Simply marking a field as @Autowired
in a GWT servlet does not work as intended. The code will compile and the web application will start up - which means Spring was successfully able to autowire the field, but when the servlet is actually hit by client-side code, it will yield a NullPointerException
- like there's a different, uninitialized copy of the servlet being hit.
I've found several ways on the web to get this working, one is by using a base servlet class that does some Spring logic but doing this means every GWT servlet must extend this base class. The other way was by using AspectJ and the @Configurable
Spring annotation. There was very little configuration involved here and it just magically worked.
My question is why doesn't just autowiring the field just work as intended? What is GWT doing that causes this to break.
The code will compile and the web application will start up - which
means Spring was successfully able to autowire the field
Not necessarily. The web container can instantiate a servlet without any help from Spring. Which you could be experiencing:
but when the servlet is actually hit by client-side code, it will
yield a NullPointerException - like there's a different, uninitialized
copy of the servlet being hit.
try overriding Servlet's init():
@Override
public void init(ServletConfig config) throws ServletException {
super.init(config);
WebApplicationContextUtils.getWebApplicationContext(config.getServletContext())
.getAutowireCapableBeanFactory().autowireBean(this);
}
When the RPC service is called from the client, the "server-side" looking at the called URL and the servlets mapping will find the class, will make the instance and it will serve the request. Meaning if you have @Autowired
annotation, or you already have an instance of the RPC class in the spring context, it does not matter. The new instance will be created and it won't "know" about Spring.
I resolve this by implementing a class which extends RemoteServiceServlet
and implements Controller
(from Spring MVC) and ServletContextAware
.
This way you can map every RPC service by URL using the Spring MVC approach, for ex:
<bean id="publicUrlMapping"
class="org.springframework.web.servlet.handler.SimpleUrlHandlerMapping">
<property name="mappings">
<props>
<prop key="/myFirstRpc">firstRpcServiceBeanRef</prop>
<prop key="/mySecondRpc">secondRpcServiceRef</prop>
</props>
</property>
</bean>
You also avoid the declarations for every single RPC servlet in web.xml
, the mappings are clean and you have the Spring injection.
You declare only a single mapping in web.xml
for org.springframework.web.servlet.DispatcherServlet
which will serve all RPC calls.
There are couple of examples on the web with explanation about GWT RPC and Spring MVC controller integration.
Hope this will help.
It turns out that when using Spring at least, there's a MUCH simpler way to do this such that you can use @Autowired and it doesn't involve massive configuration or base classes. The caveat is that you must also use AspectJ. Here's what you need for your GWT servlet:
@Configurable
public class MyGwtServiceImpl extends RemoteServiceServlet implements MyGwtService
{
@Autowired
private MyService service;
// ...
}
And in your Spring config make sure you also have:
<!-- enable autowiring and configuration of non-spring managed classes, requires AspectJ -->
<context:spring-configured/>
One final note. If you are also using Spring security with your GWT application (and in your GWT servlets), you will need to make sure you define the correct mode to ensure the AspectJ weaving is done correctly (i.e., you get both @Secured annotation processing AND the @Autowired processing) you will need:
<!-- turn on spring security for method annotations with @Secured(...) -->
<!-- the aspectj mode is required because we autowire spring services into GWT servlets and this
is also done via aspectj. a server 500 error will occur if this is changed or removed. -->
<security:global-method-security secured-annotations="enabled" mode="aspectj"/>