Some really bad code

Every now and then I find myself in the discussion of whether we should use comments in our code or not. I believe that comments are often misused as duct tape for bad code and limiting their use leads to higher quality code. Here is why:

  • Commenting code violates the DRY principle. By writing comments you are essentially duplicating what already is written in code. The comments will inevitably become out of date as the code evolves. Either by developers who do not bother to update the comments, or simply do not understand the consequences of a change well enough to update the comments.
  • Comments can be plain wrong! People use comments to explain what code does on a higher abstraction level, this could be because the code is really complicated (read: bad) so commenting it is justified to make the code more readable. There are however no guarantees that that the comments are correct, the only 100% correct description of what the code does is the code itself.

A pretty common counter argument I receive, which Jeff Atwood (the author of Coding horror) nicely put into words, is:

No matter how simple, concise, and clear your code may end up being, it’s impossible for code to be completely self-documenting. Comments can never be replaced by code alone.1

Interestingly enough, he uses the following example:

What is perfectly, transparently obvious to one developer may be utterly opaque to another developer who has no context. Consider this bit of commenting advice:

You may very well know that

string = join('',reverse(split('',$string)));

reverses your string, but how hard is it to insert

# Reverse the string

into your Perl file?

Yes! How hard can it be? Well, I might as well ask how hard can it be to refactor it to:

    reversedString = ReverseString($string)

Not only do we now have a reusable method for reversing strings, there is also no doubt what the code is doing.

Techniques to replace comments

I dug up some code I myself wrote a few years ago (it was actually written in objective-c for an iPhone application, but I rewrote it as C# now to save horizontal space) which we can analyze for this purpose.

public void DidUpdateLocation(Location newLocation, Location oldLocation)
{
    //Check if the horizontal accuracy indicates an invalid measurement
    if(newLocation.horizontalAccuracy < 0) 
    {
        return;
    }

    //Test the measurement to see if it is more accurate than the previous measurement
    if(BestEffortAtLocation == null || BestEffortAtLocation.horizontalAccuracy > newLocation.horizontalAccuracy) 
    {
        //Store location as best effort
        bestEffortAtLocation = newLocation;
        //Test the measurement to see if it meets the desired accuracy
        if(newLocation.horizontalAccuracy <= LocationManager.desiredAccuracy) 
        {
            //We have a measurement that meets our requirements and can stop updating the location
            LocationManager.StopUpdatingLocation();
            CurrentLocation = newLocation;
            PostNotification(Event.LocationFound)
        }
    }
}

The code is cluttered with comments, everything that seems the least complicated has an explanatory comment and most of them are simply restating the following line of code in plain English. So in order to improve the code, the duplicate lines can simply be removed:

//Store location as best effort
bestEffortAtLocation = newLocation;

But by far the best refactoring that can be done is to replace comments with well named functions that explain what is done:

//Check if the horizontal accuracy indicates an invalid measurement
if(newLocation.horizontalAccuracy < 0) 
{
    return;
}

Becomes

if(IsInvalidMeasurement(newLocation)) 
{
    return;
}

Now the code reads like a book, just go ahead and try this trick in any complicated function with loads of conditional statements, you will be surprised how much this improves readability. It also encapsulates errors, making stack traces point at the exact location of for instance a null exception. Additionally the methods are really easy to unit test! Because usually they have easily definable input/output data.

Finally, fully refactored the method could look like the following:

public void DidUpdateLocation(Location newLocation, Location oldLocation)
{
    if(IsInvalidMeasurement(newLocation)) 
    {
        return;
    }

    if(IsMoreAccurate(newLocation, BestEffortAtLocation)) 
    {
        bestEffortAtLocation = newLocation;
        
        if(newLocation.horizontalAccuracy <= LocationManager.desiredAccuracy) 
        {
            LocationManager.StopUpdatingLocation();
            CurrentLocation = newLocation;
            PostNotification(Event.LocationFound)
        }
    }
}
    

Don’t you agree that this is easier to read?

If comments are so bad, why do people still use them?

