Is it abusive to use IDisposable and "using" as a means for getting "scoped behavior" for exception safety?_问答_开发者

Something I often used back in C++ was letting a class A handle a state entry and exit condition for another class B, via the A constructor and destructor, to make sure that if something in that scope threw an exception, then B would have a known state when the scope was exited. This isn't pure RAII as far as the acronym goes, but it's an established pattern nevertheless.

In C#, I often want to do

class FrobbleManager
{
    ...

    private void FiddleTheFrobble()
    {
        this.Frobble.Unlock();
        Foo();                  // Can throw
        this.Frobble.Fiddle();  // Can throw
        Bar();                  // Can throw
        this.Frobble.Lock();
    }
}

Which needs to be done like this

private void FiddleTheFrobble()
{
    this.Frobble.Unlock();

    try
    {            
        Foo();                  // Can throw
        this.Frobble.Fiddle();  // Can throw
        Bar();                  // Can throw
    }
    finally
    {
        this.Frobble.Lock();
    }
}

if I want to guarantee the Frobble state when FiddleTheFrobble returns. The code would be nicer with

private void FiddleTheFrobble()
{
    using (var janitor = new FrobbleJanitor(this.Frobble))
    {            
        Foo();                  // Can throw
        this.Frobble.Fiddle();  // Can throw
        Bar();                  // Can throw
    }
}

where FrobbleJanitor looks roughly like

class FrobbleJanitor : IDisposable
{
    private Frobble frobble;

    public FrobbleJanitor(Frobble frobble)
    {
        this.frobble = frobble;
        this.frobble.Unlock();
    }

    public void Dispose()
    {
        this.frobble.Lock();
    }
}

And that's how I want to do it. Now reality catches up, since what I want to use requires that the FrobbleJanitor is used with using. I could开发者_开发问答 consider this a code review issue, but something is nagging me.

Question: Would the above be considered as abusive use of using and IDisposable?

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.

I don't think so, necessarily. IDisposable technically is meant to be used for things that have non-managed resources, but then the using directive is just a neat way of implementing a common pattern of try .. finally { dispose }.

A purist would argue 'yes - it's abusive', and in the purist sense it is; but most of us do not code from a purist perspective, but from a semi-artistic one. Using the 'using' construct in this way is quite artistic indeed, in my opinion.

You should probably stick another interface on top of IDisposable to push it a bit further away, explaining to other developers why that interface implies IDisposable.

There are lots of other alternatives to doing this but, ultimately, I can't think of any that will be as neat as this, so go for it!

Eric Gunnerson, who was on the C# language design team, gave this answer to pretty much the same question:

Doug asks:

re: A lock statement with timeout...

I've done this trick before to deal with common patterns in numerous methods. Usually lock acquisition, but there are some others. Problem is it always feels like a hack since the object isn't really disposable so much as "call-back-at-the-end-of-a-scope-able".

Doug,

When we decided[sic] the using statement, we decided to name it “using” rather than something more specific to disposing objects so that it could be used for exactly this scenario.

If you just want some clean, scoped code, you might also use lambdas, á la

myFribble.SafeExecute(() =>
    {
        myFribble.DangerDanger();
        myFribble.LiveOnTheEdge();
    });

where the .SafeExecute(Action fribbleAction) method wraps the try - catch - finally block.

It is a slippery slope. IDisposable has a contract, one that's backed-up by a finalizer. A finalizer is useless in your case. You cannot force the client to use the using statement, only encourage him to do so. You can force it with a method like this:

void UseMeUnlocked(Action callback) {
  Unlock();
  try {
    callback();
  }
  finally {
    Lock();
  }
}

But that tends to get a bit awkward without lamdas. That said, I've used IDisposable like you did.

There is however a detail in your post that makes this dangerously close to an anti-pattern. You mentioned that those methods can throw an exception. This is not something the caller can ignore. He can do three things about that:

Do nothing, the exception isn't recoverable. The normal case. Calling Unlock doesn't matter.
Catch and handle the exception
Restore state in his code and let the exception pass up the call chain.

The latter two requires the caller to explicitly write a try block. Now the using statement gets in the way. It may well induce a client into a coma that makes him believe that your class is taking care of state and no additional work needs to be done. That's almost never accurate.

A real world example is the BeginForm of ASP.net MVC. Basically you can write:

Html.BeginForm(...);
Html.TextBox(...);
Html.EndForm();

using(Html.BeginForm(...)){
    Html.TextBox(...);
}

Html.EndForm calls Dispose, and Dispose just outputs the </form> tag. The good thing about this is that the { } brackets create a visible "scope" which makes it easier to see what's within the form and what not.

