Ended: June 16, 2014
FUNCTIONS SHOULD DO ONE THING. THEY SHOULD DO IT WELL. THEY SHOULD DO IT ONLY.
If a function does only those steps that are one level below the stated name of the function, then the function is doing one thing. After all, the reason we write functions is to decompose a larger concept (in other words, the name of the function) into a set of steps at the next level of abstraction.
So, another way to know that a function is doing more than “one thing” is if you can extract another function from it with a name that is not merely a restatement of its implementation [G34].
Functions that do one thing cannot be reasonably divided into sections.
In order to make sure our functions are doing “one thing,” we need to make sure that the statements within our function are all at the same level of abstraction.
We want the code to read like a top-down narrative.5 We want every function to be followed by those at the next level of abstraction so that we can read the program, descending one level of abstraction at a time as we read down the list of functions. I call this The Step-down Rule.
The ideal number of arguments for a function is zero (niladic). Next comes one (monadic), followed closely by two (dyadic). Three arguments (triadic) should be avoided where possible. More than three (polyadic) requires very special justification—and then shouldn’t be used anyway.
Flag arguments are ugly. Passing a boolean into a function is a truly terrible practice. It immediately complicates the signature of the method, loudly proclaiming that this function does more than one thing. It does one thing if the flag is true and another if the flag is false!
In general output arguments should be avoided. If your function must change the state of something, have it change the state of its owning object.
Functions should do one thing. Error handing is one thing. Thus, a function that handles errors should do nothing else. This implies (as in the example above) that if the keyword try exists in a function, it should be the very first word in the function and that there should be nothing after the catch/finally blocks.
Writing software is like any other kind of writing. When you write a paper or an article, you get your thoughts down first, then you massage it until it reads well. The first draft might be clumsy and disorganized, so you wordsmith it and restructure it and refine it until it reads the way you want it to read. When I write functions, they come out long and complicated. They have lots of indenting and nested loops. They have long argument lists. The names are arbitrary, and there is duplicated code. But I also have a suite of unit tests that cover every one of those clumsy lines of code. So then I massage and refine that code, splitting out functions, changing names, eliminating duplication. I shrink the methods and reorder them. Sometimes I break out whole classes, all the while keeping the tests passing. In the end, I wind up with functions that follow the rules I’ve laid down in this chapter. I don’t write them that way to start. I don’t think anyone could.
The proper use of comments is to compensate for our failure to express ourself in code. Note that I used the word failure. I meant it. Comments are always failures. We must have them because we cannot always figure out how to express ourselves without them, but their use is not a cause for celebration.
Dependent Functions. If one function calls another, they should be vertically close, and the caller should be above the callee, if at all possible. This gives the program a natural flow. If the convention is followed reliably, readers will be able to trust that function definitions will follow shortly after their use.
Objects hide their data behind abstractions and expose functions that operate on that data. Data structure expose their data and have no meaningful functions.
This exposes the fundamental dichotomy between objects and data structures: Procedural code (code using data structures) makes it easy to add new functions without changing the existing data structures. OO code, on the other hand, makes it easy to add new classes without changing existing functions. The complement is also true: Procedural code makes it hard to add new data structures because all the functions must change. OO code makes it hard to add new functions because all the classes must change.
There is a well-known heuristic called the Law of Demeter2 that says a module should not know about the innards of the objects it manipulates.
More precisely, the Law of Demeter says that a method f of a class C should only call the methods of these: • C • An object created by f • An object passed as an argument to f • An object held in an instance variable of C The method should not invoke methods on objects that are returned by any of the allowed functions. In other words, talk to friends, not to strangers.
Active Record Active Records are special forms of DTOs. They are data structures with public (or bean-accessed) variables; but they typically have navigational methods like save and find. Typically these Active Records are direct translations from database tables, or other data sources. Unfortunately we often find that developers try to treat these data structures as though they were objects by putting business rule methods in them. This is awkward because it creates a hybrid between a data structure and an object. The solution, of course, is to treat the Active Record as a data structure and to create separate objects that contain the business rules and that hide their internal data (which are probably just instances of the Active Record).
In fact, wrapping third-party APIs is a best practice. When you wrap a third-party API, you minimize your dependencies upon it: You can choose to move to a different library in the future without much penalty. Wrapping also makes it easier to mock out third-party calls when you are testing your own code. One final advantage of wrapping is that you aren’t tied to a particular vendor’s API design choices. You can define an API that you feel comfortable with. In the preceding example, we defined a single exception type for port device failure and found that we could write much cleaner code. Often a single exception class is fine for a particular area of code. The information sent with the exception can distinguish the errors. Use different classes only if there are times when you want to catch one exception and allow the other one to pass through.
