[Bug tree-optimization/104475] [12 Regression] Wstringop-overflow + atomics incorrect warning on dynamic object

["amacleod at redhat dot com" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104475

--- Comment #6 from Andrew Macleod <amacleod at redhat dot com> ---
Created attachment 52564
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=52564&action=edit
possible patch

Perhaps better would be to not try to propagate NULL-ness.

    _1 = this_10(D)->d;
    _2 = &_1->m_mutex;
     if (_2 != 0B)
      goto <bb 5>; [90.00%]
    else
      goto <bb 3>; [10.00%]

2->5  (T) _1 :  struct QFutureInterfaceBasePrivate * [1B, +INF]
2->5  (T) _2 :  struct QMutex * [1B, +INF]
2->3  (F) _2 :  struct QMutex * [0B, 0B]

ON the edge 2->3 where _2 is 0, while it is also true that _1 must be 0,
perhaps that doesn't buy us much? and may lead is down the UB path of differing
expectations.     We certainly would like to preserve knowing non-null, so
instead of calling operator_addr_expr::fold_range, we could do the same
actions, just not deal with the NULL side, thus sidestepping issues that arise
with this whacky behaviour?

So if the _2 is non-null,  _1 is nonnull.  OTherwie we claim to know nothing of
_1..


The attached patch does just that, for what its owrth.