stack-protector guard location

Richard Henderson rth@redhat.com
Fri Aug 28 13:34:00 GMT 2009


On 08/27/2009 07:05 PM, Roland McGrath wrote:
> 2. We well know that in practice "implementation namespace" belongs
>     to all the libraries and so forth that want to use it and rely
>     on their mutual maintenance cooperation to decide conflicts.

Yeah, well.  In this particular case there's not even
any conflict until we get into the debugger.  Even then
I'd hope that the debugger returns a user decl before
one that's marked DW_AT_artificial.

> Granted.  My initial thought was that we might add some sort of custom
> attribute to mark this.  But that's likely to be even further from Just Works.

Do you really think DW_AT_description (currently unused
in both gcc and gdb) is a good enough attribute mark?
It would be easy enough to move DW_AT_name to DW_AT_description
on all DW_AT_artificial decls.  I have no idea what kind
of fallout that would have.  I suspect lots, now that I
think of C++ and all the auto-generated stuff it does...

> 3. It's still a lie, and that just ain't right. :-)

Sure it has a name -- it just doesn't come when called.  :-)


r~



More information about the Gcc-patches mailing list