Error handling strategies in Swift - Part 1

November 7, 2015

Initially I wanted to wrap this up in a single post but as I was writing it I decided to split it in two parts. So, this is the first part, if you're already familiar with basic error handling in Swift 2.0 feel free to jump to the second part.

When the official error handling model was updated in Swift 2.0 Apple decided to go on a more imperative path than most of us expected. Before 2.0, the established unofficial way followed Haskell's monadic approach and got a lot of love from the community. So, when Swift 2.0 came out, some happily embraced the new model, but many were utterly disappointed.

For the scope of this article, we shall not weight the advantages vs the disadvantages, but instead, we will explain the new model and later on, in the second part, focus on strategies and try to look forward to a solution that satisfies both the functional and object oriented worlds. As a side note, if you're interested in comparing the the two models, Brad Larson and Nick Lockwood wrote comprehensive articles on the matter.

That being said. Let's get started with a little bit of background. Conceptually, an error is an unexpected result to an operation. Such a result can happen at different levels. So, we have hardware, kernel, system library, application level errors and so on. Most of these can be handled from the current running process (yes, even hardware and kernel errors) however many of the lower level errors leave the process is such a dire state that it can't be recovered, leaving us with the only valid option of terminating it.

Historically, Objective-C had several ways of handling top level errors, by either keeping them in the application space (usually for failures: NSError pointer to pointer, error return) or forwarding them to lower level mechanisms (NSException, @throw). Using the latter for general flow-control was strongly discouraged by Apple because of performance issues.

What Swift brings new is that it takes throw from the second category and puts it in the first. It no longer implicitly forwards the handling to lower levels. As you will soon see, this means throw-try becomes very similar with error returns, but with its handling enforced by the compiler.

In Swift, every single function that can throw an error is marked accordingly. Adding throws or rethrows (which will be discussed later on) to a function changes the function signature. Meaning that the following will fail with error: invalid conversion from throwing function of type '(String) throws -> String' to non-throwing function type 'String -> String' and in order to fix it we will have to tell the compiler that queueOperation actually accepts throwing operations too.

func sendRequest(request:String) throws -> String
{
    return "Response"
}


func queueOperation(operation:String -> String)
//func queueOperation(operation:String throws-> String)
{
    //add the operation to the queue
}

queueOperation(sendRequest)

Now let's spice things up a little and add the function that executes all the queued operations.

//don't be intimidated by the trailing `()`
//it simply allows to partially apply the first
//parameter and be able to later call the function with
//only the second parameter

func sendRequest(request:String)() throws -> String
{
    return request + "Response"
}

func sendRequestSafe(request:String)() -> String
{
    return request + "Safe Response"
}

//an array that holds throwing functions
var array = Array<() throws->String>()
func queueOperation(operation:() throws->String)
{
    array.append(operation)
}

func flushOperations() throws
{
    for op in array
    {
        print(try op())
    }
}

queueOperation(sendRequest("First"))
queueOperation(sendRequestSafe("Second"))

try? flushOperations()

All works nice and dandy. But what if one function throws an error.

func sendRequest(request:String)() throws -> String
{
    throw NSError(domain: "com.cocoaexposed", code: 1, userInfo: nil)
    return request + "Response"
}

flushOperations() will throw as well and the operations will get executed only up to the throwing one. In our case, the throwing one is the first one, so no operations will execute.

Let's modify flushOperations() so that it accepts a callback function with an optional parameter, the result (which will be nil if the operation fails).

func flushOperations(callback:(String? throws -> ())) rethrows
{
    for op in array
    {
        try callback(try? op())
    }
}

flushOperations
{
    result in
    print(result)
}

As you can see, we didn't had to mark flushOperations with try although it could throw. That's because when using rethrows a function will throw only if at least one of the closure parameter throws as well.

Since only closures and functions can throw, rethrows can only be used on functions that take at least one closure parameter.

So, if we were to write things different:

try? flushOperations
{
    result in
    throw NSError(domain: "com.cocoaexposed", code: 1, userInfo: nil)
}

We would have had to handle the error thrown by flushOperations as well.

Bottom line, there a 4 ways of handling a thrown error:

  1. Do nothing but bubble up the error:
func fetchPersons() throws -> [AnyObject]
{
    let result = try self.executeFetchRequest(NSFetchRequest(entityName: "Person"))
    return result
}
  1. Handle and control the flow:
do
{
    let result = try self.executeFetchRequest(fetchRequest)
    //other statements that may or may not throw errors
}
catch let e
{
    //handle the error accordingly
}
  1. Use result optionals:
let result = try? self.executeFetchRequest(fetchRequest)
//result is now an optional
  1. Force unwrap. Forward the error to lower handling levels (this will eventually send an EXC_BAD_INSTRUCTION mach message, by default terminating the process):
let result = try! self.executeFetchRequest(fetchRequest)
//if `executeFetchRequest` fails, you'll get a crash

Nice enough. But how do we throw an error? That's even easier:

throw NSError(domain: "com.cocoaexposed", code: 1, userInfo: nil)
throw CWError.badAccess //as long as CWError implements ErrorType protocol

We don't have to throw a NSError, any struct, class or enum that implements ErrorType can be used with throw.

Finally, there's defer. Useful for a bunch of other things too, but probably created with this purpose in mind:

try? flushOperations
{
    result in
    let file = open(filename)
    defer {
       close(file)
    }

    //do things with the file
    //...
    throw NSError(domain: "com.cocoaexposed", code: 1, userInfo: nil)

    //do here other stuff that won't be called
    //if the closure throws
}

close(file) will be called after the closure exits, no matter if you throw an error or not.

And so we learned the basics. In our next part we will present an approach that sort of unifies both worlds, the functional and object oriented and allows us to easily handle errors application wide.

Comments

comments powered by Disqus