Eeks I wish these would have been seperate, this is going to be a long
response and not be as easily referenceable in the archives.
souravm wrote:
Hi Scott,
Thanks for the detailed answer/explanation. They were really helpful to verify
my understanding and also enriching the same.
My consolidated response to your last 2 mails are embedded below.
Regards,
Sourav
-----Original Message-----
From: Scott O'Bryan [mailto:[EMAIL PROTECTED]
Sent: Friday, April 18, 2008 12:27 PM
To: MyFaces Discussion
Subject: Re: Myfaces Portlet does not work when a bean is stored in
Requestscope ...
Souravm,
Just a clairification, the request bean you have, is it not getting
preserved between a single Action->Render or is it just not getting
preserved in subsequent renders?
<Sourav>
It does not get preserved in single Action->Render.
I'm not sure
- Whether this should be responsibility of the Portal server to preserve the
bean within the same request scope when the bean is declared to be of request
scope.
- Or it is responsibility of the bridge
Currently is it nobodies responsibility. I would certainly be
interested in enforcing consistency here at the bridge level. All I'm
saying is that in JSF, this isn't defined at all. In Portlet 1.0 it's
not defined either. So today, it works as it works.
If it is the responsibility of the bridge, then my take is the root cause of
this problem again goes back the issue#1 (replicating parameters/attributes
from ActionRequest to RenderRequest).
Your first issue and this one are two COMPLETLY different things..
Attributes are attributes and parameters are parameters. Why?
Request attributes in a portal env last though the current request while
request parameters last through the current request and subsequent
non-direct render requests.
The entire JSF lifecycle execution (except render) happens within
processAction() method which runs with the ActionRequest. So the bean creation,
execution of bean's methods (which in turn populate the result to be displayed
in the resultant response page created in render phase) also happen within this
scope. So if the bean in its latest state needs to be stored and used in the
render phase the bean has to be stored either in session (which works fine in
case of session scoped bean) or it has to be explicitly set in RenderRequest.
This is totally incorrect actually. First off, there is nothing in JSF
which says the Lifecycle.execute has to happen during an action
request. Quite the contrary it CAN'T. Portals, according to Portlet
1.0 spec make an initial call to a portlet through a render request.
This means that, at the very least, the initial call into the execute
must be a render request. When you start adding usecases for Portlet
2.0, you cannot tie specific pieces of a lifecycle to specific lifecycle
phases. That said, I don't disagree that Request Attributes should be
preserved. That's how it was spec'd in JSR-301 because pretty much
everyone agrees with you. Pre-JSR-301 beidges did not address this
usecase though. It was not a requirement of JSF and the spec simply
says that the maps reflect what is currently stored on the request.
As such, if you take an attribute, store it on the native Request
object, and then in the render try to get it, you'll find your portal is
not preseving that attribute.
</Sourav>
The first issue, in bridges before JSR-301 is actually a portal issue.
The JSR specification does not say whether request attributes set in the
action request have to be available in the render request. IMO, if they
are not, request attributes are basically pointless. Pre-JSR 301
bridges were ignorant of this fact and just did what the portal did.
The JSR-301 bridge DOES define this behavior and I believe he have
special code to handle these issues. This code is NOT in the MyFaces
1.1 bridge.
<Sourav>
I see your point.
However, going back to the comment you made in last mail (whether this is a
valid usecase or not, or should this scenario has to be handled through Render
URL), I don't think using a RenderURL is a right solution. This is because
following reasons -
a) RenderURLs are to be directly called only when there is no processing needs
to be done for a Portlet, only the previous view has to be rendered. In my
understanding, this is to be used especially for the pages with multiple
portlets. This ensures that in case one Portlet sends an ActionRequest, all
other portlets in the same page does not need to go through the action
processing for the previous request (instead they can just repeat the render
phase of Portlet Lifecycle with the result from previous action).
You are partially correct. ProcessAction is designed to be used in
response to expensive processing operations which are usually caused by
form submissions. Portal developers realized that a person will only
ever interact with one portlet at a time and that, when a person does
interact with a portlet, they have access to things (like the request
input stream), that other portlets do not.
Where you are wrong is that this HAS to be the case. Indeed during the
initial render of a portlet (which is always a render request) this is
NOT the case, because some processing has to be done. The correct way
to think about it is that you should do as little in your render request
as you can, but no less.
So why do I think the Render URL is appropriate here? Let's say you had
a normal (non-JBOSS) search portlet. In order to execute it, you would
need an initial screen (which could absolutely do some processing). If
this initial screen was a JSF application, JSF would handle all the
binding and assignment to the backing beans and everything would work.
b) Secondly, not sure how valid is the assumption that the first request to a
Portlet will always be Render Request. Even during first time bringing up the
portlet in a page there may be need of doing some processing based on the
Portlet Preference which ideally should be handled in processAction() phase of
Portlet lifecycle. So ideally this assumption should be relooked at.\
Again, according to the Portlet 1.0 specification, this CANNOT happen.
