I am generally wondering how to let swift delegates run in a dedicated thread other than the main thread.
More specifically, I am currently using the HueSDK4EDK to establish a connection of my app to a Hue bridge. Part of the process is to define state observers and connection observers (as delegates) to handle incoming events.
private func buildBridge(withBridge bridge : BridgeInfoModel) -> PHSBridge {
return PHSBridge.init(block: { (builder) in
...
builder?.bridgeConnectionObserver = self
builder?.add(self)
}, withAppName: AppBundleName, withDeviceName: DeviceName)
}
The delegates are implemented in extensions, such as the connection observer:
extension HueApiManager : PHSBridgeConnectionObserver {
func bridgeConnection(_ bridgeConnection: PHSBridgeConnection!, handle connectionEvent: PHSBridgeConnectionEvent) {
...
}
}
Since some code within the connection observer may be time intensive, I am wondering if there is a more elegant way than this:
func bridgeConnection(_ bridgeConnection: PHSBridgeConnection!, handle connectionEvent: PHSBridgeConnectionEvent) {
let apiThread = DispatchQueue.global(qos: .utility)
apiThread.async {
...
}
}
Thanks a lot!
Yes:
let apiThread = DispatchQueue.global(qos: .utility)
func bridgeConnection(_ bridgeConnection: PHSBridgeConnection!, handle connectionEvent: PHSBridgeConnectionEvent) {
apiQueue.async {
...
}
}
There's no reason to fetch the queue every time; you can just store it as a property.
I know this isn't what you're thinking of, but it is the best approach. Just perform the dispatch when you need it.
It is possible to build something like what you're actually thinking of in ObjC using forwardInvocation:, but it doesn't translate well into Swift (the underlying machinery can't be implemented in Swift), and I don't recommend it. A trampoline is an object (sometimes an NSProxy object) that can accept any message, do something with it (like move it to another queue), and then redispatch it to another object. The problem is that it's hard to tell the complier "trust me, this will implement every method you need at runtime" (because you can't actually promise that, so it might crash). Even in ObjC, these often wound up being more trouble than they were worth, and the only reason they were worth the trouble in the first place is ObjC didn't always have GCD or blocks, and when blocks were added, the syntax was a headache. In Swift, it's much simpler. Just add the .async call.
Just to show what a trampoline looks like, though, here's what one would look like that uses an operation queue (very similar to a dispatch queue). This is written in pre-ARC ObjC and shouldn't be read as an example of modern ObjC.
#import <Foundation/Foundation.h>
#interface OperationQueueInvocationTrampoline : NSObject
{
#private
NSOperationQueue *myQueue;
id myTarget;
}
- (id)initWithQueue:(NSOperationQueue *)queue;
#end
#import "OperationQueueInvocationTrampoline.h"
#interface OperationQueueInvocationTrampoline ()
#property (nonatomic, readwrite, retain) NSOperationQueue *queue;
#property (nonatomic, readwrite, assign) id target;
#end
#implementation OperationQueueInvocationTrampoline
#synthesize queue = myQueue;
#synthesize target = myTarget;
- (id)initWithQueue:(NSOperationQueue *)queue
{
self = [super init];
if (self != nil)
{
self.queue = queue;
}
return self;
}
- (void)dealloc
{
[myQueue release];
myQueue = nil;
myTarget = nil;
[super dealloc];
}
- (NSMethodSignature *)methodSignatureForSelector:(SEL)selector
{
return [self.target methodSignatureForSelector:selector];
}
- (void)forwardInvocation:(NSInvocation *)invocation
{
[invocation setTarget:self.target];
self.target = nil;
NSOperation *operation = [[[NSInvocationOperation alloc] initWithInvocation:invocation] autorelease];
[self.queue addOperation:operation];
}
- (id)prepareWithInvocationTarget:(id)target
{
self.target = target;
return self;
}
#end
You'd use it like this:
id delegate = [[OperationQueueInvocationTrampoline alloc] initWithTarget:self];
[otherThing setDelegate:delegate];
This worked pretty well back when most protocols were informal (so there was no type-checking here, and you could just pass id), but it's gotten messier and messier to make this compile without warnings, and it would definitely be a headache in Swift today. It's have to save a lot of trouble, and it doesn't.
