Implement symdiff()

Fixes #4811

NAMESPACE

@@ -166,6 +166,8 @@ S3method(summarise_bullets,"dplyr:::summarise_incompatible_size")
 S3method(summarise_bullets,"dplyr:::summarise_mixed_null")
 S3method(summarise_bullets,"dplyr:::summarise_unsupported_type")
 S3method(summarise_bullets,default)
+S3method(symdiff,data.frame)
+S3method(symdiff,default)
 S3method(tally,data.frame)
 S3method(tbl,DBIConnection)
 S3method(tbl,src_local)
@@ -440,6 +442,7 @@ export(summarize_each)
 export(summarize_each_)
 export(summarize_if)
 export(sym)
+export(symdiff)
 export(syms)
 export(tally)
 export(tally_)

NEWS.md

@@ -1,8 +1,6 @@
 # dplyr (development version)
 
-* Passing `...` to `across()` is now deprecated because the evaluation timing of
-  `...` is ambiguous. Now instead of (e.g.) `across(a:b, mean, na.rm = TRUE)`
-  you should write `across(a:b, ~ mean(.x, na.rm = TRUE))` (#6073).
+* New `symdiff()` function computes the symmetric difference (#4811).
 
 * Rowwise-`mutate()` behaves a little better with 0-row inputs (#6303).
 

R/sets.r

@@ -7,41 +7,47 @@
 #' * `union(x, y)` finds all rows in either `x` or `y`, excluding duplicates.
 #' * `union_all(x, y)` finds all rows in either `x` or `y`, including duplicates.
 #' * `setdiff(x, y)` finds all rows in `x` that aren't in `y`.
+#' * `symdiff(x, y)` computes the symetric difference, i.e. all rows in
+#'    `x` that aren't in `y` and all rows in `y` that aren't in `x`.
 #' * `setequal(x, y)` returns `TRUE` if `x` and `y` contain the same rows
 #'   (ignoring order).
 #'
-#' Note that `intersect()`, `union()` and `setdiff()` remove duplicates
-#' in `x` and `y`.
+#' Note that `intersect()`, `union()`, `setdiff()` and `symdiff()` remove
+#' duplicates in `x` and `y`.
 #'
 #' # Base functions
 #' `intersect()`, `union()`, `setdiff()`, and `setequal()` override the base
 #' functions of the same name in order to make them generic. The existing
 #' behaviour for vectors is preserved by providing default methods that call
 #' the base functions.
 #'
-#' @param x,y Pair of data frames.
+#' @param x,y Pair of compatible data frames. A pair of data frames is
+#'   compatible if they have the same column names (possibly in different
+#'   orders) and compatible types.
 #' @inheritParams rlang::args_dots_empty
 #' @name setops
 #' @examples
-#' mtcars$model <- rownames(mtcars)
-#' first <- mtcars[1:20, ]
-#' second <- mtcars[10:32, ]
+#' df1 <- tibble(x = 1:3)
+#' df2 <- tibble(x = 3:5)
 #'
-#' intersect(first, second)
-#' union(first, second)
-#' setdiff(first, second)
-#' setdiff(second, first)
+#' intersect(df1, df2)
+#' union(df1, df2)
+#' union_all(df1, df2)
+#' setdiff(df1, df2)
+#' setdiff(df2, df1)
+#' symdiff(df1, df2)
 #'
-#' union_all(first, second)
-#' setequal(mtcars, mtcars[32:1, ])
+#' setequal(df1, df2)
+#' setequal(df1, df1[3:1, ])
 #'
-#' # Note the following 3 functions also remove pre-existing duplicates in `x` or `y`:
-#' a <- data.frame(x = c(1:3, 3, 3))
-#' b <- data.frame(x = c(3:5, 5))
+#' # Note that the following functions remove pre-existing duplicates:
+#' df1 <- tibble(x = c(1:3, 3, 3))
+#' df2 <- tibble(x = c(3:5, 5))
 #'
-#' intersect(a, b)
-#' union(a, b)
-#' setdiff(a, b)
+#' intersect(df1, df2)
+#' union(df1, df2)
+#' setdiff(df1, df2)
+#' symdiff(df1, df2)
 NULL
 
 #' @name setops
@@ -82,6 +88,17 @@ NULL
 #' @export setequal
 NULL
 
+#' @rdname setops
+#' @export
+symdiff <- function(x, y, ...) {
+  UseMethod("symdiff")
+}
+#' @export
+symdiff.default <- function (x, y, ...) {
+  check_dots_empty()
+  setdiff(union(x, y), intersect(x, y))
+}
+
 #' @export
 intersect.data.frame <- function(x, y, ...) {
   check_dots_empty()
@@ -134,6 +151,18 @@ setequal.data.frame <- function(x, y, ...) {
   all(vec_in(cast$x, cast$y)) && all(vec_in(cast$y, cast$x))
 }
 
