Reconsider the deprecation of the groupOnX argument

[TengMCing]

When one tries to use the groupOnX argument to specify the grouping behaviour, the current version will give this warning:

The `groupOnX` argument of `geom_quasirandom()` is deprecated as of ggbeeswarm 0.7.1.
i ggplot2 now handles this case automatically.

I have tried many alternative methods to specify the grouping behaviour, but only one works, which is by using the coord_filp() function.

library(tidyverse)
library(ggbeeswarm)
ggplot(bind_rows(cars, cars, cars)) +
  geom_quasirandom(aes(x = dist, y = speed), width = 2) +
  coord_flip()

This loses the flexibility to control the grouping behaviour of a single layer, since coord_flip() will be applied to all layers.

In ggbeeswarm 0.6.0, I can write the following function without worrying about the content of the input plot. But in ggbeeswarm 0.7.1, I haven't found a way to do it.

add_a_quasirandom_layer <- function(p) {
    p <- p + geom_quasirandom(data = <some_data>, aes(<some_aes>), groupOnX = FALSE)
    p
}

I may be wrong about this if there is actually a way to control the per-layer grouping behaviour. However, if it is not possible, could you consider not deprecating the groupOnX argument?

[csdaw]

You're right that in v0.7.1 with the removal of groupOnX, its not possible to control on which axis the points are separated without
using coord_flip(). And I understand what you mean by per layer grouping.

However, the example you've given is a bit confusing. Can you provide a different example with v0.6.0 that demonstrates what you would like to do?

Normally, you'd want to jitter points along a categorical axis (regardless of whether the categories are on the x or y axis). But in the example you are comparing two continuous variables (in which case I wouldn't jitter but use alpha instead to preserve the point coordinates).

[eclarke]

Hi, after looking at some of the other functions in ggplot2, I've introduced an orientation parameter in v0.7.2 similar to what's used in geom_line and others. Now the main axis (e.g., the axis on which the data is grouped) gets inferred based on the data, but this can be overridden using orientation="x".

Furthermore, the groupOnX parameter remains soft-deprecated, but specifying it will have the same behavior as passing orientation="x". This means you shouldn't have to use coord_flip() now unless you want to for other reasons.

This plus other fixes will be on CRAN soon, or you can try them out by installing from GitHub.