Hi, PR 68948 shows a case which has been broken for a long time and very hard to see. So I am recommending that we introduce an assert inside the gimplifier if we process an error_mark_node and there was no errors or sorrys from the front-end; I will send a patch if there is a general feeling that it is a good idea.
Good idea or bad idea? This will prevent cases like PR 68948 from happening and make it easier to understand why the wrong code is happening. Thanks, Andrew
