Re: [Springframework-developer] new bean factory and application context features

[Colin S. <col...@ex...>]

Rod Johnson wrote:

>>W/regards to redefining TransactionAttributeSourceEditor to allow for
>>wildcards also in the classname, that's a useful feature, but the
>>internal workings of the class would have to change. Right now, using a
>>MethodMapTransactionAttributeSource, essentially all the methods of the
>>specified classes get added to the transactional method map (if they
>>match the method regex). So this approach wouldn't work for regexes at
>>the class level. You'd probably have to do matches at runtime, with
>>consequently slower performance (although I guess you could cache the
>>hits and misses, to alleviate this, making the performance pretty well
>>identical on second and subsequent calls to the same method). What do
>>you think of this idea
>>    
>>
>
>I prototyped such an approach a couple of months ago. The performance seemed
>not too bad. I haven't committed it. However, maybe I should when I get
>time, if people are interested.
>  
>
I think this would be useful to some people (although the ability to 
handle the 'all' case with another mechanism like the specialized 
TransactionalAttributeSource would cover most of my needs)..

If there are performance concerns, it could use a dual approach where 
regexes on methods only (ie everything up to and including the class 
name is specified) get handled as they do now, and other regexes get 
handled with a runtime lookup with cache. My feeling is that the runtime 
lookup with cache would perform well enough though. One consideration is 
that the config semantics would change in terms of typos. Right now 
MethodMapTransactionalAttributeSource gives you an exception if you feed 
it a FQN which can't be resolved, or there are no matching methods. I 
like the idea of detecting typos in this fashion, if at all possible. 
Considering that your code may appear to run fine even without the 
transaction, if you mess up the regex, I think the error detection is 
valuable.  So if the code is extended, maybe one way to do it would be 
to have two sets of properties, one which behaves exactly like the 
present property (config time regexes on methods only), and a new one 
which adds runtime regex matching on the package and class portion as well.

Regards,
Colin