In the comments on my blog about the CMMI Made Practical conference here, I quote Microsoft as an example of a very successful low-maturity company. I make no secret of my partiality for CMMI-style process improvement (almost 30 years in IT has convinced me that it’s far from a process-free zone) but it’s not the only way to run software development (only, perhaps, the most effective way in the longer term).
So I was interested in the insight to Microsoft’s development philosophy provided by Tim Anderson, here. In particular, I thought Greenfield’s quoted remark about the OMG’s MDA - “...[it has] CIM [Computation Independent Model], PIM [Platform Independent Model] and PSM [Platform Specific Model]. The same viewpoints apply to everything, whether you're building eBay or a mobile device application. We don't buy that. They're different things” – was just plain silly.
As a fan of MDA, I think, more or less, that both eBay and a mobile phone have a business model that doesn't change when the technology does; a model that represents a subset of this corresponding to the automated system inside the man-machine boundary; and a model corresponding to the implementation on a specific computer/phone platform. That makes perfect sense to me.
When I put this point to Tim, he said that what Greenfield was really saying was that people were forced to extend the UML behind MDA and that this led to chaos. In a hitherto unpublished quote from Tim’s interview, Greenfield says: “What we say is this: look at what people do with it. To be useful, they'll take it, they'll hold their nose and use stereotypes and tags and mark it all up, because what they're really trying to do is to enable the capture of additional information, the tags allow you to add new properties, the stereotypes give you new metaclasses more or less, and in a good tool, you can actually add some additional constraints. What you end up with is this horribly marked up hotchpotch of stuff, which is really a domain-specific language [DSL]. It's a tailoring of the fundamental abstraction of state chart or activity graph or whatever, to solve some specific purpose. Then people will go write a code generator that rummages over that stuff, spews out the code, the SQL or whatever it might be”. Emotive language, but in other words, why not do DSLs properly from the start?
Well, that’s not so silly, it’s a valid approach. But my response would be, that UML 2.0 was designed to be extended, and that a mature organisation using MDA tools from, say, Compuware and with a high-maturity development process can cope with all this rather better than Greenfield envisages. The advantage of high-maturity MDA is that it helps to avoid tying your automated business process into a particular technology platform, and avoids the temptation to build tomorrow’s “legacy”, but very efficiently. I think that a DSL risks ending up as (in MDA terms) just a Platform Specific Viewpoint. Perhaps with a high-maturity process, however, the DSL approach will work as well as MDA.
Which sort-of brings me back to Microsoft as a successful low-maturity company (in the 1990’s at least). It is certainly aware of CMMI now (see here). But Tim also commented: “I respect [Greenfield’s] views but I did get a sense that Microsoft's strategy here is driven excessively by whichever faction happens to be in the ascendant internally. For example, apparently that is how the whole "Design for deployment" thing came about. But I guess this is not uncommon”. Not uncommon, certainly, but “faction-driven development process” is probably a feature of low-maturity organisations.