[Bug c++/106389] [11/12/13 Regression] IPA modref breaks Safe Bitfields in C++

["rguenth at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106389

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |rguenth at gcc dot gnu.org
          Component|ipa                         |c++

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
There's a similar rule in C and we simply don't implement that - it's a "stupid
rule" given there's no good way to implement it.  Well.  -fno-strict-aliasing.

For GCC the allowed way is to access the common initial sequence through the
_union_ type.  Thus

union U { struct A { int i; } a; struct B { int i; float f; } b; };

union U *p;

p->a.i = 1;
.. = p->b.i;  // OK
B *b = &p->b;
... = b->i; // not OK

IPA modref simply makes this issue more visible (this across function
boundary).

Note the C rule doesn't involve unions but consider

A *a = &p->a;

  a->i = 1;
  ... = b->i;

I read the C++ rule as if that were allowed as well (*a could be inside a
union,
who knows - the union declaration might not even be visible in the TU!).

For the rule to take effect the frontend needs to mark each access that
falls under the rule as to have alias-set zero.