问题
I would like to unit test a class that acts as a CBPeripheralManagerDelegate to the CBPeripheralManager class. Typically, in order to stub out an external class dependency, I would use either a form of dependency injection by passing in via the class initializer or via a property. When dealing with singleton-based API's, I have been able to use libraries like Kiwi to stub the class level method that returns the singleton (i.e. [ClassName stub:@selector(sharedInstance) andReturn:myStubbedInstance]). The issue in the case of mocking CBPeripheralManager is that its initializer takes the delegate instance. So any code that uses my class would need to do something like this:
PeripheralManagerWrapper *wrapper = [[PeripheralManagerWrapper alloc] init];
CBPeripheralManager *peripheralManager = [[CBPeripheralManager alloc] initWithDelegate:wrapper queue:nil options:nil];
wrapper.peripheralManager = peripheralManager;
Then, for unit testing my PeripheralManagerWrapper class, I could simply instantiate it and pass in a mocked CBPeripheralManager. However, I don't like requiring any calling code of my wrapper object to have to go through this setup. Is there a better pattern for dealing with this situation? I've used both Kiwi and OCMockito, but neither seem to provide this functionality short of maybe stubbing the alloc and init methods of CBPeripheralManager and then just instantiating the instance in the PeripheralManagerWrapper
's initializer.
回答1:
IMHO, the Core Bluetooth APIs are perfect match for unit testing. All the delegate callbacks take the manager and the relevant parameters so if you follow the pattern that you use these arguments instead of the internal state, then you will be able to pass in anything you want. Using mock objects is the best way to do this. While unit testing, you shouldn't try to mock the behavior of the managers. You should focus on verifying the interaction of your code with the API and nothing more.
Wrappers may better suit integration testing. But actually, the integration testing of Core Bluetooth code is better done manually to my experience. The stack is not stable enough to allow for reliable testing, and the test code would have to be fortified against the stack errors too which is really hard as, obviously, those are not documented or predictable just by looking at the APIs. While on the other hand, your test code would have to simulate the erroneous behavior of the stack too. There may be cases when it is possible but the test code will be just as complex if not more than the code you are testing.
来源：https://stackoverflow.com/questions/21336643/how-does-one-unit-test-code-that-interacts-with-the-core-bluetooth-apis