Wrap GetProcessHeap and make it never inline

library/std/src/sys/pal/windows/alloc.rs

@@ -94,9 +94,10 @@ static HEAP: AtomicPtr<c_void> = AtomicPtr::new(ptr::null_mut());
 // a non-null handle returned by `GetProcessHeap`.
 #[inline]
 fn init_or_get_process_heap() -> c::HANDLE {
-    let heap = HEAP.load(Ordering::Relaxed);
-    if heap.is_null() {
-        // `HEAP` has not yet been successfully initialized
+    // The method is marked never inline to shrink the binary size. It won't be called many times.
+    #[cold]
+    #[inline(never)]
+    fn init_process_heap() -> c::HANDLE {
         let heap = unsafe { GetProcessHeap() };
         if !heap.is_null() {
             // SAFETY: No locking is needed because within the same process,
@@ -109,6 +110,11 @@ fn init_or_get_process_heap() -> c::HANDLE {
             // Could not get the current process heap.
             ptr::null_mut()
         }
+    }
+
+    let heap = HEAP.load(Ordering::Relaxed);
+    if heap.is_null() {
+        init_process_heap()
     } else {
         // SAFETY: `HEAP` contains a non-null handle returned by `GetProcessHeap`
         heap