Thursday, October 13, 2005

WARNING: Reference Parameters considered bad for your health

There's been a discussion going on over at the WS-Addressing public forum around Reference Parameters, which degenerated into the need for session semantics. The specification is already more complex than it probably needs to be: not a good example of Occam's Razor. I hate to sound like a broken record, but use WS-Context. The argument for it is pretty compelling.

2 comments:

William Vambenepe said...

Hi Mark. I agree w/ you that WS-Context is more appropriate for the scenario described in the current thread on the WS-Addressing mailing list, but there might be some value in being able to update an EPR as I try to explain at: http://h20325.www2.hp.com/blogs/vambenepe/archive/2005/09/29/302.html. Of course, this assumes that there is some value in reference parameters, unlike what your title suggests...

Mark Little said...

Hi William. I'm not sure about Reference Parameters at all really, but am prepared to entertain use cases; but glad to hear you agree session-semantics shouldn't be in there.

As for changing the EPR: it's an address (or should be). What needs to change in it that can't be dealt with at a higher level? For example, most distributed systems have had the concept of object/service migration and one of the more well known examples was SSP Chains (http://regal.lip6.fr/projects/sspc/): the concept is you migrate, leave the old endpoint valid, and return the new endpoint address to users of the old endpoint address lazily as required. But the address of the old endpoint doesn't change.

I'll check out your blog entry in case I'm missing something.