An operation object is a single-shot object—that is, it executes its task once and cannot be used to execute it again. You typically execute operations by adding them to an operation queue (an instance of the NSOperationQueue class). An operation queue executes its operations either directly, by running them on secondary threads, or indirectly using the libdispatch library (also known as Grand Central Dispatch). For more information about how queues execute operations, see NSOperationQueue Class Reference.

If you do not want to use an operation queue, you can execute an operation yourself by calling its startmethod directly from your code. Executing operations manually does put more of a burden on your code, because starting an operation that is not in the ready state triggers an exception. The ready property reports on the operation’s readiness.

Operation Dependencies

Dependencies are a convenient way to execute operations in a specific order. You can add and remove dependencies for an operation using the addDependency: and removeDependency: methods. By default, an operation object that has dependencies is not considered ready until all of its dependent operation objects have finished executing. Once the last dependent operation finishes, however, the operation object becomes ready and able to execute.

The dependencies supported by NSOperation make no distinction about whether a dependent operation finished successfully or unsuccessfully. (In other words, canceling an operation similarly marks it as finished.) It is up to you to determine whether an operation with dependencies should proceed in cases where its dependent operations were cancelled or did not complete their task successfully. This may require you to incorporate some additional error tracking capabilities into your operation objects.

KVO-Compliant Properties

The NSOperation class is key-value coding (KVC) and key-value observing (KVO) compliant for several of its properties. As needed, you can observe these properties to control other parts of your application. To observe the properties, use the following key paths:

• isCancelled - read-only

• isAsynchronous - read-only

• isExecuting - read-only

• isFinished - read-only

• isReady - read-only

• dependencies - read-only

• queuePriority - readable and writable

• completionBlock - readable and writable

Although you can attach observers to these properties, you should not use Cocoa bindings to bind them to elements of your application’s user interface. Code associated with your user interface typically must execute only in your application’s main thread. Because an operation may execute in any thread, KVO notifications associated with that operation may similarly occur in any thread.

If you provide custom implementations for any of the preceding properties, your implementations must maintain KVC and KVO compliance. If you define additional properties for your NSOperation objects, it is recommended that you make those properties KVC and KVO compliant as well. For information on how to support key-value coding, see Key-Value Coding Programming Guide. For information on how to support key-value observing, see Key-Value Observing Programming Guide.

Multicore Considerations

The NSOperation class is itself multicore aware. It is therefore safe to call the methods of an NSOperationobject from multiple threads without creating additional locks to synchronize access to the object. This behavior is necessary because an operation typically runs in a separate thread from the one that created and is monitoring it.

When you subclass NSOperation, you must make sure that any overridden methods remain safe to call from multiple threads. If you implement custom methods in your subclass, such as custom data accessors, you must also make sure those methods are thread-safe. Thus, access to any data variables in the operation must be synchronized to prevent potential data corruption. For more information about synchronization, see Threading Programming Guide.

Asynchronous Versus Synchronous Operations

这里提到的同步和异步operation 指的是 是否在start方法中自己开启新线程处理operation中的任务。

If you plan on executing an operation object manually, instead of adding it to a queue, you can design your operation to execute in a synchronous or asynchronous manner. Operation objects are synchronous by default. In a synchronous operation, the operation object does not create a separate thread on which to run its task. When you call the start method of a synchronous operation directly from your code, the operation executes immediately in the current thread. By the time the start method of such an object returns control to the caller, the task itself is complete. 这种默认调用就像调用一个普通的函数。

When you call the start method of an asynchronous operation, that method may return before the corresponding task is completed. An asynchronous operation object is responsible for scheduling its task on a separate thread. The operation could do that by starting a new thread directly, by calling an asynchronous method, or by submitting a block to a dispatch queue for execution. It does not actually matter if the operation is ongoing when control returns to the caller, only that it could be ongoing.

If you always plan to use queues to execute your operations, it is simpler to define them as synchronous. If you execute operations manually, though, you might want to define your operation objects as asynchronous. Defining an asynchronous operation requires more work, because you have to monitor the ongoing state of your task and report changes in that state using KVO notifications. But defining asynchronous operations is useful in cases where you want to ensure that a manually executed operation does not block the calling thread.

When you add an operation to an operation queue, the queue ignores the value of the asynchronous property and always calls the start method from a separate thread. Therefore, if you always run operations by adding them to an operation queue, there is no reason to make them asynchronous.

