On 30/06/2010 21:23, Peter Roßbach wrote:
Hi Marc,
Am 30.06.2010 um 20:03 schrieb Mark Thomas:
On 30/06/2010 17:01, Marc Guillemot wrote:
Hi,
What you write means that the freshly release Tomcat 7 doesn't fully
implement Servlet API 3.0, isn't it?
Interesting to see that nobody reacts :-(
If you were following the dev list you would have seen that
Jean-Frederic has already vetoed the change Peter made on the basis
that it breaks the TCK. That suggests that Peter's analysis of the
current situation is not 100% correct.
Which of my findings aren't correct?
I don't know - I haven't run the TCK with your patch applied. Just the
fact the the findings led to the patch which in turn caused TCKs
failures suggests at least something isn't quite right. It could also be
the analysis is OK but the patch has issues or that there is a problem
with the TCK.
I wouldn't be at all surprised to find part of the spec not covered by
the TCK. I would be surprised to find a bug in the TCK. My own
experiences with potential TCK bugs is that the TCK has always been correct.
Regardless, until the TCK is proved wrong Jean-Frederic's veto is valid
and your patch needs to be reverted.
Mark
Peter
My recollection is that the last thing that Filip implemented for
Servlet 3.0 was the necessary timeout features. I'd be surprised if
Filip, as a member of the Servlet Expert Group that wrote the spec,
missed something fundamental. I wouldn't be surprised if there were
some edge cases missed. I do know that Filip tended to leave a lot of
the Servlet 3.0 TODOs in the code, even when he had implemented the
appropriate feature. Those certainly need cleaning up.
Looking at Peter's test case - along with the other issues folks have
raised in Bugzilla for the 7.0.0 release - is on my todo list but like
the other committers I have a day job. Whilst many of the committers
are fortunate enough (me included) to be able to allocate some work
time to Tomcat development, it doesn't follow one of the other
committers will immediately have the time necessary to look at Peter's
e-mail and comment in detail. I can't speak for the others, but my
free time is currently occupied with another Tomcat 7 issue - namely
http://tomcat.markmail.org/thread/25g353eqvqfqniiz
Like any area of Tomcat, help is always appreciated. If you are
interested in this issue and would like to help out I would encourage
you to look at Peter's test case, compare the expected behaviour to
what the spec defines and offer your view on whether there is a bug
that needs fixing or not. If there is an issue, patches would be even
better.
Mark
Cheers,
Marc.
Peter Roßbach wrote:
Hi!
I detected some Todo's at AsyncContext
- AsyncContext.createListener
We don't make a resource injection or make it configurable!
Servlet 3.0 API comment:
"This method supports resource injection if the given clazz represents
a Managed Bean. See the Java EE platform and JSR 299 specifications
for additional details about Managed Beans and resource injection."
- I miss the is[Started,Completed] methods at AsyncContext interface
The only way to detect a completed state at AsyncContext inside an
AsyncListener is:
if (ac.getResponse() != null)
AsyncListener got an event, with empty state... -- Very strange!
Can't we clear all AsyncListener after send complete and before we
refresh the AsyncContext state?
How can a user handle a completed situation?
see o.a.c.core.TestAsyncListener
- AsyncContext.start() dosen't really start a thread!
We create wrapper of a Runnable, but we don't start a thread...
Error handling seam a little bit strange.
We only throw a RuntimeException(ex) and log an error!
Servlet API comment:
"Causes the container to dispatch a thread, possibly from a managed
thread pool, to run the specified Runnable. The container may
propagate appropriate contextual information to the Runnable."
How can we implement that?
Start everytime a new thread without pooling?
Use the connector executor pool from request?
Then we must extend the ProtocolHandler interface with getExecutor()
signature.
Use a separate Executor pool per Engine/Host or Context
How do we have to implemented the error handling?
- JioEndpoint has a timeout detection thread, but we don't start it!
I easy fix it with revision 958362,
but the polling strategy is slow and ineffectiv.
I didn't make a test at NIO/APR-HTTP or AJP-Connectors.
- After the timeout event is emitted, the completed method is
automatically called!
Servlet API 3.0 Spec says:
====
■In the event that an asynchronous operation times out, the container
must run
through the following steps:
■ Invoke the AsyncListener.onTimeout method on all the AsyncListener
instances registered with the ServletRequest on which the asynchronous
operation was initiated.
■ If none of the listeners called AsyncContext.complete() or any of the
AsyncContext.dispatch methods, perform an error dispatch with a status
code equal to HttpServletResponse.SC_INTERNAL_SERVER_ERROR.
■ If no matching error page was found, or the error page did not call
AsyncContext.complete() or any of the AsyncContext.dispatch
methods, the container MUST call AsyncContext.complete().
===
I miss the implementation that we call the error page!
- Currently a AsyncListener can't reinit a async cycle!
see o.a.c.connector.Request.startAsync
... L1555ff
if (asyncContext==null) {
asyncContext = new AsyncContextImpl(this);
} else if (asyncContext.isStarted()) {
throw new IllegalStateException("Already started.");
}
...
What does this spec definition really mean?
=== Ch 2 Page 18
■public void onStartAsync(AsyncEvent event) - Is used to notify the
listener that a new asynchronous cycle is being initiated via a call
to one of the
ServletRequest.startAsync methods. The AsyncContext corresponding
to the asynchronous operation that is being reinitialized may be
obtained by
calling AsyncEvent.getAsyncContext on the given event.
====
- AsyncListener doesn't receive a onStartAsync event.
We don't implement it.
At AsyncListenerWrapper fireOnStartAsync is missing
As AsyncContext.startAsync method doesn't emit an event
- ContextClassLoader at AsyncContext.dispatch is not set!
I can't see that we correcly set the application ContextClassLoader
before
we dispatch the Request. What must we do at a CrossContext dispatch?
- After first complete async cycle, an AsyncListner receive
onStartAsync as code start next async cycle?
Why doesn't the Servlet 3.0 TCK check basic AsyncContext
functionalities?
It seams that we better start a open test suite implementation at the
new Servlet 3.0 API.
Regards,
Peter
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@tomcat.apache.org
For additional commands, e-mail: dev-h...@tomcat.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@tomcat.apache.org
For additional commands, e-mail: dev-h...@tomcat.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@tomcat.apache.org
For additional commands, e-mail: dev-h...@tomcat.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@tomcat.apache.org
For additional commands, e-mail: dev-h...@tomcat.apache.org
