I see -- thanks for clarifying. I *think* that we should just change to
using InternalUtils.ensureModuleConfig() at that point. It's meant to
be used in any code path where a desired module might still be
unregistered (and as so, getModuleConfig() is actually very rarely
used). I should have switched to ensureModuleConfig(), not
getModuleConfig(), in that checkin. Does that sound alright to you?
Rich
Carlin Rogers wrote:
I haven't checked the setting or value of _moduleConfig. I've just been
comparing the old code path to the new changes while running this convoluted
scenario. I noticed that in the old code, that if _moduleConfig was null in
initModuleConfig(), it would see if it was already an attribute of the
context. Then, if not, AutoRegisterActionServlet.ensureModuleRegistered()
was called.
In turn, it called...
AutoRegisterActionServlet.registerModule(), which called...
AutoRegisterActionServlet.initModuleConfig().
This initModuleConfig() does a servletContext.setAttribute() with the new
module config, as does the AutoRegisterActionServlet.ensureModuleRegistered()
after registerModule() returns. The ModuleConfig object is returned from
ensureModuleRegistered() and assigned to _moduleConfig.
I've just noticed that in this scenario, when I'm in processMapping() that a
call to InternalUtils.getModuleConfig() for the GlobalApp module config will
now return null.
Does that make sense? Sorry if this isn't so clear.
Thanks,
Carlin
On 1/9/06, Rich Feit <[EMAIL PROTECTED]> wrote:
Hey Carlin,
Just want to make sure I'm understanding here. Are you saying that the
original call to initModuleConfig() (the one I removed) also registered
the module in the ServletContext, as a side effect? It was only
supposed to set the reference to _moduleConfig, which is now always
taken care of in getModuleConfig(). Is the problem happening because
the module isn't registered in the ServletContext, or because
_moduleConfig is somehow null?
Rich
Carlin Rogers wrote:
Hey Rich,
Thanks for the reply and the information on the
InternalUtils.avoidDirectResponseOutput() function.
I looked at the issue some more and it turns out that for the scenario
I'm
investigating, there's another revision that is also impacting the
behavior.
In revision 351812, http://svn.apache.org/viewcvs.cgi?rev=351812&view=rev,
for BEEHIVE-1017, the FlowController.reinitialize() method was modified
so
that in no longer calls initModuleConfig(), which ensured that the module
config was registered (attribute of the context).
Now, when the initial page flow of the portal scenario is opened in a
portlet, the GlobalApp module config is not added to the servlet context
attributes. Then when the unhandled action is hit and we fall into
processMapping(), the call to InternalUtils.getModuleConfig() for the
GlobalApp module config will be null and we wouldn't even be able to
check
for an action that was declared as "unknown".
Unfortunately, I'm still trying to understand why the call to from
FlowController.reinitialize() to initModuleConfig() was removed.
I've logged a JIRA issue on this and am trying to figure out what is the
best way to resolve the problem. See
http://issues.apache.org/jira/browse/BEEHIVE-1037
The bug description might be a better illustration of the scenario I'm
trying to solve.
Let me know if you have some more thoughts about how best to resolve
this.
Many thanks,
Carlin
On 1/9/06, Rich Feit <[EMAIL PROTECTED]> wrote:
Hey Carlin,
Sorry for the delay. I agree that we should be checking for an
"unknown" action mapping in the global app module, so if you're
suggesting making that change, I agree.
My answer to the rest is a little more involved:
- There's already a rough mechanism for avoiding direct response
output. In InternalUtils, there's avoidDirectResponseOutput(). If
that's called on the request, then InternalUtils.sendError() will throw
an exception instead of writing to the response. This is what I think
should be happening here -- we should be calling sendError().
- I think that two things should probably change here: 1)
InternalUtils.avoidDirectResponseOutput() should be replaced with a flag
in PageFlowRequestWrapper, and 2) strutsLookup() should just set this
flag off the bat. We shouldn't be writing to the response *ever* during
strutsLookup().
Let me know what you think (and if you have any questions about this).
Rich
Carlin Rogers wrote:
Just wanted to note that the difference in the behavior is also related
to a
struts merge where the struts module config has an action defined with
the
"unknown" attribute (making it like a default). I think the missing
condition is that we check to see if the GlobalApp has the action
config
but
we don't check any of the action configs on the global app to see if
they're
"unknown".
So, If the global app includes a Struts Merge and that struts module
config
includes an unknown action, we'll never hit it.
Carlin
On 1/5/06, Carlin Rogers <[EMAIL PROTECTED]> wrote:
Hey Rich,
Hope your work is going well!
I have a question about svn revision 356056 (
http://svn.apache.org/viewcvs.cgi?rev=356056&view=rev
) checked in as a fix for BEEHIVE-1024. It seems that it changed the
behavior of PageFlowRequestProcessor.processMapping() and how we
handle
an
unknown action. With this change, the code path for an unknown action
in
processMapping() fails the new check to see if it is in the globalApp
(...globalApp.findActionConfig(path) != null). We drop to the else
statement
and into a call to processUnresolvedAction() which uses the
DefaultExceptionsHandler class and eventually writes out the HTML of
our
action not found error message directly to the response. I think this
looks
OK. However, having the error message written to the response may not
be
the
desired behavior for something like a portal using a call to
PageFlowUtils.strutsLookup(). What do you think?
If we leave the fix as is, could we use the
PageFlowRequestWrapper.isScopedLookup() condition to determine if this
is
from strutsLookup() or not before calling processUnresolvedAction().
I.E
.
do something different for an unknown action in a strutsLookup()? Just
curious.
Thanks,
Carlin