(Side note: queues are not the same thing as threads. The issue at hand is what queue things are dispatched to, not what thread they run on.)
I don't think this is possible unless that library specifically supports it.
One thing you could do is create a "proxy" delegate class that forwards all delegate method calls on a specific DispatchQueue.
It's not a perfect solution, but at least you can keep your HueApiManager a bit cleaner because you do not need apiThread.async everywhere.
class DelegateProxy: PHSBridgeConnectionObserver {
weak var delegate: PHSBridgeConnectionObserver?
var queue = DispatchQueue.global(qos: .utility)
func bridgeConnection(_ bridgeConnection: PHSBridgeConnection!, handle connectionEvent: PHSBridgeConnectionEvent) {
queue.async {
delegate?.bridgeConnection(bridgeConnection, handle: connectionEvent)
}
}
// Other delegate methods...
}
Related
I have a wrapper class that is built around a network request.
lets call it 'WrapperNetworkRequest'
The reason I wrap it is because I process the JSON response once it comes back.
Then using the 'WrapperNetworkRequestDelegate' protocol, return the response to the delegate.
The problem I am having is that when using the 'WrapperNetworkRequest' class as an autoreleased object, it auto-releases before the network request can complete.
I came up with what I think is a horrible idea ([self retain]+[self release] at the appropriate times)
Any ideas on what is a proper way to handle this/What I am doing wrong?
Thanks
Instead of having your WrapperNetworkRequest retain itself in its NSURLConnectionDelegate implementation methods, have its delegate assume ownership. One possible way to do this:
#interface Foo : NSObject <WrapperNetworkRequestDelegate>
#property (nonatomic, retain) WrapperNetworkRequest * wrappedRequest;
#end
#implementation Foo
#synthesize wrappedRequest;
//....
- (void)bar
{
WrapperNetworkRequest * request = [WrapperNetworkRequest aNewAutoreleasedRequest];
request.delegate = self;
self.wrappedRequest = request; // Foo instance assumes ownership
[request goGetData];
}
//....
#end
Retaining self is a perfectly sane idea when you have to extend the object's lifetime. How exactly you do that is another matter: you can do [self retain] or you can add self to a shared retaining array keeping all objects of the given kind. Think NSOperationQueue. After you add an operation object to a queue, you can safely release the operation because by adding it to the queue you've transferred ownership of the operation to the queue.
If I want to pass an object that was created on the main thread onto an NSOperation object, what's the standard way of doing to so that I'm not creating any memory management issues? Should I make my object's properties not have the 'nonatomic' attribute?
Right now, I allocate the objects via [[[AClass alloc] init] autorelease], keep a copy of the instance on my main thread and then pass another copy into the NSOperation as part of an NSArray. When I try to iterate through the array list objects inside NSOperation class and access one of the AClass's properties, the debugger reports that one of the member properties of AClass's instance object is already zombied while others are not. The error I'm seeing is:
-[CFString retain]: message sent to deallocated instance 0x5a8c6b0
*** -[CFString _cfTypeID]: message sent to deallocated instance 0x5a8c6b0
*** -[CFString _cfTypeID]: message sent to deallocated instance 0x5a8c6b0
I can't tell who is releasing my string properties too early but the entire object instance has not been released.
My class looks like:
#interface AClass
{
NSString *myTitle;
NSString *myDescription;
}
#property (nonatomic, retain, readonly) NSString *myTitle;
#property (nonatomic, retain, readonly) NSString *myDescription;
#end
#implementation AClass
#synthesize myTitle, myDescription;
- (void)dealloc
{
[myTitle release];
[myDescription release];
}
#end
Here's an updated snippet for an efficient, 'thread-safe' version of AClass:
/**
AClass is an immutable container:
- category methods must never change the state of AClass
*/
#interface AClass : NSObject < NSCopying >
{
#private
NSString * title;
NSString * description;
}
/**
subclassing notes:
- do not override properties: title, description
- implement #protocol NSCopying
*/
/*
1) document copy on entry here, even though the compiler has no
additional work to do.
2) nonatomic in this case - these ivars initialized and never mutate.
