Have just spent about six hours trying to track down a bug in my code that was causing a crash in Synergy Advance. Turns out that it wasn’t a bug in my code at all, rather it was a misunderstanding of the way NSConnection works. I was using an NSConnection to communicate between threads, passing a pointer to an NSInvocation from one thread to another. The receiving thread was testing for pointer equality prompting my assertion to fail. The reason I spent so much time before finding the cause is that I assumed it was an error in my code; it wasn’t until the very end that I decided to test my assumptions about the way Cocoa works…

It turns out that because NSConnection uses NSDistantObject, a subclass of NSProxy, to do the communication it wasn’t just passing my pointer; it was actually creating a new NSInvocation on the fly, and passing a pointer to that object! I guess I should have known this if I’d’ve had more experience with distributed objects. A painful and frustrating lesson, nonetheless. I guess I’m a little surprised that it works this way; I can easily handle thread-safe access to the object, all I want to do is perform a pointer comparison!

So the "solution" is to pass something else. My first attempt was to pass an NSValue object containing the pointer, but that causes causes an NSInvalidArgumentException (cannot encode (void *) value: <170ff900 >). Seems like the way to go will be to use NSNumber to represent the pointer… Unbelievable.

Of course, once you’ve learned (the hard way, using the debugger) that this is the way NSConnection works, you quickly find some explanation in the documentation. In the description for the rootProxy method of NSConnection you click on the link to NSDistantObject, and from there to the Distributed Objects manual, where you find a heading, "Making Substitutions During Message Encoding", which tells all.

At the end of your journey you find good old NSPortCoder and see that it has two methods, isByCopy and isByRef which report whether the coder will encode by copying (the behaviour you’re trying to avoid) or by reference (but is this the desired behavior?; I suspect that what actually gets passed is an NSDistantObject proxy and not the real NSInvocation). But you’ve now got your working (but ugly) NSNumber workaround, so the question becomes, how much work will it take to force your NSConnection to pass the invocation as an unmodified pointer instead of by copying, and will it be worth it? Will you have to subclass NSConnection? Or NSInvocation? Or NSPortCoder? Might you even have to subclass NSDistantObject?

I’m going to explore these questions and see what I find out.