Before Swift, in Objective-C I would swizzle or hook methods in a class using <objc/runtime.h>.
If anyone has any info on the topic of modifying Swift's runtime and hooking functions like CydiaSubstrate and other libraries that helped in this area, please inform me.
I've succeed with method swizzling in Swift. This example shows how to hook description method on NSDictionary
My implementation:
extension NSDictionary {
func myDescription() -> String!{
println("Description hooked")
return "Hooooked " + myDescription();
}
}
Swizzling code:
func swizzleEmAll() {
var dict:NSDictionary = ["SuperSecret": kSecValueRef]
var method: Method = class_getInstanceMethod(object_getClass(dict), Selector.convertFromStringLiteral("description"))
println(dict.description) // Check original description
var swizzledMethod: Method = class_getInstanceMethod(object_getClass(dict), Selector.convertFromStringLiteral("myDescription"))
method_exchangeImplementations(method, swizzledMethod)
println(dict.description) //Check that swizzling works
}
Edited:
This code will work for any custom Swift class that inherits from NSObject (but will not work for classes that don't.) More examples - https://github.com/mbazaliy/MBSwizzler
You would likely be able to swizzle swift-generated classes that inherit from Objective-C classes with no problem, since they appear to use dynamic method dispatch all the time. You may be able to swizzle methods of swift-defined classes that exist in the Objective-C runtime by virtue of being passed across the bridge, but the Objective-C side methods are likely to just be proxies back across the bridge to the swift-side runtime, so it's not clear that it'd be particularly helpful to swizzle them.
"Pure" swift method calls do not appear to be dispatched dynamically via anything like objc_msgSend and it appears (from brief experimentation) that the type safety of swift is implemented at compile time, and that much of the actual type information is absent (i.e. gone) at runtime for non-class types (both of which likely contribute to the purported speed advantages of swift.)
For these reasons, I expect that meaningfully swizzling swift-only methods will be significantly harder than swizzling Objective-C methods, and will probably look a lot more like mach_override than Objective-C method swizzling.
I'm answering this question more than one year later because none of the other answers provide the definitive set of requirements for method swizzling for every kind of class.
What is described by other, while it will work flawlessly for extensions to foundation/uikit classes (like NSDictionary), will simply never work for your own Swift classes.
As described here, there is an additional requirement for method swizzling other than extending NSObject in your custom class.
The swift method you want to swizzle must be marked dynamic.
If you don't mark it, the runtime will simply continue to call the original method instead of the swizzled one, even if the method pointers appear to have been swapped correctly.
Update:
I've expanded this answer in a blog post.
I had a Xcode 7 iOS project written in Swift 2, using Cocoapods. In a specific Cocoapod, with Objective-C source, I wanted to override a short method, without forking the pod. Writing a Swift extension wouldn't work in my case.
For using method swizzling, I created a new Objective-C class in my main bundle with the method I wanted to replace/inject into the cocoapod. (Also added the bridging header)
Using mbazaliy 's solution on stackflow, I put my code similar to this into the didFinishLaunchingWithOptions in my Appdelegate:
let mySelector: Selector = "nameOfMethodToReplace"
let method: Method = class_getInstanceMethod(SomeClassInAPod.self, mySelector)
let swizzledMethod: Method = class_getInstanceMethod(SomeOtherClass.self, mySelector)
method_exchangeImplementations(method, swizzledMethod)
This worked perfectly. The difference between #mbazaliy 's code is that I didn't need to create an instance of the SomeClassInAPod class first, which in my case would have been impossible.
Note: I put the code in the Appdelegate because every other time the code runs, it exchanges the method for the original - it should only run one time.
I also needed to copy some assets that were referenced in the Pod's bundle to the main bundle.
I wouldn't do it that way, I think closures provide the answers (as they give you a chance to intercept, evaluate, and forward the invocation of the function, additionally it will be easy to extend when and if we have reflection.
http://www.swift-studies.com/blog/2014/7/13/method-swizzling-in-swift
I would like to extend the great answer provided by mbazaliy.
