Move {tibble} to suggests #5990
Conversation
| @@ -533,7 +533,7 @@ is_facets <- function(x) { | |||
| # but that seems like a reasonable tradeoff. | |||
| eval_facets <- function(facets, data, possible_columns = NULL) { | |||
| vars <- compact(lapply(facets, eval_facet, data, possible_columns = possible_columns)) | |||
| data_frame0(tibble::as_tibble(vars)) | |||
| data_frame0(!!!vars) | |||
There was a problem hiding this comment.
We potentially lose some tibble validation checks here, but if we would want to validate here, I think it'd be better to control the validation ourselves.
There was a problem hiding this comment.
I (and probably we all) agree with you in that the validation ggplot2 wants isn't necessarily the same as the one provided by tibble (cf. #3018 (comment))! But, I feel a bit anxious that no validation will be applied here after we move tibble to suggests.
Defining a concrete spec of the validation is a tough job and takes time. So, it might be realistic to keep relying on tibble at the moment. I think we can assume an ordinary user uses tibble anyway via tidyverse.
There was a problem hiding this comment.
I feel a bit anxious that no validation will be applied here
Yeah I empathise with this, but let me try to convince you this change wouldn't make the situation any worse than it currently is.
tibble()'s validation is essentially vec_is(column) || is.expression(column) in tibble:::is_valid_col() combined with tibble:::make_valid_col(), which is just turning the expression into a list so that it complies with {vctrs}.
The validation in ggplot2:::data_frame0() also has the vec_is() constraint (via vctrs C internals I presume), but not the expression coersion. Seeing as we don't really accept expression columns as facet specificiation, this isn't a loss.
In addition to the similarity, the current validation isn't constraining the data types to the ones we can use as facet variables. Current validation happily accepts matrix or list columns which we then struggle to use downstream:
library(ggplot2)
df <- mpg
df$mtx <- matrix(1:4, nrow = 1, ncol = 4)
df$lst <- list(4)
p <- ggplot(df, aes(displ, hwy)) +
geom_point()
p + facet_wrap(~mtx)
#> Warning in `[<-.data.frame`(`*tmp*`, , value = list(mtx = structure(c(1L, :
#> replacement element 1 has 936 rows to replace 234 rows
#> Warning in `[<-.data.frame`(`*tmp*`, , value = list(PANEL = structure(1:4,
#> levels = c("1", : replacement element 4 has 16 rows to replace 4 rows
#> Warning in mat[index] <- unlist(unname(strips), recursive = FALSE)[[position]]:
#> number of items to replace is not a multiple of replacement length
p + facet_wrap(~lst)
#> Error in sort.int(x, na.last = na.last, decreasing = decreasing, ...): 'x' must be atomicCreated on 2024-07-10 with reprex v2.1.0
There was a problem hiding this comment.
Ah, thanks for the details. Then, I think I'm now convinced! The issue I referred to was before the vctrs integration, so it seems I was outdated...
I now feel this pull request is a go, but I'd like to wait for another approval to make sure.
This PR aims to fix part of #5986 (comment).
Briefly, code didn't use {tibble} a lot, so these instances have been replaced with
data_frame0().I still think {tibble} has a place in Suggests rather than removed entirely, because it has a nicer print method for data such as
mpgand it is used in vignettes. If we wouldn't feel strongly about these use-cases, we could also remove {tibble} altogether.