In the academia I was taught that code without comments is bad because it is hard to read and understand. I used to believe in this statement in school because in that environment it was somewhat true. We programmed a lot with C and somehow the praxis when teaching C (maybe because of its heritage as an old school language) is to use short and abstract variable names. It seems like the more theoretical a subject is, the worse the code is. For instance, let us take a look at the Wikipedia page for Merge sort, it is interestingly common to use bad variable names such as j, B, i1, A and crowd the code with comments.

// left half is A[iBegin :iMiddle-1]
// right half is A[iMiddle:iEnd-1   ]
TopDownMerge(A[], iBegin, iMiddle, iEnd, B[])
{
    i0 = iBegin, i1 = iMiddle;

    // While there are elements in the left or right runs
    for (j = iBegin; j < iEnd; j++) {
        // If left run head exists and is <= existing right run head.
        if (i0 < iMiddle && (i1 >= iEnd || A[i0] <= A[i1]))
            B[j] = A[i0];
            i0 = i0 + 1;
        else
            B[j] = A[i1];
            i1 = i1 + 1;    
    } 
}
    

It makes sense to use general variable names in cases like this, because the purpose is to teach an algorithm, not how to write clean code. The problem is, that if this is the only thing students see they start to write similar code. And when the code is abstract and hard to understand, it also makes sense to write comments that explain what is happening at a higher abstraction level. To make matters worse, we were all taught to write loads of comments in our code, so much that we even failed our assignments if they were not “properly commented”. I also believe that the learning material is crowded with comments simply because of teaching purposes, the author helps the students understand important concepts within the code. Dave Thomas also made the same observation in Pragmatic Programmer:

Programmers are taught to comment their code: good code has lots of comments. Unfortunately, they are never taught why code needs comments: bad code requires lots of comments.2

I also did some digging in one of my favorite books, Code Complete to see what Steve McConnell’s view on commenting code is, and he somewhat agrees:

Comments are used to explain difficult code. Comments have an important role to play, but they should not be used as a crutch to explain bad code. The age-old wisdom is dead-on: “Don’t document bad code - rewrite it” (Kernighan and Plauger 1978).3

There you have it again, if you have read anything he has written you know how focused he is on writing good quality code.

So my point is that most programmers are taught from the beginning to become vivid commenters and therefore need to be convinced or taught these ideas.

But we write comments anyway

Do you agree at this point that comments are a code smell? I do realize that even though we have our pragmatic views and principles about coding practices, only a fool would follow them blindly. We live in a world with loads of stress and tight deadlines, the code we write is not always perfect and rightfully so, sometimes it is more important to ship on time than to write perfect code. Perhaps my favorite quote about this subject is from Clean Code:

The proper use of comments is to compensate for our failure to express ourselves in code. Note that I used the word failure. I meant it. Comments are always failures.

So given the time frame to write the code, if by flipping all the stones you still do not feel confident enough to leave the code without an explanatory comment, you should rightfully feel bad. You failed to express yourself with your craft, but at least you know it is bad and might come back some day to fix it.

Update

I just wanted to clarify a point I feel some of the readers are missing. I am by no means claiming that you should never write comments, it is obviously pretty easy to figure out an edge case which countradicts my thesis. Where not writing a comment would be plain stupid. If you for instance:

  • Find a bug in .NET 3.5 and implement an obscure work around, write a comment! Otherwise someone might remove it because it looks like bad code.
  • You make a complicated implementation, perhaps because of performance reasons that might be extremely hard to understand, do not hesitate to write a comment!

The point is, all we strive for is to make the code more readable and understandable, if it requires a comment, go for it! But it should not be an excuse for bad code.

Sources

  1. (Jeff Atwood, Code Tells You How, Comments Tell You Why, Coding Horror, 2006, URL, (accessed 05 August 2015)) 

  2. (Andrew Hunt, David Thomas. Pragmatic Programmer: From Journeyman to Master, Adison Wesley Professional, 1999) 

  3. (Steve McConnell. Code Complete: A Practical Handbook of Software Construction, 2nd Edition, Microsoft Press, 2004)