3) readonly because they are readonly
*/
#property (copy, readonly, nonatomic) NSString * title;
#property (copy, readonly, nonatomic) NSString * description;
/* prohibited: */
- (id)init;
/* designated initializer */
- (id)initWithTitle:(NSString *)inTitle description:(NSString *)inDescription;
#end
#implementation AClass
#synthesize title;
#synthesize description;
- (id)init
{
assert(0 && "use the designated initializer");
self = [super init];
[self release];
return 0;
}
- (id)initWithTitle:(NSString *)inTitle description:(NSString *)inDescription
{
self = [super init];
assert(self && "uh oh, NSObject returned 0");
if (0 != self) {
if (0 == inTitle || 0 == inDescription) {
assert(inTitle && inDescription && "AClass: invalid argument");
[self release];
return 0;
}
/* this would catch a zombie, if you were given one */
title = [inTitle copy];
description = [inDescription copy];
if (0 == title || 0 == description) {
assert(title && description && "string failed to copy");
[self release];
return 0;
}
}
return self;
}
- (void)dealloc
{
/* which could also happen when if your init fails, but the assertion in init will be hit first */
assert(title && description && "my ivars are not meant to be modified");
[title release], title = 0;
[description release], description = 0;
/* don't forget to call through super at the end */
[super dealloc];
}
- (id)copyWithZone:(NSZone *)zone
{
assert(self.title == title && self.description == description && "the subclasser should not override the accessors");
if ([self zone] == zone && [self class] == [AClass class]) {
/*
this is one possible (optional) optimization:
- avoid using this approach if you don't entirely understand
all the outlined concepts of immutable containers and low
level memory management in Cocoa and just use the
implementation in 'else'
*/
return [self retain];
}
else {
return [[[self class] allocWithZone:zone] initWithTitle:self.title description:self.description];
}
}
#end
Beyond that, avoid overusing autorelease calls so your memory issues are local to the callsite. This approach will solve many issues (although memory issues may still exist in your app).
Update in response to questions:
Justin Galzic: so basically, the copy
ensures that objects are local to the
caller and when the instance is shared
out to the thread on which
NSOperations is on, they're two
different instances?
actually, the copy call to an immutable string could perform a retain.
as an example: AClass could now implement #protocol NSCopying by simply retaining the 2 strings. also, if you know AClass is never subclassed, you could just return [self retain] when the objects are allocated in the same NSZone.
if a mutable string is passed to the initializer of AClass, then it will (of course) perform a concrete copy.
if you want objects to share these strings, then this approach is preferred (in most cases) because you (and all clients using AClass) now know the ivars will never change behind your back (what they point to, as well as the strings' contents). of course, you still have the ability to make the mistake of changing what title and description point to in the implementation of AClass - this would break the policy you've established. if you wanted to change what the members of AClass pointed to, you'd have to use locks, #synchronized directives (or something similar) - and then typically set up some observation callbacks so you could guarantee that your class works as expected... all that is unnecessary for most cases because the above immutable interface is perfectly simple for most cases.
to answer your question: the call to copy is not guaranteed to create a new allocation - it just allows several guarantees to propagate to clients, while avoiding all thread safety (and locking/synchronizing).
What if there are some cases where you
do want multiple classes (on the same
thread) to share this object? Would
you then make an implicit copy of this
object and then pass along to the
NSOperation?
now that i've detailed how copying immutable objects can be implemented. it should be obvious that properties of immutable objects (NSString, NSNumber, etc.) should be declared copy in many cases (but many Cocoa programmers don't declare them that way).
if you want to share a NSString which you know is immutable, you should just copy it from AClass.
if you want to share an instance of AClass you have 2 choices:
1) (best) implement #protocol NSCopying in AClass: - (id)copyWithZone: implementation added above.
now the client is free to copy and retain AClass as is most logical for their needs.
