Hi Nodet,
I'm not quite sure that there is a need to change the chain
dynamically at all.
I don't like the idea neither, and it's the last solution i would like
to try. (And i think DanD also agreed on it)
Why not considering a tree instead of
a simple list ?
Depending on some conditions (HTTP VERB,
operation QName, ...), a branch of the tree would be used.
I guess this may need a bit of design, but would allow a clean
separation of interceptors between, while allowing a static
interceptor chain.
I think you are right, it's could be tree, and it's not just for HTTP
verb, in some cases we need to dispatch the message in different route.
for example, CXF-147 <http://issues.apache.org/jira/browse/CXF-147> the
mixed bare/wrapped style can be benefit from this approach. and also
CXF-35 <http://issues.apache.org/jira/browse/CXF-35> Support multiple
services/bindings on the same endpoint
<http://issues.apache.org/jira/browse/CXF-35>
I think maybe we should have something call connectors, and the
connectors basically have two interfaces one for the IN interceptor, and
one for OUT interceptor(s), and we can use connectors to chain the
interceptors, so it's the connector to decide which way/exit to go
depending on the message it accepted from the IN interceptor(just like a
switch). but i'm not sure will that works? as you said we need a bit of
design.
current chain is getting more and more complex, we even have sub chain
concept now, it's getting debug more and more difficult. to be honest i
don't like. maybe we should think more about this.
Cheers,
James.
As I said in a previous mail, I don't really see how policies can
be applied on a per operation level without allowing different
interceptor chain (or all interceptors would have to check if
they should be applied, which is the current case with the GET
problem).
Would it be too complex ? Or there is no real use case for that ?
On 12/15/06, Dan Diephouse <[EMAIL PROTECTED]> wrote:
On 12/7/06, James Mao <[EMAIL PROTECTED]> wrote:
>
>
> I'm OK with the changing the chain dynamically, they both works. if we
> change the chain dynamically, then for both the SOAP binding and XML
> binding and any other binding to filter the interceptors
dynamically, i
> mean the maintenance cost is same. but this approach do have a
benefit,
> the benefit is that all the isGET logic in the same place, if we
want to
> add some configuration for this function, it'll be more easier. But
the
> other side is, it'll be harder to change the chain if the
interceptor is
> coarse-grained, that means we want some part of the logic of the
> inteceptor, but in some conditions we want to exclude the
interceptors,
> but yes, you can break down the interceptors into pieces to work
around
> the problem. So there's pros and cons.
Can you please justify the performance benefit of this if we go down
this
route? As noted in the previous email if we have a dynamic interceptor
removal, than we still have problems if a user adds an interceptor
and they
aren't aware they need to look for the isGET case.
I think we should synthesize a document, and unless you can provide some
compelling performance reason I don't see any reason not too. You
haven't
shown anything to back up your reasoning that there is a performance
issue.
- Dan
--
Dan Diephouse
Envoi Solutions
http://envoisolutions.com | http://netzooid.com/blog
