[Bug c++/118763] [12/13/14/15 regression] memory leak involving early return from statement expressions since r12-6325

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118763

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org
            Summary|[12/13/14/15 regression]    |[12/13/14/15 regression]
                   |memory leak involving early |memory leak involving early
                   |return from statement       |return from statement
                   |expressions                 |expressions since r12-6325

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Started with r12-6325-gad64a821970ef5547d77ece22d302b0e599ae0d2
  D.2863 = operator new (4);
  D.2864 = 1;
  try
    {
      if (x != 0) goto <D.2873>; else goto <D.2874>;
      <D.2873>:
      D.2869 = 0B;
      // predicted unlikely by early return (on trees) predictor.
      return D.2869;
      <D.2874>:
      MEM[(int *)D.2863] = 1;
      D.2864 = 0;
      D.2869 = D.2863;
      return D.2869;
    }
  catch
    {
      if (D.2864 != 0) goto <D.2870>; else goto <D.2871>;
      <D.2870>:
      operator delete (D.2863, 4);
      goto <D.2872>;
      <D.2871>:
      <D.2872>:
    }
I think the problem is in using CLEANUP_EH_ONLY cleanup here which is turned
into try/catch, rather than one that is turned into try/finally.  Because that
one would DTRT here already, the D.2864 guard variable is already cleared if
new int succeeded.