diff --git a/odoo/models.py b/odoo/models.py
index 84e4b16b399..fba69554872 100644
--- a/odoo/models.py
+++ b/odoo/models.py
@@ -6056,7 +6056,9 @@ Fields:
                         # diff(), therefore evaluating line._prefetch_ids with an empty
                         # cache simply returns nothing, which discards the prefetching
                         # optimization!
-                        record[name]
+                        record._cache[name] = tuple(
+                            line_snapshot['<record>'].id for line_snapshot in self[name]
+                        )
                         for line_snapshot in self[name]:
                             line = line_snapshot['<record>']
                             line = line._origin or line
@@ -6080,11 +6082,13 @@ Fields:
         nametree = PrefixTree(self.browse(), field_onchange)
 
         if first_call:
-            values.update(self.default_get([
-                name
-                for name in nametree
-                if name not in values
-            ]))
+            names = [name for name in values if name != 'id']
+            missing_names = [name for name in nametree if name not in values]
+            defaults = self.default_get(missing_names)
+            for name in missing_names:
+                values[name] = defaults.get(name, False)
+                if name in defaults:
+                    names.append(name)
 
         # prefetch x2many lines: this speeds up the initial snapshot by avoiding
         # to compute fields on new records as much as possible, as that can be