You're (Probably) Documenting That Wrong

Programmers Don't Like to Document Code

Code documentation is one of those tasks that software developers like to slack on. It's not the documentation that's important, it is the code that is important! Lots of developers either don't add documentation blocks or fill them in with only the basic amount of information. This post will look to address bad software documentation habits and how they can be improved.

Why Should We Document Code?

I've heard lots of arguments as to why writing code documentation is a waste of time. It takes too much time to write! The code is self-documenting! It will just be out of date by the next release, so why bother! Sure, documentation is time consuming. Agreed, well written code can be partially self documenting. Yes, if you don't have good habits in place the documentation will be out of date.

A lot of the arguments are perfectly valid if you are documenting the wrong things. Good code documentation provides a clear understanding of the contract that code will adhere to. Good code documentation will alert consumers of the code of the "gotchas" that can crop up from using it. Good code documentation will serve as a reminder for why something was done or not done.

What Should You Be Documenting Internally?

By internally I mean inside functions. This refers to code that is not part of an interface, but in the lowest level blocks of code that only developers on the same project will ever be able to see.

Internal Code that Is Hard to Understand

First, you don't need to be documenting each and every line of code in a function with what it does. If the code is not clear enough, it should be rewritten. One of the "code smells" from Martin Fowler's book Refactoring: Improving the Design of Existing Code is too much inline documentation. If you think you need to write a lengthy treatise, first consider renaming some variables for clarity or extracting the block into a well-named method with its own documentation. That's not to say that there aren't cases where you need internal developer documentation. If the code is difficult to understand because it is a complex formula or a particularly tricky regular expression, then by all means throw some comments above it. Just don't do this:

// Adds one and one together and saves the result in a variable
int two = 1 + 1;

The "Why" and not the "How"


So we've covered not documenting how something is done unless absolutely necessary, but sometimes comments are still needed to explain why something is being done. Often far more important than how you did something is why you did it, especially if it is not immediately clear to someone unfamiliar with the code. Sometimes comments are needed because the code looks funny, or relies on a side effect elsewhere that most readers of the code wouldn't know about. Comments like the following can be extremely useful for other team members (or yourself weeks later when you forget why you wrote the code that way):

// Subtraction must be performed first to prevent an 
// off-by-one error

// The list is guaranteed to be pre-sorted as a side 
// effect of validation. We do not need to sort the 
// list again here.

// We are adding the initial database entry here 
// because we need the unique ID for <reason> before 
// we will have status. The actual database entry will
// be updated after processing with the actual status.

What Should You Be Documenting Externally?

Far more important than what you document internally is what you document externally. By externally, I mean an interface exposed to consumers (either internal through private methods, or external through public methods). While the code's interface provides a contract between the code and its consumer, the external documentation provides the finer details of that contract.

What You Should Definitely NOT Document

Before we dive in to what you should document, I think it is very important to call out exactly what you should not document. Never expose implementation details in your external documentation. Doing so can lead to a lot of the issues that developers complain about. Internal details are far more likely to change and there's a good chance someone will forget to update the documentation. Even worse, a consumer of the code might make assumptions based on the details, so if the implementation does change the user's code might be broken. Consider the following:

/**
 * Returns an iterable object that can be used to access the 
 * individual line items in the order that have been sorted 
 * using a bubble sort.
 *
 * @return a LinkedHashSet object that provides access to 
 *         the sorted line items in the order as an iterator.
 */
public Iterable<LineItem> getSortedLineItems();

Not the best interface, but it serves a point. The only method you can call on Iterable is iterator(), so that is all that is really exposed to the consumer. Except, the documentation explicitly calls out which Iterable will be returned. Since the documentation is a contract, if you change the implementation of the Iterable from LinkedHashSet you will be breaking that contract. The sorting method mentioned is purely fluff; all the consumer really needs to know is that the items in the iterator are sorted. Now consider what a consumer can do with this:

LinkedHashSet<LineItem> sortedLineItems = 
  (LinkedHashSet<LineItem>) obj.getSortedLineItems();
sortedLineItems.add(new LineItem(...));

Most likely we don't want users adding new line items to the collection of sorted ones, but by documenting the actual type we allow users who want to be a little risky to try to anyway. I say risky because a few months down the road you may realize that line items don't have to be unique and so a set is not appropriate, so you switch to a LinkedList instead. You may also realize that the bubble sort is not as efficient as a quick sort and change that too. In both cases the documentation would have to be updated, and in the first case you may have just broken a consumer relying on the original documentation.