This is called the SPECIAL CASE PATTERN [Fowler]. You create a class or configure an object so that it handles a special case for you. When you do, the client code doesn’t have to deal with exceptional behavior. That behavior is encapsulated in the special case object.
There is a natural tension between the provider of an interface and the user of an interface. Providers of third-party packages and frameworks strive for broad applicability so they can work in many environments and appeal to a wide audience. Users, on the other hand, want an interface that is focused on their particular needs. This tension can cause problems at the boundaries of our systems.
Learning the third-party code is hard. Integrating the third-party code is hard too. Doing both at the same time is doubly hard. What if we took a different approach? Instead of experimenting and trying out the new stuff in our production code, we could write some tests to explore our understanding of the third-party code. Jim Newkirk calls such tests learning tests.
Interesting things happen at boundaries. Change is one of those things. Good software designs accommodate change without huge investments and rework. When we use code that is out of our control, special care must be taken to protect our investment and make sure future change is not too costly. Code at the boundaries needs clear separation and tests that define expectations. We should avoid letting too much of our code know about the third-party particulars. It’s better to depend on something you control than on something you don’t control, lest it end up controlling you. We manage third-party boundaries by having very few places in the code that refer to them. We may wrap them as we did with Map, or we may use an ADAPTER to convert from our perfect interface to the provided interface. Either way our code speaks to us better, promotes internally consistent usage across the boundary, and has fewer maintenance points when the third-party code changes.
First Law You may not write production code until you have written a failing unit test. Second Law You may not write more of a unit test than is sufficient to fail, and not compiling is failing. Third Law You may not write more production code than is sufficient to pass the currently failing test. These three laws lock you into a cycle that is perhaps thirty seconds long. The tests and the production code are written together, with the tests just a few seconds ahead of the production code. If we work this way, we will write dozens of tests every day, hundreds of tests every month, and thousands of tests every year. If we work this way, those tests will cover virtually all of our production code. The sheer bulk of those tests, which can rival the size of the production code itself, can present a daunting management problem.
1. When you add one month such that the last day of that month has 31 days, then the date should be the 30th, not the 31st. Stated like this, you can see that there is a general rule hiding amidst the miscellaneous tests. When you increment the month, the date can be no greater than the last day of the month. This implies that incrementing the month on February 28th should yield March 28th. That test is missing and would be a useful test to write. So it’s not the multiple asserts in each section of Listing 9-8 that causes the problem. Rather it is the fact that there is more than one concept being tested. So probably the
Clean tests follow five other rules that form the above acronym: Fast Tests should be fast. They should run quickly. When tests run slow, you won’t want to run them frequently. If you don’t run them frequently, you won’t find problems early enough to fix them easily. You won’t feel as free to clean up the code. Eventually the code will begin to rot. Independent Tests should not depend on each other. One test should not set up the conditions for the next test. You should be able to run each test independently and run the tests in any order you like. When tests depend on each other, then the first one to fail causes a cascade of downstream failures, making diagnosis difficult and hiding downstream defects. Repeatable Tests should be repeatable in any environment. You should be able to run the tests in the production environment, in the QA environment, and on your laptop while riding home on the train without a network. If your tests aren’t repeatable in any environment, then you’ll always have an excuse for why they fail. You’ll also find yourself unable to run the tests when the environment isn’t available. Self-Validating The tests should have a boolean output. Either they pass or fail. You should not have to read through a log file to tell whether the tests pass. You should not have to manually compare two different text files to see whether the tests pass. If the tests aren’t self-validating, then failure can become subjective and running the tests can require a long manual evaluation. Timely The tests need to be written in a timely fashion. Unit tests should be written just before the production code that makes them pass. If you write tests after the production code, then you may find the production code to be hard to test. You may decide that some production code is too hard to test. You may not design the production code to be testable.
Public functions should follow the list of variables. We like to put the private utilities called by a public function right after the public function itself. This follows the stepdown rule and helps the program read like a newspaper article.
The first rule of classes is that they should be small. The second rule of classes is that they should be smaller than that. No, we’re not going to repeat the exact same text from the Functions chapter. But as with functions, smaller is the primary rule when it comes to designing classes. As with functions, our immediate question is always “How small?” With functions we measured size by counting physical lines. With classes we use a different measure. We count responsibilities.
An optimal system architecture consists of modularized domains of concern, each of which is implemented with Plain Old Java (or other) Objects. The different domains are integrated together with minimally invasive Aspects or Aspect-like tools. This architecture can be test-driven, just like the code.