
#> * Column `expression` is bench_expr

rlang::last_trace()
#>  36. └─bench::mark(x = 1)
#>  37.   ├─base::summary(...) /Users/davis/Desktop/r/packages/bench/R/mark.R:120:2
#>  38.   ├─bench:::bench_mark(tibble::as_tibble(results, validate = FALSE)) /Users/davis/Desktop/r/packages/bench/R/mark.R:120:2
#>  39.   │ └─base::unique(c("bench_mark", class(x))) /Users/davis/Desktop/r/packages/bench/R/mark.R:124:2
#>  40.   ├─tibble::as_tibble(results, validate = FALSE) /Users/davis/Desktop/r/packages/bench/R/mark.R:120:2
#>  41.   └─tibble:::as_tibble.list(results, validate = FALSE)
#>  42.     └─tibble:::lst_to_tibble(x, .rows, .name_repair, col_lengths(x))
#>  43.       └─tibble:::check_valid_cols(x)

^{Created on 2019-09-30 by the reprex package (v0.2.1)}

[krlmlr]

I took a stab at tweaking vctrs so that this eventually becomes a non-issue.

[krlmlr]

It's not that simple to fix in vctrs.

Options, in order of my preference:

1. Fix in vctrs
2. Cast to a list or a list_of when inserting into a tibble
3. Declare this as a breaking change

I think the second option wouldn't be too much of an effort to implement. We still can strive for the first option.

CC @lionel-.

[lionel-]

Option 2 would also be a breaking change, but if it doesn't break bench it'd be a reasonable workaround until we figure this out in vctrs.

[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.