I was originally going to talk about API: Design Matters and Cocoa, but, and I believe the title of this post may give this away, I’m not going to now. That’s made its way into OmniFocus though, so I’ll do it sooner or later. No, today I’m more likely to talk about The Cathedral and the Bazaar, even though that doesn’t seem to fit the context of requirements gathering.
So I’ve been reading a few papers on Requirements Engineering today, most notably Goguen’s The Dry and the Wet. One of the more interesting and subtle conclusions to draw from such a source (or at least, it’s subtle if you’re a physics graduate who drifted into Software Engineering without remembering to stop being a physicist) is the amount of amount of political influence in requirements engineering. Given that it costs a couple of orders of magnitude more to mend a broken requirement in maintenance than in requirements-gathering (Boehm knew this back in 1976), you’d think that analysts would certainly leave their own convictions at the door, and would try to avoid the "write software that management would like to buy" trap too.
There are, roughly speaking, three approaches to requirements elicitation. Firstly, the dry, unitarian approach where you assume that like a sculpture in a block of marble, there is a single "ideal" system waiting to be discovered and documented. Then there’s the postmodern approach, in which any kind of interaction between actors and other actors, or actors and the system, is determined entirely by the instantaneous feelings of the actors and is neither static nor repeatable. The key benefit brought by this postmodern approach is that you get to throw out any idea that the requirements can be baselined, frozen, or in any other way rendered static to please the management.
[That’s where my oblique CatB reference comes in – the Unitary analysis model is similar to ESR’s cathedral, and is pretty much as much of a straw man in that ‘purely’ Unitary requirements are seldom seen in the real world; and the postmodern model is similar to ESR’s bazaar, and is similarly infrequent in its pure form. The only examples I can think of where postmodern requirements engineering would be at all applicable are in social collaboration tools such as Facebook or Git.]
Most real requirements engineering work takes place in the third, intermediate realm; that which acknowledges that there is a plurality among the stakeholders identified in the project (i.e. that the end-user has different goals from his manager, and she has different goals than the CEO), and models the interactions between them in defining the requirements. Now, in this realm software engineering goes all quantum; there aren’t any requirements until you look for them, and the value of the requirements is modified by the act of observation. A requirement is generated by the interaction between the stakeholders and the analyst, it isn’t an intrinsic property of the system under interaction.
And this is where the political stuff comes in. Depending on your interaction model, you’ll get different requirements for the same system. For instance, if you’re of the opinion that the manager-charge interaction takes on a Marxist or divisive role, you’ll get different requirements than if you use an anarchic model. That’s probably why Facebook and Lotus Notes are completely different applications, even though they really solve the same problem.
Well, in fact, Notes and Facebook solve different problems, which brings us back to a point I raised in the second paragraph. Facebook solves the "I want to keep in contact with a bunch of people" problem, while Notes solves the "we want to sell a CSCW solution to IT managers" problem. Which is itself a manifestation of the political problem described over the last few paragraphs, in that it represents a distortion of the interaction between actors in the target environment. Of course, even when that interaction is modelled correctly (or at least with sufficient accuracy and precision), it’s only valid as long as the social structure of the target environment doesn’t change – or some other customer with a similar social structure comes along ;-)
This is where I think that the Indie approach common in Mac application development has a big advantage. Many of the Indie Mac shops are writing software for themselves and perhaps a small band of friends, so the only distortion of the customer model which could occur would be if the developer had a false opinion of their own capabilities. There’s also the possibility to put too many “developer-user” features in, but as long as there’s competition pushing down the complexity of everybody’s apps, that will probably be mitigated.