Re: PATCH: PR c++/39060: [4.4 regression] ICE with lots of invalid member functions

[Jason Merrill]

H.J. Lu wrote:
> On Sat, Mar 7, 2009 at 1:23 PM, Jason Merrill <jason@xxxxxxxxxx> wrote:
> > H.J. Lu wrote:
> > > This patch returns NULL if default_argument is error_mark_node.
> > So if parsing the default argument fails, we pretend the parameter itself
> > didn't exist?  Why does that work better than just setting default_argument
> > to NULL_TREE?

> I still got

> /tmp/x.ii:19: internal compiler error: canonical types differ for
> identical types void (A::)(void*) and void (A::)(void*)

Yes, I expected that it wouldn't work better, I was curious why it makes 
a difference.  It seems that the problem is coming in when we build the 
method type.

Jason