Another way of doing swizzling in Swift is by providing an implementation using an Objective-C block.
e.g. to replace descriptionmethod on class NSString we can write:
let originalMethod = class_getInstanceMethod(NSString.self, "description")
let impBlock : #objc_block () -> NSString =
{ () in return "Bit of a hack job!" }
let newMethodImp = imp_implementationWithBlock(unsafeBitCast(impBlock, AnyObject.self))
method_setImplementation(originalMethod, newMethodImp)
This works as of Swift 1.1.
A safe, easy, powerful and efficient hook framework for iOS (Support Swift and Objective-C). https://github.com/623637646/SwiftHook
For example, this is your class
class MyObject {
#objc dynamic func noArgsNoReturnFunc() {
}
#objc dynamic func sumFunc(a: Int, b: Int) -> Int {
return a + b
}
#objc dynamic class func classMethodNoArgsNoReturnFunc() {
}
}
The key words of methods #objc and dynamic are necessary
The class doesn't have to inherit from NSObject. If the class is written by Objective-C, Just hook it without any more effort
Perform the hook closure before executing specified instance's method.
let object = MyObject()
let token = try? hookBefore(object: object, selector: #selector(MyObject.noArgsNoReturnFunc)) {
// run your code
print("hooked!")
}
object.noArgsNoReturnFunc()
token?.cancelHook() // cancel the hook
Perform the hook closure after executing specified instance's method. And get the parameters.
let object = MyObject()
let token = try? hookAfter(object: object, selector: #selector(MyObject.sumFunc(a:b:)), closure: { a, b in
// get the arguments of the function
print("arg1 is \(a)") // arg1 is 3
print("arg2 is \(b)") // arg2 is 4
} as #convention(block) (Int, Int) -> Void)
_ = object.sumFunc(a: 3, b: 4)
token?.cancelHook() // cancel the hook
The key word #convention(block) is necessary
For hook at before and after. The closure's args have to be empty or the same as method. The return type has to be void
Totally override the mehtod for specified instance. You can call original with the same parameters or different parameters. Don't even call the original method if you want.
let object = MyObject()
let token = try? hookInstead(object: object, selector: #selector(MyObject.sumFunc(a:b:)), closure: { original, a, b in
// get the arguments of the function
print("arg1 is \(a)") // arg1 is 3
print("arg2 is \(b)") // arg2 is 4
// run original function
let result = original(a, b) // Or change the parameters: let result = original(-1, -2)
print("original result is \(result)") // result = 7
return 9
} as #convention(block) ((Int, Int) -> Int, Int, Int) -> Int)
let result = object.sumFunc(a: 3, b: 4) // result
print("hooked result is \(result)") // result = 9
token?.cancelHook() // cancel the hook
For hook with instead. The closure's first argument has to be a closure which has the same types with the method. The rest args and return type have to be the same as the method.
Perform the hook closure before executing the method of all instances of the class.
let token = try? hookBefore(targetClass: MyObject.self, selector: #selector(MyObject.noArgsNoReturnFunc)) {
// run your code
print("hooked!")
}
MyObject().noArgsNoReturnFunc()
token?.cancelHook() // cancel the hook
Perform the hook closure before executing the class method.
let token = try? hookClassMethodBefore(targetClass: MyObject.self, selector: #selector(MyObject.classMethodNoArgsNoReturnFunc)) {
// run your code
print("hooked!")
}
MyObject.classMethodNoArgsNoReturnFunc()
token?.cancelHook() // cancel the hook
After spending some time on it... Wake up this morning.... beta 6 is out and
Problem Fixed in beta6!
From release notes
"Dynamic dispatch can now call overrides of methods and properties introduced in class extensions, fixing a regression introduced in Xcode 6 beta 5. (17985819)!"