2) (BAD) expect that all clients will keep their code up to date with changes to AClass, and to use copy or retain as required. this is not realistic. it is a good way to introduce bugs if your implementation of AClass needs to change because clients will not always update their programs accordingly. some people consider this acceptable when the object is private in a package (e.g., only one class uses and sees its interface).
in short, it's best to keep the retain and copy semantics predictable - and just hide all the implementation details in your class so your clients' code never breaks (or is minimized).
if your object is truly shared and its state is mutable, then use retain and implement callbacks for state changes. otherwise, keep it simple and use immutable interfaces and concrete copying.
if an object has an immutable state, then this example is always a lock free thread safe implementation with many guarantees.
for an implementation of an NSOperation subclass, i find it best (in most cases) to:
- create an object which provides all the context it needs (e.g., an url to load)
- if the something needs to know about the operation's result or to use the data, then create a #protocol interface for the callbacks and add a member to the operation subclass which is retained by the NSOperation subclass, and which you've prohibited from pointing to another object during the lifetime of the NSOperation instance:
#protocol MONImageRenderCallbackProtocol
#required
/** ok, the operation succeeded */
- (void)imageRenderOperationSucceeded:(AClass *)inImageDescriptor image:(NSImage *)image;
#required
/** bummer. the image request failed. see the #a error */
- (void)imageRenderOperationFailed:(AClass *)inImageDescriptor withError:(NSError *)error;
#end
/* MONOperation: do not subclass, create one instance per render request */
#interface MONOperation : NSOperation
{
#private
AClass * imageDescriptor; /* never change outside initialization/dealloc */
NSObject<MONImageRenderCallbackProtocol>* callback; /* never change outside initialization/dealloc */
BOOL downloadSucceeded;
NSError * error;
}
/* designated initializer */
- (id)initWithImageDescriptor:(AClass *)inImageDescriptor callback:(NSObject<MONImageRenderCallbackProtocol>*)inCallback;
#end
#implementation MONOperation
- (id)initWithImageDescriptor:(AClass *)inImageDescriptor callback:(NSObject<MONImageRenderCallbackProtocol>*)inCallback
{
self = [super init];
assert(self);
if (0 != self) {
assert(inImageDescriptor);
imageDescriptor = [inImageDescriptor copy];
assert(inCallback);
callback = [inCallback retain];
downloadSucceeded = 0;
error = 0;
if (0 == imageDescriptor || 0 == callback) {
[self release];
return 0;
}
}
return self;
}
- (void)dealloc
{
[imageDescriptor release], imageDescriptor = 0;
[callback release], callback = 0;
[error release], error = 0;
[super dealloc];
}
/**
#return an newly rendered NSImage, created based on self.imageDescriptor
will set self.downloadSucceeded and self.error appropriately
*/
- (NSImage *)newImageFromImageDescriptor
{
NSImage * result = 0;
/* ... */
return result;
}
- (void)main
{
NSAutoreleasePool * pool = [NSAutoreleasePool new];
NSImage * image = [self newImageFromImageDescriptor];
if (downloadSucceeded) {
assert(image);
assert(0 == error);
[callback imageRenderOperationSucceeded:imageDescriptor image:image];
[image release], image = 0;
}
else {
assert(0 == image);
assert(error);
[callback imageRenderOperationFailed:imageDescriptor withError:error];
}
[pool release], pool = 0;
}
#end
If there is something I do not recommend, it is that you are keeping your reference unretained by the knowledge that it will be retained by someone else already. Because the reference is unretained, when it is removed from the array it will be released, retain count might drop to zero and object might be dealloc and now you are holding time bomb i.e. invalid reference.
I would suggest that do not autorelease the reference, and try to do break point at your dealloc and see the stack call to see who cause your object to be dealloc.
I am a little confused by this snippet of code (presented in the CocoaFundamentals guide) that overrides some of the methods when creating a singleton instance.
static id sharedReactor = nil;
+(id)sharedInstance {
if(sharedReactor == nil) sharedReactor = [[super allocWithZone:NULL] init];
return sharedReactor;
}
.
+(id)allocWithZone:(NSZone *)zone {
return[[self sharedInstance] retain];
}
-(id)retain {
return self;
}
In the code where the singleton instance is created the +sharedInstance method calls [super allocWithZone:NILL] from the superclass (which in my case is NSObject) The allocWithZone above is only called if you attempt to use it to create a new singleton.
The bit I am confused about is the use of retain, especially seeing as retain is also overridden to return self. Can anyone explain this, could it not be written:
+(id)allocWithZone:(NSZone *)zone {
return [self sharedInstance];
}
-(id)retain {
return self;
}
EDIT_001:
Based on comments and reading various posts on the web I have decided to go with the following (see below) I have chosen to go for a shared singleton approach where if needed I would have the option of creating a second or third instance. Also at this stage as I am only using the singleton for the model portion of MVC for a simple iPhone app I have decided to leave thread safety out. I am aware its important and as I get more familiar with iPhone programming I will likely use +initialize instead (keeping in mind the subclass issue where it can be called twice) Also I have added a dealloc, firstly to log a message should the singleton be released, but also to clean things up properly should the singleton be no longer required.