In general, make sure your code documentation is a black box that is only exposing information that its consumers really need. Avoid mentioning specific classes used for interfaces (I see ArrayList called out a lot for the List return type). Also avoid describing the algorithms used, since you might want to change them later.

Document At Minimum What the Documentation Tool Supports

So what do you want to document? Most programming languages have a documentation system that can be used to generate detailed developer documentation. Java has Javadoc, there is XML documentation for .NET, and C++ and C code can use Doxygen. In most of those cases the documentation will include a description of the class or method, any parameters it takes, the return value, and any exceptions it might throw. Make sure that you document all of these things at a very minimum.

Document Units of Measure

Here's something I see developers trip up on all the time. Consider that you're going to sell your software and I come up and hand you a contract offering to buy it for one million! Would you accept that offer? You shouldn't! One million dollars? One million pesos? One million bottle caps? There's a very vital piece of information missing from that contract: the units. At least one ~$200M space craft was lost due to using the wrong units of measure. Now considering the following piece of code:

/**
 * Calculates the amount of time it takes to travel the 
 * distance at the speed provided.
 *
 * @param distance the distance that will be travelled.
 * @param speed the speed at which is being travelled.
 *
 * @return the amount of time it will take to travel the
 *        distance at the provided speed.
 */
public static double calculateTravelTime(double distance, 
                                         double speed);

Does that look fairly standard to you. I've seen code like it at least a half of a dozen times. Since the person who wrote that was probably also the consumer it made perfect sense to him at the time. That is not a good specification though. What is the distance measured in? What is the speed measured in? What unit of time is being returned? Now consider the following:

/**
 * Calculates the amount of of time it takes to travel the 
 * number of kilometers at the speed provided.
 *
 * @param distanceKm the distance (in kilometers) that will 
 *        be travelled.
 * @param speedKph the speed (in kilometers per hour) at which 
 *        the distance is being travelled.
 *
 * @return the amount of time (in minutes) it will take to 
 *         travel the distance at the provided speed.
 */
public static double calculateTravelTime(double distanceKm, 
                                         double speedKph);

That looks a lot better and provides the missing information that consumers will need to call the method. Common places to look for missing units of measure include times, distances, sizes, amounts, weights, temperatures, and screen measurements (em vs. pixel).

Assumptions About Parameters

When we write new functions we automatically make a lot of unconscious assumptions about the parameters, especially if we are also writing the code that will be consuming that function. A lot of common assumptions we make are about the format or expected values of parameters. We expect that certain parameters will never be null. We expect that the string provided representing a phone number will be in the form "XXX-XXX-XXXX" and not "(XXX) XXX-XXXX". We expect that the number of items the user wants to add to their cart is a positive number. We expect the user's GPS latitude value to be between the range of -90 and 90. Many times we test for these assumptions, and some times we don't. The import thing is to document those assumptions though, so that consumers can know why something went wrong when they pass in bad values.

I've found it is a good practice to always document any restrictions on parameters next to the parameters themselves. I usually do it in parenthesis following the parameter description. Here is an example:

/**
 * Records the GPS location of a phone at a specific time.
 *
 * @param trackingTime a timestamp indicating when the phone's 
 *        GPS data was recorded (must not be null, must be in 
 *        the format "MM/dd/yyyy hh:mm:ss" with hours in 
 *        24-hour (0-23) military time).
 * @param phoneNumber the number of the phone being tracked 
 *        (must not be null, must be in the format
 *        "XXX-XXX-XXXX").
 * @param gpsLatitude the latitude of the phone at the time it 
 *        was recorded (must be in the valid range of -90 to 
 *        90 degrees).
 * @param gpsLongitude the longitude of the phone at the time 
 *        it was recorded (must be in the valid range of -180 
 *        to 180 degrees).
 *
 * @throws NullPointerException if any parameter is null.
 * @throws IllegalArgumentException if trackingTime 
 *         or phoneNumber are not formatted 
 *         properly, or if gpsLatitude or 
 *         gpsLongitude is not in the valid range 
 *         of degrees.
 */
public static void trackPhoneLocation(String trackingTime,
                                      String phoneNumber,
                                      double gpsLatitude,
                                      double gpsLongitude);

The Reason for Thrown Exceptions

I see this one quite a lot. Take a look at this example:

/**
 * Updates the note about a customer in the database.
 *
 * @param customerId the ID of the customer for which 
 *        the note will be updated (must be a valid
 *        customer ID).
 * @param note the new note for the customer (must 
 *        not be null, must not exceed 255 characters).
 *
 * @throws NullPointerException
 * @throws IllegalArgumentException
 * @throws NamingException
 * @throws SQLException
 */
public void updateCustomerNote(int customerId, 
                               String note);

