Have str_subset() drop NAs from string.

The "missings never match" example from ?str_subset broke in commit d681e45, which added name preservation.
The current commit keeps names while not matching NAs in string.
Also added a test for missings never to match.

> # stringr 1.5.2
> stringr::str_subset(c("a", NA, "b"), ".")
[1] "a" "b"

> # pak::pkg_install("tidyverse/stringr@d681e45")
> stringr::str_subset(c("a", NA, "b"), ".")
[1] "a" NA  "b"
> stringr::str_subset(c(First = "a", Second = NA, Third = "b"), ".")
First  <NA> Third
  "a"    NA   "b"

> # Current version
> stringr::str_subset(c(First = "a", Second = NA, Third = "b"), ".")
First Third
  "a"   "b"

Author: jonovik

R/subset.R

@@ -30,13 +30,18 @@ str_subset <- function(string, pattern, negate = FALSE) {
   check_lengths(string, pattern)
   check_bool(negate)
 
-  switch(type(pattern),
-         empty = no_empty(),
-         bound = no_boundary(),
-         fixed = string[str_detect(string, pattern, negate = negate)],
-         coll  = string[str_detect(string, pattern, negate = negate)],
-         regex = string[str_detect(string, pattern, negate = negate)])
+  if (type(pattern) == "empty") no_empty()
+  if (type(pattern) == "bound") no_boundary()
 
+  idx <- str_detect(string, pattern, negate = negate)
+  # str_detect() returns NA for NAs in string, but str_subset() should drop them
+  idx[is.na(idx)] <- FALSE
+  out <- string[idx]
+  # Work around the fact that as.character() drops names
+  nm <- names(out)
+  out <- as.character(out)
+  names(out) <- nm
+  out
 }
 
 #' Find matching indices

tests/testthat/test-subset.R

@@ -52,3 +52,8 @@ test_that("str_subset() preserves names of retained elements", {
   out <- str_subset(x, "[12]")
   expect_equal(names(out), c("B", "A"))
 })
+
+test_that("str_subset() never matches missing values", {
+  expect_equal(str_subset(c("a", NA, "b"), "."), c("a", "b"))
+  expect_identical(str_subset(NA, "."), character(0))
+})