Thursday, April 21, 2011

Industrial Strength Requirements

A while ago I was in a discussion with one of my friends. He said something interesting that really caught my attention. His code was in testing and had turned up some corner-cases. While developing the code he became aware of these issues, but as he said “they weren’t in the requirements”, so he didn’t address them.

I found myself surprised. There is -- I strongly believe -- for all code an inherent set of basic requirements that goes without saying. The users don’t need to specify them, they just come in with the expectation that the code will be able to withstand normal wear and tear. That it will be usable.

We don’t, for instance, need to specify that a GUI should be easy to use. That requirement is fundamental, it is right there in the definition of GUI, and we certainly don’t need people to iterate what that means in general.

We don’t need people to specify that a batch script should be restartable and idempotent. Both qualities go back decades and are important to insuring that the system is operationally sound.

We don’t need people to specify that installation should be simple and repeatable. Those were lessons learned a long time ago.

I could go on, there is a massive amount of these well-known ‘best practices’, ‘conventions’, ‘acceptable behavior’, etc. We already know a huge amount about the proper way to build software so that it lives up to its usage; so that it is industrial strength.

A set of requirements from the users, managers, or whoever, is a set of things that are merged with this existing knowledge base. These default requirements. And for experienced developers, although there are no official standards for these necessities, we know what they are and why they are important.

When we are commissioned to write or enhance a code base,  these fundamentals need to be in place. They don’t need to be iterated by non-experts, they just need to be there.

On occasion, a few of the stakeholder requirements may contradict the default ones. That happens, particularly when stakeholders don’t have enough experience with software. In such collisions, the default requirements hold precedence, and it is up to the programmers to bring this issue to the stakeholders. If there are good solid reasons for altering the defaults and someone make a reasonable case, then it can be done in that way, but otherwise we’re professional software developers so we need to build things that work. So the defaults are important.

For me, in my friends situation, I would have gone back early with the corner-cases and the way I intended to solve the problems. I’d be open for debate, but I prefer to go in with a viable solution, it makes it faster to get an agreement. If it’s a domain issue, since I’m not the expert, I would be extremely flexible to what they need. With a technical issue, I would insure that any alternatives don’t compromise the system. But in either case, one should never violate good practices unless there is a valid reason to do so. The most important aspect is that the issue should come up for discussion, and that the resolution should be noted and followed.

Requirements in that sense, are those things above and beyond the fact that we are building some industrial-strength software. Even before the project was a glimmer in some-one’s eye there is a huge amount of basic work necessary to make it successful. We shouldn’t dump the responsibility for knowing what that is on the stakeholders, it’s part of our job.