[PATCH, C++] Make Canonical ICE instead of just warn when mismatching

[Andrew_Pinski@PlayStation.Sony.Com]

Daniel Jacobowitz <drow@false.org> wrote on 01/04/2007 04:03:39 PM:

> On Thu, Jan 04, 2007 at 07:00:22PM -0500, Andrew Pinski wrote:
> > Except what Mike is talking about is not really an issue we should
> > be worrying about right now because again this is an internal check.
> > If GCC is broken, why try to continue?
> 
> You haven't thought enough about the reason the warning was added as a
> warning.  Please go read it again, and then think about it (and don't
> copy me about it again).  Thanks.

The only thing the warning says to me, is that GCC is broken.  And if an 
user
gets it, what can they do except complain or ignore it.  With an internal
compiler error, they are more willing to complain about that than a 
warning
being there.  If it was an internal error, we would have caught the issue 
with
tramp3d earlier.

The warning is useless to the common user of GCC really.  I don't see why
I am getting push back on a simple change that is actually obviously the 
correct
thing to do for an internal check.  Can I now then change all the internal 
checking
ICEs into warnings because that is what the real issue here?  To me, 
Canonical
types was not thought through enough if this was a warning and the 
reviewer of
the patch should have caught this issue when reviewing the patch in the 
first place.

There are multiple issues with Canonical types right now, since it causes 
memory usage
to go up around 1-2% (in C and other langauge code) which seems like not a 
good thing
when most of the problems are really just internal to the C++ front-end 
rather than 
in generic code.

Thanks,
Andrew Pinski