It is nice that at least you know what checked and unchecked exceptions can be thrown from the method. Unfortunately, that's all you know. If a consumer does catch a NullPointerException thrown from this method, will they know why (especially if the message is the wonderful default "null")? Just as important as listing the exceptions is to list why they are thrown so consumers can troubleshoot their code. Consider the following revised version instead:

/**
 * Updates the note about a customer in the database.
 *
 * @param customerId the ID of the customer for which 
 *        the note will be updated (must be a valid
 *        customer ID).
 * @param note the new note for the customer (must 
 *        not be null, must not exceed 255 characters).
 *
 * @throws NullPointerException if the note 
 *         parameter is null.
 * @throws IllegalArgumentException if the 
 *         note parameter exceeds its maximum
 *         length or if customerId is not a 
 *         valid customer ID.
 * @throws NamingException if there is an looking up the 
 *         database connection details from the naming 
 *         context.
 * @throws SQLException if there is an error connecting 
 *         to the database or updating the customer
 *         record.
 */
public void updateCustomerNote(int customerId, String note);

That adds some clarity as to why an exception would be thrown and gives the user something to look for in their own code.

Null Return Values

I encountered this issue with a fairly well known API a few weeks ago. According to the documentation, I provide a URL to the function and I get an InputStream back with the contents of the file located at the URL. For the protection of the offender I offer this version:

/**
 * Opens a connection to the provided URL and returns 
 * an InputStream that can be used to read the 
 * contents of the file located at the URL.
 *
 * @param url the URL pointing to the file location 
 *        that the input stream will read (must not 
 *        be null).
 *
 * @return an InputStream that can be used to read 
 *         the contents of the URL.
 *
 * @throws NullPointerException if url 
 *         is null.
 * @throws IOException if there was an error 
 *         connecting to the file location specified 
 *         by the URL.
 */
public static InputStream openUrl(URL url);

That seems pretty good. If the URL can't be reached, exceptions will be thrown. Ok, I can handle that. I write my code, wrap it in a try...catch block, and use the InputStream. During testing we get a NullPointerException and we trace it back to the InputStream returned from the method being null. What? Nothing in the documentation says that the InputStream returned can be null. It turns out that if the URL can be reached but the contents of the document are blank then null is returned instead of an empty InputStream. Well, instead of having to figure that out through trial and error, it would have been nice for the documentation to have read more like this:

/**
 * Opens a connection to the provided URL and returns an 
 * InputStream that can be used to read the contents of 
 * the file located at the URL.
 * 
 * @param url the URL pointing to the file location that 
 *        the input stream will read (must not be null).
 *
 * @return an InputStream that can be used to read the 
 *         contents of the URL, or null if the contents 
 *         of the InputStream would be empty.
 *
 * @throws NullPointerException if url 
 *         is null.
 * @throws IOException if there was an error connecting 
 *         to the file location specified by the URL.
 */
public static InputStream openUrl(URL url);

The general rule is that if your methods can return null, make sure that the user knows that so they can null check the response. A common offender is "search" methods that don't find a result.

All Values and Meanings of "Return Codes"

Return codes, which are integer values that represent the response from a function, have a way of slipping in at the last minute. Often a developer will realize after writing a method that the caller needs to know the result of the operation (especially if it fails), so they change "void" to "int" and return some magic number. The programmer then updates his code to make use of this number and moves on with his day, and the actual meaning of it is lost in time. Consider the following:

/**
 * Scans a dropbox for new files and processes the files it 
 * locates.
 *
 * @return a return code indicating the result of the dropbox 
 *         scan.
 */
int scanDropbox();

I've seen code like that too many times before. I'm certain that the author knows what magic codes can be returned from that function, but I don't. Consider this revision:

/**
 * Scans a dropbox for new files and processes the files it 
 * locates.
 *
 * @return a return code indicating the result of the 
 *         dropbox scan. Possible return values include:
 *         0 -- Dropbox was empty
 *         1 -- Dropbox contained files that were processed
 *         2 -- Dropbox did not exist
 *         3 -- Dropbox could not be read
 *         4 -- An error occurred while processing a file
 */
int scanDropbox();

That's a nice start. The next step is a bit more refactoring than it is documentation, but it really adds clarity. With a few constants declared for dropbox return values the comments and code will be much clearer:

/**
 * Represents the return code indicating that the dropbox 
 * was empty and did not contain any files to process.
 */
const int EMPTY = 0;

/**
 * Represents the return code indicating that the dropbox 
 * contained files and that they were processed successfully.
 */
const int SUCCESS = 1;

// Other return codes here
...

