[Bug c++/56481] [4.8 Regression] endless loop compiling a C++ file

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56481

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2013-02-28
                 CC|                            |jakub at gcc dot gnu.org,
                   |                            |jason at gcc dot gnu.org
   Target Milestone|---                         |4.8.0
     Ever Confirmed|0                           |1

--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> 2013-02-28 11:59:57 UTC ---
Reduced testcase:
struct S
{
  bool foo () const;
#define A(n) , f##n##0, f##n##1, f##n##2, f##n##3
#define B(n) A(n##0) A(n##1) A(n##2) A(n##3)
#define C B(0) B(1) B(2) B(3)
  bool f C;
};

bool
S::foo () const
{
#undef A
#define A(n) && f##n##0 && f##n##1 && f##n##2 && f##n##3
  const bool ret = f C;
  return ret;
}


Started with my http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=192862
It actually isn't endless, just the compile time complexity is bad.

The problem is I think in:
        if (!potential_constant_expression_1 (op, rval, flags))
          return false;
        if (!processing_template_decl)
          op = maybe_constant_value (op);
in TRUTH_{AND,OR}*_EXPR handling in potential_constant_expression_1, because
maybe_constant_value calls potential_constant_expression (op) again.
They are called with different second/third argument (false, tf_none inside
maybe_constant_value, or true (== rval), flags above) though, so I think the
fix wouldn't be as easy as inlining maybe_constant_value by hand here and
avoiding the potential_constant_expression there.
Jason, any ideas?
For large && or || expressions perhaps we could just handle the cases where
op is the same code specially, still we wouldn't get rid of the large
complexity if say TRUTH_AND*_EXPR is only in every second code and there is
some other code in between.