Fix iv_count for too_complex generic ivar

Previously this branch forgot to assign iv_count when used with a
generic ivar which is too_complex, which could result in it incorrectly
returning 0.

test/ruby/test_object.rb

@@ -970,6 +970,34 @@ class TestObject < Test::Unit::TestCase
     assert_not_include(s, "@password=")
   end
 
+  def test_inspect_too_complex
+    kernel_inspect = Kernel.instance_method(:inspect)
+
+    klasses = [
+      Class.new,
+      Class.new(String),
+      Class.new(Array),
+      Class.new(Hash),
+      Struct.new(:x),
+      Class.new(Thread::Mutex),
+      # It's very difficult to get a too_complex T_CLASS, to that isn't tested here
+    ]
+
+    klasses.each_with_index do |klass, idx|
+      8.times do |i|
+        klass.new.instance_variable_set(:"@sib_#{rand(999999)}", 1)
+      end
+
+      obj = klass.new
+      obj.instance_variable_set(:@a, 1)
+      obj.instance_variable_set(:@b, 2)
+
+      s = kernel_inspect.bind_call(obj)
+      assert_include(s, "@a=1")
+      assert_include(s, "@b=2")
+    end
+  end
+
   def test_singleton_methods
     assert_equal([], Object.new.singleton_methods)
     assert_equal([], Object.new.singleton_methods(false))

variable.c

@@ -2402,7 +2402,7 @@ rb_ivar_count(VALUE obj)
             VALUE fields_obj = rb_obj_fields_no_ractor_check(obj);
             if (fields_obj) {
                 if (rb_shape_obj_too_complex_p(fields_obj)) {
-                    rb_st_table_size(rb_imemo_fields_complex_tbl(fields_obj));
+                    iv_count = rb_st_table_size(rb_imemo_fields_complex_tbl(fields_obj));
                 }
                 else {
                     iv_count = RBASIC_FIELDS_COUNT(obj);