/**
 * Scans a dropbox for new files and processes the files it 
 * locates.
 *
 * @return a return code indicating the result of the 
 *         dropbox scan. Possible return values include:
 *         EMPTY            -- Dropbox was empty
 *         SUCCESS          -- Dropbox contained files that 
 *                             were processed
 *         ERR_NOT_FOUND    -- Dropbox did not exist
 *         ERR_CANNOT_READ  -- Dropbox could not be read
 *         ERR_FILE_PROCESS -- An error occurred while 
 *                             processing a file
 */
int scanDropbox();

That makes the documentation and the code make more sense than the magic number return codes presented before.

Side Effects and Relationships

Code that has "side effects" (i.e. affects something other than it was intended to or that is not obvious) should definitely have the side effect documented. Side effects, in general, should be avoided. To reduce possible confusion about the side effects, always make sure that they are well documented and stand out (preferably in bold text). Additionally, if there is a relationship between two calls, such as one call cannot be made until another one is made, those relationships need to be listed. In those cases, always make sure to point out the required order of the calls and indicate exactly which method must be called first.

/**
 * Scans the database for outstanding orders and notifies the 
 * shipping system. 
 *
 * NOTE: This assumes that a connection to the database has 
 * already been established. Ensure that "init_db_connection()" 
 * is called prior to calling this function.
 *
 * SIDE EFFECT:

 * Any error that occurs will be recording in the global 
 * "error_code" variable and a textual description of the error 
 * will be set in the global "error_msg_ptr" variable.
 */
void prepareOutstandingOrders();

Use Examples for Tricky Interfaces


Someone actually asked me the other day if it was appropriate to put code examples in the documentation. Yes! They say a picture is worth a thousand words, and sometimes a simple example code snippet can be be worth a few paragraphs. I have encountered many classes where I have read the class documentation and method documentation and still have had no idea how to use it. Usually, I end up searching Google for an example to use for clarification. For classes that are complex to use (for instance, have a lot of associations between calls, calls that have to be in specific orders, have specific "states", methods with lots of parameters, etc.) it is a good idea to show an example how of the class or is used in its own documentation. The same can be true for methods that require very specific sets of parameters. This is one of those areas where the documentation can easily become out of sync with the code, so make sure to always check for examples when you change an interface.

Conclusion

I hope that these tips help you to better document your code. Code documentation can take a lot of time, but really good code documentation can pay off big in providing clarity and understanding of how the code works and should be used. If you can think of anything I missed, please feel free to drop a comment. CodeProject

Confessions of a Programming Language Bigot

Hi. My Name is Michael and I'm a Programming Language Bigot.

One of the things that I've come to realize is that I'm a one trick pony. It's a very good trick though, at least for the moment. I'm a Java developer. I've been doing Java development professionally for over five years, and semi-professionally for several more years than that. I have mostly worked on user interfaces, and in the past few years have migrated to Java EE and Google Web Toolkit. I taught myself Java in college while most of the formal training was in Pascal (yes, Pascal, it was a great learning language), C, and C++. I had a brief stint with a company where I did C# and ASP.NET work, but for the most part I'm a Java guy.

So where does my Java bias come from? Why Java and not Python or C#? Well, some of it comes from experience, some from politics, but most of it comes from stupidity and laziness. The following expresses my feelings about other languages over the years and why I've come to feel I was wrong. In the end, I'll summarize the root of my language bigotry and how I plan to overcome it.

Python: It Wasn't Java

Early on in my semi-professional career I encountered Python. My first paid programming job was a Java project and consisted of a desktop client and a back-end server. My job was going to be to add some features to the client and server. I would finally get to test those Java skills I honed in college! One problem: the individual whose code I inherited was a Python fan and wrote his client code using Jython (Python interpreted by Java) and the server using a Python server called Zope. None of that code was documented either. The result was a frankensteined monster of Java and Python code. I had no experience with scripting languages, so the concept of no semicolons and indentation based blocks was utterly foreign to me. So in my first "Java" job I started behind the curve and was forced to learn Python on my own time very rapidly. That left a bitter taste in my mouth for the language that still exists today.

I know several people who are very fond of the language, and a lot of great tools (like Mercurial) are built using it. I've since gone back and looked into it more and it doesn't seem that bad, but still I don't want to learn it.

