[PATCH] PR debug/20161

[Daniel Berlin]

On Fri, 2005-02-25 at 07:25 +0100, Steven Bosscher wrote:
> Hi,
> 
> Given this piece of invalid C,
> 
> struct XX;
> typedef void *(*F) (struct XX []);
> 
> we ICE when generating debug info, because drarf2out.c does not
> expect to find error_mark_nodes in the formal argument list. So
> I just make gen_formal_types_die a little more roboust.
> 
> The patch below was bootstrapped and tested on x86_64-suse.
> OK for mainline?

Wouldn't it be much easier and more robust to not call any of the
debug_hooks from toplev/final if error_count > 0, then it is to try to
add checks for error_mark_node all over the debug writers?