[PATCH] Fix PR c++/28266: ICE on invalid default variable

[Steven Bosscher]

On Sunday 10 September 2006 09:53, Simon Martin wrote:
> The attached patch fixes this problem the following way: if a default
> argument is not valid (i.e. is parsed into error_mark_node), then we act as
> if there was no default argument in the first place by setting
> TREE_PURPOSE(...) to NULL_TREE instead of error_mark_node.

Seems to me that you should just robustify the gimplifier.  These
error_mark_nodes mark errors, and the rest of the compiler should
just handle them.

Gr.
Steven