#interface SharedManager : NSObject
+(id)sharedInstance;
#end
#implementation SharedManager
static id myInstance = nil;
+(id)sharedInstance {
if(myInstance == nil) {
myInstance = [[self alloc] init];
}
return myInstance;
}
-(void)dealloc {
NSLog(#"_deal: %#", [self class]);
[super dealloc];
myInstance = nil;
}
#end
In testing I found that I had a set the static variable to nil in the dealloc or it maintained its pointer to the original object. I was initially a little confused by this as I was expecting the scope of the static to be the instance, I guess its the class instead, which makes sense.
cheers gary
First, don't use this code. There is almost never a reason to do all this for a simple singleton. Apple is demonstrating a "Forced Singleton," in that it is impossible to create two of them. It is very rare to really need this. You can almost always use the "shared singleton" approach used by most of the Cocoa objects that have a singleton constructor.
Here's my preferred way of implementing shared singleton:
+ (MYManager *)sharedManager
{
static MYManager *sharedManager = nil;
if (sharedManager == nil)
{
sharedManager = [[self alloc] init];
}
return sharedManager;
}
That's it. No other code is required. Callers who use +sharedManager will get the shared instance. Callers who call +alloc can create unique instances if they really want to. This is how such famous "singletons" as NSNotificationCenter work. If you really want your own private notification center, there is no reason the class should forbid it. This approach has the following advantages:
Less code.
More flexible in cases where a non-shared instance is useful.
Most importantly: the code does what it says it does. A caller who thinks he's making a unique instance with +alloc doesn't encounter surprising "spooky action at a distance" behavior that requires him to know an internal implementation detail of the object.
If you really need a forced singleton because the object in question maps to a unique resource that cannot be shared (and it's really rare to encounter such a situation), then you still shouldn't use +alloc trickery to enforce it. This just masks a programming error of trying to create a new instance. Instead, you should catch the programming error this way:
+ (MYManager *)sharedManager
{
static MYManager *sharedManager = nil;
if (sharedManager == nil)
{
sharedManager = [[self alloc] initSharedManager];
}
return sharedManager;
}
- (id)init
{
NSAssert(NO, #"Attempting to instantiate new instance. Use +sharedManager.");
return nil;
}
// Private method. Obviously don't put this in your .h
- (id)initSharedManager
{
self = [super init];
....
return self;
}
There is a good example of different singleton methods with comments here on SO:
What does your Objective-C singleton look like?
If it helps, the example has a different approach to allocWithZone: which returns nil.
What's the best way to handle memory management with nested factory methods, such as in the following example?
#interface MyClass : NSObject {
int _arg;
}
+ (MyClass *) SpecialCase1;
+ (MyClass *) SpecialCase2;
+ (MyClass *) myClassWithArg:(int)arg;
- (id) initWithArg:(int)arg;
#property (nonatomic, assign) int arg;
#end
#implementation MyClass
#synthesize arg = _arg;
+ (MyClass *) SpecialCase1
{
return [MyClass myClassWithArg:1];
}
+ (MyClass *) SpecialCase2
{
return [MyClass myClassWithArg:2];
}
+ (MyClass *) myClassWithArg:(int)arg
{
MyClass *instance = [[[MyClass alloc] initWithArg:arg] autorelease];
return instance;
}
- (id) initWithArg:(int)arg
{
self = [super init];
if (nil != self) {
self.arg = arg;
}
return self;
}
#end
The problem here (I think) is that the autorelease pool is flushed before the SpecialCaseN methods return to their callers [Edit: apparently not - see comments below]. Hence, the ultimate caller of SpecialCaseN can't rely on the result having been retained. (I get "[MyClass copyWithZone:]: unrecognized selector sent to instance 0x100110250" on trying to assign the result of [MyClass SpecialCase1] to a property on another object.)