上面说的是，如果你使用了queue，就不必在start方法中自己建立线程了，系统会自动建立新线程运行start函数。

注意，这里自动异步执行的是start方法，如果start方法中调用了 [self performSelector:@selector(executeOperation) onThread:[[self class] globalAPIOperationNetworkThread] withObject:nil waitUntilDone:NO]; 之类的方法，而真正的业务逻辑在executeOperation函数中，那么start函数返回后，真正的业务逻辑会在globalAPIOperationNetworkThread中执行！所以说，nsoperation真正的逻辑到底执行在哪里是要看operation start中的具体的实现！

For information on how to define both synchronous and asynchronous operations, see the subclassing notes.

Subclassing Notes

The NSOperation class provides the basic logic to track the execution state of your operation but otherwise must be subclassed to do any real work. How you create your subclass depends on whether your operation is designed to execute concurrently or non-concurrently.

Methods to Override

For non-concurrent operations, you typically override only one method:

• main

Into this method, you place the code needed to perform the given task. Of course, you should also define a custom initialization method to make it easier to create instances of your custom class. You might also want to define getter and setter methods to access the data from the operation. However, if you do define custom getter and setter methods, you must make sure those methods can be called safely from multiple threads.

If you are creating a concurrent operation, you need to override the following methods and properties at a minimum:

• start

• asynchronous

• executing

• finished

In a concurrent operation, your start method is responsible for starting the operation in an asynchronous manner. Whether you spawn a thread or call an asynchronous function, you do it from this method. Upon starting the operation, your start method should also update the execution state of the operation as reported by the executing property. You do this by sending out KVO notifications for the executing key path, which lets interested clients know that the operation is now running. Your executing property must also provide the status in a thread-safe manner.

Upon completion or cancellation of its task, your concurrent operation object must generate KVO notifications for both the isExecuting and isFinished key paths to mark the final change of state for your operation. (In the case of cancellation, it is still important to update the isFinished key path, even if the operation did not completely finish its task. Queued operations must report that they are finished before they can be removed from a queue.) In addition to generating KVO notifications, your overrides of the executing and finishedproperties should also continue to report accurate values based on the state of your operation.

For additional information and guidance on how to define concurrent operations, see Concurrency Programming Guide.

Even for concurrent operations, there should be little need to override methods other than those described above. However, if you customize the dependency features of operations, you might have to override additional methods and provide additional KVO notifications. In the case of dependencies, this would likely only require providing notifications for the isReady key path. Because the dependencies property contains the list of dependent operations, changes to it are already handled by the default NSOperation class.

Maintaining Operation Object States

Operation objects maintain state information internally to determine when it is safe to execute and also to notify external clients of the progression through the operation’s life cycle. Your custom subclasses must maintain this state information to ensure the correct execution of operations in your code. Table 1 lists the key paths associated with an operation’s states and how you should manage that key path in any custom subclasses.

Responding to the Cancel Command

Once you add an operation to a queue, the operation is out of your hands. The queue takes over and handles the scheduling of that task. However, if you decide later that you do not want to execute the operation after all—because the user pressed a cancel button in a progress panel or quit the application, for example—you can cancel the operation to prevent it from consuming CPU time needlessly. You do this by calling the cancelmethod of the operation object itself or by calling the cancelAllOperations method of the NSOperationQueueclass.

Canceling an operation does not immediately force it to stop what it is doing. Although respecting the value in the cancelled property is expected of all operations, your code must explicitly check the value in this property and abort as needed. The default implementation of NSOperation includes checks for cancellation. For example, if you cancel an operation before its start method is called, the start method exits without starting the task.

You should always support cancellation semantics in any custom code you write. In particular, your main task code should periodically check the value of the cancelled property. If the property reports the value YES, your operation object should clean up and exit as quickly as possible. If you implement a custom start method, that method should include early checks for cancellation and behave appropriately. Your custom start method must be prepared to handle this type of early cancellation.这里说的是，cancel完全是依赖我们写的代码，不自己写出检测条件，就不可能拥有cancel机能。

In addition to simply exiting when an operation is cancelled, it is also important that you move a cancelled operation to the appropriate final state. Specifically, if you manage the values for the finished and executingproperties yourself (perhaps because you are implementing a concurrent operation), you must update those properties accordingly. Specifically, you must change the value returned by finished to YES and the value returned by executing to NO. You must make these changes even if the operation was cancelled before it started executing. 这和operation的移除有关！不正确设置标志位，系统无法对operation进行操作！