Tuesday, November 20, 2012

The Untouchables

There are some technologies that have such strong "street cred", they are effectively untouchable. Anyone who dares imply that they are deficient in any way is instantly labelled an uneducated moron. It is impossible to have a realistic conversation about these technologies, and it's not very common for enthusiasts of these technologies to admit their failings. Let me give some examples:

  • Linux is the best operating system.
  • Common Lisp is the best programming language ever invented.
  • Haskell code is shorter, more elegant, and more correct than code written in any other language.
  • Emacs is the best editor for real programmers.
  • Apple creates the most user friendly user-interfaces.
  • Tiling window managers lead to the highest degree of user productivity.
  • Ruby on Rails is the best web framework written in any language.
  • The only way to write good software that meets its users needs and is delivered on time is to embrace agile software development and use test-driven development.

Of course, there are problems with each of these things:

  • It can be very difficult to get hardware accelerated 3D, wireless drivers, and suspend working reliably in Linux depending on the hardware you have and which version of which distro you are using.
  • Common Lisp has a lot of historical baggage, and it lacks the breadth of community library support that some languages such as Python, Ruby, and C have.
  • Haskell code can be very terse, and advanced Haskell code can be difficult for even intermediate Haskell programmers to understand.
  • IntelliJ has many advantages over Emacs when it comes to editing Java.
  • Not everyone prefers Apple user interfaces. My wife always complains when I make her use OS X.
  • Many of the tiling window manager fans that I know spend an inordinate amount of time configuring and tweaking their window manager, presumably because it doesn't yet do exactly what they want.
  • Ruby on Rails isn't the best approach for real-time applications or applications that must have very, very low latency.
  • Agile software development is not a good fit when it is impossible to iterate, such as when you're building software that must be completely done and completely correct the first time it is used (e.g. pacemakers and satellite software).

There are also some technologies that have such negative street cred that it's difficult to praise them in certain circles. For instance:

  • It can be hard for a Linux advocate to admit that Microsoft has ever done anything good or innovative.
  • It can be difficult for a Python or Ruby enthusiast to admit that Java has any advantages whatsoever.

Linux, Common Lisp, Haskell, Emacs, Apple, tiling window managers, Ruby on Rails, agile, and TDD each have amazing amounts of street cred, and I've spent a lot of time learning playing with all of them. What I discovered is that the world isn't always so black-and-white. It's really helpful when people can honestly admit the weaknesses in technologies they like as well as the strengths in technologies they don't like. I'm not saying that all technologies are equally good. I'm just saying that it would help if we could be more realistic.

Saturday, November 17, 2012

Humor: Metamocking

There are a lot of very powerful concepts in the world of software engineering that have "meta" in their name. Just consider metaprogramming and metaclasses. I've spent some time thinking about what "metamocking" might look like. What would it mean to mock a mock?

First, some background. Let's say I have the following function (written in Dart):

void helloWorld() {
  print("Hello, World!");
}

How do I know if it works correctly? Since it doesn't return anything, but rather changes the state of the outside world (specifically, the stuff printed to my terminal), it's not so easy to test.

Mocking to the rescue! I need to add a bit of dependency injection so that the test can pass in its own code for printing and so that I can mock things properly:

library hello_world;

const defaultPrinter = const Printer();

class Printer {
  const Printer();
  printValue(String s) => print(s);
}

void helloWorld({Printer printer: defaultPrinter}) {
  printer.printValue("Hello, World!");
}

main() {
  helloWorld();
}

Now, let's write a unittest that makes use of mocking to test the helloWorld function:

import 'package:unittest/unittest.dart';

import 'hello_world.dart';

class PrinterSpy extends Mock implements Printer {
  PrinterSpy() {
    when(callsTo('printValue', "Hello, World!")).thenReturn(true);
  }
}

void main() {
  group('helloWorld', () {
    test('prints "Hello, world!"', () {
      var printerSpy = new PrinterSpy();
      helloWorld(printer: printerSpy);
      printerSpy.getLogs(callsTo('printValue'))
        .verify(happenedExactly(1))
        .verify(alwaysReturned(true));
    });
  });
}

Woah! That's a lot of code to test such a simple function! I feel pretty comfortable about "class PrinterSpy" because it's pretty short, but I'm really worried about the test itself. How do I know that the test itself is using printerSpy correctly?

