All Articles

Getting a Handle on Errors

People mostly suck at programming.

It is often said (by my college math professor, for instance), computers are the world’s fastest idiots. They can only ever do what you tell them to, and much like Amelia Bedelia, aren’t exactly going to pick up on “what you mean”.

Programming is therefore a form of communication! Between the human and the computer, sure, but first and foremost between a human and their fellow programmers. And much like communication between people, it’s important to be able to correct yourself when errors occur.

In this short post, I take a look at how some popular programming languages give the author tooling to recover from expected (or even unexpected) errors while your program runs.

Let’s start with the simplest possible example, the undefeated systems programming champion of 50 years running…

C

This section will be short. C has no official support for error handling whatsoever. Not cool, C.

Not to say that C doesn’t have errors. Beyond the usual errors that can be thrown at you due to syntax, semantics, and linking, after a program is compiled, the only official “errors” you’ll encounter in runtime are those explicitly written into the control flow, or a very small set of illegal hardware operations.

Basically, if you:

  1. have a binary with invalid opcodes (should never happen unintentionally),
  2. you divide by zero (can happen if you’re not careful),
  3. you read/write invalid memory locations (literally all the time),

…the CPU will perform its duties as a dirty snitch and switch context to the OS, who will most likely choose to reward your program with capital punishment.

Otherwise, you must anticipate edge cases yourself. For expected errors, one idiomatic approach for many standard library functions is to return NULL, or even sometimes -1 (seemly at random).

FILE *filePointer;
filePointer = fopen("file.txt", "w");

if(filePointer == NULL)
    // handle error here

The key issue here is that if you choose to ignore the possibility of an error, your code written in C is still going to compile just fine. Now though, your code is subject to the dreaded and infamous undefined behavior.

This is not an ideal developer environment.

C++

Does the world’s greatest and easiest to use language, C++, make any improvement on its strict subset? Actually, yes. C++ at least supports exceptions.

try {
    // throw an exception by either triggering a C error..
    int a = 5 / 0;
    // ...throwing it explicitly...
    throw new Exception("Arbitrary Error");
    // or by calling another function that throws an exception
    functionThatThrowsAnException();
} catch (Exception e) {
    // handle here
}

You utilize this error-handling behavior by wrapping the code that might trigger an error in a try block. All try blocks have an accompanying catch statement (or perhaps multiple) that handles an error of a certain type.

Side note: If an error is thrown outside a try block, the normal C behavior (ie getting merc’d) takes over.

One of the (many) criticisms of C++ stems from the complexity of the try-catch block. It has two major issues:

  1. try-catch is the most convoluted and atypical of all control flow structures, and has been shown to be very difficult for programmers to reason about.
  2. throwing an exception in a function does not force the caller of that function to handle it, which may lead to unexpected program crashes.

So, while an improvement over C, C++‘s approach to error handling clearly does not solve all of its problems, and even creates some new ones.

Java

Java also supports exceptions via try-catch, with essentially identical syntax, so I won’t go over it. What Java introduces aside from this is checked exceptions.

openFile(String path) throws IOException {
    //
}
//...
main() {
    // will throw a compilation error because the IOException isn't wrapped in a try-catch
    openFile("file.txt");
}

Checked exceptions are a way to add potentially thrown exceptions to a function signature. Then, the Java compiler will refuse to compile your program if you do not handle that exception (via a try-catch) in the places where that function is called.

The value of this point is contentious among less experienced programmers, who are often annoyed when they are forced to write error handlers for cases that they don’t want to consider. That said, over time and with experience, most programmers appreciate this behavior, as it avoids unexpected issues in the long run.

Even so, the most common exceptions you’ll face in Java are the unchecked ones, such as ArrayIndexOutOfBounds and NullPointerException. These errors must be checked for manually, or they will cause a runtime crash.

The latter pops up in some places where instead of throwing an exception for an unhandled case, the function will instead simply return null, which is essentially a mechanism by which certain reference type checks can be neglected. Posthumously, this has been referred to as the billion dollar mistake due to how null references tend to become a vehicle by which dealing with an unhandled case (and most exceptions) can be deferred.

Omitting such error handling logic leads to expensive bugs and program instability in the long run. It is much more productive to use the compiler to force the programmer to consider all cases when the program is being written.

Overall, Java provides good ideas on top of C++‘s constructs, but not requiring their usage ends up being a big problem.

Golang

Golang takes a unique approach. The designers of Golang (which happens to include one of the original developers of C, Ken Thompson) decided to not add support for exceptions in Golang, for some reasons mentioned in the C++ section above.

