[Bug c++/77553] [6/7 Regression] wrong code with post-increment operator in constexpr

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=77553

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2016-09-10
                 CC|                            |jakub at gcc dot gnu.org,
                   |                            |jason at gcc dot gnu.org
   Target Milestone|---                         |6.3
     Ever confirmed|0                           |1

--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
The bug is that for POINTER_PLUS_EXPR we do:
    case POINTER_PLUS_EXPR:
      r = cxx_eval_pointer_plus_expression (ctx, t, lval, non_constant_p,
                                            overflow_p);
      if (r)
        break;
      /* fall through */
...
      r = cxx_eval_binary_expression (ctx, t, lval,
                                      non_constant_p, overflow_p);
      break;
where both cxx_eval_pointer_plus_expression and cxx_eval_binary_expression
calls cxx_eval_constant_expression on both operands of the POINTER_PLUS_EXPR. 
That unfortunately means if the first one returns NULL_TREE, the side-effects
in the two subexpressions happen multiple times.
So, either we should remove cxx_eval_pointer_plus_expression and fold what it
does into cxx_eval_binary_expression, or cxx_eval_pointer_plus_expression
should copy what cxx_eval_binary_expression does, or add some helper function
which will be called only after both operands are processed through
cxx_eval_constant_expression.