+#' @export
+symdiff.data.frame <- function(x, y, ...) {
+  check_dots_empty()
+  check_compatible(x, y)
+
+  cast <- vec_cast_common(x = x, y = y)
+  only_x <- vec_slice(cast$x, !vec_in(cast$x, cast$y))
+  only_y <- vec_slice(cast$y, !vec_in(cast$y, cast$x))
+
+  out <- vec_unique(vec_rbind(only_x, only_y))
+  dplyr_reconstruct(out, x)
+}
 
 # Helpers -----------------------------------------------------------------
 

tests/testthat/_snaps/sets.md

@@ -28,6 +28,13 @@
       ! `...` must be empty.
       x Problematic argument:
       * z = 3
+    Code
+      symdiff(df1, df2, z = 3)
+    Condition
+      Error in `symdiff()`:
+      ! `...` must be empty.
+      x Problematic argument:
+      * z = 3
 
 # incompatible data frames error (#903)
 
@@ -55,6 +62,12 @@
       Error in `setdiff()`:
       ! `x` and `y` are not compatible.
       x Different number of columns: 1 vs 2.
+    Code
+      symdiff(df1, df2)
+    Condition
+      Error in `symdiff()`:
+      ! `x` and `y` are not compatible.
+      x Different number of columns: 1 vs 2.
 
 # is_compatible generates useful messages for different cases
 

tests/testthat/test-sets.R

@@ -6,6 +6,7 @@ test_that("x used as basis of output (#3839)", {
   expect_equal(union(df1, df2), tibble(x = 1:4, y = 1))
   expect_equal(union_all(df1, df2), tibble(x = c(1:4, 4, 2), y = 1))
   expect_equal(setdiff(df1, df2), tibble(x = c(1, 3), y = 1))
+  expect_equal(symdiff(df1, df2), tibble(x = c(1, 3), y = 1))
 })
 
 test_that("set operations (apart from union_all) remove duplicates", {
@@ -16,7 +17,9 @@ test_that("set operations (apart from union_all) remove duplicates", {
   expect_equal(union(df1, df2), tibble(x = c(1, 2)))
   expect_equal(union_all(df1, df2), tibble(x = c(1, 1, 2, 2)))
   expect_equal(setdiff(df1, df2), tibble(x = 1))
+  expect_equal(symdiff(df1, df2), tibble(x = 1))
 })
+
 test_that("standard coercion rules are used (#799)", {
   df1 <- tibble(x = 1:2, y = c(1, 1))
   df2 <- tibble(x = 1:2, y = 1:2)
@@ -25,6 +28,7 @@ test_that("standard coercion rules are used (#799)", {
   expect_equal(nrow(union(df1, df2)), 3)
   expect_equal(nrow(union_all(df1, df2)), 4)
   expect_equal(nrow(setdiff(df1, df2)), 1)
+  expect_equal(nrow(symdiff(df1, df2)), 2)
 })
 
 test_that("grouping metadata is reconstructed (#3587)", {
@@ -35,13 +39,17 @@ test_that("grouping metadata is reconstructed (#3587)", {
   expect_equal(group_vars(union(df1, df2)), "g")
   expect_equal(group_vars(union_all(df1, df2)), "g")
   expect_equal(group_vars(setdiff(df1, df2)), "g")
+  expect_equal(group_vars(symdiff(df1, df2)), "g")
 })
 
 test_that("also work with vectors", {
   expect_equal(intersect(1:3, 3:4), 3)
   expect_equal(union(1:3, 3:4), 1:4)
   expect_equal(union_all(1:3, 3:4), c(1:3, 3:4))
   expect_equal(setdiff(1:3, 3:4), 1:2)
+  expect_equal(symdiff(1:3, 3:4), c(1, 2, 4))
+  # removes duplicates
+  expect_equal(symdiff(c(1, 1, 2), c(2, 2, 3)), c(1, 3))
 })
 
 test_that("extra arguments in ... error (#5891)", {
@@ -53,6 +61,7 @@ test_that("extra arguments in ... error (#5891)", {
     union(df1, df2, z = 3)
     union_all(df1, df2, z = 3)
     setdiff(df1, df2, z = 3)
+    symdiff(df1, df2, z = 3)
   })
 })
 
@@ -65,6 +74,7 @@ test_that("incompatible data frames error (#903)", {
     union(df1, df2)
     union_all(df1, df2)
     setdiff(df1, df2)
+    symdiff(df1, df2)
   })
 })
 
@@ -118,3 +128,4 @@ test_that("setequal checks y is a data frame", {
 test_that("setequal checks for extra arguments", {
   expect_snapshot(setequal(mtcars, mtcars, z = 2), error = TRUE)
 })
+