The initial request in a portlet is ALWAYS a render request. It's
spec'd that way. Apparently JBoss has added some extensions to change
that, but it does not fit with JSR-168.
I surely feel this usecase should be supported (standard struts-portlet bridges
support it). I'll really appreciate if you can discuss this in next JSR 301
meeting.
I will, I'll get it added to the agenda..
</Sourav>
As for the second issue, this is also something that is now handled by
JSR-301, but the original attempt at JSF to define a bridge did NOT make
this a requirement. In order to maintain compatibility with existing
applications, the 301 bridge will preserve request attributes on
subsequent "non-direct" render requests, but we also had to add a way to
disable this functionality for beans that did not expect to be preserved.
<Sourav>
I've not really tested preserving the request for subsequent non-direct render
request. As I mentioned above, I found problem even in storing the same bean
within the single Action->Render sequence.
However, my view is, if request parameters (in a managed bean) needs to be
stored for subsequent render requests, it crosses the boundary of a single http
request. Then the managed bean has to be scoped at session level.
</Sourav>
Yeah, I know. This went back and forth as well. However, with JSF this
doesn't make sense. Let's say you have 2 JSF portlets. Portlet #1 has
a search box. You type in a value into the search box and JSF stores
the value into a request scoped bean and displays the results. You then
interact with another portlet. When your page refreshes, the item you
were searching for is no longer there. We've gone though quite a few
iterations on this and the most efficient use on this is for the request
attributes to follow the same lifecycle as the render parameters unless
they are excluded. The problem with storing everything on the session
is that it never goes away and this will eat up tons of memory. If your
application explicitly handled this storing and removing of objects,
that's one thing. But JSF does not allow you to easily remove a managed
bean from a scope.
For issue #1, I think it would probably be appropriate to add some code
to fix this. What it would entail is storing the RequestMap in a global
map with a key that you would set as a render parameter. You'll need to
be careful to clean up anything that might "leak".
<Sourav>
I agree with you on this. I'm planning to create this map in actionProcess()
method in case the VIEW_ID request parameter is null (the VIEW_ID null is the
flag to identify that it is a non-JSF action request).
</Sourav>
For issue #2, existing portlet applications in the 1.1 space DEPEND on
this behavior. Changing it would break those applications. We chose to
break it for JSR-301 because we though it more appropriate to preserve
these parameters, but we added several mechanisms (one annotation based
and one FacesConfig based) to allow these attributes to be easily excluded.
<Sourav>
I see your point. Hope JSR 301 and JSR 286 together can bring more predictable
and intuitive behavior for Portal-JSF combination.
</Sourav>
Well it's shaping up to be interesting. More predictable, I doubt it.
What 286 will do is add a bunch of functionality, like the ability to
support AJAX in a standardized fashion.
Is there any reason you can't move to JSF 1.2? I would be very
interested in your opinions of the JSR-301 bridge which should run on
Portlet 1.0 and JSF1.2 just fine. The spec's are not yet final and so
there is still time to influence some of the usecases or, at the very
least, get your head around what will be the Java standard soon.
In the mean time, I'll ask the EG if we need to support an initial
request being an action request. I know we've got some JBOSS guys on
the Expert Group so we may be able to get them to comment. For now
though, try generating a render url and I think you'll find that the
bridge will let it work.
Hope that helps,
Scott
souravm wrote:
Hi All,
I have a simple JSF application exposed as Portlet (in JBoss Portal Server
2.6.4) using MyFacesGenericPortlet. The JSF application has a managed bean with
Request scope.
The application works perfectly when it is run outside Portal environment.
But within Portal environment it does not work as the Manages Bean, though gets
initiated and do all the processing properly during the initial lifecycle
phases, during the render phase it further gets initiated and the previous
instance gets lost.
The same works perfectly fine in Portal environment when the Managed Bean is
declared in session scope.
Not sure whether this is the problem of MyFacesGenericPortlet or the Portal
Server where it is running. Or is it by design ?
Any insight/viewpoint on this would be highly appreciated.
Regards,
Sourav
**************** CAUTION - Disclaimer *****************
This e-mail contains PRIVILEGED AND CONFIDENTIAL INFORMATION intended solely
for the use of the addressee(s). If you are not the intended recipient, please
notify the sender by e-mail and delete the original message. Further, you are
not to copy, disclose, or distribute this e-mail or its contents to any other
person and any such actions are unlawful. This e-mail may contain viruses.
Infosys has taken every reasonable precaution to minimize this risk, but is not
liable for any damage you may sustain as a result of any virus in this e-mail.
You should carry out your own virus checks before opening the e-mail or
attachment. Infosys reserves the right to monitor and review the content of all
messages sent to or from this e-mail address. Messages sent to or from this
e-mail address may be stored on the Infosys e-mail system.
***INFOSYS******** End of Disclaimer ********INFOSYS***