The convention then is to return error values from the function alongside the function’s results. Fortunately, unlike C, Golang at least provides a standard error interface rather than just throwing around magic numbers. In addition, the sugary syntactical support for tuples in Golang makes this paradigm simple to declare and work with:

// in the errorable function:
func Sqrt(f float64) (float64, error) {
    if f < 0 {
        return 0, errors.New("math: square root of negative number")
    }
// ...
// in the caller:
    f, err := Sqrt(-1)
    if err != nil {
        fmt.Println(err)
    }
}

Unfortunately, much like C, this does not force the programmer to check for errors before proceeding with the values returned from such a function (although it will raise a warning in certain circumstances, thank goodness).

In addition, especially in the above circumstances, Golang employs the null reference by another name: the nil reference. This suffers from a lot of the same downsides discussed in the Java section.

After some time and a bit of feedback from developers, Golang also implemented an exception-like control flow called panic-defer-recover. However, while this construct can be used to mirror the control flow of exceptions, the designers of Golang have made it very clear this is not the intended use. Instead, it’s only intended to be used for resource cleanup. Read the linked article for more information.

Overall, Golang avoids much of the original issues with exceptions by simply not including them. Unfortunately, it only makes small improvements on C’s error handling, so it does not quite achieve my ideal either.

JavaScript

JavaScript, in all its wondrous dialects and in typical fashion, offers the naive developer many an option for dealing with unexpected behavior. It’s first class treatment is that of our familiar foe, exceptions. These are handled via the usual try-catch syntax, so I choose to not review that here.

JavaScript has an interesting relationship with asynchronous code, which also needs to handle errors. In the Olden Times, this was facilitated by the use of callbacks.

By convention, a word I’ve learned to scorn and dread, these callbacks would accept one parameter, the “result”, if the operation was successful. If unsuccessful, these callbacks would receive two parameters, one of which was the “err” object storing information about any issues that were encountered. This “result/error return value” convention is similar to Golang’s.

Then, while revolutionizing the way asynchronous code was written in JavaScript, promises were also a mechanism by which failable operations could fail more elegantly:

asyncFunction()
    .then((result) => doSomething(result))
    .catch((err) => console.log(err));

This syntax supports transparent error propagation. If either asyncFunction or doSomething throws an error, both will call the function provided to the catch block with the corresponding error. Internally, this Promise functionality uses a simple try-catch block.

The major downside to this flow is the same ones discussed in previous examples: if the possible error cases are not explicitly handled via a catch, nothing will happen when the code is run, all the way up until an error actually occurs, which will result in a crash. That said, it is common for JS linters such as ESLint to forbid so-called floating promises. This can go a long way to making your code more robust against errors.

All around, with certain tools, JavaScript can provide a more robust traditional error handling strategy, particularly when it comes to asynchronous code. Unfortunately, with its weak typing and null / undefined references that can cause runtime crashing, it probably has too many other sources of error waiting in the shadows.

TypeScript

But how could we forget JavaScript’s younger, but much cooler brother, TypeScript?

By itself, it actually does solve most of the problems I mention JavaScript. TypeScript introduces a strong(er) type system and undefined checking, with a robust support for type narrowing detection so your case handling code can be rather organic.

Combined with linters to enforce traditional error handling relating to exceptions, TypeScript is a surprisingly correct language, which means it forces you to be less ambiguous with your code. Overall, this is a great approach to development. It takes a bit longer to write your code, but often debugging time is shrunk drastically or removed all together.

Rust

Ah yes, the “new” kid on the systems programming block. But beyond that, Rust goes to great pains to enable the writing of extremely “correct” code.

Rust handles and avoids errors in a way that is actually similar to TypeScript’s narrowing system. Most functions that can error will return a Result object that can either be Ok or Err.

fn main() {
    let f = File::open("hello.txt");

    let f = match f {
        Ok(file) => file,
        Err(error) => panic!("Problem opening the file: {:?}", error),
    };
}

This might look similar to Golang’s approach, but the key difference is that Rust actually requires verifying the error did not occur before using the result.

It does this through a system of type-narrowing via the match statement. While this might look clunkier than TypeScript’s type-narrowing, experienced programmers tend to prefer Rust’s approach since its more idiomatic.

Overall, I believe Rust goes the extra mile required to ensure maximum correctness in the program while also providing flexibility to developers in how they choose to resolve errors. Combine that with a strong type system and Rust tends to work as expected the first time its run, which is the sign of a great environment for programmers.