NSCache auto-removal policy

Question

What are some of the NSCache's auto-removal policies? Apple's documentation does not mention them, and I experimentally discovered that NSCache does not respond to memory warning.

Answer 1

NSCache does not respond to UIApplicationDidReceiveMemoryWarningNotification, but it does automatically evict its objects in low memory situations, obviously using some other mechanism.

While I previously suggested observing UIApplicationDidReceiveMemoryWarningNotification, this is not the case. No special handling for low memory situations is needed, as NSCache handles this automatically.

---

Update:

As of iOS 7, the NSCache not only doesn't respond to memory warnings, but it also does not appear to properly purge itself upon memory pressure, either (see NSCache crashing when memory limit is reached (only on iOS 7)).

I subclass NSCache to observe UIApplicationDidReceiveMemoryWarningNotification, and purge the cache upon memory warning:

@interface AutoPurgeCache : NSCache
@end

@implementation AutoPurgeCache

- (id)init
{
    self = [super init];
    if (self) {
        [[NSNotificationCenter defaultCenter] addObserver:self selector:@selector(removeAllObjects) name:UIApplicationDidReceiveMemoryWarningNotification object:nil];
    }
    return self;
}

- (void)dealloc
{
    [[NSNotificationCenter defaultCenter] removeObserver:self name:UIApplicationDidReceiveMemoryWarningNotification object:nil];

    // if not ARC, also
    //
    // [super dealloc];
}

@end

Answer 2

You're best off treating NSCache as a black box, as much as you can.

From Caching and Purgeable Memory (emphasis mine):

When adding items to a cache, you can specify a cost value to be associated with each key-value pair. Call the setTotalCostLimit: method to set the maximum value for the sum of all the cached objects’ costs. Thus, when an object is added that pushes the totalCost above the totalCostLimit, the cache could automatically evict some of its objects in order to get back below the threshold. This eviction process is not guaranteed, so trying to manipulate the cost values to achieve specific behavior could be detrimental to the performance of the cache. Pass in 0 for the cost if you have nothing useful, or use the setObject:forKey: method, which does not require a cost to be passed in.

Note: The count limit and the total-cost limit are not strictly enforced. That is, when the cache goes over one of its limits, some of its objects might get evicted immediately, later, or never, all depending on the implementation details of the cache.