Re: Proposal for Describing Web Services that Refer to Other Web Services: R085

Hi Sanjiva,

On Tue, Apr 29, 2003 at 01:00:42AM +0600, Sanjiva Weerawarana wrote:
> I think we're missing the point of the discussion. What Arthur
> proposed was a way for one to point to a URI in a message and say
> "see that URI? That's not just a URI - it points to a Web service".
> That's very RESTful and hopefully you like it Mark.

For that small part of the proposal, yes.

There's a lot I'm arguing with in my response, but under the covers,
it's that services need application interfaces which are different
than one another.  The rest of the apparent confusion around this
requirement & proposal derives from that, such as the role of the URI
scheme (was it not clear that "wsep:" was a tongue-in-cheek proposal?
8-O).

> The second thing Arthur did is to say that there is this other
> thing called WS-Addressing that some companies have developed
> which they seem to like as a way to point to Web services. So his
> proposal allows one to point to a piece of XML typed by a WS-Addressing
> endpoint reference and say "see that endpoint refernce type thing?
> That's not just a piece of XML; it points a Web service".
> 
> It seems to me that you're arguing against the 2nd part, right?
> If so, the real argument is with those who created WS-Addressing
> and endpoint references within them.
> 
> As I'm one of those, I can give some rationale (which you may or
> may not agree with). Basically, I agree that we could've just
> defined a new URI scheme instead (wsep:*). However, that would've
> meant we have to define some funky rules to encode all the
> stuff we want to contain in an endpoint reference: a URI for the
> address, the WSDL type/service info, any "instance ID info", any
> policies that may apply or any other data that the creator of
> the reference would like to have you send back. Mathematically,
> that's a trivial thing. Practically, that would've meant that we
> would have to write a custom software to handle all this encoding /
> decoding stuff.
> 
> Alternatively, we chose to define an XML Schema type and just make
> each of the parts explicit. Why? We thought it was simpler to
> use, read and to love.

Thanks for the explanation.

But what about the drawback I mentioned about not being able to
distinguish between an identifier and "any old markup".  A client would
have to know the data format in order to know which it was.  Related to
that is that the identifier isn't uniform (like URIs), as its
identification semantics may change, or cease to exist, as it's used in
different contexts.  IMO, this is a huge architectural failing,
especially when an alternative exists within the Web/REST.

As for the cost of URIfying, I guess I don't see any problem in encoding
the type of information you need into them.  Do you have an example?
Sorry, but the "wsep" joke may have thrown you off.  8-) This is more
what I had in mind;

http://lists.w3.org/Archives/Public/www-ws-desc/2003Apr/0107.html

MB
-- 
Mark Baker.   Ottawa, Ontario, CANADA.        http://www.markbaker.ca
Web architecture consulting, technical reports, evaluation & analysis

Received on Monday, 28 April 2003 16:03:18 UTC