I am trying to implement the IDisposable pattern on a derived class, and it's not working as expecting to work,
Suppose I have two classes and I want to call the Dispose method of the derived class:
Below is the code of my base class
public class BaseClass : IDisposable
{
// To detect redundant calls
private bool _disposedValue;
// Public implementation of Dispose pattern callable by consumers.
public void Dispose()
{
Dispose(true);
GC.SuppressFinalize(this);
}
// Protected implementation of Dispose pattern.
protected virtual void Dispose(bool disposing)
{
if (!_disposedValue)
{
if (disposing)
{
// TODO: dispose managed state (managed objects)
}
// TODO: free unmanaged resources (unmanaged objects) and override finalizer
// TODO: set large fields to null
_disposedValue = true;
}
}
}
And I've something like this as a derived class
public class DerivedClass : BaseClass
{
// To detect redundant calls
private bool _disposedValue;
// Protected implementation of Dispose pattern.
protected override void Dispose(bool disposing)
{
if (!_disposedValue)
{
if (disposing)
{
}
_disposedValue = true;
}
// Call base class implementation.
base.Dispose(disposing);
}
}
And I've some wrapper class
public class WrapperClass : IDisposable
{
public ReadOnlyCollection<BaseClass> Items { get; set;}
// To detect redundant calls
private bool _disposedValue;
// Public implementation of Dispose pattern callable by consumers.
public void Dispose()
{
Dispose(true);
GC.SuppressFinalize(this);
}
// Protected implementation of Dispose pattern.
protected virtual void Dispose(bool disposing)
{
if (!_disposedValue)
{
if (disposing)
{
foreach (var item in Items)
{
item.Dispose();
}
}
_disposedValue = true;
}
}
}
My issue is in the wrapper class, it doesn't call the Dispose method of the DerivedClass, and it calls the Dispose method of the BaseClass instead.
Update
It's my bad, I forget to say that the collections of items were created by NSubstitute like below:
// Having this will not call the Dispose method of the derived class
Items = new ReadOnlyCollection<BaseClass>(new List<BaseClass>
{
Substitute.For<DerivedClass>()
})
// Having this will call the Dispose method of the derived class
Items = new ReadOnlyCollection<BaseClass>(new List<BaseClass>
{
new DerivedClass>()
})
The public implementation of the Dispose() pattern in the documentation is there as an example to handle every case and contingency. It's always "safe" to do the whole thing, but it's also true that most of the time you can skip most of the pattern.
Especially regarding finalizers: you should only need a finalizer if you are creating an original wrapper implementation for a brand new kind of unmanaged resource. Each unmanaged resource type only needs one finalizer, at the root of the IDisposable inheritance tree. And if you don't add a finalizer, you don't need to worry about GC.SuppressFinalize(). Remove that, and some other balls drop as well.
In short, we can reduce the pattern for your DerivedClass all the way down to this:
public class DerivedClass : BaseClass
{
}
This DerivedClass type still provides the Dispose() method and IDisposable implementation inherited from it's parent, and if it doesn't introduce any other unmanaged resources that's all it needs.
My issue is in the wrapper class, it doesn't call the Dispose method of the DerivedClass, and it calls the Dispose method of the BaseClass instead.
When the base.Dispose() calls Dispose(true) the method override in the derived class is executed, which in turn will call base. Dispose(true).
DerivedClass doesn't appear to override Dispose(). Consequently, the base class version is invoked.
Add a public void Dispose() method that calls your protected void Dispose(bool disposing) implementation in DerivedClass.
Related
I work on a huge project in C# .NET 4.0. There is a custom class inherited from System.Net.Sockets.SocketAsyncEventArgs class. Something like the following:
public class SocketTaskArgs : SocketAsyncEventArgs
{
public SocketTaskArgs()
{
Completed += someEventhHandler;
}
public void CleanUp()
{
Completed -= someEventhHandler;
}
/*
There is a lot of code here that is unimportant at the moment.
*/
}
So, I wanted to move the content of CleanUp() method to Dispose(bool) method.
As first, I checked the source code of the base class - SocketAsyncEventArgs (using Go To Definition so that I saw metadata as source). I found out, this class implements IDisposable interface. Nice, I just need to override the Dispose(bool) method, don't I? (See IDisposable Interface on MSDN, the "IDisposable and the inheritance hierarchy" section, for more details). Nothing new for me... Unfortunately, the SocketAsyncEventArgs class is implemented as following:
public class SocketAsyncEventArgs : EventArgs, IDisposable
{
public void Dispose();
//some other stuff here
}
That means, there is no way how to override Dispose(bool) method, as it's implemented as private instead of protected... What is the reason for this?
