Should facet + empty spec be a noop?

[lionel-]

Should these be equivalent to ggplot(mtcars) instead of failing?

ggplot(mtcars) + facet_wrap(vars())
#> Error: Can't subset with `[` using an object of class NULL.

ggplot(mtcars) + facet_wrap(NULL)
#> Error: Must specify at least one variable to facet by

Allowing empty specs would make sense in functions:

my_plot <- function(.data, .mapping, ...) {
  ggplot(.data, .mapping) +
    geom_point() +
    facet_wrap(vars(...))
}

Then both of these would work, instead of just the latter:

my_plot(starwars, aes(mass, height / 100))

my_plot(starwars, aes(mass, height / 100), gender)

[thomasp85]

I think the error messages should be better but I don’t think facet_wrap should change itself to facet_null if the input is lacking.

[lionel-]

Why is an error better? I think it makes sense not to error because NULL/empty facets should only occur in programmatic cases. Handling the empty case sensibly makes the layer more composable and generally useful.

[yutannihilation]

I think this makes sense, considering facet_grid() generates at least 1 row/col even when there are no variables for row/col.

[thomasp85]

But in facet_grid you ask for this explicitly by providing a . at either side of the formula...

I think my major gripe is that this can lead to situation where users mistype the name of the faceting variable and then get this single panel view without any errors or warnings... but @hadley should be the judge of that...

[lionel-]

You should normally get an "object not found" error in the case of an expression, or an "At least one layer must contain all faceting variables" error in the case of a bare symbol.

Related issue: #2986

[thomasp85]

Ah - this is purely for NULL...

[thomasp85]

I don't think

my_plot <- function(.data, .mapping, ...) {
  facet_vars <- vars(...)
  p <- ggplot(.data, .mapping) +
    geom_point()

  if (length(facet_vars) != 0) {
    p <- p + facet_wrap(facet_vars)
  }
  p
}

is too complicated, but of course accepting NULL will be more magical. I just tend to dislike objects changing class (FacetWrap -> FacetNull) based on the input data... ¯\(o_O)/¯

[lionel-]

I don't think it's changing class. An empty FacetWrap behaves as a FacetNull, which makes sense to me. Just like character(0) behaves as NULL in some contexts such as concatenation.

[thomasp85]

For all intend and purpose it will change class as it will behave just like FacetNull... I'd expect FacetWrap to create panels with strips etc., which will not happen as there is not anything to put in them...

I dunno... I'm against, but not vehemently, so if anyone else weighs in in favour I'll concede...

[yutannihilation]

But in facet_grid you ask for this explicitly by providing a . at either side of the formula...

You know, it's "for compatibility with the classic interface" as the doc says. Now we have rows and cols and their defaults are NULL.

[thomasp85]

fair point...

[lock]

This old issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with reprex) and link to this issue. https://reprex.tidyverse.org/