Hello,

My question is where I can find a specification of how exactly
object networks are serialized (when sent over RMI).

My real problem is that I am planning to use Hibernate in multi-tired architecture:
- Presentation (web)
- Business (EJB)
- Session Facade
- ...
- Hibernate
- Database

The question is how and in what form do I transport objects between presentation and business layers.

There are basicly two options:
1. Create a parallel structure of Data Transfer Objects (DTO) and a set of helpers to convert from Hibernate Domain Model Objects (DMO) to DTO and back.
2. Pass DMO's directly.
The first option is standard (in CMP/JDBC world), but has this annoying feature of having to enter the same information twice (DTO & DMO).

In order to evaluate the second option I need to know how serialization of DMO networks works.

Finally, I have one idea I would like to get comments about:
----------------
The basic difference between DTO & DMO is that DMO has relations, whereas DTO usually keeps just foreign key.

This brought me to idea of avoiding using Hibernate relations and map DTO's theirselves to the tables. In this case I would have to use Hibernate ehanced SQL (but not HQL), because Hibernate cannot use information about relations any more.

What do You think of such solution?
-------------

Finally, a couple of proposals:
- Add serialization section to Hibernate manual
- Write a systematic account of multi-layered architecture which explores these issues.

There are some docs, but I think they are not sufficient:
http://www.hibernate.org/124.html
Most of examples are oriented towards a situation when business layer and presentation layer are in the same JVM.

With best wishes,
Oleg M

2. is definetly the thing to do
and not using relation in hibernate sounds really strance, like driving the highway using a 747 (a non-sense).
What is you pb about serialization ? Any specific questions ?

Hi!



Well, first it is not so black-and-white. Making DTO's is not really my idea:
http://www.hibernate.org/124.html
I have seen it quite many times on different pages/postings related to Hibernate.

About serialization:
* Type:
class X {
Long getId();
String getName();
X getParent();
Y getY();
}

* Operation in EJB:
X loadX(id) {
X x = (X)session.get(id, X.class);
x.getY(); // What changes if i remove this line?
x.getY().getName(); // And this?
x.getParent(); // And this?
return x;
}

saveX(X x) {
}

* Operation in web/client:
X x = ejb.loadX(xId); // What object network did i just receive?
x.getY().getId(); // What happens here?
x.getY().getName(); // And here?
x.getParent().getParent().getParent(); // And here?

x.setName("sxx");
Y y = ejb.getY(yId);
x.setY(y); // If I only wanted to change a relation between X & Y without modifying Y, why do I have to load Y?
y.setName("x");
ejb.saveX(x); // What object network did I just send back?

Now regarding 747: basicly, from the viewpoint of EJB interface, what I really need is not an object model, but relational model. I need to pass around records which have foreign keys and stuff. Can You suggest a better tool to do the job?

Yes, I don't use 70% of Hibernate capabilities, but if it does the job without the overhead, who cares? Besides, if situation changes (which I doubt) I can start using objects.

Now, theoretically speaking: what are advantages of passing around real objects instead of (relational)records aka DTO? I understand, that when Your data access layer is in the same web container as the servlet itself - it is convenient to use real objects. But why would I want to do this way when working with EJB over remote inerfaces?

OM

If you want a record orientated approach, why would you use an ORM and not something more table-orientated like iBatis ?

Well, I still need a concept of mapping Record <-> Table. Otherwise I would have to duplicate infomration in every:
select
update
insert
...

So can anyone answer my serialization questions?

OM

I have absolutely no idea what your actual question is. If your question is: "can you serialize graphs of Hibernate objects?" then, as Emmanuel already said, the answer is: "yes".

How exactly? See my questions in the comments of the source code from my second post:

* Type:


* Operation in EJB:


* NB! Remote calls, Hibernate objects are serialized, no lazy-loading proxies any more!

* Operation in web/client:

I'm sorry, you'll have to ask an actual question, written in English, not post some bizarre psuedo-code that not even a compiler could understand, let alone a mere human.

Or how about you just try Hibernate out, experiment with different things, and learn for yourself.

* Let us have an object:


* If I load an object "x"

and then access a related object of type Y with getter

and after that send this object to the web tier (which implies serialization), what network of serialized objects will I receive?

Note that last two objects were not lazy-loaded.
* So in short when I load object "x", access a part of its lazy-loaded object network, and then serialize/deserialize it, which part of object network is passed?
* If I have object "x" in web tier and I want to change its relation to another object "y", should I do it like the following code or is there any way without loading "y"?



Sure, that's an option that I will definitely use. But I would expect to get such information from Hibernate reference.

Whatever network of objects you had on the server side .... I mean, this is a reasonably intuitive expectation .... and one you could easily have discovered by experimentation.

Serializing the graph does not cause more objects to be fetched, nor does it trim off any objects that are fetched. I'm not sure why you would expect this explicit statement in the manual.




I would advise that code.

Hmmmm.... I think oleg_myrk's question is this: If we return a proxy from an EJB call, will it load all of it's lazy properties before being serialized and returned to the EJB client.

oleg_myrk, is that a good interpretation of your question?

If so, methinks the answer is "NO" :)

Yes, partially this was my question.

See my next post

First of all thanks for the answer :)


What happens, if I loaded only X and then ask in web tier x.getY().getId()?
I know that in EJB "x.getY().getId()" would not actually result in loading the object.


Speak for Yourself :)


I would still suggest adding information to the reference. Hey, its *reference* :)

One more question: how would You organize a way to let a user edit a record like that:


Would I have to:


This seems to be very inefficient:
* When loading Z, I have to load 4 extra objects
* When saving Z, I have to load 4 extra objects again (+ 4 round-trips)!

All this stuff with Hibernate works perfectly well when Hibernate is in web container. But in EJB it smells like I still need Data Transfer Objects with foreign keys. Am I wrong?

Hope I am not getting too annoying ;)

M'kay, funny thing is I saw no hibernate code in your last example. How does any of that relate to hibernate?