Always be shipping

Building software is fun.

It’s great to develop something from nothing, or extend something to do more than it did before.

I think coders want to build things the best way they can. But in practice, that doesn’t always happen. Shortcuts can be taken, perhaps due to time pressures, or major requirements popping up late into a piece of work. Sometimes, you need to make changes to software that are quite different to what it was built for to begin with.

But software changes, and we can’t always anticipate how it’s going to change. In fact, we probably shouldn’t worry too much about how it might change in ways we simply don’t know about yet. (It’s a bit different if it’s your product, such as a hobby project, and you know some of what you’ll be adding later – and can build for those things accordingly.)

Perfect code

We shouldn’t be aiming for perfect code. Equally, we shouldn’t be happy with crappy code, but that’s not the point here.

Delivery – shipping a product or feature – is vital. It doesn’t matter how beautiful your code is if you never ship.

It definitely matters if you build a mess of a system that will be painful to work with in the future. But we shouldn’t be agonising over making everything perfect before anything is shipped.

I know there are developers who will find some of my code from years ago and say “what was he thinking?!” And to be honest, if I read it today I’d probably think something similar. I’m sure I could think of ways to improve the code I worked on back then that I didn’t think of at the time.

But the important thing is that the code shipped, we could monitor how things went in production, and we could get stuck into other projects.

Continuous improvement

Continuous improvement is a wonderful thing. Some of my favourite projects have been framework migrations. I’ve migrated from a proprietary framework to Zend v1; I’ve migrated from a basic home-grown set of PHP classes to an MVC framework; I’ve managed a migration from Drupal to Laravel. More recently, I’ve worked with a team who have been migrating from a FuelPHP monolith to a service-oriented architecture.

With large-scale migrations or any major refactoring, breaking off components of the application and migrating them piece by piece is an excellent way to minimise risk, keep momentum going, and avoid a seemingly infinite end date. The concept of “dropping” a massive new upgrade on a given date for an entire customer base is extremely risky, and runs the risk of being rolled back if it all goes pear-shaped.

Testing

There’s a phrase I have in my head at all times as a reminder of why “big bang” launches are so dangerous:

It is difficult to know when testing is complete.

Testing is not something you start and finish. Testing is an ongoing process, a discipline, that can incorporate a selection of the following at various times:

Automated unit tests
Automated integration tests
Manual exploratory tests
Manual functional tests
Manual or automated regression tests
Manual UI tests
Non-functional tests (a large category)

Testing helps to highlight everything from catastrophic failures to tiny UX quirks. But having tests doesn’t mean your big bang implementation will be fine – especially in a big, complex system.

Scrum

In Scrum, the output of a sprint is a potentially shippable product increment. Whether you ship or not is up to you. But the point is that you can ship if you want to. This capability should be available to the Product Owner.

When a development team thinks they have 6-12 months, or years, to deliver a project (or maybe there’s no end date at all), it’s can be very tricky to keep the team focused. Agile accepts that requirements change frequently. This in itself is not an instant issue. However, when you have a long-running project that isn’t shipping anything, and constantly changing requirements, especially when development is already in progress – that’s a recipe for disaster.