Once again, mocking to the rescue! Let me start by pulling out the heart of the test into a separate function so that I can make use of dependency injection again:

import 'package:unittest/unittest.dart';

import 'hello_world.dart';

class PrinterSpy extends Mock implements Printer {
  PrinterSpy() {
    when(callsTo('printValue', "Hello, World!")).thenReturn(true);
  }
}

void testPrintsHelloWorld({PrinterSpy printerSpy}) {
  if (!?printerSpy) {  
    printerSpy = new PrinterSpy();
  }
  helloWorld(printer: printerSpy);
  printerSpy.getLogs(callsTo('printValue'))
    ..verify(happenedExactly(1))
    ..verify(alwaysReturned(true));
}

void main() {
  group('helloWorld', () {
    test('prints "Hello, world!"', () => testPrintsHelloWorld());
  });
}

Now, I can create a mock for my mock to make sure that I wrote it correctly. Unfortunately, this is fairly challenging since you can't use a Mock to mock a method named getLogs since Mock has its own definition of that function. Nonetheless:

import 'package:unittest/unittest.dart';

import 'hello_world.dart';

class PrinterSpy extends Mock implements Printer {
  PrinterSpy() {
    when(callsTo('printValue', "Hello, World!")).thenReturn(true);
  }
}

void testPrintsHelloWorld({PrinterSpy printerSpy}) {
  if (!?printerSpy) {  
    printerSpy = new PrinterSpy();
  }
  helloWorld(printer: printerSpy);
  printerSpy.getLogs(callsTo('printValue'))
    ..verify(happenedExactly(1))
    ..verify(alwaysReturned(true));
}

class PrinterSpySpy extends PrinterSpy {
  LogEntryList getLogs([CallMatcher logFilter,
                        Matcher actionMatcher,
                        bool destructive = false]) {
    log.add(new LogEntry(null, "getLogs",
        [logFilter, actionMatcher, destructive], null));
    return super.getLogs(logFilter, actionMatcher, destructive);
  }
}

void testTestPrintsHelloWorld() {
  var printerSpySpy = new PrinterSpySpy();
  testPrintsHelloWorld(printerSpy: printerSpySpy);
  printerSpySpy.getLogs(callsTo('getLogs')).verify(happenedExactly(2));

  // Verifying the number of times verify is called is left as an
  // exercise for the reader.
}

void main() {
  group('helloWorld', () {
    test('prints "Hello, world!"', () => testPrintsHelloWorld());
  });
  
  group('testPrintsHelloWorld', () {
    test('uses printerSpy correctly', () => testTestPrintsHelloWorld());      
  });
}

As you can see, I was able to verify that my mock executed correctly. Metamocking is certainly an interesting, albeit advanced technique for software engineers who consider themselves true experts in the art of software testing.

However, I should mention that it does have a couple flaws. For instance, who's to say that the mock for my mock is itself correct? In theory, running a mock should serve to test the mock, but one can never be too sure! Furthermore, despite the fact that I've written a lot of test code, I have no idea if helloWorld will truly print "Hello, World!" to my terminal since I never tested the actual print function. Mocking is like that sometimes ;)

At this point, I should probably wrap up. I'll finish with a proper definition of metamocking. The astute reader may have guessed by now that metamocking (i.e. mock mocking) actually has two meanings:

  • To create mock objects in order to test your mock objects.
  • To mock (i.e. poke fun at) those people who rely on mocking too much :)

My next blog post will be on metametamocking. In it, I plan on mocking this blog post and explaining that it was just a hopelessly bad and overly elaborate pun ;)