I'm for adding position argument to geom_vline() and geom_hline(). I don't see any reason to prohibit tweaking on the position, even if not so many people need this.

Use after_scale mapping evaluation.

I think it doesn't work. Values mapped to a discrete scale is still discrete.

library(ggplot2)

mydata <-
  data.frame(
    x=factor(c("1", "2")),
    y=3:4
  )

ggplot(mydata, aes(x, y)) +
  geom_point() +
  geom_vline(
    aes(xintercept = stage(x, after_scale = xintercept - 0.5)),
    data = data.frame(x = "2"),
    colour = "red"
  )
#> Error in xintercept - 0.5: non-numeric argument to binary operator

I hadn't heard of after_scale() before this. Thanks for pointing it out. Also, the first way that I thought would work didn't. What is incorrect about using after_scale() directly where using stage() works?

With scale() and @yutannihilation's example, I was able to get the code working (see below), but if my method is the best way to do it, then it would not easily match the intent of the code. (Where did the 1.5 come from?) While the intent with position_nudge() from the original example is clear, to me. (I want this shifted to be a half-step below "2".)

Is there a better way to use after_scale than what I came up with below?

I agree with @yutannihilation, adding a position argument is a simpler solution, and it would be consistent with many other geom_*()s.

library(ggplot2)

mydata <-
  data.frame(
    x=factor(c("1", "2")),
    y=3:4
  )

ggplot(mydata, aes(x, y)) +
  geom_point() +
  geom_vline(
    aes(xintercept =  after_scale(1.5)),
    colour = "red"
  )
#> Error: geom_vline requires the following missing aesthetics: xintercept

ggplot(mydata, aes(x, y)) +
  geom_point() +
  geom_vline(
    aes(xintercept =  stage(x, after_scale = 1.5)),
    colour = "red"
  )

^{Created on 2020-12-06 by the reprex package (v0.3.0)}

the below works not sure what I am missing ?

library(ggplot2)
mydata <-
  data.frame(
    x=factor(c("1", "2","3")),
    y=3:5
  )
ggplot(mydata, aes(x, y)) +
  geom_point(size=3) +
  geom_vline(
    aes(xintercept =  as.numeric(x) - 0.5),alpha=0.2,size=5,
    colour = "red"
  )

^{Created on 2020-12-06 by the reprex package (v0.3.0)}

@smouksassi, I am surprised that works because it's providing a continuous value to a discrete scale. Thanks for pointing out that it does work.

Your example does rely on x being an explicit factor and the factor levels being the same while the original geom_vline() example was able to use a simpler character value. For my real use case, I have datasets that will have different factor levels, but there will always be one named "Baseline". I want to put the vertical line 0.5 points away from the location of "Baseline", so I wanted to automatically get the location of "Baseline", and nudge from there.

For my real use case, I have datasets that will have different factor levels, but there will always be one named "Baseline". I want to put the vertical line 0.5 points away from the location of "Baseline", so I wanted to automatically get the location of "Baseline", and nudge from there.

If your x axis is virtually continuous, it might be easier to use numeric values and add the corresponding labels.

What is incorrect about using after_scale() directly where using stage() works?

The after_scale mapping is evaluated after a variable has been mapped to a value by the scale. after_scale allows you to modify the mapped value, but the variable has to come from somewhere. Either mapped from the dataframe or calculated by the stat. Using stage allows you to map the variable in the dataframe (as geom_vline requires it), then modify it after the scale has handled it.

@smouksassi, I am surprised that works because it's providing a continuous value to a discrete scale.

At that point the type of scale does not matter, the categorisation between discrete scale and continuous scale is at a higher level of abstraction. Real (continuous) numbers get plotted.

I don't think of the axis as "virtually continuous". It is a discrete axis where I want to separate the times that are before the discrete value of "Baseline" from the times at and after "Baseline".

To me, it is discrete with two layers of discrete categories: The value shown on the x-axis and a grouping of values shown on the x-axis (before and at or after "Baseline"). The vline is to illustrate the second layer.

this has always worked numeric on top of discrete ( as mentioned above at the end it is 1 ,2 ,3 that are plotted) but the reverse is not true when you supply discrete on numeric scale this will error out

@smouksassi is right. It is okay and helpful to think of the axis. The discrete scale provides "macro organisation" for the groups e.g this boxplot/violin is centered at 1, that one is centered at 2 ..., then the boxes/violins are calculated in continuous form around those central values.

Ah, now I understand better what you were both meaning, thank you. When you were saying that it was "virtually continuous" you were referring to the way that ggplot handles it and not the way that my data is logically considered. Thanks!

Looping back to the original request, I do think that a position argument to geom_vline() and geom_hline() could be helpful, but I do have several other ways to get what I'm wanting if others do not agree.

I agree that a position argument could be helpful see below when faceting and with free_scales the above will not work.
I get that you expect the geom_vline positioning would operate like the geom point/ line positions ?

library(ggplot2)
mydata <-
  data.frame(
    x=factor(c("1", "2","3")),
    y=c(3,4,5)
  )
mydata$splits <- factor(mydata$x, labels=c("baseline","Cycle 1","Cycle 1"))

ggplot(mydata, aes(x, y)) +
  geom_point(size=3,position=position_nudge(x=-0.1)) +
  geom_vline(
    aes(xintercept = as.numeric(x)+0.1) ,
    alpha=0.2,size=5,
    colour = "red"
  )+
  facet_grid(~splits,switch = "x", scales = "free_x", space = "free_x")+
  theme(strip.placement = "outside")

^{Created on 2020-12-07 by the reprex package (v0.3.0)}