Comments on: Spring and load-time weaving of Neo4j-based domain objects

By: activey

activey — Wed, 28 Oct 2009 09:44:48 +0000

Matias,

Can you attach project source code?

By: open source testing tools

open source testing tools — Sat, 05 Sep 2009 08:12:35 +0000

wow….good effort buddy

By: tk

tk — Sat, 30 May 2009 03:34:16 +0000

Jan: that’s very helpful. I like that approach, thanks!

By: Jan Kronquist

Jan Kronquist — Thu, 28 May 2009 08:05:53 +0000

I haven’t used neo4j, but maybe you could do something similar to Hibernate, ie keep the transaction open in the view using a servlet Filter.

See the Hibernate docs here: https://www.hibernate.org/43.html

By: tk

tk — Wed, 27 May 2009 23:39:25 +0000

In my case, the controller does not actually handle the rendering of the view (JSP) so annotating the controller does not help.

I might just end up plucking property values into request attributes for my JSP to pick up. I really just want to pass in my domain objects but creating a value object from my (node wrapper) domain object just defeats the point doesn’t it?

By: Mattias Ask

Mattias Ask — Wed, 27 May 2009 08:23:20 +0000

Nice to hear that you’ve solved the problem in a similar way, tk

I see two solutions to your dilemma:
1. Create simple view-VOs that hold the property values of the Node.
2. Set the transaction annotation on the controller instead of the service.

The first alternative is some work. But if you want to pass your domain around, another alternative is to create a DetachedNode which just creates a HashMap of the property values of the Node and that implements the Node interface. This has a whole bunch of other problems (like traversers, relations etc.), but it’s less code to write if you want to pass detached properties out of the TX scope.

The second alternative can be a bit risky since you can make changes to the DB in your controller, but it’s sure to be fast since you only need one TX per request.

Personally, I put the TX around the controller that populates the view. I don’t know if it’s correct, but it feels like the lowest hanging fruit

Does anyone else have any thoughts on this?

By: tk

tk — Tue, 26 May 2009 23:57:58 +0000

I have been playing around with neo4j and so far, have independently done more or less the same thing as you, including the NodeDelegate wrapper. It is good to see someone else thinking the same.

How do you plan on getting around the fact that pretty much all operations on a node object require a transaction block? It is easy enough inside the service layer with annotation markers but what about when you pass your domain model to your view?

I thought about introducing the transaction block in the getProperty method in the node delegate base class but that means I have to inject a NeoService reference into every node.

Your thoughts?

By: Mattias Ask

Mattias Ask — Tue, 26 May 2009 10:25:57 +0000

I forgot that, Emil. The only change I made was that I used the empty constructor instead of sending in “null” in the one that takes a Node. But on the other hand, “new TweetNode(null)” works equally fine The setting of the Node is done after the new TweetNode is instantiated.

By: Emil Eifrem

Emil Eifrem — Tue, 26 May 2009 10:10:41 +0000

Good stuff.

Any chance you could post the final UserNode.createTweet() implementation?

-EE