Hello Matej

After testing adding a call to `bean.getMsg` in the catch block, the behaviour is unchanged. I did some further digging and here's what I found:

On a fresh start of the sever I ping the url with a nonsense cid.

ConversationBean will call conversation.begin() in the try block. This triggers a codepath that leads to LazyHttpConversationContextImpl.checkContextInitialized() line 124, that line will throw an exception. We go back out to ConversationBean where the exception is caught. Then when the catch block calls conversation.begin() it will once again reach LazyHttpConversationContextImpl.checkContextInitialized() but this time the if statement on line 121 returns true and so we never call initialize(). In both cases it is the same LazyHttpConversationContextImpl.

I also put a breakpoint in the observer and pinged the URL without manually specificing a cid. From inside the observer method I can see that LazyHttpConversationContextImpl.checkContextInitialized() line 128 is on the stack.

So to summarise. When I call the url with a nonsense cdi: The try block reaches checkContextInitialized and gets an exception on line 121. Then the catch block reaches checkContextInitialized and does nothing because isInitialized() returns true. Thus neither attempt reaches line 128 and the observer method is never fied.

This feels like a bug to me, not just because the observer isn't fired but also because if the initilization method had an exception half way though is it left in a good state? I don't know enough about these weld internals to check.

It occurs to me that one possible fix is to swap line 89 with line 90 so that the exception takes place before initilized is set to true. Of course that assumes that running initilized twice won't cause worse problems.

What do you think? Is this a bug?

From:        Matej Novotny <manovotn@redhat.com>
To:        Benjamin Confino <BENJAMIC@uk.ibm.com>
Cc:        Takayuki T Ishii <EBB0F3L@jp.ibm.com>, weld-dev@lists.jboss.org
Date:        28/01/2020 15:03
Subject:        [EXTERNAL] Re: [weld-dev] Question about conversations scope initilization        obeserver


I think I know what is the "problem" here.
Weld uses lazy conversation init - that means we don't activate context until you try and access a conversation scoped bean.

Now, in your example, the ConversationBean tries to begin() a conversation, then calls the bean (which initializes the context and notifies the observer).
However, in the situation where you try and pass in a non-existing conversation, the invocation to conversation.begin()[1] will blow up with NonExistingConversationException and
you will jump right into the catch block[2] where you begin a conversation with given ID, but you no longer invoke the bean, hence the context won't get activated.
Try adding the `bean.getMsg()` call to the catch block and see if that helps.

Note that CDI spec sets no requirements on how/when to activate the conversation context, so the lazy behaviour is compliant with spec (and this is also why you saw no such test in TCKs).



----- Original Message -----
> From: "Benjamin Confino" <BENJAMIC@uk.ibm.com>
> To: "Matej Novotny" <manovotn@redhat.com>
> Cc: "Takayuki T Ishii" <EBB0F3L@jp.ibm.com>, weld-dev@lists.jboss.org
> Sent: Tuesday, January 28, 2020 12:06:14 PM
> Subject: RE: [weld-dev] Question about conversations scope initilization                 obeserver
> Hello
> Thanks for the link. I had a look but I couldn't find any TCK tests
> checking to see if an observer method will catch the new
> ConversationContext being created for the "new transient conversation". To
> check if a new conversation was activated I created an entirely fresh
> server and ran the test application on it, the behaviour was the same, the
> first url I pinged on this new server ended with "cid=" and the observer
> didn't . Normally I've just been restarting the old server but restarting
> frequently.
> I've attached the recreate you requested. it consists of the three classes
> attached to my previous email as well as a minimal html page. To run it
> load it onto your server and ping
http://localhost:9080/ConversationContextTest/index.xhtml  to see the
> observer fire, and
http://localhost:9080/ConversationContextTest/index.xhtml?cid=99999  to see
> the observer fail to fire.
> Best regards
> Benjamin
> From:   Matej Novotny <manovotn@redhat.com>
> To:     Benjamin Confino <BENJAMIC@uk.ibm.com>
> Cc:     weld-dev@lists.jboss.org, Takayuki T Ishii <EBB0F3L@jp.ibm.com>
> Date:   27/01/2020 11:39
> Subject:        [EXTERNAL] Re: [weld-dev] Question about conversations
> scope initilization     obeserver
> Hello,
> I'd start by pointing you to CDI TCK as that's a good starting point to
> see what's covered.
> For your question, that would be this test -
> And possibly few more in the same test class.
> As for the linked classes - your `ConversationBean` is listening for
> @Initialized event. Can you verify that a new conversation was activated
> for your request instead of verifying that context was activated?
> E.g. check IDs or something along those lines? I suppose that will hold
> true and in that case it works just as spec requires it to.
> From the top of my head I don't really know how we activate/deactivate
> ConversationContext, I'd need to dig that up, but looking at CDI spec, it
> doesn't mandate that it is activated every time again and it could already
> be active for given request.
> Plus from just the classes you linked, I cannot know if you test this with
> no existing conversation or maybe with some long running one before you
> try to send a request for non-existing one...and so on.
> So if the above doesn't is not enough to answer your question, then we're
> going to need a complete reproducer so that we both talk about the same
> scenario :)
> Matej
> ----- Original Message -----
> > From: "Benjamin Confino" <BENJAMIC@uk.ibm.com>
> > To: weld-dev@lists.jboss.org
> > Cc: "Takayuki T Ishii" <EBB0F3L@jp.ibm.com>
> > Sent: Monday, January 27, 2020 11:42:14 AM
> > Subject: [weld-dev] Question about conversations scope initilization
> obeserver
> >
> > Hello
> >
> > I have a customer who's sent me a sample application, I have attached
> the
> > source to it below.
> >
> > When the customer visits index.xhtml they see the following output:
> >
> > Conversation initialized.
> > Conversation begun.
cid:1 , timeout:3600000
> > Conversation destroyed.
> >
> > However when they append "?cdi=" or a non-existnant identifier like
> > "?cdi=10000" to the url they do not see "Conversation initialized."
> >
> > The CDI spec says that: If the propagated conversation cannot be
> restored,
> > the container must associate the request with a new transient
> conversation
> > and throw an exception of type
> > javax.enterprise.context.NonexistentConversationException.
> >
> > I'm wondering if this should apply here? Or would it only apply if the
> cid
> > pointed to an existing conversation that could not be restored? And is
> there
> > anything in the spec that covers this specific situation?
> >
> > Unless stated otherwise above:
> > IBM United Kingdom Limited - Registered in England and Wales with number
> > 741598.
> > Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6
> 3AU
> >
> > _______________________________________________
> > weld-dev mailing list
> > weld-dev@lists.jboss.org
> >
> Unless stated otherwise above:
> IBM United Kingdom Limited - Registered in England and Wales with number
> 741598.
> Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU

Unless stated otherwise above:
IBM United Kingdom Limited - Registered in England and Wales with number 741598.
Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU