NSObject+BlockObservation.m: -dealloc swizzling should be smarter to avoid redundant swizzling

[ipmcc]

In the current implementation, if you have a hierarchy like this:

NSObject (implements dealloc) -> Foo (implements dealloc) -> Bar (doesn't implement dealloc)

the current dealloc swizzling will wrap Foo's dealloc, then wrap it again when Bar is swizzled, because the check is by classname. This will mean that when a Bar is dealloced, removeAllBlockObservers will be called twice. This isn't the end of the world since -removeAllBlockObservers appears idempotent, but it does take a lock (because of @synchronized), so it's not free.

A better check would be to keep track of the pointers to all the -dealloc IMPs that bk_addObserverForKeyPaths generated, and then only swizzle if the current IMP is not in that set. (i.e. disregard the classname, and look only at the implementation.)

Incidentally, it's probably not a big deal, but -removeAllBlockObservers could probably avoid taking that lock when called in dealloc if it had an argument that told it that it was in dealloc.

[zwaldowski]

The swizzled dealloc doesn't call removeAllBlockObservers, it calls bk_removeAllAssociatedObjects, which saves the lock you mention, however _BKObserver could and should not @synchronized during dealloc so that will be changed! Thanks.

[zwaldowski]

The way I describe it is as it is today in next. Sorry about the confusion I might've caused.