g++.abi/vtable2.C aliasing safe?

[Mark Mitchell]

>>>>> "Franz" == Franz Sirl <Franz.Sirl-kernel@lauterbach.com> writes:

    Franz> Hi,

    Franz> this testcase is failing on powerpc-linux-gnu with -O2. Out
    Franz> of curiosity I looked a bit closer today on it and noticed
    Franz> that the FAIL goes away with -fno-strict-aliasing. But I
    Franz> can't believe the C++ people put in a non-aliasing aware
    Franz> testcase, so do we have an aliasing bug somewhere still?

It's possible that I did.  Poking around in vtables the way these
tests do is definitely bad manners.  It may be that they should just
be compiled with -fno-strict-aliasing.  It's also possible that
there's an obscure aliasing bug.

Please file a report in GNATS, and make it high priority.

--
Mark Mitchell                   mark@codesourcery.com
CodeSourcery, LLC               http://www.codesourcery.com