"stage()" does not work properly when prefixed with "ggplot2::"

[HMU-WH]

When I use a function from ggplot2 in my own R package, I need to prefix it with "ggplot2::". I found that, in this case, "stage()" does not work properly!

Here is the code to reproduce the bug:
This is the original example in ggplot2. I just changed "stage" to "ggplot2::stage".

ggplot(mpg, aes(class, displ)) +
  geom_violin() +
  stat_summary(
    aes(
      y = ggplot2::stage(displ, after_stat = 8),
      label = after_stat(paste(mean, "±", sd))
    ),
    geom = "text",
    fun.data = ~ round(data.frame(mean = mean(.x), sd = sd(.x)), 2)
  )

Throws the following error：

Error in `stat_summary()`:
! Problem while mapping stat to aesthetics.
ℹ Error occurred in the 2nd layer.
Caused by error:
! object 'displ' not found

[yutannihilation]

I need to prefix it with "ggplot2::"

While I agree ggplot2::stage() should work, I think you can import it to avoid the "Undefined global functions or variables" warnings.

#' @importFrom ggplot2 stage

[HMU-WH]

I need to prefix it with "ggplot2::"

While I agree ggplot2::stage() should work, I think you can import it to avoid the "Undefined global functions or variables" warnings.

#' @importFrom ggplot2 stage

I agree with your suggestion, but I still hope that ggplot2::stage() can work properly, because according to the R language specification, this should not cause an error. After all, this is not a symbolic function.

[teunbrand]

I think what is happening is that the stage() function is masked here to evaluate the after_stat component, but this masking doesn't work due to the ggplot2::-prefix, which uses the original unmasked stage() and thus evaluates the wrong component. I'm not sure yet what would be the best way to fix this.

[HMU-WH]

I think what is happening is that the stage() function is masked here to evaluate the after_stat component, but this masking doesn't work due to the ggplot2::-prefix, which uses the original unmasked stage() and thus evaluates the wrong component. I'm not sure yet what would be the best way to fix this.

Sorry, with my current ability, I am not capable of understanding the source code you pointed out in a short period of time.

However, it can be confirmed that there are indeed some flaws in the evaluation operation of the stage( ) function.

In addition to the issues I mentioned earlier, I have also noticed the following situations:

library(ggplot2)
library(patchwork)

by <- "cut"
p <- ggplot(diamonds, aes(x = cut, y = price))

p1 <- p + geom_violin(aes(colour = .data[[by]], fill = after_scale(color)), trim = FALSE)
p2 <- p + geom_violin(aes(colour = .data[[by]], fill = stage(!!sym(by), after_scale = color)), trim = FALSE)
p3 <- p + geom_violin(aes(colour = .data[[by]], fill = stage(.data[[by]], after_scale = color)), trim = FALSE)

p1 | p2 | p3

The legend of "P3" is inconsistent with "P1" and "P2". Of course, I understand that it can be corrected using p3 + labs(fill = by), but in my opinion, these three methods should produce the same image.

[teunbrand]

I have also noticed the following situations:

This particular situation should already be fixed in the current development version of ggplot2.

[yutannihilation]

To be comprehensive, I think we have three options about the level of fix.

1. Not fix. We can say, for example, after_scale() is a syntax rather than a usual function, so qualified form (ggplot2::) is not intended. cf. tidyselect didn't even export where() for a long time (Export where() r-lib/tidyselect#201).
2. Fix only ggplot2::.
3. Fix all cases including when stage() is re-exported.

I'd vote for 3 or 1. So, I like #6107 more because the approach of #6108 probably cannot handle the re-export cases.

[teunbrand]

Yeah you make a good point. So if we want this:

library(ggplot2)

ggplot(mpg, aes(displ, hwy)) +
  geom_point(
    aes(fill = stage(drv, after_scale = alpha(fill, 0.3))),
    shape = 21
  )

We cannot do.call() or otherwise construct stage(). Note that the points are not transparent.

ggplot(mpg, aes(displ, hwy)) +
  geom_point(
    aes(fill = do.call("stage", list(start = drv, after_scale = quote(alpha(fill, 0.5))))),
    shape = 21
  ) +
  labs(fill = "drv")

Nor can we use stage wrappers. Again, points are not transparent.

restage <- function(...) stage(...)

ggplot(mpg, aes(displ, hwy)) +
  geom_point(
    aes(fill = restage(drv, after_scale = alpha(fill, 0.3))),
    shape = 21
  ) +
  labs(fill = "drv")

^{Created on 2024-09-16 with reprex v2.1.1}

[thomasp85]

I personally holds the position that stage(), after_stat(), the old stat(), etc are all decorators, more than actual functions, that instructs aes() how to interpret the mapping.

I know this is a distinction we cannot necessary expect the users to know or understand, but I do think we can make a case for not allowing alternate call constructions

[teunbrand]

I don't think that is unreasonable. But if these are just syntax instead of functions, then I think the following should also work:

ggplot2::ggplot(ggplot2::mpg, ggplot2::aes(displ, hwy)) +
  ggplot2::geom_point(
    ggplot2::aes(fill = stage(drv, after_scale = scales::alpha(fill, 0.3))),
    shape = 21
  )
#> Error in `ggplot2::geom_point()`:
#> ! Problem while computing aesthetics.
#> ℹ Error occurred in the 1st layer.
#> Caused by error in `stage()`:
#> ! could not find function "stage"

^{Created on 2024-09-16 with reprex v2.1.1}

[thomasp85]

That I agree on

[yutannihilation]

Sorry for replying late. While I support @thomasp85's argument, I was thinking about the consistency with the rest of tidyverse. As dplyr allows using these special syntax with dplyr::, it might be a bit confusing if ggplot2 doesn't allow ggplot2::. (My personal preference is not to allow either dplyr:: or ggplot2::, but probably it's a bit too late).

data.frame(x = 1) |> 
  dplyr::select(dplyr::everything())
#>   x
#> 1 1

^{Created on 2024-09-23 with reprex v2.1.1}

[thomasp85]

@yutannihilation I agree. The solution is to have ggplot2:stage() and stage() work at all times

[yutannihilation]

Thanks for confirming. Then, should we keep this open (although it's not very high priority considering this is the first time we got the request since ggplot2 introduced stage())?

[teunbrand]

I'm fine with keeping this closed. The argument I made in #6104 (comment) was mostly academic and in practise I think it is fine to support ggplot2::stage() and stage() when ggplot2 is loaded, but not stage() when ggplot2 is not loaded. I'm comfortable with stage()/after_stat()/after_scale() blurring the lines between what are functions and what is syntax as long as they don't give any unexpected problems.

[yutannihilation]

Ah, sorry, I misunderstood what #6104 fixed (I thought it was in line with your argument in #6104 (comment)). Then, that should be mostly fine and I agree with keeping this closed, considering the original problem reported in this issue is definitely solved.

However, to be precise, "stage() when ggplot2 is not loaded" should work if ggplot2 aims to be in line with the rule of dplyr. It seems all the cases work.

d <- data.frame(x = 1)

# without attaching the namespace and without dplyr::
dplyr::select(d, everything())
#>   x
#> 1 1

# without attaching the namespace and with dplyr::
dplyr::select(d, dplyr::everything())
#>   x
#> 1 1

library(dplyr, warn.conflicts = FALSE)

# with attaching the namespace and without dplyr::
select(d, everything())
#>   x
#> 1 1

# with attaching the namespace and with dplyr::
select(d, dplyr::everything())
#>   x
#> 1 1

^{Created on 2024-09-23 with reprex v2.1.1}