Related
I am starting to learn Swift, and have been following the very good Stanford University video lectures on YouTube. Here is a link if you are interested or it helps (although it isn't required to understand my problem):
Developing iOS 8 Apps with Swift - 2. More Xcode and Swift, MVC
While following the lectures I got to a point where (as far as I could tell) my code was identical to the code in the video but on my system I got a compiler error. After a lot of trial and error I have managed to reduce my code to two examples, one of which generates an error, the other or which doesn't, but I have no idea what is actually causing the error or how to resolve it.
The code which creates the error is:
import UIKit
class BugViewController: UIViewController
{
func perform(operation: (Double) -> Double) {
}
func perform(operation: (Double, Double) -> Double) {
}
}
This creates the following compiler error:
Method 'perform' with Objective-C selector 'perform: ' conflicts with previous declaration with the same Objective-C selector
By simply removing the sub-classing of UIViewController the code compiles:
import UIKit
class BugViewController
{
func perform(operation: (Double) -> Double) {
}
func perform(operation: (Double, Double) -> Double) {
}
}
Some other information which may or may not be relevant:
I have recently upgraded to Yosemite.
When I installed Xcode, I ended up with a Beta version (Version 6.3 (6D543q)) because (if I remember correctly) this was the version I needed to run on my version of OS X.
I am half hoping this is a bug in the compiler because otherwise this doesn't make any sense to me. Any help very gratefully received!
I myself am also taking the Standford course and I got stuck here for a long time too, but after some searching, I found something from here: Xcode release notes and it mentioned something below:
Swift 1.2 is strict about checking type-based overloading of #objc
methods and initializers, something not supported by Objective-C.
// Has the Objective-C selector "performOperation:".
func performOperation(op: NSOperation) { /* do something */ }
// Also has the selector "performOperation:".
func performOperation(fn: () -> Void) {
self.performOperation(NSBlockOperation(block: fn))
}
This code would work when invoked from Swift, but could easily crash
if invoked from Objective-C. To solve this problem, use a type that is
not supported by Objective-C to prevent the Swift compiler from
exposing the member to the Objective-C runtime:
If it makes sense, mark the member as private to disable inference of #objc.
Otherwise, use a dummy parameter with a default value, for
example: _ nonobjc: () = (). (19826275)
Overrides of methods exposed
to Objective-C in private subclasses are not inferred to be #objc,
causing the Swift compiler to crash. Explicitly add the #objc
attribute to any such overriding methods. (19935352)
Symbols from SDKs are not available when using Open Quickly in a
project or workspace that uses Swift. (20349540)
what i did was just adding "private" in front of the override method like this:
private func performOperation(operation: Double -> Double) {
if operandStack.count >= 1 {
displayValue = operation(operandStack.removeLast())
enter()
}
}
Objective-C does not support method overloading, you have to use a different method name. When you inherited UIViewController you inherited NSObject and made the class interopable to Obj-C. Swift on the other hand does support overloading, that's why it works when you remove the inheritance.
As it has already been answered, ObjC doesn't support method overloading (two methods with the same name) and In swift 2 under Xcode 7 there are two options to solve this kind of problems. One option is to rename the method using the attribute: #objc(newNameMethod:)
func methodOne(par1, par2) {...}
#objc(methodTwo:)
func methodOne(par1) {...}
another option to solve this problem in Xcode 7+ is by applying #nonobjc attribute to any method, subscript or initialiser
func methodOne() {...}
#nonobjc
func methodOne() {...}
The problem is UIViewController is an #objc class. When inheriting from UIViewController, BugViewController is also a #objc class.
This means it must conform to the rules of Objective-C selectors (the name of a method). The methods func perform(operation: (Double) -> Double) and func perform(operation: (Double, Double) -> Double) both have the same selector #selector(perform:). This is not allowed.
To resolve this, use different names: like func perform1(operation: (Double) -> Double) and func perform2(operation: (Double, Double) -> Double).
