draw_quantiles param deprecated from geom_violin() with no replacement?

[marcuslehr]

Hi, I just updated ggplot, reran some code, and was greeted with a warning that the draw_quantiles parameter has been deprecated from geom_violin(). However, I'm a bit confused as the new parameters only specify the linetype and color. I'm happy to have new parameters controlling the quantiles, but this does not replace the ability to specify which quantiles should be drawn. Why regress the functionality? Doesn't make any sense to me. Looks like the .25, .5, and .75 quantiles are drawn.. which is a great default but you're really going to get rid of any control over this? What If I want to see .1 and .9? Or .05 and .95? Or just .5? I'd really appreciate if you didn't retire the draw_quantiles param. How about just updating it to default to c(.25,.5,.75)? Please thanks.

[teunbrand]

If you're using geom_violin(stat = "ydensity") (the default), you can use the quantiles argument which is directed at the stat, not the geom. So you still have the control you'd need when you want to draw the .1 and .9 quantiles, by using geom_violin(quantiles = c(0.1, 0.9)). The reasoning was communicated in the release blogpost here: https://www.tidyverse.org/blog/2025/09/ggplot2-4-0-0/#violin--quantiles.

[marcuslehr]

Thanks for the tip. I just tried it and it works but the lines are not drawn unless quantile.linetype is specified. This doesn't seem super intuitive to me. So how about the following suggestions:

• explicitly define the quantiles param in geom_violin() for clarity to benefit us less informed folk. It can be passed to the stat function just the same.
• If the quantiles param is passed without linetype or color specs, default them to 1/'black'
• Support passing T/F to quantiles would be nice, where T is converted to c(0.25,0.5,0.75) before being passed to the stat function

[teunbrand]

I get that the change in behaviour is confusing if you're used to the old way. In one way or another, all those suggestions blur the lines between stats and geoms. We like to keep these separate out of the principle that layers can be composed by freely combining stats and geoms. In some more detail:

1. We don't want a stat parameter in the geom constructor. If a different stat is chosen, we'd get in trouble with unused parameters.
2. We also don't want a stat parameter dictate values passed to geom parameters.
3. Assuming passing T/F will also inform geom parameters, this runs into the same problem as (2).

[marcuslehr]

Ugh okay okay. Final suggestion- replace the geom param draw_quantiles with quantiles.draw as a logical on/off switch. linetype could then be NULL by default (and I presume would inherit a value of 1 by default).

Also, it would be helpful to mention specifying quantile values with the quantiles param in the quantiles.* documentation description. Certainly would've prevented my post. And under examples the documentation still shows draw_quantiles.

[teunbrand]

I think having quantile.linetype = TRUE/FALSE should work, because FALSE should numerically evaluate to 0 (blank line) and TRUE should be 1 (solid line). There'd be no need for a separate argument like quantile.draw.
Adapting the documentation is a good suggestion.

[marcuslehr]

Thanks. I tested and passing F is fine but T returns "invalid line type".

Anyway, I suggest the separate param only for clarity. When looking at the parameters my immediate confusion in seeing only linetype and color was "okay but how do I turn them on and/or specify which ones to show". Perhaps this is being applied in other geoms and should be understood as standard ggplot operation. However, to my knowledge a similar thing was done recently in the boxplot geom. That is, a logical outliers param was added so users could specify outliers=F, where before one had to set the outlier shape or size to NA or alpha to 0. The latter options felt like a work around vs the direct switch which is nice and intuitive imo.

[teunbrand]

Thanks. I tested and passing F is fine but T returns "invalid line type".

Yes, and this should change to work correctly.

The geom_boxplot(outliers) parameter also avoids training scales with outliers, which is not exactly the same thing as setting outlier.shape = NA, which just hides the points.