Next, I read about SocketAsyncEventArgs.Dispose() method on MSDN. The funny thing is that, it contains the following section:
Notes to Inheritors
Dispose can be called multiple times by other
objects. When overriding Dispose(Boolean), be careful not to reference
objects that have been previously disposed of in an earlier call to
Dispose. For more information about how to implement Dispose(Boolean),
see Implementing a Dispose Method.
Wait... what?
When overriding Dispose(Boolean), ...
How am I supposed to override Dispose(Boolean)?
What is the recommended way to implement IDisposable interface in this case?
There doesn't seem to be anything stopping you from implementing IDisposable on your child class, take this example:
public class DisposableParent : IDisposable
{
public void Dispose()
{
Console.WriteLine("The parent was disposed.");
}
}
public class DisposableChild : DisposableParent, IDisposable
{
public new void Dispose()
{
base.Dispose();
Console.WriteLine("The child was disposed.");
}
}
public class Program
{
public static void Main()
{
using (DisposableChild c = new DisposableChild()) { }
Console.ReadKey(true);
}
}
Gives the following output:
The parent was disposed.
The child was disposed.
The compiler warns about hiding the dispose of the parent class in the child, so using the new operator gets rid of that warning, just make sure to call the base class Dispose from the child class (and implement it the right way).
The dispose for the child would become something like:
public class DisposableChild : DisposableParent, IDisposable
{
private bool _disposed = false;
public new void Dispose()
{
Dispose(true);
}
protected virtual void Dispose(bool disposing)
{
if (disposing)
{
if (!_disposed)
{
base.Dispose();
Console.WriteLine("The child was disposed.");
_disposed = true;
}
}
}
}
And yes, this still works if you do something like:
using (DisposableParent p = new DisposableChild())
{
}
But something like this can break it:
public class Program
{
public static void Main()
{
DisposableChild c = new DisposableChild();
DisposeOfIt(c);
Console.ReadKey(true);
}
public static void DisposeOfIt(DisposableParent p)
{
p.Dispose();
}
}
Only prints out that the parent was disposed. So if you used this method you would have to be careful about controlling the lifetime of your objects.
Say a factory for SomeDisposable actually is creating/returning a sort of watch dog Wrapper
public class Wrapper : SomeDisposable
{
public new /*:(*/ Dispose() { ... };
}
and the caller uses like
using (SomeDisposable sd = SomeDisposableFactory.Create(...))
{
} // Wrapper.Dispose() never called.
The Wrapper.Dispose() is never called. If Dispose() were virtual then Wrapper.Dispose() would be called.
The IDisposable interface does not guarantee that the other best practice method virtual Dispose(bool) actually exists or enforce that either be virtual so it cannot be generally relied on to exist (it is only a recommended pattern). Interfaces currently do not allow constraints on virtual.
What are some pros and cons for not making the recommended Dispose() pattern virtual which would have solved this particular dilemma. Should C# allow a way of forcing virtual methods via an interface (since abstract classes aren't popular as contract definitions).
No. The pattern actually says that Dispose() (non-virtual) should call a protected virtual void Dispose(bool) method. This guarantees that the base class Dispose call can pass up the hierarchy properly.
This is spelled out in the documentation for IDisposable:
It should provide one public, non-virtual Dispose() method and a protected virtual Dispose(Boolean disposing) method.
The Dispose() method must call Dispose(true) and should suppress finalization for performance.
The base type should not include any finalizers.
This is already solved.
Disposable types which are not sealed should use the common dispose pattern:
public class DisposableResourceHolder : IDisposable
{
private SafeHandle resource; // handle to a resource
public DisposableResourceHolder()
{
this.resource = ... // allocates the resource
}
public void Dispose()
{
Dispose(true);
GC.SuppressFinalize(this);
}
protected virtual void Dispose(bool disposing)
{
if (disposing)
{
// dispose managed resources.
if (resource != null) resource.Dispose();
}
// free unmanaged resources.
}
}
I've looked in SO about a question like this one, and even that I've found quite a few, any of those threw any light into this matter for me.