The reason for wanting the SpecialCaseN factory methods is that in my actual project, there are multiple parameters required to initialize the instance and I have a pre-defined list of "model" instances that I'd like to be able to create easily.
I'm sure there's a better approach than this.
[Edit: #interface added per request.]
Why do you think the autorelease pool is being flushed?
The autorelease pool is not flushed in cocoa-touch unless either you flush it, or control is returned back to the event loop.
From the iPhone Memory Management Guide
The Application Kit automatically
creates a pool at the beginning of an
event cycle (or event-loop iteration),
such as a mouse down event, and
releases it at the end, so your code
normally does not have to worry about
them.
The error describes the problem exactly: copyWithZone is called and it's not implemented in MyClass. Check out the attributes or the implementation of the property it's assigned to (see if it's copying). Also, copyWithZone will be called if the instance of MyClass is used as a key to an NSDictionary. Here's some info on implementing copyWithZone.
I'm just getting into iPhone development after many years doing Java development. I'm looking for the Objective-C equivalent to Java's BlockingQueue. Is there something like that?
In case I'm going about things the wrong way, here's what I'm trying to achieve:
I want to display, one at a time, chunks of data pulled from a network server. To keep the user from noticing network lag, I want to always have a few chunks of data pre-fetched. In Java-land, I'd use a thread-safe queue between my fetching thread and my display thread.
Here's an implementation of a blocking queue with a queue and dequeue method. The expectation would be that one thread goes into a loop calling dequeueUnitOfWorkWaitingUntilDate: and processes units of work while a second thread is calling queueUnitOfWork:.
#interface MyBlockingQueue : NSObject {
NSMutableArray *queue;
NSConditionLock *queueLock;
}
- (id)dequeueUnitOfWorkWaitingUntilDate:(NSDate *)timeoutData;
- (void)queueUnitOfWork:(id)unitOfWork;
#end
enum {
kNoWorkQueued = 0,
kWorkQueued = 1
}
#implementation MyBlockingQueue
- (id)init {
if ((self = [super init])) {
queueLock = [[NSConditionLock alloc] initWithCondition:kNoWorkQueued];
workItems = [[NSMutableArray alloc] init];
}
return self;
}
- (void)dealloc {
[queueLock release];
[workItems release];
[super dealloc];
}
- (id)dequeueUnitOfWorkWaitingUntilDate:(NSDate *)timeoutDate {
id unitOfWork = nil;
if ([queueLock lockWhenCondition:kWorkQueued beforeDate:timeoutDate]) {
unitOfWork = [[[queue objectAtIndex:0] retain] autorelease];
[queue removeObjectAtIndex:0];
[queueLock unlockWithCondition:([workItems count] ? kWorkQueued : kNoWorkQueued)];
}
return unitOfWork;
}
- (void)queueUnitOfWork:(id)unitOfWork {
[queueLock lock];
[queue addObject:unitOfWork];
[queueLock unlockWithCondition:kWorkQueued];
}
#end
You can simply spin off an NSOperation and post a notification when the data has come back (finished loading). Take a look at Dave Dribin's blog post on concurrency with NSOperation that shows how to encapsulate an NSURLConnection session:
http://www.dribin.org/dave/blog/archives/2009/05/05/concurrent_operations/
If you are not talking about accessing a web service or site where NSURLConnection is appropriate, you can instead use Cocoa Async Socket if it's straight TCP/IP or UDP:
http://code.google.com/p/cocoaasyncsocket/
Best Regards,
I don't think such a thing exists natively - you're probably going to have to write your own class that maintains a queue of network objects. Your header might look something like:
#interface ObjcBlockingQueue : NSObject {
// The objects that you're holding onto
NSArray *objects;
}
#property(nonatomic,retain) NSArray *objects;
- (ServerData *)getNextChunk;
Then you can implement getNextChunk to pop and return the top object off your objects array, and if [objects count] is less than a certain value, launch a thread to fetch some more objects (probably using NSURLConnection with ObjcBlockingQueue being the delegate). You can also have that thread/connection launched inside an overridden init method to prefill the queue.
You might also want to think about adding a
- (BOOL)isChunkAvailable;
method that will let your display thread know whether it can display something new right away or if it has to display a loading message. Depending on where you're displaying the data and how your app is structured, it may also be worth your while to make ObjcBlockingQueue a singleton class.