Swift doesn’t support throwing exceptions, nor does it support catching them. This wouldn’t be a problem if you could develop iOS apps in pure Swift, but unfortunately at the moment you cannot. When developing an app most of the APIs you’ll be working with are Objective-C APIs that have been bridged to Swift. These APIs can, and do throw exceptions. So how do you catch them?
In this blog post I’ll take a look at how you can handle exceptions that are thrown by Objective-C APIs.
I’ll also explain why I think Swift should support exceptions ‘natively’:
I believe that exceptions would allow us to write simpler, more readable code with fewer branches, whilst maintaining all the security that optionals currently offer.
But more of that later … first we’ll look at how to handle the more immediate issue of exceptions thrown by Objective-C APIs.
The Problem - KVO
I am currently working on a project that makes use of key-value-observing, where observers are used to execute logic when various object properties change. In order to use KVO with Swift types, you must subclass NSObject
and declare any properties you wish to observe as dynamic
:
On a side note, a few people have proposed a pure Swift alternative to KVO, where properties are wrapped in objects that allow subscription (e.g. Observable-Swift and ReactiveCocoa 3.0), however I find this technique to be rather cumbersome. It also doesn’t allow for truly dynamic observing, where the property being observed isn’t know until runtime.
Anyhow, back to the problem at hand - with an NSObject
subclass you can add an observer as follows:
However, this method throws an NSUnknownKeyException
exception if the given key is not valid for the object being observed.
Various other KVC API methods throw exceptions also, for example removeObserver
helpfully throws NSRangeException
if you try to remove an object that isn’t actually a current observer; valueForKeyPath
and setValue(_:forKey:)
can also throw exceptions.
So if you want to avoid the possibility of your application blowing up at runtime, what do you do?
This had me scratching my head for a while, until thankfully a colleague of mine, Nicholas Wolverson, pointed out the seemingly obvious - you catch the exception with Objective-C!
Catching Exceptions in Swift
The simple solution to the problem is to create a small Objective-C ‘shim’ that catches any exceptions that might be thrown, returning them via the method signature directly.
Here’s a category that adds a tryAddObserver
method to NSObject
:
@interface NSObject (KVOHelper)
- (NSString *)tryAddObserver:(NSObject *)anObserver
forKeyPath:(NSString *)keyPath
options:(NSKeyValueObservingOptions)options
context:(void *)context;
@end
Which is implemented as follows:
@implementation NSObject (KVOHelper)
- (NSString *)tryAddObserver:(NSObject *)anObserver
forKeyPath:(NSString *)keyPath
options:(NSKeyValueObservingOptions)options
context:(void *)context {
NSString *result;
@try {
[self addObserver:anObserver forKeyPath:keyPath
options:options context:context];
}
@catch (NSException *exception) {
result = exception.name;
}
return result;
}
@end
Any exception that is thrown when trying to add an observer, is now caught, with the method returning the exception name.
When bridged to Swift this results in the following method being added to NSObject
:
That can be used as follows:
This is a nice simple solution, which can also be rolled out to removeObserver
, and setValue(_:forKey:)
, changing each method from having a void return type to returning any error that might have occurred.
The messy bits …
Unfortunately things get a bit messier with valueForKeyPath
; this method
already returns a value, so in order to add exception reporting it would either have to be adapted to include an in-out parameter (yuck), or multiple returns values.
Opting for the later, the following type holds either the return value, or the exception that was thrown:
@interface ValueWrapper: NSObject
@property id propertyValue;
@property NSString *exception;
@end
Which can be used to build a shim for getValueForKeyPath
:
- (AVValueWrapper *)tryGetValueForKeyPath:(NSString *)keyPath {
AVValueWrapper *result = [AVValueWrapper new];
@try {
result.propertyValue = [self valueForKeyPath: keyPath];
}
@catch (NSException *exception) {
result.exception = exception.name;
}
return result;
}
Whilst this solves the problem, it results in a pretty nasty method signature! Multiple returns values are widely considered a ‘code smell’
Fortunately Swift enums types, with their associated values, are an excellent way to represent multiple return values.
Renaming the above method to _tryGetValueForKeyPath
(which of course means other developers can’t see it :-P), allows the bridges method to be further adapted as follows:
The enum that is returned clearly encodes the success and failure scenarios, and is used as follows:
Using the basic concepts outlined above it should be possible to create a shim that adapts any Objective-C API that throws exceptions.
The above techniques worked just fine for me, and allowed me to get on with my current pet project (which is pretty cool, more on that another time!). However, this all feels like an unnecessary and cumbersome workaround.
Which brings me onto the final part of this blog post …
Swift Needs Exceptions!
Most modern languages support exceptions - the lack of support in Swift does seem to make it be something of an exception (ba-dum-tish). However, despite the fact Objective-C fully supports exceptions, they are often frowned upon.
As an example, the Google Objective-C style guide simply states “Don’t throw exceptions”.
However, I personally feel that Swift would benefit from the introduction of exceptions. To see why, we’ll return back the everyone’s favourite topic - optionals!
This is the all-to-familiar problem of optional variables - if you have an algorithm that depends on a number of optionals, each must be explicitly unwrapped:
Note the shadowing of each optional as per the Ray Wenderlich Swift Style Guide.
Although if you’re feeling reckless you can use forced unwrapping, but this will result in a non-recoverable runtime error should any of the optionals be nil:
Things get even worse if you want to execute some logic if any of the values are nil …
In practice, and in most of the code that I find myself writing, optionals are rarely nil. For example when parsing JSON, the data you are supplied with is rarely malformed.
The need to continually nil check in the ‘success scenario’ really clutters your code. It reminds me of a small child on a car journey, but instead of repeating;
My code reads as follows:
Considering that nils are often the exception, I think that ‘exceptions’ are a the appropriate tool in this context.
In my opinion, if you use forced unwrapping to access an optional that is nil, it should throw an exception. You would then be able to catch these exceptional failures in a single location (which would not necessarily be within the same block of code).
Imagine if JSON parsing were this carefree …
This would give us the best of both worlds, where if-let unwrapping can be employed to make detail-level decisions regarding nil optionals, whereas forced unwrapping and exceptions could be used to write logic that relies on numerous optionals which we expect to almost always be non-nil.
I feel there are other benefits that exceptions provide, such as their ability to more clearly define the contract of an API method. This is something which I think the Java language does very well. However, Swift does have viable alternatives, such as enums with associated values as shown earlier.
Put simply, I believe that exceptions would allow us to write simpler, more readable code with fewer branches, whilst maintaining all the security that optionals currently offer.
Regards, Colin E.