Dare Obasanjo had an interesting post yesterday on the Second System Effect in software development. For those who are unaware, the second system effect is a term first coined by Frederick Brooks in The Mythical Man Month. It deals with (in general) the idea that the second system designed/implemented by anyone is typically over-architected, with more bells and whistles added then need be.
Dare goes on to describe a number of common factors that keep systems from falling into this trap. Factors that my experience do contribute greatly to the success of a second version. I do have a couple of factors to add.
Only One Driver
There is a fundamental clash between marketing and developers when it comes to the priority of items added to a product. Marketing is looking for features that will help to drive sales. Developers are looking for features that will improve the architecture, stabilize the product, ease any future enhancements and simply be cool to implement. Frequently, these two groups will not agree 100% at the features which should be in the next release.
Successful projects have a single driver. That is, there is one person who is responsible for determine which features do and don't make the cut. They will listen to both sides of the argument and make a decision, with their ultimate responsibility being to drive the successful shipping of the release. It doesn't matter which discipline the person comes from, although it helps if the driver has the respect of both groups). The important element is to have someone who is making the decision and ensuring that the process doesn't become a continual stream of requests for new features.
Rewriting is not Right
The title should probably read "Rewriting is not Right unless you have extraordinary unit test coverage...and probably not even then", but that wasn't catchy enough.
After you come up for air at the end of a project, it is usual to have some sort of post mortem. Even the name indicates how developers look at this part of the software development process. It is not in our nature (generally speaking) to sit back and admire the good things that were accomplished. Instead, we concentrate on the warts of the system. How many times have you said, immediately after completing a task, that you wished you could rewrite it from scratch? Take solace that you're not alone in that feeling...it is quite common among your brethren.
The problem is that the feeling to rewrite needs to be resisted. There is, whether you realize it or not, more invested in the development of a particular feature than you might expect. There is more than the code that implements the feature set that is visible. There are also all of the bug fixes associated with the application. The one or two lines of code that were necessary to make the system load the poorly-formatted file that arrives monthly from your biggest client. The use of a semaphore to ensure that a timing problem was corrected. All of those little things that had to be done to take the first pass of code and make it ready for use in the real world.
When you're thinking about rewriting, your mind is focused on reworking the architecture. It is not thinking about the many hours of effort that went into identifying, replicating and correcting the bugs. We know that it's easier to write new code than to read old code, but we don't consider all of the the knowledge embedded in the old code. While throwing out old code is sometimes useful, we tend to fall back on that choice too quickly, believing that 'new' is faster than spending the time to understand the impact of enhancements and changes. If you have a set of unit tests that covers the vast majority of functionality, then you might be able to make a case. But if you don't, then rewriting part of your system should be the last choice, not the first one.