[jsr342-experts] Re: request scope for Web Sockets?

From: Pete Muir <pmuir_at_bleepbleep.org.uk>
Date: Wed, 22 May 2013 11:18:39 +0100

Regardless of what approach is decided on, as CDI spec lead I would prefer that this should be done in WebSockets, not CDI. This follows the precedent that we made with @TransactionScoped - that any additional contexts should be defined in the spec that "own" them, not in CDI. IMO this is a good thing, as the hard part of designing a context is understanding when it is active, and how it propagates, not understanding how to implement it. The experts in technology have a much better understanding of how it should propagate than the CDI EG does. As usual, of course, the CDI EG is happy to help any other group with the details and explain how contexts work.

On 16 May 2013, at 19:21, Bill Shannon <bill.shannon_at_oracle.com> wrote:

> Experts,
> An issue has come up about the definition of the CDI request scope and how
> it applies to Web Sockets applications. The issue is reported here:
> https://issues.jboss.org/browse/CDI-370
> We're trying to decide whether this is a simple oversight that could be
> corrected with an errata to the existing spec(s), or whether it's a missing
> requirement that would require a new revision of the spec(s). Since this
> involves the interaction of three specs, I'm starting the conversation here.
> Danny, Pete, Shing Wai, please forward this message to your expert groups
> for their input as well.
> Here's the definition of when a request scope is active and when it is destroyed:
>> The request scope is active:
>> - during the service() method of any servlet in the web
>> application, during the doFilter() method of any servlet filter and
>> when the container calls any ServletRequestListener or AsyncListener,
>> - during any Java EE web service invocation,
>> - during any remote method invocation of any EJB, during any
>> asynchronous method invocation of any EJB, during any call to an EJB
>> timeout method and during message delivery to any EJB message-driven
>> bean, and
>> - during any message delivery to a MessageListener for a JMS
>> topic or queue obtained from the Java EE component environment.
>> The request context is destroyed:
>> - at the end of the servlet request, after the service() method, all
>> doFilter() methods, and all requestDestroyed() and onComplete()
>> notifications return,
>> - after the web service invocation completes,
>> - after the EJB remote method invocation, asynchronous method invocation,
>> timeout or message delivery completes, or
>> - after the message delivery to the MessageListener completes.
> It would be easy to "fix" the first bullet in each list above by saying
> "oops, we forgot to include the work done by a protocol handler in
> Servlet 3.1". Since all this other work done by Servlet applications
> is part of the same request scope, adding the work done by protocol
> handlers would make sense.
> But, we have to decide if that's the fix we want.
> Adding bullet items to each list to cover specific Web Socket operations
> might be more what people are expecting, resulting in a request scope for
> Web Sockets that's "smaller" than the request scope for the corresponding
> http request. Even if we did that, we would still need to define clearly
> whether or not a request scope is active during any arbitrary protocol
> handler operation (not just Web Socket protocol handlers). Defining it
> for Web Sockets but not defining it for protocol handlers in general might
> be acceptable. Defining it one way for Web Sockets and a different way
> for other protocol handlers would not be acceptable.
> Should we fix this as an errata by saying that obviously protocol handler
> operations should've been included in those lists of Servlet operations?
> Or should we add items to each list to cover specifically Web Socket
> operations? (In which case what do we say about protocol handlers in
> general?) This would clearly require a new version of either the CDI
> spec or the Web Sockets spec.
> If we defined all Web Socket operations for a single http request to be
> part of the same request scope (the "errata" approach), we could later
> define a "message" scope or something similar to cover individual Web Socket
> operations.
> Let us know what you think.