I think the best way to handle this is to give your perform() methods more descriptive names. What do these methods do? How do they change the state of the view controller? Look at the other UIViewController methods to get a feel for the style of method naming, or read Method Names Should Be Expressive and Unique Within a Class
From https://developer.apple.com/library/ios/releasenotes/DeveloperTools/RN-Xcode/Chapters/xc6_release_notes.html under "Xcode 6.3 Release Notes" -> "Swift Language Changes" you find
Swift now detects discrepancies between overloading and overriding in the Swift type system and the effective behavior seen via the Objective-C runtime.
I got the same error due to having having two methods with the same Obj-C signature:
static func prepareForUpSyncing(obj : NSManagedObject!) -> Bool
static func prepareForUpSyncing(objs : [NSManagedObject]!) -> Bool
I didn't want to mark one of them as #nonobjc due to possibility of unforseen consequences at runtime. (Someone can correct me if there is no possibility)
Resolved it by using Swift's external parameter name feature (I made external name same as local name) to the second method, which effectively changes the Obj-c method signature:
static func prepareForUpSyncing(objs objs : [NSManagedObject]!) -> Bool {
I am trying to trap keypresses using a CGEventTapCreate etc in Swift. I want to do this so that when my toddler is using a program the Mac, I can limit which keys do something. I appreciate that I need to enable assistive technology to get CGEventTap to work
I am having trouble understanding what I need to do in Swift to get the final parameter in the call. Essentially, I do not understand part of the documentation at quartz event documentation
Specifically the final parameter which is described by
refcon: A pointer to user-defined data. This pointer is passed into
the callback function specified in the callback parameter.
Here is my code missing the final parameter from the call
#IBAction func interceptKeyboard(sender: NSButton) {
CGEventTapCreate(kCGSessionEventTap, kCGHeadInsertEventTap, kCGEventTapOptionDefault, CGEventMask((kCGEventKeyDown) | (kCGEventKeyUp)), MyCallBack: kCGHIDEventTap, )
}
func MyCallBack(myLocation:CGEventTapLocation) -> Void{
//do Nowt
}
Bit of a while ago, but I think when you look at the type of CGEventTapCallback, you see that it is a CFunctionPointer:
typealias CGEventTapCallBack = CFunctionPointer<((CGEventTapProxy,
CGEventType, CGEvent!, UnsafeMutablePointer<Void>) -> Unmanaged<CGEvent>!)>
At the moment you can't create these in Swift, so you have to dip into objective-c, create a function there, which can then call an equivalent function in Swift, see here:
Objective-C Wrapper for CFunctionPointer to a Swift Closure
I ended up writing the whole of this stuff in Objective-C...
I've got some code that runs a fairly complex algorithm. I want to put together a fairly simple UI that will allow me to monitor the values of the various variables in the algorithm in some graphical ways -- think of it like a dashboard of sorts.
So, for simplicity's sake, let's say I have an algorithm like what follows. It searches a vector of values for the two values that most closely sum to a target value:
import Foundation
class algorithm {
var numbers = [Double]()
let numberOfRandoms = 1000
dynamic var a: String
dynamic var b: String
init () {
// Load initial vector with some random numbers between 0 and 1
for _ in 1...numberOfRandoms {
numbers.append(Double(arc4random()) / Double(UINT32_MAX))
}
a = " "
b = " "
}
func findTheTwoNumbersThatAddUpTheClosestToTarget(target: Double) {
//Initializing this to a very large value
var currentBestSum = 1000.0
//Begin brute force search for the optimal solution
for i in 0...numbers.count-2 {
for j in i+1...numbers.count-1 {
//Check to see if the current candidate exceeds the best solution
if abs(numbers[i] + numbers[j] - target) < currentBestSum {
//If it does, store the new champion
a = String(i)
b = String(j)
//And reset the current top score to match
currentBestSum = abs(numbers[i] + numbers[j]-target)
}
}
}
}
}
Now, this is just a simple (and silly) example, but it suits these purposes. I basically want to create a simple UI that displays the important values in the process as it runs (dynamically).
In this example, let's say that I just want to display two labels that contain the index values of the two leaders as the algorithm executes.
I created the labels in the storyboard.
