[C++ PATCH] Fix ICE with NOP_EXPRs in check_case_bounds (PR c++/68362)

[Marek Polacek]

This patch ought to fix a regression introduced by the C++ delayed folding
merge.  What happens here for this testcase is that in c_add_case_label we
convert "0" to NOP_EXPR: "(const A) 0" in convert_and_check.  We pass this
to check_case_bounds which only expects INTEGER_CSTs.  Since we can't use
maybe_constant_value or fold_simple, I decided to try to use just fold (but
just STRIP_SIGN_NOPS should work as well).

Bootstrapped/regtested on x86_64-linux, ok for trunk?

2015-11-16  Marek Polacek  <pola...@redhat.com>

        PR c++/68362
        * c-common.c (check_case_bounds): Fold low and high cases.

        * g++.dg/delayedfold/switch-1.C: New test.

diff --git gcc/c-family/c-common.c gcc/c-family/c-common.c
index f8ccb6d..03c90f7 100644
--- gcc/c-family/c-common.c
+++ gcc/c-family/c-common.c
@@ -3769,6 +3769,10 @@ check_case_bounds (location_t loc, tree type, tree 
orig_type,
   min_value = TYPE_MIN_VALUE (orig_type);
   max_value = TYPE_MAX_VALUE (orig_type);
 
+  /* We'll really need integer constants here.  */
+  case_low = fold (case_low);
+  case_high = fold (case_high);
+
   /* Case label is less than minimum for type.  */
   if (tree_int_cst_compare (case_low, min_value) < 0
       && tree_int_cst_compare (case_high, min_value) < 0)
diff --git gcc/testsuite/g++.dg/delayedfold/switch-1.C 
gcc/testsuite/g++.dg/delayedfold/switch-1.C
index e69de29..302da23 100644
--- gcc/testsuite/g++.dg/delayedfold/switch-1.C
+++ gcc/testsuite/g++.dg/delayedfold/switch-1.C
@@ -0,0 +1,19 @@
+// PR c++/68362
+// { dg-do compile { target c++11 } }
+
+enum class A { foo };
+enum E { bar };
+
+void
+fn1 (const A a)
+{
+  switch (a)
+  case A::foo:;
+}
+
+void
+fn2 (E e)
+{
+  switch (e)
+  case bar:;
+}

        Marek