Don’t Let Your Code Marry An Axe Murderer

It’s a lesson I learned from the school of hard knocks: be careful about the shady characters you let into your software.

There are lots of software frameworks, components, and tools in the world – and they all want you to think you can live with them happily ever after. They seduce you with the promise of productivity, then hack you to death once they’ve gained your trust.

Here are a few questions I’ve learned to ask when trying to spot next year’s problems:

1. How easy is it to deploy?

I’ve been burned by software that is difficult to move from one environment to the next. Some products, particularly server products, assume you never need to move software into different environments (like development, test, productions). Avoid anything you can’t automate!

Software should be as easy to deploy as possible, otherwise it doesn’t get deployed often enough.

2. Will it work in source control?

Some tools and designers generate files that don’t work well under source control. If you can’t diff and merge you lose a tremendous amount of flexibility when it comes time to branch, patch, and support concurrent development. Database tools are historically notorious for having problems in this area because they tend to design for DBAs instead of developers.

It is, however, perfectly reasonable to check in binary dependencies.

3. Is it needy?

Speaking of dependencies – they are a touchy subject. You can’t invoke the “not invented here” rule in every decision without writing your own compiler. At the same time, every dependency you introduce to a product is a calculated risk. How many dependencies does this new thing have? Will you ever delay shipping because you are waiting for a new version of a dependency? Will you ever delay shipping because you are waiting for a new version of a dependency’s dependency? How many things will break when the next version arrives? Are you willing to run beta code, or code from the trunk? The answer is different for every application and business.

In the end you need to make a decision from the gut, which means research into the history and pedigree of a dependency might be required.

4. When will it be obsolete?

If you implement the solution to a problem in a mainstream programming language like C#, you can be reasonably sure it will still work and compile in 5 -10 years, even with new versions of the compiler. But, if you implement the same solution with Data Transformation Services (wait, that’s obsolete), or Notification Services (wait, that disappeared), or SQL Distributed Management Objects (oops, it’s only a matter of time) - then you got some rewritin’ to do!

Remember – nothing lasts forever … unless you have all the source code.