Then, I created IBOutlets in the ViewController (Actually, storyboards did it for me when I Ctrl-dragged):
class ViewController: NSViewController {
#IBOutlet weak var a: NSTextField!
#IBOutlet weak var b: NSTextField!
override func viewDidLoad() {
super.viewDidLoad()
// Do any additional setup after loading the view.
}
override var representedObject: AnyObject? {
didSet {
// Update the view, if already loaded.
}
}
}
Then, I ctrl-dragged the labels to the a and b in the algorithm class to create the bindings.
Finally, I create an class variable in the view controller and instantiate it in the viewDidLoad method. This doesn't seem like the right thing to do -- maybe it is. Seems like you would want to keep separate your interface and data...
The labels do, in fact, show up -- but they never show any values of a and b. They just show the default text.
Not sure what I'm doing wrong.
Help!?
P.S., in response to Anthony Kong, I do recognize that I could manually synchronize all the view elements in the code, but I thought the whole point of using bindings was to avoid having to do this manual synchronization. I just can't figure out how to set it up.
Without commenting on your specific code I think I have experienced (and solved) the problem you describe. I was able to write an app that had two targets, one NIB-based and one Storyboard-based. As much as I was able I duplicated the code in each and shared the common data instance that I was trying to display in a TableView. The NIB-based app worked using the stock Cocoa Bindings that I set in IB. But the Storyboard-based app did not, the array controller did not see the data.
My solution was simply to add the binding for contentArray programmatically in viewDidLoad. The one line that fixed it for me is:
ac.bind("contentArray", toObject: cd, withKeyPath: "people", options: nil)
ac is the IBOutlet for the ArrayController in the Storyboard. cd is the class instance that contains the people array.
This is using XCode 6.2 (6C107a) which is Beta 3 I think.
This was the only binding that I had to set myself, the TableView to ArrayController (arrangedObjects) and TableViewCell to TableView (objectValue) didn't need any tweaking.
There are several problems with your code
1) In your code,
func applicationDidFinishLaunching(aNotification: NSNotification?) {
// Insert code here to initialize your application
var test = algorithm()
test.findTheTwoNumbersThatAddUpTheClosestToTarget(0.5)
}
The variable test goes out of scope when the function exits. Based on the wording of your question, you are expecting it to be a long running process. So this will not do what you want.
You should
a) Add test as class variable to the ViewController class
b) instantiate the variable in viewDidLoad method.
2) In your algorithm it does not actually provide any feedback to the labels. Maybe you think because the class has the ivar a and b so they are hooked to the IBOutlet by the same names. But of course it is not the case. And you do not need the keyword dynamic too.
What you should do is:
a) provide a method in the View Controller class to update the labels. It will serve as a callback function to be used by algorithm class to feedback the calculation result.
It may look like this:
func update_value_callback(vala: String, valb: String) {
a.text = vala; // updating the label here
b.text = valb;
}
b) make the algorithm class calls the callback function e.g.
func findTheTwoNumbersThatAddUpTheClosestToTarget(target: Double, viewController: ViewController) {
// do your stuff
...
// execute the callback
viewController.update_value_callback(a, b)
}
I'm trying to understand IOKit and how it allows me to access serial ports in a Swift program.
The class I'm manipulating at the moment is as follows:
import Foundation
import Cocoa
import IOKit.serial.IOSerialKeys
class Serial {
init() {
}
#IBOutlet var serialListPullDown : NSPopUpButton!
func refreshSerialList(defaultprompt: String) {
// remove everything from the pull down list
serialListPullDown?.removeAllItems()
// ask for all the serial ports
IOServiceGetMatchingServices(kIOMasterPortDefault, IOServiceMatching(kIOSerialBSDServiceValue), io_iterator_t)
}
}
Based on what I've read I think I've setup IOServiceMatchingServices correctly but I'm getting several errors such as "Expected member name or constructor call after type name" and "'o_iterator_t.Type' is not convertible to 'UnsafeMutualPointer'"
What does this mean?
