NewUndo: Fix (studio-reported) discrepency in proxies when undoing.

Took me an unreasonable amount of time to understand what was happening
here... Our beloved proxies, as usual, need some specific careful
handling.

1 files changed, 14 insertions, 0 deletions

diff --git a/source/blender/blenloader/intern/readfile.c b/source/blender/blenloader/intern/readfile.c
index 898f318f60b..799330a07bf 100644
--- a/source/blender/blenloader/intern/readfile.c
+++ b/source/blender/blenloader/intern/readfile.c
@@ -9556,6 +9556,20 @@ static void read_libblock_undo_restore_identical(
   /* Recalc flags, mostly these just remain as they are. */
   id_old->recalc |= direct_link_id_restore_recalc_exceptions(id_old);
   id_old->recalc_undo_accumulated = 0;
+
+  /* As usual, proxies require some special love...
+   * In `blo_clear_proxy_pointers_from_lib()` we clear all `proxy_from` pointers to local IDs, for
+   * undo. This is required since we do not re-read linked data in that case, so we also do not
+   * re-'lib_link' their pointers.
+   * Those `proxy_from` pointers are then re-defined properly when lib_linking the newly read local
+   * object. However, in case of re-used data 'as-is', we never lib_link it again, so we have to
+   * fix those backward pointers here. */
+  if (GS(id_old->name) == ID_OB) {
+    Object *ob = (Object *)id_old;
+    if (ob->proxy != NULL) {
+      ob->proxy->proxy_from = ob;
+    }
+  }
 }
 
 /* For undo, store changed datablock at old address. */