Blog: Capturing and testing Business Policy.
Perhaps the biggest issue with legacy is separating out the business policies it embodies
I've just had an interesting email from Bill Nicholls in Washington State prompted by my piece on legacy systems here.
He wants me to put more emphasis on the “testing” phase, although he doesn’t like calling it “testing” because (I think) “testing” no longer implies (if it ever did) any particular attention to test design, coverage or completeness.
I have some sympathy with this view – I’ve met people who think that running a few million more-or-less sanitised “live production data” through a system is an adequate test before going into production. Well, it tests something, but it may only represent a very few equivalence classes repeated many times and the sanitising process (usually about making account numbers, names and personal data unrecognisable) may hide many data-dependent defects. So, after such testing, what can you really say about the risk associated with going live? Not much – so perhaps formal “quality assurance” is a better goal.
Bill identifies a real issue, which I think may be addressed in “model driven development processes” (such as the OMG’s MDA) used properly, as these should help to abstract business policy from the generated code. He says:
“The real killer problem in systems design and programming IMNSHO is dealing with policy changes which happen after a program is in production, usually encapsulating in hard code the then current version of policy. This particular bear trap can be devilishly difficult to get out of, sometimes requiring major code changes for a minor policy change. Policy really has to be isolated in systems. It is a part of the design process that requires high level input from management, rarely gotten in my experience. And it should be isolated in the software so that it may be easily changed, as parameters if possible, by small [discrete] code changes if not. The actual processing steps are relatively simple, it's all the policy and exception handling that generates the hardest code to verify.”
Amen to that – and it is especially relevant in modern loosely-coupled “service oriented” systems, where much of the complexity and opportunity for expensive defects may lie on the “dark path” of compensating transactions, invoked when something goes wrong. Just what is the business’ policy concerning compensating customers for errors? And, remember, that developers tend to concentrate on the positive, the “light path” through their systems; it’s the nasty cynical QA people who say “but what if…”.
Bill suggests that top (business) manager of the group being automated (or updated), someone empowered to make policy decisions which will stick, should to be included in design and reviews of the policy part of an automated system - because that is the part that they will change on minimal notice. That’s a familiar idea, because in a past life I was involved in introducing a JAD process (Joint Application Design, originally promoted by IBM, I think), which (as I understood it) involved locking the top managers (empowered to make policy decisions) with the developers, a "scribe" and a "CASE" tool in a hotel room, until the essential business policy framework for the development was captured.
Top managers couldn't be "captured" by the analysts on an ad hoc basis as needed; but they could schedule a concentrated requirements capture session. As I met it, the process was soon compromised as IT management allowed top managers to send along non-empowered representatives, which then allowed IT to get its own agenda "rubber stamped". However, I did learn something from the experience, if only that you can achieve nothing without informed top-management buy-in.
And I think that if you think that capturing policy separate from code is important, you should have a look at JAD. Or MDA. Or, perhaps the DSDM (Dynamic System Development) process; it's not anything much like JAD, but I think it has learned both from the JAD legacy and from Agile practice...