Jason Merrill <jason@redhat.com> writes:
On Fri, 19 Dec 2003 14:51:53 -0800, Richard Henderson
<rth@redhat.com> wrote:
On Fri, Dec 19, 2003 at 12:49:50AM -0500, Daniel Berlin wrote:
That's 99.2% of the PTA queries returning no alias, where TBAA only
got
23.2% of it's queries resolved to no alias.
IIRC, tbaa is crippled on c++ due to some MI wierdness.
For some mysterious reason, the C++ frontend says that a class with
virtual
bases can alias anything. Dan, you seem to have been involved with
that
code; can you explain why?
I am the other Dan :-), but I can explain.
That patch as a consequence of the long thread that started at:
http://gcc.gnu.org/ml/gcc/2002-01/msg01478.html
Somewhere in that thread Mark Mitchell said that virtual base classes
are not safe for aliasing, so the patch was written that way. That is
the only reason.