`geom_vline()` and friends cannot use `mapping` when `xintercept` is set

[arcresu]

I wanted to make geom_vline() adopt the theme's accent colour. Below was my sequence of attempts:

library(ggplot2)

ggplot() + geom_vline(xintercept = 0, colour = from_theme(accent))
#> Error:
#> ! object 'accent' not found
ggplot() + geom_vline(aes(colour = from_theme(accent)), xintercept = 0)
#> Warning: `geom_vline()`: Ignoring `mapping` because `xintercept` was provided.

ggplot() + geom_vline(aes(xintercept = 0, colour = from_theme(accent)))

^{Created on 2026-04-03 with reprex v2.1.1}

The first is obviously wrong because from_theme() only works inside aes(). The second seems like it should work, but the geoms are being too strict.

In the past I understand why it might have been discouraged to use mapping, but I think with the newer features like from_theme() the situation changed a bit and the limitation seems unnecessary.

Would it be worth making geom_vline(aes(colour = from_theme(accent)), xintercept = 0) work?

[EvaMaeRey]

I think there is a bit more complexity in this case as well as with geom_hline and geom_abline because they were created to act as an annotation layers when xintercept is provided, #6559.

I think that given use of unmapped xintercept triggers the annotation behavior, perhaps there shouldn't be the expectation to be able to do more than can be done with annotate(), where mapping isn't in play? e.g. something like the following?

get_accent <- function(){ theme_get()$geom$accent } 

ggplot() +
     annotate(geom = "point", x = 0, y = 0, color = get_accent())

But, only works if theme that you are using is set, not applied with +.

[arcresu]

The issue is that when any intercept/slope is provided, the provided mapping is discarded completely and replaced with one that only uses the internally constructed data frame. My proposal would be to still create that data frame but to update the existing mapping rather than replacing it. I can't see an inherent problem with that approach but might be overlooking some way it would create confusion. On the contrary it feels like it would improve consistency with other geoms where constant aesthetics can be passed as params instead of mappings.

I don't see any need to remove the functionality that already exists if that's what you're suggesting? The documentation examples make use of the mapping for these geoms.

[EvaMaeRey]

Below looks at a way ahead - I'd played with this a little yesterday...

But, I do feel like we could find ourselves thinking that geom_vline would be responsive to globally mapped aesthetics even when xintercept unmapped, but we'd be disappointed (the last plot). In general, a good thing, since the constructed one-row dataframe wouldn't have other globally referenced aesthetics like x and y for example.

library(ggplot2)

geom_vline <- function(mapping = NULL, 
                       data = NULL,
                       ...,
                       xintercept,
                       na.rm = FALSE,
                       show.legend = NA,
                       stat = "identity",  # new exposed argument
                       inherit.aes = NULL){   # I like this better than current implementation...         

  inherit.aes <- inherit.aes %||% missing(xintercept) # or similar to address nullness
  
  # Act like an annotation
  if (!missing(xintercept)) {
    # Warn if supplied mapping and/or data is going to be overwritten
    if (!is.null(mapping)) {
      cli::cli_warn("{.fn geom_vline}: Ignoring {.arg mapping} because {.arg xintercept} was provided.")
    }
    if (!is.null(data)) {
      cli::cli_warn("{.fn geom_vline}: Ignoring {.arg data} because {.arg xintercept} was provided.")

  # add a warning about stat being ignored if xintercept is provided ?
  
    }

    data <- ggplot2:::data_frame0(xintercept = xintercept)
    if(is.null(mapping)){
          mapping <- aes(xintercept = xintercept)} else{
          mapping <- modifyList(mapping, aes(xintercept = xintercept), keep.null = T)  # adjustment to combine mappings
    }
    show.legend <- FALSE
  }

  layer(
    data = data,
    mapping = mapping,
    stat = stat,                   # not fixed
    geom = GeomVline,
    position = PositionIdentity,
    show.legend = show.legend,
    inherit.aes = inherit.aes,       # This change or similar
    params = rlang::list2(
      na.rm = na.rm,
      ...
    )
  )
}

library(ggplot2)


ggplot() + 
  geom_vline(aes(colour = from_theme(accent)), xintercept = 0)
#> Warning: `geom_vline()`: Ignoring `mapping` because `xintercept` was provided.

ggplot(cars) + 
  geom_vline(aes(colour = from_theme(accent), 
                 linetype = I("dashed")), 
             xintercept = 0)
#> Warning: `geom_vline()`: Ignoring `mapping` because `xintercept` was provided.

ggplot(cars) +
  aes(colour = from_theme(accent), 
         linetype = I("dashed")) +
  geom_vline(xintercept = 0) 

^{Created on 2026-04-03 with reprex v2.1.1}

[arcresu]

Just for clarity since there are a few related threads, this is specifically about whether to support using both mapping and xintercept on these geoms. Currently if you use xintercept, the mapping is thrown away with a warning.

It seems like there's a bit of tension where on one hand these geoms are like annotations so should be separate from the plot data (not interacting with the mapping, dataset, or legend) but on the other hand want to be like normal geoms that affect the scale limits etc. Currently !missing(xintercept) signals that it should act like an annotation. I think that behaviour is a bit subtle since it results in e.g. what I tried in the description. The documentation does explain this though.

It might take some more experimentation to work out an end state that doesn't break anything. Just wanted to confirm first whether others agree that geom_vline(aes(colour = from_theme(accent)), xintercept = 0) "should" work.