When a filter aesthetic doesn't leave behind any data, some stats fail, e.g. adapted from the geom_edge_elbow() example:

irisDen <- hclust(dist(iris[1:4], method = 'euclidean'), method = 'ward.D2')

# this is basically the example from the docs, it runs fine:
ggraph(irisDen, 'dendrogram', circular = TRUE) + geom_edge_elbow()

# this errors:
ggraph(irisDen, 'dendrogram', circular = TRUE) + geom_edge_elbow(aes(filter = FALSE))
# Warning message:
# Computation failed in `stat_edge_elbow()`.
# Caused by error in `if (!data$circular[1] && n%%2 == 0) ...`:
# ! missing value where TRUE/FALSE needed

# this makes an empty plot with no errors:
ggraph(irisDen, 'dendrogram', circular = TRUE) +
  geom_edge_elbow(data = function(x) { data <- get_edges()(x); data[FALSE, names(data) != 'filter'] })

I'm guessing this is related to #324 and its fix (945bdf1).

With some print statement debugging, I found that StatEdgeElbow$compute_panel is called with an empty data frame when aes(filter=FALSE), but isn't called at all with the version that filters using data=, presumably because of the way it's invoked in Stat$compute_layer. That means filtering with data= is a workaround.

I imagine the fix would be either to return early in $compute_panel() when there is no data, or assume a default circularity value when there is no data. This would apply to many stats, not just StatEdgeElbow.