Let's assume I have this code:
public class SuperObject : IDisposable
{
public void Dispose()
{
Dispose(true);
GC.SuppressFinalize(this);
}
protected virtual void Dispose(bool disposing) { }
}
Do I need the protected virtual void Dispose(bool) on SuperObject? Since there is really nothing to dispose there.
public interface ICustom : IDisposable { }
public class Custom : ICustom
{
public SuperObject Super { get; protected set; }
public Custom()
{
Super = new SuperObject();
}
public void Dispose()
{
Dispose(true);
GC.SuppressFinalize(this);
}
public virtual void Dispose(bool disposing)
{
if (!disposing) return;
if (Super != null)
Super.Dispose();
}
}
public class Foo
{
public Foo()
{
using (var c = new Custom())
{
//do magic with c
}
}
}
Now what happens if I want/need/try to use Custom on a class like System.Web.Mvc.Controller which already implements and has implemented IDisposable?
public class Moo : Controller
{
Custom c;
public Moo()
{
c = new Custom();
}
// Use c throughout this class
}
How to properly dispose c in Moo?
The normal approach is to apply the standard IDisposable implementation - HOWEVER this is really only necessary if your class or some class that derives from it will use UNmanaged resources - this case is infact VERY rare (and when this case does apply it is better to wrap the unmanaged resource in its own class that has a full standard IDisposable implementation).
So assuming you are not dealing with UNmanaged resources (raw file handles, globally alloced memeory etc) and are only dealing with members that are disposable (i.e that have managed resources and implement IDisposable) then you can safely get a way with a mimimal implimentation of IDispose - that is:
Just have a single void Dispose() method. In that method just call dispose on dispoable members and then Dispose on the base class if its disposable. If you have a class hierachy its ok to make this Dispose virtual. There is no need to have a Dispose(bool) method. Nor is there any need to check if the object is disposed - because all your doing is calling dipsose on other objects and those implementation will do that check.
If you don't like the mimimal appraoch then apply the standard full implimentation (but it is not strictly necessary). I.e either do a standard implimentation because your a stickler for following the recommended approach OR do a simple minimal (but correct) implementation - but don't do something in between (i.e not standard, not simple or not correct)!
See this question for more details: Minimal IDispose implimenation for managed resources only
So in your case the following is the mimimal implimentation:
public class SuperObject : IDisposable {
public void Dispose() {
// Dispose code...just call dispose on dispoable members.
// If there are none then no need to implement IDisposable!
}
}
public interface ICustom : IDisposable { }
public class Custom : ICustom {
public SuperObject Super { get; protected set; }
public Custom() {
Super = new SuperObject();
}
public void Dispose() {
if (Super != null)
Super.Dispose();
}
}
public class Moo : Controller {
Custom c;
public Moo() {
c = new Custom();
}
public Dispose() {
if (c!=null)
c.Dispose()
base.Dispose();
}
}
Note that if Super object does not have any disposable resources then there is no point in implementing IDisposable and having a Dispose method. If Customs only disposable object is SuperObject then the same applies there, and again the same logic rocks through to Moo. Finally then if all the above applies and there are no other disposable objects around all you need really need is:
public class Moo : Controller {
Custom c;
public Moo() {
c = new Custom();
}
public Dispose() {
base.Dispose();
}
}
How to properly dispose c in Moo?
public class Moo : Controller
{
Custom c;
public Moo()
{
c = new Custom();
}
// Use c throughout this class
protected override Dispose(bool disposing)
{
base.Dispose(disposing);
if (disposing)
c.Dispose()
}
}
And that also answers your first question, Controller needs to make its Dispose(bool) method protected virtual or the above would not be possible.
But a few notes:
you don't have any isDisposed logic. It is a good idea to do the disposing only once, and you may want to trap usage-after-dispose.
omitting destructors (finalizers) is in itself a good idea but you now have the extra constraint that no derived class should own an unmanaged resource.
I have an implementation of an interface, and that interface extends IDisposable. In my particular implementation of the interface, I don't need to dispose anything, so I just have an empty Dispose() method.
public interface IMyStuff : IDisposable
{
}
public MyStuffImpl : IMyStuff
{
public void Dispose()
{
}
}
Now in FxCop, this results in a CA1063:
Error, Certainty 95, for ImplementIDisposableCorrectly
{
Resolution : "Provide an overridable implementation of Dispose(
bool) on 'MyStuffImpl' or mark the type as sealed.
A call to Dispose(false) should only clean up native
resources. A call to Dispose(true) should clean up
both managed and native resources."