C#: Java Ripped Off by the Devil

When C# first came out I loathed the language. Not that there is anything inherently anything wrong with it, but I was a serious Microsoft hater back in the day and the initial version of C# looked like a blatant Java clone that would only run on Windows. Basically, it was built by Microsoft so I wanted nothing to do with it.

Years after C#'s debut I would work for about half a year as a C# developer. I learned to kind of like C#, especially how property getters and setters are paired. C# was easy to learn coming from a Java background. I've read up a bit on LINQ and think that it is a pretty solid idea, and something I wish that would come to Java. C# also seems to evolve much faster than Java, which is a great thing. Last year I received some training in C# and was actually looking forward to starting work on a C# project, but just before my transfer I was placed on another team that needed a seasoned Java developer.

Objective-C: Square Brackets Go Where?

Like everyone and their mother and their grandmother I wanted to learn how to make apps for my iPhone. There were two great hurdles to learning how to program for the iPhone: Objective-C and XCode. Even though Objective-C is considered a "C like" language and is compatible with both C and C++, its syntax, object system, and memory model differ greatly from those languages. When I first saw Objective-C code I couldn't make heads or tails of it. I quickly got over "message passing" using square brackets, strings with "@" signs, and the weird "release/retain" memory model. I never did get the handle of the way the named parameters worked though. XCode and Interface Builder also led to my confusion. The concept of doing work in one or the other and switching back and forth, and dragging and dropping visual elements to link objects was just a bit much to me. Needless to say, I gave up.

Since XCode 4 released things seem much simpler. At least it is all in one user interface. Someday I still want to write an app for the iPhone, but it's going to take some cramming on Objective-C to get there.

Ruby: A Language About Duck Enclosures on Tracks...Or Something

As Ruby, and especially Rails, started to gain traction I began to get interested in the language. I actually started a book club for Dave Thomas' Programming Ruby book. As we progressed through the book, however, the book club became more and more lost. We all had backgrounds in C, C++, and Java, so the concepts of a weakly typed language, duck typing, and closures confused us. It takes a large shift in thought to go from strongly typed and structured code to a more dynamic language, and due to a lack of time I lost interest in Ruby.

Later on I would lead another book club about Groovy, which is a scripting language similar to Ruby with Java as its foundation. I found that by relating Groovy's implementations of weak typing and closures to facets of Java I already understood that my understanding of those features in Ruby finally snapped into place. I actually respect Ruby a lot more now that I understand what is going on. The problem for me was the large leap from Java to Ruby, but with Groovy as a stepping stone Ruby makes a lot more sense now. I'm actually looking at learning Ruby and Rails to play with on side projects.

JavaScript: Only Useful for Annoying Popups and Browser Tricks

I do have to say that if there is one language I have truly looked down on it is JavaScript. Part of it stems from the name. When I was first learning Java I tried to learn JavaScript as well because, you know, it had "Java" in it. Within the first paragraph of the JavaScript book I was reading at the time I learned that the two languages were completely unrelated and the "Java" moniker was all for marketing. Back then, there was very little that JavaScript could do. It could pop up windows, move the browser window, scroll text in the status bar, and a few other "tricks" that didn't seem to have any use to me. On top of that, the weak typing and strange classless objects were too different from what I was used to.

Fast forward to today and HTML5. My belief is no longer that JavaScript is for silly scripts or pointless web tricks. JavaScript is the way forward for most applications. More and more applications are moving into the cloud and executing the browser: image editing, word processing, full-blown software IDEs. JavaScript now stands to be the most important of all of the programming languages, which is why I've been cramming my head with as much JavaScript know-how as I can.

Lisp and Clojure: I Can't Wrap My Head Around Reverse Reverse Polish Notation

I'm lumping these two into one category because they are both functional languages. The "functional" languages have just never made sense to me. I just don't think that way. I mean, in a sense I get it. In English we say "Add one, two, and three together.", but mathematically we're using to seeing "1 + 2 + 3" though. Writing "(+ 1 2 3)" to express that same concept just seems foreign.

I haven't worked on any code that would need to make use of the functional languages, so I'm sure I'm just missing the point. Personally, I just can't imagine trying to write a user interface or any other extensive amount of code in them.

Scala and Clojure: What Part of Java Virtual Machine Did They Not Understand

Clojure gets to show up twice, because it is both a functional language and one that runs on the JVM. My initial reaction to hearing about these languages was "why?" They are designed to run on the Java Virtual Machine (JVM). Isn't there already a language that runs fine on the Java Virtual Machine? Did they miss the "J" in JVM?

