On Nov 29, 2007 9:08 AM, Simon Nash <[EMAIL PROTECTED]> wrote: > > Simon Laws wrote: > > > On Nov 28, 2007 11:31 AM, Simon Nash <[EMAIL PROTECTED]> wrote: > > > > > >>I'm trying to work out what domain level autowiring would mean. > >>Having a target assigned dynamically according to a set of rules > >>opens up the possibility that the service chosen would not be the > >>one expected. The other approach is to have a service registry > >>that could be queried using some other metadata. My first reaction > >>was that an explicit query would be more flexible, but perhaps > >>these approaches come to the same thing and it's just a matter of > >>where the lookup metadata is held. This would be a good topic to > >>raise in the OASIS SCA spec group. > >> > >> Simon > >> > >>Simon Laws wrote: > >> > >> > >>>Should autowiring work at the domain level? It doesn't at the moment. > >> > >>It's > >> > >>>only considered within a contributed composite. To autowire at the > >> > >>domain > >> > >>>level we have to adjust the way that service endpoints and references > >> > >>are > >> > >>>matched, i.e. extend the process to looking at interfaces and policies. > >>> > >>>In line with moving the registration and lookup code out of the SCA > >> > >>binding > >> > >>>[1] I'm thinking about tidying the code in this area to make future > >>>reliability scenarios easier to cope with (by making sure all the > >> > >>processing > >> > >>>is in one place with a clear services of associated events) so now > would > >> > >>be > >> > >>>a good time to cover other outstanding issues. I think the outstanding > >>>things are > >>> > >>>Domain level wire elements in stand alone composites. > >>>Domain level autowiring > >>> > >>>Are there any more that people can think of? > >>> > >>>Regards > >>> > >>>Simon > >>> > >>>[1] > >> > >>http://www.mail-archive.com/tuscany-dev%40ws.apache.org/msg25636.html > >> > >> > >> > >>--------------------------------------------------------------------- > >>To unsubscribe, e-mail: [EMAIL PROTECTED] > >>For additional commands, e-mail: [EMAIL PROTECTED] > >> > >>Yes, I'm not particularly a fan of autowiring myself as it seems to > bring > > > > potentially non-deterministic behaviour but the specs include this > feature > > at 1.0 so it would be interesting to hear the justification/scenarios. > > > The 1.0 specs include it but only within a composite.
Does this include the domain level composite into which all deployed composites are included? If no then we can treat each composite independently. If yes then we are discussing autowiring at the domain level. > In that context > it is deterministic and is effectively a shorthand for writing explicit I'm a little skeptical of this as in the case where more services are found than the multiplicity of an autowired reference is able to cope with the spec says that "the SCA runtime selects one of the target services in a runtime-dependent fashion". How do we translate our rule into something that make sense to the user contributing the composite. Choose the first component/service in alphabetical order? > > wires. At the domain level it could be non-deterministic depending on > what services have been deployed to the domain. I'd like to understand > more about the use case that requires this, so that we can explore what > issues the nondeterminism raises. > > Simon > > > > --------------------------------------------------------------------- > To unsubscribe, e-mail: [EMAIL PROTECTED] > For additional commands, e-mail: [EMAIL PROTECTED] > > Simon
