Is it abusive to use IDisposable and “using” as a means for getting “scoped behavior” for exception safety?

I consider this to be an abuse of the using statement. I am aware that I’m in the minority on this position.

I consider this to be an abuse for three reasons.

First, because I expect that “using” is used to use a resource and dispose of it when you’re done with it. Changing program state is not using a resource and changing it back is not disposing anything. Therefore, “using” to mutate and restore state is an abuse; the code is misleading to the casual reader.

Second, because I expect “using” to be used out of politeness, not necessity. The reason you use “using” to dispose of a file when you’re done with it is not because it is necessary to do so, but because it is polite — someone else might be waiting to use that file, so saying “done now” is the morally correct thing to do. I expect that I should be able to refactor a “using” so that the used resource is held onto for longer, and disposed of later, and that the only impact of doing so is to slightly inconvenience other processes. A “using” block which has semantic impact on program state is abusive because it hides an important, required mutation of program state in a construct that looks like it is there for convenience and politeness, not necessity.

And third, your program’s actions are determined by its state; the need for careful manipulation of state is precisely why we’re having this conversation in the first place. Let’s consider how we might analyze your original program.

Were you to bring this to a code review in my office, the first question I would ask is “is it really correct to lock the frobble if an exception is thrown?” It is blatantly obvious from your program that this thing aggressively re-locks the frobble no matter what happens. Is that right? An exception has been thrown. The program is in an unknown state. We do not know whether Foo, Fiddle or Bar threw, why they threw, or what mutations they performed to other state that were not cleaned up. Can you convince me that in that terrible situation it is always the right thing to do to re-lock?

Maybe it is, maybe it isn’t. My point is, that with the code as it was originally written, the code reviewer knows to ask the question. With the code that uses “using”, I don’t know to ask the question; I assume that the “using” block allocates a resource, uses it for a bit, and politely disposes of it when it is done, not that the closing brace of the “using” block mutates my program state in an exceptional cirumstance when arbitrarily many program state consistency conditions have been violated.

Use of the “using” block to have a semantic effect makes this program fragment:

}

extremely meaningful. When I look at that single close brace I do not immediately think “that brace has side effects which have far-reaching impacts on the global state of my program”. But when you abuse “using” like this, suddenly it does.

The second thing I would ask if I saw your original code is “what happens if an exception is thrown after the Unlock but before the try is entered?” If you’re running a non-optimized assembly, the compiler might have inserted a no-op instruction before the try, and it is possible for a thread abort exception to happen on the no-op. This is rare, but it does happen in real life, particularly on web servers. In that case, the unlock happens but the lock never happens, because the exception was thrown before the try. It is entirely possible that this code is vulnerable to this problem, and should actually be written

bool needsLock = false;
try
{
    // must be carefully written so that needsLock is set
    // if and only if the unlock happened:

    this.Frobble.AtomicUnlock(ref needsLock);
    blah blah blah
}
finally
{
    if (needsLock) this.Frobble.Lock();
}

Again, maybe it does, maybe it doesn’t, but I know to ask the question. With the “using” version, it is susceptible to the same problem: a thread abort exception could be thrown after the Frobble is locked but before the try-protected region associated with the using is entered. But with the “using” version, I assume that this is a “so what?” situation. It’s unfortunate if that happens, but I assume that the “using” is only there to be polite, not to mutate vitally important program state. I assume that if some terrible thread abort exception happens at exactly the wrong time then, oh well, the garbage collector will clean up that resource eventually by running the finalizer.