Fix fallout from merge.

Regressions are the same as the last merge on June 27.  They are repeated here for the
record:

    Differences from trunk at merge point are:

    > FAIL: gcc.dg/tree-ssa/rvrp09.c scan-tree-dump-times rvrp "Branch rewritten" 4

            Expected.  Fails because I have disabled irange_adjust_bit_and_mask() to keep
            the verification code from tripping.

    > FAIL: gcc.dg/uninit-pred-6_c.c bogus warning (test for bogus messages, line 25)

            Expected.  Long-standing regression in our branch.  Jeff has mentioned he has
            a work-in-progress to fix this.

    < XFAIL: gcc.dg/pr80776-1.c  (test for bogus messages, line 22)
    < XFAIL: gcc.dg/Walloca-13.c  (test for bogus messages, line 11)
    < XFAIL: gcc.dg/Walloca-6.c (test for excess errors)

            Expected.  Ranger is smarter than mainline.

    And finally, the only non-expected regression:

    > FAIL: gfortran.dg/char_result_14.f90   -O3 -fomit-frame-pointer -funroll-loops -fpeel-loops -ftracer -finline-functions  (test for excess errors)
    > FAIL: gfortran.dg/char_result_14.f90   -O3 -g  (test for excess errors)

    This is some -Wprintf thing.  It may or may not be a bug.  I'll look at it:

            Warning: '__builtin_memset' writing between 1 and 2147483640 bytes into a region of size 0 overflows the destination [-Wstringop-overflow=]

From-SVN: r274622

gcc/tree-vrp.c

@@ -223,7 +223,6 @@ value_range_base::check ()
 	break;
       }
     case VR_UNDEFINED:
-      gcc_assert (!min () && !max ());
       break;
     case VR_VARYING:
       gcc_assert (m_min && m_max);

gcc/vr-values.c

@@ -187,7 +187,7 @@ vr_values::get_lattice_entry (const_tree var)
       return vr;
     }
 
-  vr->set_undefined ();
+  vr->set_undefined (TREE_TYPE (var));
 
   /* If VAR is a default definition of a parameter, the variable can
      take any value in VAR's type.  */