Nulls Are Not The Problem

I recently saw this promoted tweet:

screen-shot-2017-01-03-at-10-31-28-am

Of course, this isn’t the first time Null has been brought up as a source of woe for programmers. There’s a wiki on why Null is Considered Harmful, and the most recognizable argument against null is from its inventor:

I call it my billion-dollar mistake. It was the invention of the null reference in 1965. At that time, I was designing the first comprehensive type system for references in an object oriented language (ALGOL W). My goal was to ensure that all use of references should be absolutely safe, with checking performed automatically by the compiler. But I couldn’t resist the temptation to put in a null reference, simply because it was so easy to implement. This has led to innumerable errors, vulnerabilities, and system crashes, which have probably caused a billion dollars of pain and damage in the last forty years. — Tony Hoare, 2009

Since ‘null’ is clearly the source of our problems (it’s right there in the name “NullReferenceException”), then we should get rid of Nulls. Problem Solved!

Or, as H.L. Mencken once said:

Explanations exist; they have existed for all time; there is always a well-known solution to every human problem — neat, plausible, and wrong.

Absolutist positions in programming immediately make me question the source. The absolute ‘badness’ of NULL is a sacred cow that needs to be slaughtered for the good of our industry.

Null, in its simplest form, is the absence of a value.  In the same way Evil is the absence of Good, or a donut hole is the absence of a donut, null is the absence of a value.

That absence, however, has meaning.  And this is where we as programmers falter.  We give that meaning several distinct meanings that depend on context. We make decisions based on that meaning without clearly communicating our intent, and we forget to check whether there is an absence of a value.  How many times have you heard a programmer say “There’s no possible way that method could fail!” only to have it fail in production? I’ve personally lost count.

If we banished null to the “never use” bin of object oriented programming because we use it incorrectly, then there are a whole lot of things we need to get rid of with that reasoning.

The second issue with banning null is that the absence of a value has meaning contextually.  In a purely closed system it’s not an issue, but any system that interfaces with the outside world (meaning any program other than yours) is going to have times where it doesn’t work, or times where (for whatever reason), it doesn’t return the data you need.

If we tell programmers to write code to handle failure, why wouldn’t we tell programmers to check for data failure as well? That’s what the basis of Object Oriented Protocol is, after all (check the name, “Object Oriented Programming”): it’s data and behavior mixed together in one cohesive unit.

So, if you write code that will never interface with anything else, feel free to ignore anything I’ve said. After all, you can program against that uncertainty.  But for the rest of us, Null is an important part of our toolkit.

In three different systems I’ve built, I had good reason to use Null:

  • Sometimes I don’t get data back from the source.  Not getting this data back means I shouldn’t reason about it negatively; but its absence is a decision point for other paths in my program
  • Parsing text is a common failure;  should I always and forever represent that as something other than what it is? Is the failure to create an object from a parse an actual object itself?  It’s the absence of an object. After all, I failed to create it!
  • They haven’t set a date for an action.  Should I treat that as a value?  Or the absence of a value?

In each case; I used null (wrapped in a method to adequately describe what null means). According to the Null detractors, I should have returned a constant, or thrown an exception, or something.

Sadly, none of the ways to get around null fix the problem, they only paper over the issue. Let’s take his first example, returning a constant instead of a null, using the null object pattern:

public Employee getByName(String name) {
  int id = database.find(name);
  if (id == 0) {
    return Employee.NOBODY;
  }
  return Employee(id);
}

If we don’t find an employee, it’s purely semantical to return a constant Employee.NOBODY, and even worse is that it’s potentially harmful downstream. If our payment processing software (as a purely contrived example) gets an employee object back, but doesn’t check that that ‘nobody’ isn’t an actual employee, then they’ll go ahead and use the default values for the properties in their calculations.  A salary of $0 really skews reports.

“But, they’ll check against Employee.Nobody”, you say.  That’s the same as checking against Null.  The only difference is one will throw a runtime exception and the other one won’t — they both still have logical errors if you insist on filling it with default values (and not null).

The second argument against null is to use exceptions as flow control:

public Employee getByName(String name) {
  int id = database.find(name);
  if (id == 0) {
    throw new EmployeeNotFoundException(name);
  }
  return Employee(id);
}

Except, of course, Exceptions should be used for handling Exceptional things (it’s in the name), and not for common problems.Not finding an employee (for instance, in the context of searching) should not be considered exceptional.  If anything, it’s normal. People mistype names all the time. Humans make mistakes.

