I came across this tweet from Oracle Developers (Oracle Developers Tweet) which picked up on a post from the NoBugsProject about common errors with the use of exceptions. One of the first errors the article described is one of my pet hates – the use of standard exceptions for application specific errors.
This took me onto one of my other pet hates – code without any documentation. I’m not advocating the days of waterfall Development where reams of documentation had to be produced before a single line of Code was written. In fact this is in my option worse than nothing as the docs would often not match the coded reality. But I absolutely agree with the agile manifesto statement:
We value working code over documentation
This doesn’t say no documentation, despite the fact that I have encountered more times than I care to recall the use of this statement to justify not documenting code. So what is the right balance?
We want to save effort from Code reviews and get clean code by using static code analysis but it doesn’t have the ability to apply smarts as what needs documenting? Pair programming is rarely practised, and there is plenty of psychology about group behaviour that can undermine documentation in a pair working approach effort. So what is the answer?
Well, I’ve always applied a couple of personal rules of thumb that can be measured with static code analysis particularly if you use conventional documentation tags. The rules are:
- Interfaces warrant an interface level description of the interface purpose. It’s always helpful to describe/illustrate with use example. This is code equivalent to a good API Blueprint or swagger doc.
- Provide a class level description of what the class is for – if it is a DAO then just say what the entity is.
- If a class is part of a pattern, name the pattern. This is most important when relating to supporting a composite or solution pattern. Remember there will always situations where a newbie will get asked to extend or change your code, help them. Remember not every developer is as experienced or clever as you. If in doubt, give your code to someone who doesn’t know what you’re working on and ask them to explain what your code is doing and why. I had once, had to create a JDBC abstraction layer as we needed to support multiple databases. But if you know JDBC you’ll be aware of there are some subtle but important differences in implementation of connectors. I took the time to explain it in the interface header. I know a couple of developers appreciated the investment of 5 minutes.
- If you have a function that has a code analytics score such as cyclometeric then describe the function. Use the comment to convey why the high score is justifiable.
- If the code has specific dependencies or has to perform in a very specific sequence a short comment will help, and anyone going through refactoring code.
With these guidelines it becomes possible to then use javadoc tools to generate your documentation. It doesn’t require you to go find a word document or a wiki page to update the documentation. Of course then reviewing the generated documentation will soon help you finesse the process of documenting in a manner that is whilst light also supports readability without needing the code.
For those, who still disagree I would say …
- Do you want to be maintaining and updating the same code for the rest of your career to meet new minor changes etc?
- Not everyone is a great coder like you, do you want someone less capable who may have to make a change messing up your elegant code?
- Sooner or later someone will ask you to fix or enhance some code that in your eyes is a chaotic unintelligible mess, I’m sure you’d appreciate some comments that will help you understand what the developer was trying to do? We can’t expect those not so good at the craft to document if the best of us are not prepared to do so.
If you don’t agree, or have found different approaches that ensure enough accurate documentation, please share.