On 6/16/06, jhakim <[EMAIL PROTECTED]> wrote:
Clearly both setObject(name, map) and setObject(name, mapMsg) work. As you correctly point out, using a hierarchical naming scheme would allow the client to specify nesting and works with any MOM. However, I would argue that forcing clients to write code to create/parse a hierarchical naming scheme defeats the key goal of ease of use. For instance, suppose one wants to create a framework for marshaling arbitrary beans to JMS messages. A logical implmentation would be to use reflection to discover bean properties and create a corresponding MapMessage. Now, suppose that a bean contains other beans as properties. One elegant approach would to marshal each bean property to a nested MapMessage. This exact strategy is used by many systems on Wall Street and by open-source projects (messageforge.sourceforge.net). BTW - the underlying JMS provider can, beneath the covers, use hierarchical naming schemes and strip off properties from nested messages. As far as the client is concerned, this should just be an implementation detail and not the required way for clients to use the MOM.
Agreed. Obviously if you have a nested array of beans you can obviously use ObjectMessage too - but I totally understand the motivation for having a typesafe hierarchial MapMessage that other languages can parse too. I've created an issue to track this feature request http://issues.apache.org/activemq/browse/AMQ-757 -- James ------- http://radio.weblogs.com/0112098/
