[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]
Subject: Re: [sca-assembly] issue 243: is default channel configurable? Proposals
Hi Eric,Thanks for pointing out the flaw in proposal 1. If we were to go down that path then we'll have to fix it. We did discuss this issue on the call today, but didn't take a decision as folks wanted more time to think. I took an AI to continue discussion on this list and summarize (hopefully correctly) the concall discussion here.
I think we need to figure out which direction we want to go: do we want to allow one to configure the default channel in a standard portable way or not? Disallowing configuration of the default channel in the spec, I don't think, is a good idea. Either we say it is undefined (or stay silent) OR we specify exactly how it is done.
Mike has made an argument that we ought to provide a standard way to configure the default channel (for things like JMS, policies etc.) Personally, I could go either way with a slight preference for some version of proposal 1. I find the argument, default channel configuration is going to be similar to non-default channel configuration and therefore providing standard way to do this would allow the same tools to configure the default channel without having to invent new ways for every implementation, persuasive.
If we decide we want to go with some modified version of proposal 1. We'll have to figure out how to conform to the production rules for NCName. One choice is to use __DEFAULT__ (as suggested by Eric) or some other reserved value (that doesn't include ':', '/' etc). If we do that we'll have to figure out whether this reserved value applies only to global channels or all channels. Another choice (proposed by Danny) is to add another attribute (boolean) called 'default_channel' (or something similar).
We informally decided on resolving this issue next week on the concall, so please send your opinions/comments on the ML.
Thanks. -Anish -- On 11/8/2011 1:39 AM, Eric Johnson wrote:
Hi Anish, On 11/8/11 8:52 AM, Anish Karmarkar wrote:I took an AI to provide two competing proposals for resolving issue 243. Proposal 1: allow the default channel to be configurable. In section 5.9 of WD06, add the following -- The default domain channel, like any domain channel, can be configured. Configuring the default domain channel requires one to deploy a domain composite containing a channel with the 'name' attribute value set to "/". Note that since the default domain channel always exists, configuring the default domain channel in a running system is equivalent to reconfiguring an existing channel and will be subject to restrictions and conditions imposed by a runtime on reconfiguration of domain channels.I was wondering whether you'd fall into the trap/pitfall you hit. I meant to send you an email reminder about that, but I forgot. This proposal doesn't work, because channel names are declared as "xs:NCName <http://www.w3.org/TR/1999/REC-xml-names-19990114/#NT-NCName>", and you can't name a channel "/". We had an issue <http://www.osoa.org/jira/browse/ASSEMBLY-254> about "/" in the names of channels. Perhaps something like this? The default domain channel, like any domain channel, can be configured. Note that how it is configured is determined by the implementation, because it cannot be deployed like other channels ("/" is not a valid channel name). Also note that since the default domain channel always exists, configuring the default domain channel in a running system is equivalent to reconfiguring an existing channel and will be subject to restrictions and conditions imposed by a runtime on reconfiguration of domain channels. Or, more simply: "This specification does not define any way to configure the default domain channel." [In any spec, what the spec explicitly or implicitly chooses not to define is, by implication, left to an implementation. I'm not sure we need to bludgeon readers with more words, or that they add value here.] I think, instead, if we wish for the default channel to have some spec'd way to be configured, then we have to come up with some spec'd way to name it, perhaps (borrowing from C) such as "__DEFAULT__".Proposal 2: do not allow the default channel to be configured. In section 5.9 of WD06, add the following -- Since the default channel always exists, one is not allowed to deploy and configure the default channel using a domain composite. This does not prohibit runtimes from defining implementation-specific ways of statically or dynamically specifying configuration of the default channel.Given the fact that you cannot, as currently spec'd, name the default channel in order to deploy and configure, this new proposal just seems like a variation, "configuring the default channel is outside the SCA Assembly spec."Comments?Nope, none. ;-) -Eric.-Anish -- --------------------------------------------------------------------- To unsubscribe, e-mail: firstname.lastname@example.org For additional commands, e-mail: email@example.com