Legacy projects & the maintainer

Aug 14, 2015

Code has an incredible ability to survive far beyond the expectations of its original authors. There are a multitude of reasons for this, but foremost is simply that at some point software becomes so feature-rich that trying to reimplement it would be a costly endeavor, and would provide no guarantee that the new software would be superior quality to its predecessor. There are many reasons that your bank still programs modules in COBOL, but the language’s technical merits aren’t one of them.

By extension, almost all of us will at some point (as we start a new job or take over maintenance of an open-source project) have to work in a codebase that we would consider to be “legacy”, a term that we apply because in our minds it’s not up to the same exacting standards that we would demand were we to build it again today. Simultaneously though, most of us understand that this is a falsehood; poor quality software rarely got the way it is due to malicious intent or sheer incompetence by its original authors, but rather from slow degradation over time as it continued to accumulate new capabilities that didn’t fit well into the original framework.

Developers understand this on some level, but the natural instinct of the majority of us is still to treat any less-than-optimal projects that we come across as a dumping ground for low quality patches. This causes these projects to succumb to the tragedy of the commons and accelerate their free fall in quality. This is detrimental to any company or organization – if a project is worthy of attention and contribution, it’s probably also relevant to the bottom line.

The obvious advice here is to encourage everyone to treat it like it’s their own and to try and make sure that every new patch leaves the code in a better state than it’d previously been. This is a good general guideline, but also simultaneously a platitude. Although unquestionably an admirable aim, a little like the case of a popular summertime park that’s left with excess trash at the end of the day, it’s unlikely that every developer in your organization will take it upon themselves equally to help improve things.

A more pragmatic step would be elect a maintainer for the project. This might also seem like an obvious step, but keep this in mind: you’re not just looking for somebody to merge pull requests, you’re looking for an overseer who cares about the project on a deep level, and who strive continuously to help build a better future for it. Once other contributors realize that patches are now being rejected on the basis of quality, it doesn’t take long for things to start improving.

Finding this person is the first step; empowering them to do their job effectively is the second. This involves a few things:

  1. Give them the slack time necessary to build improvements.
  2. Give them the political capital necessary to apply backpressure even in the face of overwhelming force.

It’s the second point here which is challenging. Especially in the private world where hard deadlines are around every corner, it’s tempting to justify inferior quality with a line like, “We just don’t have time to do this the right way right now. We’ll improve it later.” As most engineers with a few years of experience under their belts can tell you, that “later” will never come. The best way to stop that debt from piling up is to stop it in the present, not to offload it to an imaginary future of relative idleness. A maintainer must be able to call out problems and apply resistance even in the face of important company directives.

The key to improving that legacy codebase probably isn’t a rewrite. It’s a reversal of its downward trend in quality to one of slow but steady improvement. If you keep it going long enough, one day you can even drop the “legacy” moniker entirely and be left with something that’s pleasurable to work with.

Did I make a mistake? Please consider sending a pull request.