I wouldn't overuse it, but essentially IDisposable is just a way to say "You HAVE to call this function when you're done with it". MvcForm uses it to make sure the form is closed, Stream uses it to make sure the stream is closed, you could use it to make sure the object is unlocked.

Personally I would only use it if the following two rules are true, but thet are set arbitarily by me:

Dispose should be a function that always has to run, so there shouldn't be any conditions except Null-Checks
After Dispose(), the object should not be re-usable. If I wanted a reusable object, I'd rather give it open/close methods rather than dispose. So I throw an InvalidOperationException when trying to use a disposed object.

At the end, it's all about expectations. If an object implements IDisposable, I assume it needs to do some cleanup, so I call it. I think it usually beats having a "Shutdown" function.

That being said, I don't like this line:

this.Frobble.Fiddle();

As the FrobbleJanitor "owns" the Frobble now, I wonder if it wouldn't be better to instead call Fiddle on he Frobble in the Janitor?

We have plenty of use of this pattern in our code base, and I've seen it all over the place before - I'm sure it must have been discussed here as well. In general I don't see what's wrong with doing this, it provides a useful pattern and causes no real harm.

Chiming in on this: I agree with most in here that this is fragile, but useful. I'd like to point you to the System.Transaction.TransactionScope class, that does something like you want to do.

Generally I like the syntax, it removes a lot of clutter from the real meat. Please consider giving the helper class a good name though - maybe ...Scope, like the example above. The name should give away that it encapsulates a piece of code. *Scope, *Block or something similar should do it.

Note: My viewpoint is probably biased from my C++ background, so my answer's value should be evaluated against that possible bias...

What Says the C# Language Specification?

Quoting C# Language Specification:

8.13 The using statement

[...]

A resource is a class or struct that implements System.IDisposable, which includes a single parameterless method named Dispose. Code that is using a resource can call Dispose to indicate that the resource is no longer needed. If Dispose is not called, then automatic disposal eventually occurs as a consequence of garbage collection.

The Code that is using a resource is, of course, the code starting by the using keyword and going until the scope attached to the using.

So I guess this is Ok because the Lock is a resource.

Perhaps the keyword using was badly chosen. Perhaps it should have been called scoped.

Then, we can consider almost anything as a resource. A file handle. A network connection... A thread?

A thread???

Using (or abusing) the `using` keyword?

Would it be shiny to (ab)use the using keyword to make sure the thread's work is ended before exiting the scope?

Herb Sutter seems to think it's shiny, as he offers an interesting use of the IDispose pattern to wait for a thread's work to end:

http://www.drdobbs.com/go-parallel/article/showArticle.jhtml?articleID=225700095

Here is the code, copy-pasted from the article:

// C# example
using( Active a = new Active() ) {    // creates private thread
       …
       a.SomeWork();                  // enqueues work
       …
       a.MoreWork();                   // enqueues work
       …
} // waits for work to complete and joins with private thread

While the C# code for the Active object is not provided, it is written the C# uses the IDispose pattern for the code whose C++ version is contained in the destructor. And by looking at the C++ version, we see a destructor which waits for the internal thread to end before exiting, as shown in this other extract of the article:

~Active() {
    // etc.
    thd->join();
 }

So, as far as Herb is concerned, it's shiny.

I believe the answer to your question is no, this would not be an abuse of IDisposable.

The way I understand the IDisposable interface is that you are not supposed to work with an object once it has been disposed (except that you're allowed to call its Dispose method as often as you want).

Since you create a new FrobbleJanitor explicitly each time you get to the using statement, you never use the same FrobbeJanitor object twice. And since it's purpose is to manage another object, Dispose seems appropriate to the task of freeing this ("managed") resource.

(Btw., the standard sample code demonstrating the proper implementation of Dispose almost always suggests that managed resources should be freed, too, not just unmanaged resources such as file system handles.)

The only thing where I'd personally worry is that it's less clear what's happening with using (var janitor = new FrobbleJanitor()) than with a more explicit try..finally block where the Lock & Unlock operations are directly visible. But which approach is taken probably comes down to a matter of personal preferences.

I think you did it right. Overloading Dispose() would be a problem the same class later had cleanup it actually had to do, and the lifetime of that cleanup changed to be different then the time when you expect to hold a lock. But since you created a separate class (FrobbleJanitor) that is responsible only for locking and unlocking the Frobble, things are decoupled enough you won't hit that issue.

I would rename FrobbleJanitor though, probably to something like FrobbleLockSession.

It's not abusive. You are using them what they are created for. But you might have to consider one upon another according to your needs. For example, if you are opting for 'artistry', then you can use 'using', but if your piece of code is executed a lot of times, then for performance reasons, you might use the 'try'..'finally' constructs, because 'using' usually involves creations of an object.