`Inf` does not maintain annotation within boundaries of plot anymore

[rempsyc]

When adding an annotation to a plot, Inf used to keep the annotation within the frame of the plot. Recently however I noticed that this behaviour has changed with a recent update (of, probably, ggplot2 or one of its dependencies), such that the text is now almost completely hidden outside the borders of the frame when using this historically common "trick". I noticed this change of behaviour in my pkdown website and personal documents (old vs new).

Reprex of the new behaviour:

packageVersion("ggplot2")
#> [1] '3.5.1'
library(ggplot2)
ggplot(mtcars, aes(wt, mpg)) +
  geom_point() +
    annotate("text", label = "text", size = 15, x = Inf, y = Inf)

^{Created on 2024-09-17 with reprex v2.1.1}

Same with I(1):

library(ggplot2)
ggplot(mtcars, aes(wt, mpg)) +
  geom_point() +
    annotate("text", label = "text", size = 15, x = I(1), y = I(1))

^{Created on 2024-09-17 with reprex v2.1.1}

vjust and hjust used not to be necessary for this, or putting them at zero would put them at the edge, but now 0 completly hides:

library(ggplot2)
ggplot(mtcars, aes(wt, mpg)) +
  geom_point() +
    annotate("text", label = "text", size = 15, x = Inf, y = Inf,
             vjust = 0, hjust = 0)

^{Created on 2024-09-17 with reprex v2.1.1}

Whereas the old behaviour can be recovered using vjust and hjust of 1:

library(ggplot2)
ggplot(mtcars, aes(wt, mpg)) +
  geom_point() +
    annotate("text", label = "text", size = 15, x = Inf, y = Inf,
             vjust = 1, hjust = 1)

^{Created on 2024-09-17 with reprex v2.1.1}

Or with "inward":

library(ggplot2)
ggplot(mtcars, aes(wt, mpg)) +
  geom_point() +
    annotate("text", label = "text", size = 15, x = Inf, y = Inf,
             vjust = "inward", hjust = "inward")

^{Created on 2024-09-17 with reprex v2.1.1}

Is this a bug or feature introduced by some new changes to ggplot2?

[teunbrand]

Thanks for the report!
I don't recall the behaviour of automatically adjusting for proximity to the panel boundaries.
AFAIK you would've always needed to use hjust/vjust to get the positioning right.
When I use 3.4.3, I get the same thing as in 3.5.1, so it can't have been that recent of a change:

library(ggplot2)
packageVersion("ggplot2")
#> [1] '3.4.3'

ggplot(mtcars, aes(wt, mpg)) +
  geom_point() +
  annotate("text", label = "text", size = 15, x = Inf, y = Inf)

^{Created on 2024-09-17 with reprex v2.1.1}

[rempsyc]

Thanks. While I have not changed my function rempsyc::nice_density in over a year, my previous rmarkdown analysis report shows this (as I remember my package website do as well in the examples):

But when I run the script again after a few weeks (and updating my packages), I now get:

That function relies on ggrepel, so... perhaps it was a change there? I tried testing on ggrepel first and got the same thing and so tried to see if I could nail it down further down the line to ggplot2, thus opening this issue. But perhaps ggrepel was adding automatic hjust and vjust and stopped doing that?

In any case, if that was always the behaviour, I think we can close the issue? Although I would have imagined that it would be sensible to have default values within the boundaries of the plot without having to make custom adjustments, but perhaps this is not easily generalizable to all the possible cases.

[smouksassi]

ggrepel geom_text_repel is not the same as geom_text indeed it used to put the inf into the panel a recent bug broke that but I think it was fixed in the last release:
slowkow/ggrepel#252

[teunbrand]

If this issue occurs in {ggrepel}, this is probably best fixed over there and we can close this issue here.
As a side-note, if you don't need to repel any points, it surprises me that you'd opt for a repelling text layer.