I admit, this is extremely shortsighted of me. Especially since I took an interest in Groovy, which also runs on the JVM. The difference was that Groovy code makes sense to me and for the most part looks like Java code. Once again, the functional languages are hard for me to read and understand, so I was biased against Scala and Clojure because they ran on the JVM and did not look like Java code.

To be honest, I still don't know much about Scala, other than it must be fairly good since it is running one of the most popular websites of our time (Twitter). It is supposedly more succinct than Java and is completely object oriented, and both of those aspects would be welcomed by any Java developers working today.

So Why be a Bigot?

I have to say, for the most part it is not intentional. There are a number of factors that have led to my bigotry.

First, Java became my preferred language because of my ideals. I was totally down for with "write once, run anywhere" slogan. I didn't want to write in a language that only ran on Windows/Macintosh/Linux.  Back when I was learning Java I hated Windows, revered Linux, and dreamed of being able to afford a Mac. I believed that code should be portable and run on whatever operating system I had available or was forced to use. Not being a fan of Microsoft, I had no interest in the .NET languages. I didn't agree with Microsoft's philosophies so I shunned their languages. Quite honestly, I'm starting to feel the same way about my beloved Java now that Oracle has the reins.

Second, there is the matter of tooling. Java was easy to learn because there were some powerful IDEs with great tools such as code completion. IDEs and editors are another area where I have a bias (Eclipse for development, Vi for text editing). Having tools that are cheap, easy to obtain, and easy to work with made Java a better choice for me. Another black mark against the Microsoft's languages was that you needed Visual Studio, which is anything but free (I do realize there are free learning versions now). Tools for other languages also have different layouts, commands, and keyboard shortcuts.

Third, the other languages don't look like the ones I know. Learning something new can be hard, especially if it differs greatly from what you've been taught is normal. New languages come with all new syntax and semantic rules which can be completely foreign, like Python's indention based blocks, JavaScript's blocks not defining scope, Objective-C's message passing and unique memory model, or functional programming in general.

Fourth, I don't need the other languages to do my job. My company doesn't use Ruby or Python or Scala or Lisp or many of the other languages I mentioned. Having skills in those languages provides me with no immediate benefits.

Lastly, the other languages pose a threat to my preferred language. Every week I see another blog post with the headline "Java Is Dead: Long Live Scala/C#/Ruby/Etc." What if one of those languages does manage to surpass Java? Java is my favorite language and the one I use professionally to put food on the table. All the years of knowledge I've gained about Java could all be thrown out if one of these other languages does gain a significant foothold.

The Ultimate Root of My Bigotry

With the exception of the ideals, there is one root to all of my bigotry: time. I lack it. Learning a new language takes time, especially if that language is vastly different from the languages you already know. It takes time to learn new concepts. It is not that I'm too dumb to learn new languages and concepts, it is that I'm too dumb to take the time try to learn them. It is far easier to downplay a language than to take the time to learn it.

With the exception of perhaps the functional languages, the worst part of learning a new language isn't even the language itself (syntax is easy); the worst part about learning a new language is learning the huge set of libraries. I know a ton of different Java libraries, and it has taken a long time to gain that knowledge. Learning the equivalent libraries in a different language is a time consuming prospect, so its easier to just dismiss learning the language, especially if you have no immediate use for it.

Learning new IDEs and editors also suffers from the same time constraints. I know Eclipse's shortcuts by heart. Every time I use NetBeans I am completely lost and frustrated. That doesn't mean that NetBeans is a bad IDE, it just means that I don't want to take the time to relearn everything I already know how to do very well. The same is true for Vi vs. Emacs. I learned Vi first and I am decently adept at it, and every time I'm forced to use Emacs I can't figure anything out, including how to exit it (I looked this up after writing this and now know how). I know people who swear by Emacs, but I just haven't taken the time to learn it.

The Cost of Programming Language Bigotry

So what is the downside to all of my bigoted ways? This whole article was spurned by my renewed interest in JavaScript. Like I said, it looks like web applications are the way forward, so going forward means casting aside my old feelings about that language and taking a fresh look at it. My fear is that if I don't, I'll get left behind when software development moves in that direction.

In the beginning I mentioned that I've come to view myself as a one trick pony (that trick being Java development). Right now that is a solid position to be in, and I don't see Java going away any time soon. Some day that trick won't be as useful though, and if I don't learn any new ones then I'm afraid I'll be like the developers who stuck with COBOL or FORTRAN because those were the only languages they needed. Those people are most likely out of jobs and can't find new ones with their skill set (and with the recent layoffs in the space program, I actually know that some of them are in that position).

