Should theme_void() include paper argument? ink argument not working with geoms

[EvaMaeRey]

Thank you for the excellent work in the theming arena! 🚀

Just opening an issue to note that having the 'paper' argument for theme_void() might be misleading.

It also seems that the ink declaration doesn't seem to propagate to the geoms.

library(ggplot2)
p <- ggplot(mtcars, 
            aes(x = factor(cyl), 
                fill = factor(cyl))) +
  geom_bar() + 
  geom_point(stat = StatCount, size = 7) + 
  geom_tile(aes(y = -2, fill = NULL))

p +
  theme_void(ink = "plum", paper = "cornsilk", base_size = 18)

last_plot() + 
  theme(geom = element_geom(ink = "plum"))

^{Created on 2024-12-18 with reprex v2.1.0}

[clauswilke]

Is the problem that theme_void() doesn't have a background element at all and is fully transparent? I think the first question is whether this theme should respond to the paper setting. My sense is it should. We could address this while keeping backwards compatibility (I hope) by adding a background element and giving it a fill color of full transparency.

[teunbrand]

I recommend the following for transparent paper, as in theme_transparent():

ggplot2/R/theme-defaults.R

Line 510 in efc53cc

ink = "black", paper = alpha(ink, 0)) {

It prevents some interpolation artefacts. @EvaMaeRey is there any chance you're interested in preparing a PR for this? 😁

[EvaMaeRey]

Yes. I'd only lean towards a theme that looks like theme_void (remove axes and gridlines), but is fully opaque, white paper as a default, called something like theme_blank(), theme_void2(), or theme_poster(). Reading p + theme_transparent(paper = "blue") still seems like it should still be transparent somehow.

[teunbrand]

If we adopt Claus' solution of setting a background element (i.e. plot.background), then an opaque version would be as easy as setting theme_void(paper = "white"). I don't think an opaque variant merits a separate theme in this case.

[EvaMaeRey]

I like Claus' proposal, setting plot.background element - and paper default of alpha(ink, 0). I guess the main worry is backward compatibility issues. Could we revisit introducing a new theme (and naming and default decision), if compatibility issues arise?

[teunbrand]

Why would backwards compatibility pose a problem do you think?

[EvaMaeRey]

I'm not aware of any problems this might cause, but maybe @clauswilke could elaborate if there are any specific things that came to mind when he wrote:

We could address this while keeping backwards compatibility (I hope) by adding a background element and giving it a fill color of full transparency.

[clauswilke]

It should work fine, but my concern is that somebody somewhere has written code that assumes theme_void() has no background element rather than a background with transparent color, and then something goes sideways. I would still vote for going ahead with the change though. I'm also planning to make it in all my themes in the cowplot package.