Describe why we don't have to guard against PL_dirty in vmg_mgwiz_free()

diff --git a/Magic.xs b/Magic.xs
index e336abe648bb8efc737b08bb2378bdaaa433ddac..9aabb411d9a158cd52f5ed0127f3fad00ab1226d 100644 (file)
--- a/Magic.xs
+++ b/Magic.xs
@@ -449,6 +449,15 @@ STATIC void vmg_mgwiz_free(pTHX_ MGWIZ *w) {
  if (!w)
   return;
 
+ /* We reach this point in dirty state when ptable_free() is called from the
+  * atexit cleanup callback, and that the global table still holds a live
+  * wizard. This happens before all the SV bodies are freed, so all the wizard
+  * callbacks are still alive (as they are referenced by the undead wizard).
+  * Hence it is safe to decrement their refcount. Later on, the wizard free
+  * callback itself will trigger when the wizard body is reaped, but it will
+  * be skipped as it guards against dirty state - which is good since nothing
+  * has to be done anymore at that point. */
+
  SvREFCNT_dec(w->cb_data);
  SvREFCNT_dec(w->cb_get);
  SvREFCNT_dec(w->cb_set);