[Bug tree-optimization/108416] False negative -Wdangling-pointer

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108416

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|False positive              |False negative
                   |-Wdangling-pointer          |-Wdangling-pointer
     Ever confirmed|0                           |1
                 CC|                            |msebor at gcc dot gnu.org
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2023-01-16

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed, this is a false negative.  Simpler testcase follows

class HoldsNonOwning {
 public:
  void reset(int* nonOwning)  { mNonOwning = nonOwning; }
  int compute(int input)
  {
    if (!mNonOwning)
      return -1;
    return input * *mNonOwning;
  }
  int* mNonOwning;
};

class HolderTest {
 public:
  HoldsNonOwning holder;
  int compute();
};

void foo(int *);

int __attribute__((noipa)) HolderTest::compute() {
  int nonOwning = 2;
  holder.reset(&nonOwning);
  int i = holder.compute(42);
  // If we uncomment this we don't get the warning
  // foo(&i);
  return i;
}

int main(int, char**)
{
  HolderTest ht;
  int i = ht.compute();
  foo (&i);
  return 0;
}


Uncommenting the call to 'foo' likely makes us believe the reference in
the global object might be pruned by it?