From ff5da1f3770872595369ff8de54d742e7a4fdad5 Mon Sep 17 00:00:00 2001
From: "Guillem L. Jara" <4lon3ly0@tutanota.com>
Date: Fri, 31 Oct 2025 16:13:11 +0100
Subject: [PATCH] perf(uucore): fix bottleneck on hot loop

Fixes #9099 by replacing a sleep() on a spin loop waiting on a process with a yield to the OS' scheduler.
---
 src/uucore/src/lib/features/process.rs | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

diff --git a/src/uucore/src/lib/features/process.rs b/src/uucore/src/lib/features/process.rs
index 55e8c36482b..a9fd0b2e780 100644
--- a/src/uucore/src/lib/features/process.rs
+++ b/src/uucore/src/lib/features/process.rs
@@ -141,10 +141,11 @@ impl ChildExt for Child {
                 break;
             }
 
-            // XXX: this is kinda gross, but it's cleaner than starting a thread just to wait
-            //      (which was the previous solution).  We might want to use a different duration
-            //      here as well
-            thread::sleep(Duration::from_millis(100));
+            // Yield back to the OS' scheduler; this is better than just arbitrarily
+            // waiting and the usually preferred solution, spinning with
+            // [`std::hint::spin_loop()`], because the operations here are all
+            // OS-related and orders of magnitude slower than single CPU operations.
+            thread::yield_now();
         }
 
         Ok(None)