It’s even another ‘best practice’ to not use Exceptions as Flow Control. While I happen to agree with that practice, I put best practice in scare quotes believe all best practices are conditional on context. I haven’t yet encountered an occasion where exceptions as flow control is good; but that doesn’t mean there isn’t one, and I won’t tell someone “don’t do that” unless I’m looking at the code with them when I’m saying it, and I’m saying it for that particular circumstance.

Matt Sherman, a development manager at Stack Overflow (and someone I greatly respect) has said this about Nulls:

I submit that we can avoid a large majority of null-related errors if we follow these two principles:

  • Passing a null is a programming error
  • Branching on null is a programming error

I think most of our problems trace back to imbuing null with meaning, i.e., we use it as a signal.

I submit that null is the absence of meaning. It is not a de facto Boolean describing existence. Rather, it should be treated non-deterministically.

This implies that most functions should not “handle” null. Because null is the absence of meaning, we can not use it to make a decision.

These are great ideas, but they only work for a closed system, which is a system where you control all inputs and outputs, and aren’t dependent upon any external systems or processes that might fail for unknowable reasons.

Passing a null in a system is contextual (and maybe it shouldn’t be): If an external system passed me null; I may have to make a decision about that; I can’t simply say, “It doesn’t exist”, because it may exist (I argue that if you want a certainty for truthiness, then you need true or false; but if there is a degree of uncertainty, null is useful for conveying that).

For instance, if I get passed null for an Employee (to go back to our previous reason, because another system couldn’t find an employee), then I have to make a decision based on that, so I have to branch on null.

My point is that we can’t wish away null or just chide programmers for using it. It’s here, and it has meaning.  What I believe we should do is to reduce the amount of meanings it has; and that doesn’t mean not using it or programming it away completely; but it does mean reducing uses of null to mean “I don’t know, and don’t make any assumptions on why I don’t know“.

You should read all of Matt’s post. He brings up some great common sense approaches for reducing null and where it could be a code smell:

A common idiom in C# is to use optional parameters, with null as the default value:

void Foo(int id, Bar bar = null, Baz baz = null)

When I see a method signature like this, I have $5 that says the method’s implementation branches on those parameters. Which means that parameters are no longer just values, but are control flow. Bad semantics.

Instead, consider creating several methods with good names, and which accept and require only the parameters they use – no more optionals. “Branching” is done via method resolution, about which we can more easily reason.

Also, you should follow Matt on Twitter; he’s pretty awesome.

But, back to slaying sacred cows. Another sacred cow for Yegor are that static methods are bad:

A dry summary of [the above listed arguments] is that utility classes are not proper objects; therefore, they don’t fit into object-oriented world. They were inherited from procedural programming, mostly because we were used to a functional decomposition paradigm back then.

And much like the arguments against null, none of this is prefaced with context. So if you say  statics are bad in OOP, you should then say why they’re good:

[Stack Overflow] goes to great lengths to reduce garbage collection costs, skipping practices like TDD, avoiding layers of abstraction, and using static methods. While extreme, the result is highly performing code. When you’re doing hundreds of millions of objects in a short window, you can actually measure pauses in the app domain while GC runs. These have a pretty decent impact on request performance.

Null, much like static; has its place.  We should not minimize its importance, rather we should write our software to handle it. For developers this means:

  • Don’t shy away from using null/returning null if it fits your business case.
  • Do ensure you’re documenting (preferably through great naming) what null means in the context.
  • Don’t abstract null away to a ‘default’ reference type if your system has consumers; they may treat a default value differently from not knowing the value; and that’s a source of bugs.
  • In your own closed systems, reduce the amount of knowledge someone needs to interact with your code. You can architect out chaos in a truly closed system and it’s worth the extra time to do that.  This may mean throwing exceptions or having a default type. This works because you control the inputs and outputs. It works because you’re in a closed system.
  • In a system where you communicate with the outside world (read: anyone other than your own program), you will encounter failure. You will encounter the absence of data. Prepare for it.
  • If you have downstream clients, then don’t pass them null for multiple reasons (if you choose to use null); and if you can possible pass something that conveys more information, do so.  If null means “I don’t know” in your system, then it should mean that everywhere.

Programming is a means of representing and automating life interactions with computers, and life is not nearly as well ordered as programmers want it to be.  To make better software, we should prepare ourselves and our code for external forces, not chastise them for existing.