Issue with new ggplot2 version

[thomasp85]

Hi

We see an issue with the latest ggplot2 version. The issue is related to this function

ggdistribute/R/geom_posterior.R

Lines 195 to 228 in 3efffa1

	get_posterior_data <- function(data, lower_cut=NULL, upper_cut=NULL, interp_thresh=NULL,
	mirror=FALSE, colour="#000000", fill=NA, warn=TRUE) {
	dt <- compute_post_seg_data(data, lower_cut, upper_cut, interp_thresh, warn)
	# find the bottom and top parts of the distribution to make a complete line
	assert_names(c("ymin", "ymax", "grp_max", "grp_min"), data)
	if (mirror) {
	dt[, `:=`(ylower=grp_max / 2 + grp_min - y / 2, yupper=grp_max / 2 + y / 2)]
	adj <- 0
	adj <- dt[
	, .(adj=mean(c(ymax, ymin)) - mean(c(grp_min, grp_max))),
	.(grp_min, grp_max, ymin, ymax)
	] %>%
	.[, adj]
	dt[, `:=`(ylower=ylower + adj, yupper=yupper + adj)]
	} else {
	dt[, `:=`(ylower=grp_min, yupper=y)]
	}
	rbind(
	copy(dt) %>%
	.[, y := ylower] %>%
	.[order(x)],
	copy(dt) %>%
	.[, y := yupper] %>%
	.[order(-x)]
	) %>%
	rm_temp_cols(c("colour", "fill")) %>%
	.[, `:=`(colour=colour, fill=fill)] %>%
	rbind(.[1, ])
	}

where data.table fails to properly rbind things together. The change in ggplot2 that provoked this is that position aesthetics that originated as discrete values now has an additional class added to them so that they are easily distinguishable from continuous data. It appears data.table does not coerce this class down to its numeric base, so you might have to do that manually.