Who knows, if I stuck with Objective-C I might be making the big bucks as an app developer (I seriously doubt this).

Really good programmers are lifelong learners and keep their skills up to date. While I've definitely done this in regard to Java, I've spurned other languages in the process. It turns out, those other languages are quickly becoming more applicable. Java 8 is set to include closures. When I first encountered closures in Ruby they were foreign and made no sense, so I didn't care for them. After seeing them in Groovy I started to understand the power of closures, and now I'm excited about the fact they will be coming to Java. Without that exposure from other languages I probably would have a hard time understanding closures when they do come to the one language I'm most familiar with as it evolves.

Conclusion

So how I am going to overcome my past views? Well, part of it is just carving out time to learn new things and being more open to learning them. Using an e-reader and my phone I've already read a couple of great JavaScript books (Douglas Crockford's JavaScript: The Good Parts being a fantastic quick read). I've started experimenting with JavaScript to learn it better. I'm ramping up to start learning Ruby and Rails and challenged myself with a simple project. I'm even challenging myself to learn TextMate just to have a different editor under my belt. I just don't want my old views to affect my future value as a developer going forward.

Still, don't expect me to be writing Lisp or Python code in Emacs any time soon...I only have enough time for my open mindedness to go so far :)

How to Fix a Broken Software Patent System

Oracle vs. Google
Apple vs. HTC
Lodsys vs. App Developers
Microsoft vs. Barnes & Noble


Recent headlines are filled with cases of patent battles between technology and industry titans. It seems like more and more patent cases (and especially software patent cases) make the news every day. NPR recently did a fantastic piece in their This American Life program called When Patents Attack. It is definitely worth the time to listen to or read the transcript.

As a software engineer, I have to agree with many of my colleagues in the field that the patent system is broken. I have actually witnessed it firsthand to a certain extent. Several years ago I worked on a project for what my team thought at the time was a novel idea. After we had a working prototype, our company of course wanted us to look into patenting the idea. So I was tasked with doing some patent research. What I found kind of surprised me. It turns out, our idea wasn't novel. In fact, it had already been patented three times! Two of the patents were for recent years, but one of them was over a decade old. Ours wasn't a new idea, it was just one that hasn't been capitalized on yet.

Now the real question is how did the exact same idea get patented three times in the first place. Well, having read the patent applications in full (and they were quite lengthy) it was kind of easy to understand. Each application was filled with technical terms and legal mumbo-jumbo that made the applications nearly unreadable. Many of the claims in one patent used synonyms for terms in the other. When you boiled it down to layman's terms though, the patents were identical. Heck, with enough big words obfuscating what you really mean, you can patent anything (the When Patents Attack story mentions Patent 6,080,436 titled "Bread Refreshing Method", which is better known to the rest of the world as "toast").

In truth, to find the patents matching our idea it took some searching and using combinations of terms that were similar to the ones we used to describe our "novel idea." I knew which terms to use because I was familiar with the subject matter. In the case of the United States Patent and Trademark Office (USPTO), the patent applicant reviewers probably are not as familiar with the field and will be less likely to make the connection.

Something else I found while Googling for the patents were articles on the subject of our "novel idea." Some of these dated back well beyond a decade when the first patent was filed. Technically, these should have been considered as "prior art" since they described the same concept before the first patent application was filed. In all reality, none of the patents should ever have been valid. It is easy to understand how the first patent could have been granted. Back when it was issued, the USPTO probably didn't have the search capabilities to identify works of prior art (before the days of Google), but the subsequent patent applications should have turned up this prior art when they were being reviewed.

The point is, there are numerous patents that exist that are either identical or similar enough to be considered the same, and many of them probably aren't valid because they weren't novel ideas. These patents slipped through the cracks, but are now valid and are now viable weapons.

Weapons?

The NPR piece describes an arms race among patent holders. Apple and Google and Microsoft are amassing caches of patents to use "for defensive purposes." (updated: such as Google's announcement today to purchase Motorola and its 17,000+ patents). The NPR piece compares it to the term "Mutually Assured Destruction," which in nuclear terms means "if you nuke me then I'll nuke you and we'll both be destroyed." In patent terms it translates to "if you sue me because I infringe on your patent then I'll sue you because you infringe on mine." The idea is that if you hold enough patents that your opponent infringes on then they can't sue you because you can counter-sue them back.

