Responses inline...for some reason my email client is mishandling the formatting, so I'll mark my responses with <dab> </dab>
Dave ----- Original Message ----- From: ant elder To: scabooz Cc: tuscany-dev@ws.apache.org Sent: Tuesday, May 20, 2008 2:03 PM Subject: Re: [DISCUSS] Declaring extensions as being available in the domain Thanks thats helpful. Let me take this to the extreme so you can point out where it loses the plot to help me understand what you're saying: The idea is eventually there might be standardized interfaces for all the SCA runtime plug points, many of the Tuscany things for the various SCDL and SPI interfaces - tuscany-assembly, tuscany-spi etc - that right now we have in org.apache.tuscany.sca namespace might one day be standardized interfaces, just like the org.osoa.sca.* interfaces for application Java implementations to use, and there would be a standard way to define new binidng and implementation extensions in the definitions.xml. <dab> Yes, that is the idea/theory. </dab> Anyone could make an SCA contribution that would add a binding or implementation type which once contributed to an SCA domain then other contributions in the domain could use in their SCA assemblies. And this new binding contribution would work in any runtime that supported SCA just like application SCA contributions should. <dab> yes </dab> So as a simple example, say a file binding that component service or references could use to read and write to a file system. That could be packaged in a regular jar which has a meta-inf/definitions.xml along the lines of: <definitions ... > <bindingType type="binding.file" ... > ...elements and attributes defining the binding.file schema and implementation classes ... </bindingType> </definitions> and all the file binding implementation classes within that file binding contribution jar would implement only standard interfaces, nothing at all Tuscany specific in the contribution. <dab> Good. Only thing to add at this point is that the spec might decide that the "jar" should be an OSGI bundle. Would such a estriction help, or get in the way? </dab> The splitting out all the various Tuscany SCDL classes and SPIs points into interfaces happened when we did the 0.90 rewrite so binding and implementation type extensions can now be written using only interfaces, but the pluggablity mechanism is still quite Tuscany specific. So is whats being suggested with this definitions.xml work starting on the road to coming up with a less Tuscany specific way of doing that, it would still be in a Tuscany namespace for now just as the various Tuscany SPI interfaces are, but in the future all this would be defined in some SCA spec , but it would just be a refactor in Tuscany to move to that spec namespace instead of a complete rewrite? <dab> Bingo. </dab> Is that anything like what you had in mind? ...ant On Thu, May 15, 2008 at 7:32 PM, scabooz <[EMAIL PROTECTED]> wrote: I might be the 'they', not sure. This is a very hard email thread to follow so let me try to articulate some ideas. First, bindingType and implementationType were introduced by the SCA Policy FW spec because that spec was the first to need a bit of metadata describing bindings from a typing perspective so that policy could be provided by the type and didn't have to be repeated on every binding instance. Implementation types was an obvious next stop on the journey. The text was driven into the Assembly spec as the result of a realization that these two concepts would be needed by SCA extenders who wish to add bindings and new kinds of components into any SCA runtime that was implemented by some another party. The assembly spec group decided (rightly IMHO) NOT to dive more deeply into filling out these concepts in V1.0. It would be more appropriate to tackle these sorts of issues in a future revision of the technology. Second, Tuscany is an open community where the plethora of talent enables new innovative ideas to be explored. These extensibility points are a great place for Tuscany to experiment with different approaches. I think it would be fantastic if Tuscany could "finish" the work of the spec group and come up with a reasonable (and workable) model for extensibility and then propose it to the spec group. Building on this thought then is the assertion that bindingType is the starting point for 3rd parties to describe a binding implementation to a runtime not authored by the binding implementer. It would be naive to think that the current structure of bindingType, it's containment within a defintions.xml file or any other related aspect is off limits from change in this experiment. I'm not clear on what all the issues are that have been raised. I see a lot of mis-information and some emotion that's hard to sort through. I saw a strawman that looked like a good start in one of the emails but there seemed to be resistance that I didn't understand. The SCA bindingType and implementationType concepts have nothing to do with policy. Policy was the first functional area that happened to see the need for them. Dave ----- Original Message ----- From: "ant elder" <[EMAIL PROTECTED]> To: <tuscany-dev@ws.apache.org> Sent: Tuesday, May 13, 2008 2:40 AM Subject: Re: [DISCUSS] Declaring extensions as being available in the domain On Tue, May 13, 2008 at 7:12 AM, Venkata Krishnan <[EMAIL PROTECTED]> wrote: Agreed to all that ONLY IF definitions.xml is going to contain things related to policies only. Though it is at the present moment my belief is that it could evolve to represent information more than just policy related things. This belief of mine is based on the following : - - the name of the file is 'definitions.xml' and is not 'policy-definitions.xml' - this is defined in the assembly model specs and not in the PolicyFwk specs. If its just for policies, I'd reckon that it would have been defined completely in the PolicySpecs and only referred to in the Assembly Model specs. Right now its vice-versa. Maybe some Specs folks should give us their perspective on this. - Venkat Those are good points. Especially the last one as its they that ask for this in an offline discussion so it would be really good to get their input here. ...ant