| Re: [buckminster-dev] Action output? |
Ah, ok. I misread you. You're back to what I was on to in my initial posting when I wrote "so that transitive invocations can be mapped on the fly". I thought that was considered too complex and hard to grasp. I'm still not convinced of the opposite. One might argue that there is only one namespace (the property namespace) and using it, sure, then you cause pollution. The somewhat harsh solution to it is to say that if you want action specific properties, then define them in the cspec.
Defining them in the cspec is meaningless. 1) they're essentially static, and 2) if you give a prop on the commandline you'll override them.
So yes, I would argue that the props namespace we're talking about here (what has so far been called the context props) should not be polluted and used for all. Rather, each action should get a fresh context props instance with only the props defined there that are specifically intended for it - this obviously includes 1) the ones listed in the cspec, 2) the ones 'manually' specified by the user and *targeted for this action*, and 3) the buckminster autogenerated ones.
We could of course introduce a new XML format, a local runtime complement to the cspec. But I'd rather postpone that for later.
Sure - this was just to allow one to easily can a perform invocation with a 'complex' commandline (where 'complex' probably is anything beyond, say, two lines in a shell). Sure, a temporary solution is to just can it as a batch/shell script, but that's not quite the same.
ken1