A few different issues going on in there -- let's get your imports squared away first. It looks like you need these two:
import IOKit
import IOKit.serial
For the parameters, it'll be easier to see what we're working with if we define them as local variables, like so:
// this one's easy, just grabbing a constant from IOKit.serial
let masterPort: mach_port_t = kIOMasterPortDefault
// here we get back an Unmanaged<CFMutableDictionary> and need to convert it to
// a CFDictionary using takeRetainedValue()
let classesToMatch: CFDictionary = IOServiceMatching(kIOSerialBSDServiceValue).takeRetainedValue()
// the iterator that will contain the results of IOServiceGetMatchingServices
var matchingServices: io_iterator_t = 0
And lastly, you call the method:
// note that the last parameter is an UnsafeMutablePointer<io_iterator_t>
// so we need to prefix matchingServices with an ampersand (&)
let kernResult = IOServiceGetMatchingServices(masterPort, classesToMatch, &matchingServices)
if kernResult == KERN_SUCCESS {
// success
} else {
// error
}
This feels pretty near the edge of what Swift can handle right now -- definitely read these two pages well before going further:
Interacting with C APIs (Pointers section)
Working with Cocoa Data Types (Unmanaged Objects section)
Lastly, make sure you can get into the converted Swift declarations for the IOKit framework. There are a lot of useful comments and you'll be able to see which parameters and return values are unmanaged or pointers (since I don't think this framework's official documentation has been updated yet).
I'm working on iOS 8 custom keyboard extension right now, and there are some issues in using UITextInputDelegate Methods.
Does this right: selectionWillChange: and selectionDidChange: methods should be called when user long-presses typing area? And textWillChange: and textDidChange: methods should be called whenever the text is literally changing?
Actually, what I observed is that, when I changed selection in text input area, textWillChange: and textDidChange: are called, and I cannot get a clue that the other two methods are called in what condition. If anyone knows about the usage of these delegate methods, please let me know.
It's not working for me either... what I am currently doing is just using textWillChange and textDidChange which does get called, as you mentioned, when you change your selection... (they get called BEFORE and AFTER) And then comparing the: self.textDocumentProxy.documentContextBeforeInputself.textDocumentProxy.documentContextAfterInput From BEFORE (textWillChange) to the AFTER (textDidChange) to see if selection range or length changed at all.
Something like this (set the 4 NSStrings below in your .h file of course... haven't tested this exact snippet because I wrote it from scratch just now on SO.com but I'm sure the principle works if I made any errors)
- (void)textWillChange:(id<UITextInput>)textInput {
beforeStringOfTextBehindCursor = self.textDocumentProxy.documentContextBeforeInput;
beforeStringOfTextAfterCursor = self.textDocumentProxy.documentContextAfterInput;
}
- (void)textDidChange:(id<UITextInput>)textInput {
afterStringOfTextBehindCursor = self.textDocumentProxy.documentContextBeforeInput;
afterStringOfTextAfterCursor = self.textDocumentProxy.documentContextAfterInput;
BOOL didSelectionChange = NO;
if (![beforeStringOfTextBehindCursor isEqualToString:afterStringOfTextBehindCursor]) {
didSelectionChange = YES;
}
if (![beforeStringOfTextAfterCursor isEqualToString:afterStringOfTextAfterCursor]) {
didSelectionChange = YES;
}
if (didSelectionChange) {
NSLog(#"Selection Changed!");
}
}
I had the same problem with the functions specified in the UITextInput protocol not being called. The reason as far as I can discern is due to the fact that the inputDelegate is set at runtime. According to the ios docs:
The UIKit provides a private text input delegate, which it assigns at runtime to the inputDelegate property of the object whose class adopts the UITextInput protocol. link
The fix which works in my case is to reset the inputDelegate in the function:
textField:(UITextField *)textField shouldChangeCharactersInRange:(NSRange)range
replacementString:(NSString *)string
by the following line:
[myUITextField setInputDelegate:self];
where self implements the UITextInputDelegate protocol.