Hi Mark,
Yes, 6.2 is going to need changing :( The javadoc I think is clear, but,
I did not update 6.2 when we made the last change to the
ServerApplicationConfig class in the *Configuration refactoring. My bad
and thanks for catching it.
The reason it doesn't work to scan for ServerEndpointConfigs is that
they have to carry the basic endpoint path and class information. Up to
the last refactoring, the only way to put this in was to the information
in the no-arg constructor, meaning you had to make a subclass of
subclass ServerEndpointConfig just to put the path in. And if the
container were supposed to scan for them an instantiated them, who would
put in the path information ?
Here is the context that summarizes the issues better:
http://java.net/projects/websocket-spec/lists/jsr356-experts/archive/2013-02/message/39
The consequence of this is that a programmatic endpoint requires the
developer to provide an ServerEndpointConfig to deploy it. (The
annotated endpoints have all their config in them, so that's easy thank
goodness).
So explicitly, and referring to the PFD javadoc for
ServerApplicationConfig, the SCI needs to do two scans: one for subtypes
of Endpoint, and one for classes annotated @ServerEndpoint.
So ServerApplicationConfigs are going to look something like this:-
public class MyAppConfig implements ServerApplicationConfig {
public Set<ServerEndpointConfig> getEndpointConfigs(Set<Class<?
extends Endpoint>> endpointClasses) {
// for each endpointClass you want to deploy, create a
ServerEndpointConfig instance
// e.g. ServerEndpointConfig sec =
ServerEndpointConfig.Builder.create(MyEndpoint.class, "/foo").build();
return serverEndpointConfigInstances;
}
Set<Class<?>> getAnnotatedEndpointClasses(Set<Class<?>> scanned) {
// for each scanned annotated endpoint class, return it if you
want it deployed.
return annotatedEndpointClasses;
}
}
HTH,
- Danny
On 3/7/13 3:06 AM, Mark Thomas wrote:
> I think it would help if section 6.2 clarified exactly what the SCI is
> expected to scan for. My understanding is that the list is:
> - Implementations of ServerApplicationConfig
> - Classes annotated with ServerEndpoint
> - Implementations of ServerEndpointConfig
>
> As a follow up to this, would it make more sense to pass a set of
> ServerEndpointConfig instances to the ServerApplicationConfig instances?
>
> The reason for raising this is that I originally thought that the SCI
> needed to scan for Endpoint implementations but I quickly came to the
> conclusion that this was incorrect since there is no way to identify the
> path to use in order to create the necessary ServerEndpointConfig if no
> ServerApplicationConfigs are defined.
>
> I do wonder if I have understood this last bit correctly. An alternative
> option would be to scan for implementations of ServerEndpointConfig and
> Endpoint passing the endpoints to the ServerApplicationConfig instances
> if any are found and using the ServerEndpointConfigs if no
> ServerApplicationConfig are present but that seems a little to
> complicated not to be mention explicitly in the spec since it would be
> rather unexpected for users.
>
> Mark
--
<http://www.oracle.com> *Danny Coward *
Java EE
Oracle Corporation