shared: implement c_list_sort() as non-recursive merge-sort

This is still the very same approach (in the way the array is split
and how elements are compared). The only difference is that the
recursive implementation is replaced by a non-recursive one.

It's (still) stable, top-down merge-sort.

The non-recursive implementation better, because it avoids the overhead
of the function call to recurse.

1 files changed, 39 insertions, 13 deletions

diff --git a/shared/nm-utils/c-list-util.c b/shared/nm-utils/c-list-util.c
index 509f4e12ab..44ca26a558 100644
--- a/shared/nm-utils/c-list-util.c
+++ b/shared/nm-utils/c-list-util.c
@@ -111,26 +111,52 @@ _c_list_srt_merge (CList *ls1,
 	return head.next;
 }
 
+typedef struct {
+	CList *ls1;
+	CList *ls2;
+	char ls1_sorted;
+} SortStack;
+
 static CList *
 _c_list_sort (CList *ls,
               CListSortCmp cmp,
               const void *user_data)
 {
-	CList *ls1, *ls2;
-
-	if (!ls->next)
-		return ls;
-	ls1 = ls;
-
-	ls2 = _c_list_srt_split (ls1);
-
-	ls1 = _c_list_sort (ls1, cmp, user_data);
-	if (!ls2)
-		return ls1;
+	/* reserve a huge stack-size. We need roughly log2(n) entries, hence this
+	 * is much more we will ever need. We don't guard for stack-overflow either. */
+	SortStack stack_arr[70];
+	SortStack *stack_head = stack_arr;
+
+	stack_arr[0].ls1 = ls;
+
+	/* A simple top-down, non-recursive, stable merge-sort.
+	 *
+	 * Maybe natural merge-sort would be better, to do better for
+	 * partially sorted lists. */
+_split:
+	stack_head[0].ls2 = _c_list_srt_split (stack_head[0].ls1);
+	if (stack_head[0].ls2) {
+		stack_head[0].ls1_sorted = 0;
+		stack_head[1].ls1 = stack_head[0].ls1;
+		stack_head++;
+		goto _split;
+	}
 
-	ls2 = _c_list_sort (ls2, cmp, user_data);
+_backtrack:
+	if (stack_head == stack_arr)
+		return stack_arr[0].ls1;
+
+	stack_head--;
+	if (!stack_head[0].ls1_sorted) {
+		stack_head[0].ls1 = stack_head[1].ls1;
+		stack_head[0].ls1_sorted = 1;
+		stack_head[1].ls1 = stack_head[0].ls2;
+		stack_head++;
+		goto _split;
+	}
 
-	return _c_list_srt_merge (ls1, ls2, cmp, user_data);
+	stack_head[0].ls1 = _c_list_srt_merge (stack_head[0].ls1, stack_head[1].ls1, cmp, user_data);
+	goto _backtrack;
 }
 
 /**