Re: c++ failures

[Mark Mitchell <mark at codesourcery dot com>]

> > This patch is incorrect, and I can't find any mention of it in
> > gcc-patches -- which doesn't mean it wasn't there, just that I can't
> > find it.  Who approved this patch?
> 
> Moi?  Of course, I got approval.  I just forgot to commit the 
> cp/ChangeLog ;-).
> rth approved the patch.  For that matter, I believe you were CCed in 
> the discussion :).  Search for the thread entitled "disallow 
> initializing certain aggregates".

Well, that just shows how out of it I am. :-)  I seem to context-switch
about every ten minutes now, which means that if it ain't in front of my
nose, I lose it.  It's bad.  Sorry!

> No problem.  I am committing the patch below.

Thanks!!
 
> I will submit a follow up patch shortly taking this approach.
> 
> BTW, did my patch cause problems?  I can't see how.

Sometimes decl is NULL_TREE in that fragment.  Try:

  ~/dev/gcc-mainline/gcc/testsuite/g++.dg/other/error2.C 

for an example, at least on a hppa2.0w-hp-hpux11 target.

The other key point is that you want to check this at
template-instantiation time as well, and that doesn't go through the
parser -- but it does go through check_initializer.  So, the change will
make it more correct in that case.

When you get a patch that passes regression tests, send it to me, and I
promise to review it ASAP.

Thanks,

-- 
Mark Mitchell
CodeSourcery, LLC
mark at codesourcery dot com