Fix segfault when using method reference operator without using result

Fixes [Bug #15985]

compile.c

@@ -8146,8 +8146,11 @@ iseq_compile_each0(rb_iseq_t *iseq, LINK_ANCHOR *const ret, const NODE *node, in
 	break;
       }
       case NODE_METHREF:
-        CHECK(COMPILE_(ret, "receiver", node->nd_recv, popped));
+        CHECK(COMPILE(ret, "receiver", node->nd_recv));
         ADD_ELEM(ret, &new_insn_body(iseq, line, BIN(methodref), 1, ID2SYM(node->nd_mid))->link);
+        if (popped) {
+            ADD_INSN(ret, line, pop);
+        }
         break;
       default:
 	UNKNOWN_NODE("iseq_compile_each", node, COMPILE_NG);

test/ruby/test_syntax.rb

@@ -93,6 +93,13 @@ class TestSyntax < Test::Unit::TestCase
     assert_valid_syntax("tap (proc do end)", __FILE__, bug9726)
   end
 
+  def test_methodref_literal
+    assert_separately [], <<-EOS
+      eval 'nil.:|;1'
+      1000.times{eval 'nil.:|;1'}
+    EOS
+  end
+
   def test_normal_argument
     assert_valid_syntax('def foo(x) end')
     assert_syntax_error('def foo(X) end', /constant/)