don't copy arguments, just talloc_steal() them

because copying could require lots of memory allocations.
So even if talloc_steal() is inefficient, it's likely not much
worse than allocating and copying entire data structures

diff --git a/src/lib/unlang/xlat_expr.c b/src/lib/unlang/xlat_expr.c
index f890e92ab16bb64aad1090498b5c3be12619b741..63dabfef49978e3ce4752717e05ab4c9d68ca555 100644 (file)
--- a/src/lib/unlang/xlat_expr.c
+++ b/src/lib/unlang/xlat_expr.c
@@ -2300,14 +2300,8 @@ static fr_slen_t tokenize_field(xlat_exp_head_t *head, xlat_exp_t **out, fr_sbuf
 
                MEM(arg = xlat_exp_alloc(head, XLAT_GROUP, vpt->name, strlen(vpt->name)));
 
-               /*
-                *      This is less efficient than just stealing the
-                *      xlat into the context of the new node...
-                *
-                *      But talloc_steal can be extremely inefficient O(N)
-                *      where N is chunk siblings.
-                */
-               xlat_copy(arg->group, arg->group, xlat);
+               talloc_steal(arg->group, xlat);
+               arg->group = xlat;
                arg->quote = quote;
                arg->flags = xlat->flags;