That's not a healthy strategy for innovation and the patent system. One, it means that many patents are being cached not for their innovation value but instead for their litigation value. Sadly, the patents that are worth the most are really the least innovative ones: far reaching, overly vague patents that cover general concepts. Two, it means that smaller players such as small businesses and startups can't participate. Back to the nuclear analogy, it means that a country like San Marino (the startup of countries) better not try to build a weapon (product that may infringe a patent), or better pray it doesn't get noticed, because any of the bigger countries could take it off the map at any point.

There is one more problem with this patent approach: patent trolls (i.e. non-practicing entities). Patent trolls collect patents and use them for litigation or licensing, but do not build any products themselves. Which means all of their patents are offensive. Google can threaten Apple with a patent claim that Apple infringes on if Apple tries to sue Google for one of Apple's patents that Google infringes on. In the end, it could just be a stalemate. That doesn't work with the patent trolls. If a patent troll sues Google for patent infringement, there can be no counter suit since Google can't be holding any patents the troll is infringing.

Some people are calling for software patents to be abolished again (software could originally not be patented until a 1998 ruling, just in time for the original web gold rush). I think that once the software patent genie is out of the bottle, there's no turning back the clock to undo what was done. There are too many big established companies with too many lobbyists to let that happen. So the best we can do is reform the broken system and fix the problem going forward. 

Part of the problem with the patent system seems to be that the USPTO is overworked and underfunded. A recent proposal called the "America Invents Act" helps to push some reform through the senate. One of the benefits is that there will be additional USPTO offices and that the USPTO will get to set its own fees and keep all of those fees (previously it did not set fees and only received a portion of them).

With that in mind, I have a few suggestions for how these fees should be set and how the patent system should be reformed.

First, raise the barrier to entry. It currently costs as low as $110 to file for a patent (non-refundable whether the patent is issued or not), and $755 if the patent is issued. Raise those values to $1,000 to file (non-refundable still) and another $1,000 if the patent is issued. This would probably cause the number of bogus or far-fetched patents to be lowered and improve the quality of the patents that are submitted. This would mean more money for the USPTO and hopefully fewer patents to review (leading to more thorough and better reviews).

Second, reduce the duration of the patents (at least for software and business process patents). Patents currently last 20 years from the date of filing. When the patent system was first formed, technology (like the cotton gin) didn't advance at nearly the rate that it does now. A 20 year monopoly in the software field is well beyond the useful lifetime of most software products. Five years, or at most seven, should be more than sufficient, if not still excessive for software patents. Can you imagine if Apple refused to license their multi-touch patents for 20 years? Only the iPhone would be able to pinch zoom for that long? That number needs to go down.

Third, raise the maintenance fees and make them annual. Currently, at 3 1/2 years it costs $490 to renew a patent. At 7 1/2 years that number goes up to $1,240, and at 11 1/2 it is $2,055. I propose that a more exponential scale be used.

Year 1: $1,000
Year 2: $10,000
Year 3: $50,000
Year 4: $100,000
Year 5: $500,000
Year 6+: $1,000,000

There would be numerous benefits to this approach. First, the smaller values initially would allow startups a few years to cement the patent and get something off the ground. Second, patents with lower value and less utility would probably not be renewed after a few years, making the technology available to other companies sooner. Companies would probably only pay the upkeep on their most valuable patents and the ones that differentiate them from their competition. Third, the patent trolls will have a hell of a time paying for the upkeep on the thousands of patents they are hoarding. Unfortunately, they would probably still pay for the broadest reaching patents they can use to sue with, but it would probably cause them to reduce their patent portfolios and their threat to small companies. Fourth...that would be one well funded USPTO! With that kind of money they could hire field experts to examine patents and have multiple reviewers for patent applications. The best possible solution would be to apply those fees starting when they change (i.e. next year if your patent is 6+ years old then there's a $1M cost to renew, only 2 years then it would be $10,000).

In addition to the pricing changes, the USPTO should provide an easy system for third parties to challenge patents. The America Invents Act actually is looking to implement this by providing a one year post-grant review period where third parties can submit prior art and challenge a patent. There are multiple benefits to this as well. One, it gives competitors a chance to defend themselves against a patent by striking it down early. Additionally, if the fees are upped as I mentioned above, then a lot fewer bogus patents will likely be filed. If an idea is not really new or radical and is likely to be shot down in the post-grant phase, the filer will be less likely to file the patent at all since they will lose out on the non-refundable application fee. This too would likely improve the overall quality of patents that are granted. The hard part on the USPTO is to make this fair for both parties, as large companies could easily file endless reviews against startup competitors and put pressure on them with already granted patents past the review period. The system will need to be affordable so that startups can defend themselves and file reviews as well.

What do you think?