Hi,

Sorry, forget the AuthModule for now ;) I just put it there because it's often forgotten that it also gets to see the request and can wrap it, but for a change (I know, it's rare with me ;)) the issue here is primarily unrelated to auth.

Kind regards,

Arjan Tijms

On Thu, Sep 8, 2016 at 2:14 PM, Romain Manni-Bucau <rmannibucau@gmail.com> wrote:

Well supposes @Current means something for the component getting the injection, I can see where you go starting from AuthModule but more your browse the chain less it makes sense IMO -> maybe something to work on auth side more than CDI?

That said CDI has a notion for conversions which is to define a provided filter. Maybe that's the easiest here: a user could define one or multiple filters ("CDI Filter" or whatever name) which would set CDI position in the filter chain, defining it N times would update "current" request and the produces one so we keep current default and enable the user to control it himself finely when needed.

Romain Manni-Bucau
@rmannibucau |  Blog | Old Wordpress Blog | Github | LinkedIn | Tomitriber | JavaEE Factory

2016-09-08 14:10 GMT+02:00 arjan tijms <arjan.tijms@gmail.com>:

Hi,

I see what you mean.

There's perhaps 2 solutions.

1. Specify @Inject HttpServletRequest to inject the current one (the top one from the stack)

2. Let @Inject HttpServletRequest as it's now (you get an arbitrary one, which is hopefully at least always the same for a given server), and introduce a qualifier @Current orso.

I don't think you'd ever need @Inject @Web(FilterX.class) Request, which would mean searching back through the stack. If users would want that, they have to make their own producer.

What we could do now is:

AuthModule

|

|

Filter1 (calls) ---> BeanX.foo() #1

|

|

Filter2 

|

|

Servlet1 (calls) ---> BeanX.foo() #2

|

|

(forward) ---> Servlet2 (calls) ---> BeanX.foo() #3

If BeanX is injected with @Inject HttpServletRequest or @Inject @Current HttpServletRequest, then:

In #1, BeanX.foo() sees the HttpServletRequest as it was passed in to Filter1

In #2, BeanX.foo() sees the HttpServletRequest as it was passed in to Servlet1

In #3, BeanX.foo() sees the HttpServletRequest as it was passed in to Servlet2

Since the request processing is strictly sequential here, a given thread always has the concept of the request that was passed it to the module/filter/servlet that's highest up the call stack. That one would IMHO be the most logical one that you want to work with.

Kind regards,

Arjan Tijms

On Thu, Sep 8, 2016 at 1:53 PM, Romain Manni-Bucau <rmannibucau@gmail.com> wrote:

AFAIK the server propagates by designt he request/response pair so even if there is a way to know which one it is at each moment you still have the issue you don't know what @Inject Request; means. @Inject @Web(FilterX.class) Request; is possible but not sure the real meaning and IMO it breaks the loose coupling of CDI so not sure it does worth to spec-ed it.

Romain Manni-Bucau
@rmannibucau |  Blog | Old Wordpress Blog | Github | LinkedIn | Tomitriber | JavaEE Factory

2016-09-08 13:31 GMT+02:00 arjan tijms <arjan.tijms@gmail.com>:

Hi,

On Thu, Sep 8, 2016 at 1:09 PM, Romain Manni-Bucau <rmannibucau@gmail.com> wrote:

Hit that issue as well several times.

It is more vicious than it looks like IMO cause CDI will *never* get *the* right request for everybody, it is as simple as that. Any part of the app can rely on the wrapper level N (of course N being different for each mentionned parts of the app).

What I was thinking, but maybe I'm wrong, is that the application never "just" wraps the request and uses it for itself. It always passes it to the container, which then passes it on to the next Filter, Servlet, or whatever. So at that point the container code has the opportunity to store the request as being the "current" one.

E.g. if you do:

RequestDispatcher dispatcher = servletContext().getRequestDispatcher(...);

dispatcher.forward(wrap(request), response);

Then the RequestDispatcher, which is a container class, receives the wrapped request and can make it available.

The other way around, eventually every AuthModule, Filter or Servlet has to be called by the container at some point. E.g. the "protected void service(HttpServletRequest req, HttpServletResponse resp)" is called by the container.

So just before the container invokes the HttpServlet#service method, the container can store the request, and CDI (via an SPI) can pick it up from there.

That way in every context you can always have the *current* request (the request that's passed in to the last Servlet or Filter call on the stack).

Kind regards,

Arjan Tijms

 

Best CDI can do is to provide the request it has (already the case and cost pretty much nothing) and enable the user to produces very easily its own request from its filter (or equivalent) for its usage IMO - which is IMO already doable but maybe there is another shortcut we can introduce I didnt think about. If you look one step further any web framework built on top of CDI does it and therefore runs in a well known context.

Romain Manni-Bucau
@rmannibucau |  Blog | Old Wordpress Blog | Github | LinkedIn | Tomitriber | JavaEE Factory

2016-09-08 13:03 GMT+02:00 arjan tijms <arjan.tijms@gmail.com>:

Hi,

On Thu, Sep 8, 2016 at 12:40 PM, Martin Kouba <mkouba@redhat.com> wrote:

that's a good question. In Weld, the request object is captured during request context activation, i.e. during ServletRequestListener.requestInitialized() notification and before any filter or servlet is invoked. So wrappers are ignored and the original/first request is used.

Indeed, although do note that some servers (Liberty and WebLogic I think) send the ServletRequestListener.requestInitialized() notification rather late, and do that after the application already has seen the request and has had a chance to wrap it. This by itself is a separate problem. So on these servers, Weld would receive an early request but not necessarily the earliest.

 

But TBH I don't think we can fix this easily as I'm not aware of any portable way to listen for "wrapping actions".

This would have to happen with Server specific code I guess, just as Weld now requires an SPI to obtain the current principal for injection.

You could say that the Servlet container could store the request "somewhere" on a stack structure, just before it invokes the ServerAuthModule, Filter, Servlet and anything else I may have forgotten, and then when control flows back to each Servlet, Filter, etc unwind that stack.

At the very least the spec for now should perhaps clarify this?

Kind regards,

Arjan Tijms

 

Martin

Dne 8.9.2016 v 11:02 arjan tijms napsal(a):

Hi,

The CDI spec defines a built-in bean for the type HttpServletRequest. In
3.8 it says:

"A servlet container must provide the following built-in beans, all of
which have qualifier @Default:

a bean with bean type javax.servlet.http.HttpServletRequest, allowing
injection of a reference to the HttpServletRequest"

An HttpServletRequest however can be wrapped multiple times and by
multiple artefacts. I.e. by a ServerAuthModule, Filter and a
RequestDispatcher.

The question now is; which version of the HttpServletRequest is supposed
to be injected?

* The first one in the chain?
* The last one in the chain?
* The current one at a given point in the chain?

A little bit of experimenting seems to indicate it's now often "one of
the first ones", e.g. the one that happened to be current when e.g. a
ServletRequestListener that initialises a specific CDI implementation is
called.

I think this is a little confusing, as working with an injected request
can now totally ignore the request wrapping that has been done and break
an application badly.

Thoughts?

Kind regards,
Arjan Tijms

_______________________________________________
cdi-dev mailing list
cdi-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/cdi-dev

Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.

--
Martin Kouba
Software Engineer
Red Hat, Czech Republic

_______________________________________________
cdi-dev mailing list
cdi-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/cdi-dev

Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.