Add comment about what expansion is for

Author: DavisVaughan

R/across.R

@@ -616,7 +616,22 @@ dplyr_quosures <- function(...) {
   quosures
 }
 
-# Always guaranteed to be 1 quosure in, 1 quosure out, unlike `expand_across()`
+# Expand an `if_any()` or `if_all()` call
+#
+# Always guaranteed to be 1 quosure in, 1 quosure out, unlike `expand_across()`.
+#
+# For the dplyr backend, the main reason we expand at all is to evaluate
+# tidyselection exactly once (rather than once per group), because tidyselection
+# is rather slow.
+#
+# At one point we believed `if_any()` and `if_all()` could be implemented as
+# "pure expansion" that would run before dispatching to other backends, like
+# dbplyr. In theory this could expand to a chain of `&` and `|` operations that
+# dbplyr would already know how to translate (so dbplyr itself would not have to
+# know how to implement `if_any()` and `if_all()`), but in practice we need more
+# error checking than what `x & y & z` gets us, so we actually expand to a
+# vctrs-backed implementation since the "pure expansion" ideas have never played
+# out.
 expand_if_across <- function(quo) {
   if (quo_is_call(quo, "if_any", ns = c("", "dplyr"))) {
     variant <- "any"