process: Fix drop_kills test when running on macOS with a single thread

tests/stdio.rs

@@ -85,6 +85,18 @@ fn feed_a_lot() {
     assert_eq!(status.code(), Some(0));
 }
 
+// FIXME: delete this test once we have a resolution for #51
+// This test's setup is flaky, and setting up a consistent test is nearly
+// impossible: right now we invoke `cat` and immediately kill it, expecting
+// that it didn't write anything, but if there's something wrong with the
+// command itself (e.g. redirection issues, it doesn't actually print anything
+// out, etc.) this test can falsely pass. Attempting a solution which writes
+// some data, *then* kill the child, write more data, and assert that only the
+// first write is echoed back seems like a good approach, however, due to the
+// ordering of context switches or how the kernel buffers data we can get
+// inconsistent results. We can keep this test around for now, but as soon as
+// we have a solution for #51, we may have a better avenue for testing this
+// functionality.
 #[test]
 fn drop_kills() {
     let mut child = cat().spawn_async().unwrap();
@@ -96,9 +108,12 @@ fn drop_kills() {
     let writer = write_all(stdin, b"1234").then(|_| Ok(()));
     let reader = read_to_end(stdout, Vec::new());
 
-    let future = writer.join(reader).map(|(_, (_, out))| out);
+    let (_, output) = support::CurrentThreadRuntime::new()
+        .expect("failed to get rt")
+        .spawn(writer)
+        .block_on(support::with_timeout(reader))
+        .expect("failed to get output");
 
-    let output = support::run_with_timeout(future).unwrap();
     assert_eq!(output.len(), 0);
 }