MDA => Agile!?

Came across this article regarding the MDA. It took me long enough to get my head around the idea that it's the Model Driven Architecture, rather than just any old architecture that's driven by models. Except that Haywood tells us in the piece that actually there are two MDAs depending on which tool chain you buy. The difference is that one set of vendors builds tools to support an elaborative workflow from the Computation Independent Model through to a Platform Specific Model and another a translational one. Eh? Translation vs elaboration was one of the big arguments back in the days when there was an active marketplace for OO methodologies. Seems as if the MDA is intended to be transformational. Or possibly elaborationalist, depending it seems on whom you ask.

Steve Cook has this[pdf] (amongst other things) to say about MDA. Now, Steve was partly responsible for the Syntropy method, which features three kinds of model: Essential, Specification and Implementation--reflecting the view that one single modelling vocabulary won't do for capturing knowledge about the world, specifying a software system to deal with it or figuring out how to build such a system. That's a little bit like the MDA's distinction between Computation Independent Models, Platform Independent Models and Platform Specific Models. Kinda. But it's not clear what these models are really supposed to be written in. UML? The MDA is an OMG offering, and UML is the OMG's modelling language.

Unfortunately, it's sometimes quite hard to know what a given UML model is supposed to mean (without, ironically enough, showing the corresponding code), and so it's hard to see how automated translation of UML models is going to go anywhere interesting. During the last run at this sort of thing, the CASE tool boom of around ten years ago, I had a job in a C++ shop where all code was round-tripped through a leading tool. It wasn't unusual to have to go through certain hoops to get this tool to generate code that compiled at all, never mind do what I wanted it to do. Maybe the tools are a lot better now, although "OTI" Dave Thomas's fairly recent assessment of the MDA space makes me think not.

Meanwhile, Haywood makes this rather provocative statement:

Adopting MDA also requires a move towards agile development practices. While like many I'm an advocate of agile processes, it's still foreign to many organizations. The need for agile development follows from the fact that MDA requires models to be treated as software, hence they are the "as-is" view. Those organizations that used UML only for blueprints or sketches (Fowler's analysis [4.]) will find that MDA does not permit the use of UML in that way.

I suspect that it would come as a surprise to many Agile practitioners that adopting the MDA would make an organisation more agile (and that this is because you can't use UMLAsSketch!) But let's examine the claim in a little more detail.

Certainly the idea of keeping your whole model and code stack in sync at all times seems like an Agile idea. I'd expect that to be more of a strain on those folks who in the past have taken to producing huge UML models that no-one ever looks at, and not so much the sketchers, so maybe that does drive you in the Agile direction.

Extremely Agile

I have to be careful here, because I'm not just an Agilist, but an Extreme Programmer, so even other Agilists sometimes think I have some very strange ideas about development. The principles of Agile development compiled by the authors of the Agile Manifesto don't seem to say anything that would stop you from using MDA, but they do suggest that, not only do you maintain your UML CIM in-sync with your code in the reverse direction (which is what I understand Haywood to be talking about--don't let the code run away from the model) but that you regenerate the whole system CIM->PIM->PSM->code->deploy frequently. The principles suggest:

Deliver working software frequently, from a couple of weeks to a couple of months, with a preference to the shorter timescale.

Well, yeah. Actually, my preference is for a much shorter timescale: weekly releases do not seem too frequent to me. And I want to be able to release on any given day. And I want all my changes (to whatever artifact) reflected in a deployable system ASAP. By which I mean, within minutes. Beck recommends a Ten Minute Build, which in MDA terms would seem to mean running the whole workflow (modulo dependency analysis) all the way through and executing a suite of tests in ten minutes. Should make the hardware vendors happy.