Fix overlap identification in position_dodge2() (#5939)

* fix and vectorise `find_x_overlaps()`

* add news bullet

* add test

* deal with `NA`s more properly

* pass `end` correctly

Author: teunbrand

NEWS.md

@@ -125,6 +125,8 @@
 * `geom_hline()` and `geom_vline()` now have `position` argument
   (@yutannihilation, #4285).
 * New function `get_strip_labels()` to retrieve facet labels (@teunbrand, #4979)
+* Fixed bug in `position_dodge2()`'s identification of range overlaps 
+  (@teunbrand, #5938, #4327).
 
 # ggplot2 3.5.1
 

R/position-dodge2.R

@@ -132,14 +132,19 @@ pos_dodge2 <- function(df, width, n = NULL, padding = 0.1) {
 
 # Find groups of overlapping elements that need to be dodged from one another
 find_x_overlaps <- function(df) {
-  overlaps <- numeric(nrow(df))
-  overlaps[1] <- counter <- 1
 
-  for (i in seq_asc(2, nrow(df))) {
-    if (is.na(df$xmin[i]) || is.na(df$xmax[i - 1]) || df$xmin[i] >= df$xmax[i - 1]) {
-      counter <- counter + 1
-    }
-    overlaps[i] <- counter
-  }
-  overlaps
+  start   <- df$xmin
+  nonzero <- df$xmax != df$xmin
+  missing <- is.na(df$xmin) | is.na(df$xmax)
+  start   <- vec_fill_missing(start, "downup")
+  end     <- vec_fill_missing(df$xmax, "downup")
+
+  # For end we take largest end seen so far of previous observation
+  end <- cummax(c(end[1], end[-nrow(df)]))
+  # Start new group when 'start >= end' for non zero-width ranges
+  # For zero-width ranges, start must be strictly larger than end
+  overlaps <- cumsum(start > end | (start == end & nonzero))
+  # Missing ranges always get separate group
+  overlaps[missing] <- seq_len(sum(missing)) + max(overlaps, na.rm = TRUE)
+  match(overlaps, unique0(overlaps))
 }

tests/testthat/test-position-dodge2.R

@@ -118,3 +118,11 @@ test_that("groups are different when two blocks have externall touching point",{
   )
   expect_equal(find_x_overlaps(df1), seq_len(2))
 })
+
+test_that("overlaps are identified correctly", {
+  df <- data.frame(
+    xmin = c(1, 2, 3, 5),
+    xmax = c(4, 3, 4, 6)
+  )
+  expect_equal(find_x_overlaps(df), c(1, 1, 1, 2))
+})