Re: ResourceBundle optimisation

[Bryce McKinlay]

Thomas Zander wrote:

> I'm pretty sure its not an API design to disallow this. So its 
> implementation specific and your argument that Sun does not do it is 
> indeed a pretty strong one.  I'm not sure I agree with it, but its a good 
> argument nontheless.
>  
...

> My original point still stands;  loading "foo.properties" while 
> "foo_nl_NL.properties" was requested will effectively make it impossible 
> to find foo_nl* if its placed in the classpath later.  Which is 
> definitely not what I expect.
>  

Tomcat and other servers implement re-loading of existing classes by 
loading them into a new ClassLoader. This way, old instances of the 
classes will be garbage collected when their ClassLoader becomes 
unreachable.

Resources work in the same way: ResourceBundles are cached 
per-ClassLoader, so modified bundles will be reloaded if getBundle() is 
called from a class loaded by a new ClassLoader.

I have not confirmed that this is the behaviour you get from Sun's 
implementation, but I suspect it is. I also think it provides a good 
balance between caching performance and the flexibility to reload things 
in a long-running server. getBundle() is called a lot, so it should be 
very fast in the common case - ideally avoiding allocations or making 
system calls.

Bryce