Altrmi questions

[Richard Wallace <rw...@a--i--m.com>]

Hey guys,

I got everything working and looked at and played with the callback
tests.  It brought up one question, though.  I wanted to make sure the
server didn't need the generated classes from the client in it's
classpath or anything so I created two generation directories,
client-genclasses and server-genclasses.  The test went well, so I
assume that means the server doesn't need the generated classes from the
client.  

So, my next test was to determine if the client needed to have the
proxies pregenerated or if it could/would do it dynamically.  So, I
commented out the client generation stuff in callback.xml and tried
running the tests.  Sure enough, there was an exception thrown when
trying to add the callback to the server:

org.apache.excalibur.altrmi.server.ClassRetrievalException: Generated
class for AltrmiGeneratedCallBackTestListenerImpl_Main not found in
specified classloader

Is there a way, when the call is made, to auto-generate the classes
needed if they are not present already?  Something that would be
transparent?  Obviously, I can probably generate the needed classes by
hand, but I'd prefer something more flexible.  For instance, wouldn't it
be possible when the above exception is thrown for the host to generate
the stuff it needs?  I understand that would mean the tools.jar would
need to be in the classpath on the client, and I don't have a problem
with that.  Any ideas?

Thanks
-- 
Richard Wallace
AIM, Inc. (www.a--i--m.com)
Information Systems Consultants

"Providing New Technology,
     the Old-Fashioned Way"


--
To unsubscribe, e-mail:   <ma...@jakarta.apache.org>
For additional commands, e-mail: <ma...@jakarta.apache.org>

Re: Altrmi questions

[Paul Hammant <Pa...@yahoo.com>]

Richard,

>I've got another question you might be able to enlighten me on.  It
>appears that currently, if an object was looked up on the server, and
>one of the methods called, if the method returns something it is always
>returned by value.  Is that correct?
>
No, look at the tests, particularly TestInterface.class and one of the 
methods that has a param of TestInterface2.  That is a pass by reference 
param (because it came from the server in the first place in that manner).

>I have many things that I don't want this to happen with.  I've looked
>through the tests, but they seem to all return Strings or primitives.  I
>run into the problem of needed to be able to return objects that are not
>Serializable and need to be passed by reference, not value.  Anything I
>might be missing or is this a current limitation of Altrmi?
>
See above.  Should be OK :-)

- Paul



--
To unsubscribe, e-mail:   <ma...@jakarta.apache.org>
For additional commands, e-mail: <ma...@jakarta.apache.org>

Re: Altrmi questions

[Richard Wallace <rw...@a--i--m.com>]

Hello,

I've got another question you might be able to enlighten me on.  It
appears that currently, if an object was looked up on the server, and
one of the methods called, if the method returns something it is always
returned by value.  Is that correct?

I have many things that I don't want this to happen with.  I've looked
through the tests, but they seem to all return Strings or primitives.  I
run into the problem of needed to be able to return objects that are not
Serializable and need to be passed by reference, not value.  Anything I
might be missing or is this a current limitation of Altrmi?

Thanks again.

-- 
Richard Wallace
AIM, Inc. (www.a--i--m.com)
Information Systems Consultants

"Providing New Technology,
     the Old-Fashioned Way"


--
To unsubscribe, e-mail:   <ma...@jakarta.apache.org>
For additional commands, e-mail: <ma...@jakarta.apache.org>

Re: Altrmi questions

[Richard Wallace <rw...@a--i--m.com>]

> What are your needs re stability now?  Is this a commercial use you are 
> putting it to ?  Can you do without callback for the time being (i.e. do 
> polling)?

I'm not really in a bind right now.  It will be for commercial use, but
the project is just in the initial design stages and I'm just examining
all of our options and doing some testing of ideas.  So, I think we can
wait on some of the stuff for a while.

Rich


--
To unsubscribe, e-mail:   <ma...@jakarta.apache.org>
For additional commands, e-mail: <ma...@jakarta.apache.org>

Re: Altrmi questions

[Paul Hammant <Pa...@yahoo.com>]

Richard,

>I got everything working and looked at and played with the callback
>tests.  It brought up one question, though.  I wanted to make sure the
>server didn't need the generated classes from the client in it's
>classpath or anything so I created two generation directories,
>client-genclasses and server-genclasses.  The test went well, so I
>assume that means the server doesn't need the generated classes from the
>client.  
>
>So, my next test was to determine if the client needed to have the
>proxies pregenerated or if it could/would do it dynamically.  So, I
>commented out the client generation stuff in callback.xml and tried
>running the tests.  Sure enough, there was an exception thrown when
>trying to add the callback to the server:
>
>org.apache.excalibur.altrmi.server.ClassRetrievalException: Generated
>class for AltrmiGeneratedCallBackTestListenerImpl_Main not found in
>specified classloader
>
This is an issue yes.  Currently for the callback demo, there are two 
generations of <altrmiproxies> in terms of ant task invocations during 
the build.

We need to be able to change this to one invocation... and we are 
currently pondering things.

As it happens, we are doing a lot of thinking for callbacks at the 
moment and if there is an area of AltRMI that is likely to change some 
more before AlttRMI gets to v1.0, that is it.  Then again, it may not 
change...

>Is there a way, when the call is made, to auto-generate the classes
>needed if they are not present already?  Something that would be
>transparent?  Obviously, I can probably generate the needed classes by
>hand, but I'd prefer something more flexible.  For instance, wouldn't it
>be possible when the above exception is thrown for the host to generate
>the stuff it needs?  
>
The ideal world is that the compiled stubs know exactly shat to do 
without an exception having to be thrown to communicate the fact.  The 
reason I say that is the same bug has come in in the past in 
non-callback AltRMI and it has been fixed relatively easily enough.

>I understand that would mean the tools.jar would
>need to be in the classpath on the client, and I don't have a problem
>with that.  Any ideas?
>
Well Vinay is going to put some BCEL using proxy generators into CVS in 
the next few weeks (way to difficult for me).  In that time I think we 
may have a more stable AltRMI.

What are your needs re stability now?  Is this a commercial use you are 
putting it to ?  Can you do without callback for the time being (i.e. do 
polling)?

Regards,

- Paul H


--
To unsubscribe, e-mail:   <ma...@jakarta.apache.org>
For additional commands, e-mail: <ma...@jakarta.apache.org>