nav search
Data Center Software Security Transformation DevOps Business Personal Tech Science Emergent Tech Bootnotes BOFH

Sweet, sweet smell of comments in code?

Extreme Programming breeds overreaction

By Matt Stephens, 18 Mar 2008

Arguments rage over the importance of adding comments to your code versus the importance of writing clear code that speaks for itself, thereby potentially eliminating the need for comments. The dichotomy boils down to this: writing comments versus writing self-commenting code, as if comments and clear code are somehow mutually exclusive.

Before I get into the cut and thrust of this debate, it's worth stating that the subject of comments is less of an important issue than writing good code. While comments have value, if the code is crufty and convoluted, adding a few paragraphs of comments - while well intentioned - are only likely to add to the noise.

The Extreme Programming movement, back in its heyday, popularized the notion that comments in code are bad, taking the position that if you see a comment then it must be a "code smell" and you should grab your pair-programming buddy and refactor the stinky code immediately. The term used was "coding by intention" - writing code that so obviously communicates its method and purpose that it doesn't need to be commented.

So instead of this:

int x=0;   // day of month, 0 if not found
for (Rabbit rabbit : warren) { // check for a white rabbit
    if (Color.WHITE.equals(rabbit.getColor()) {
        x=1;   // First day of month,
        break; // so hop out of loop
// Show msg indicating start of month or not:
System.out.println( (x==1)?
    "It's the first day of the month!" :
    "Sorry, it isn't the first day of the month."

You have this:

boolean firstDay = isFirstDayOfMonth();

(and the obvious extracted methods...)

Note that isFirstDayOfMonth() isn't called something like checkForWhiteRabbit() because the "how" (or method) of determining whether it's the first day of the month is encapsulated into the method itself. This goes some way towards explaining why these are called methods.

The notion seems reasonable, and in fact it's a good guideline to follow - up to a point. As with many of its practices, XP spoiled a good guideline by taking it to extremes, turning the knob all the way up to 10: comments are a "code smell" and must be refactored out; if you spot a code smell, refactor it immediately.

But however vocally XPers argue against commenting - or over commenting - your code, the opposite camp can argue pretty loudly too.

Here are some of the arguments for comments in code:

  • Months later, someone will look at your unclear or uncommented code, they will go out, buy a shotgun and hunt you down.
  • Clear code conveys the "what" and the "how", but not the "why" - the crucial missing context that comments can and should supply. My personal answer to this argument is - if you'll excuse the book plug: write lightweight use cases (or user stories) and tie them closely to the objects. That gives you the "why" with traceability and a method of driving functional/unit tests from your use cases, for very little extra effort.
  • Sometimes you need to work around an API glitch, or your code steps outside the obvious, and you want to communicate why.
  • If you're writing an API, the public methods need to be liberally commented so, for example, you can give the Javadoc tool something decent to munch on.

Sounds reasonable? Here, then, are some of the arguments for not adding comments:

  • They are not provably correct - if you write a unit test instead of a comment, though, the test is correct. Note this argument is logical bunkum, because a test's pass/fail isn't the same as provable correctness. It's provable passing/failing - also important, but different.
  • A comment in code is a sign that the code isn't clear enough.
  • Comments become out of date because they don't fail at compile- or run-time like out-of-date code.
  • Sometimes comments are copied and pasted along with copied and pasted code. That means the code is then changed but the comments aren't, resulting in comments that essentially tell lies about the code.
  • Comments are often redundant, simply paraphrasing what the code already says ("// increment x").

Clearly, there are compelling arguments on both sides. But you should also be able to see a pattern emerge from the arguments. They're talking about particular conditions when the argument holds true.

So if it's only sometimes vitally important to comment your code or to write clearer code/tests instead of comments, why do the proponents of either side argue so vehemently for their cause? I'll delve further next time.®

Matt Stephens is co-author of Extreme Programming Refactored, which skewers XP like a rodent on a toasting fork.

The Register - Independent news and views for the tech community. Part of Situation Publishing