}
CriticalWarning, Certainty 75, for CallGCSuppressFinalizeCorrectly
{
Resolution : "Change 'MyStuffImpl.Dispose()' to call 'GC.SuppressFinalize(
object)'. This will prevent derived types that introduce
a finalizer from needing to re-implement 'IDisposable'
to call it."
}
Error, Certainty 95, for ImplementIDisposableCorrectly
{
Resolution : "Modify 'MyStuffImpl.Dispose()' so that it
calls Dispose(true), then calls GC.SuppressFinalize
on the current object instance ('this' or 'Me' in Visual
Basic), and then returns."
}
So, it looks like I can resolve this in one of 2 ways:
Make the class sealed:
public sealed MyStuffImpl : IMyStuff
{
public void Dispose()
{
}
}
Implement part of the typical pattern:
public MyStuffImpl : IMyStuff
{
public void Dispose()
{
Dispose(true);
GC.SuppressFinalize(this);
}
private void Dispose(bool disposing)
{
}
}
In my case, I don't plan on this implementation ever being extended, so I will probably resolve it by making it sealed, but I admit I don't really understand why it matters if it is sealed or not.
Also, just because my class is sealed, FxCop no longer tells me that Dispose() should call GC.SupressFinalize(this); but is that really true? Is it "better" in .NET to just always call SupressFinalize in Dispose regardless?
SuppressFinalize() is meaningless unless your instance has a finalizer.
If your class doesn't have a finalizer, but is not sealed, you should still SuppressFinalize, in case an inherited class adds a finalizer.
Both of your options are correct, except that Dispose(bool) needs to be protected virtual.
In your "implement part of the typical pattern" option, you should make your Dispose(bool) method protected virtual:
protected virtual void Dispose(bool disposing)
{
}
That will provide subclasses an opportunity to handle disposal of any resources they manage. That's the meaning of "overridable" in "Provide an overridable implementation of Dispose(bool)"
Of course, public virtual would also satisfy FxCop.
Forgive me in advance if this question is a little too open-ended, but I've seen similar language discussion posts here so I figured I'd take the plunge.
Anyway, I have read several MSDN help pages and various other blogs on the subject of properly implementing IDisposable classes. I feel like I understand things pretty well, but I have to wonder if there's a flaw in the suggested class structure:
public class DisposableBase : IDisposable
{
private bool mDisposed;
~DisposableBase()
{
Dispose(false);
}
public void Dispose()
{
Dispose(true);
GC.SuppressFinalize(this);
}
protected virtual void Dispose(bool disposing)
{
if (!mDisposed)
{
if (disposing)
{
// Dispose managed resources
mManagedObject.Dispose();
}
// Dispose unmanaged resources
CloseHandle(mUnmanagedHandle);
mUnmanagedHandle = IntPtr.Zero;
mDisposed = true;
}
}
}
Anytime the above is supposed to serve as a base class, you rely on the implementer of the subclass to properly override the Dispose(bool) method where necessary. In short, derived classes must ensure they invoke the base Dispose(bool) method from within their overridden version. If not, the base class' unmanaged resources may never get freed, defeating the primary purpose of the IDisposable interface.
We all know the benefits of virtual methods, but it seems like in this case their design falls short. In fact, I think this particular shortcoming of virtual methods manifests itself frequently when trying to design visual components and similar base/derived class structures.
Consider the following change, using a protected event rather than a protected virtual method:
public class DisposeEventArgs : EventArgs
{
public bool Disposing { get; protected set; }
public DisposeEventArgs(bool disposing)
{
Disposing = disposing;
}
}
public class DisposableBase : IDisposable
{
private bool mDisposed;
protected event EventHandler<DisposeEventArgs> Disposing;
~DisposableBase()
{
Dispose(false);
}
public void Dispose()
{
Dispose(true);
GC.SuppressFinalize(this);
}
// This method is now private rather than protected virtual
private void Dispose(bool disposing)
{
if (!mDisposed)
{
// Allow subclasses to react to disposing event
AtDisposing(new DisposeEventArgs(disposing));
if (disposing)
{
// Dispose managed resources
mManagedObject.Dispose();
}
// Dispose unmanaged resources
CloseHandle(mUnmanagedHandle);
mUnmanagedHandle = IntPtr.Zero;
mDisposed = true;
}
}
private void AtDisposing(DisposeEventArgs args)
{
try
{
EventHandler<DisposeEventArgs> handler = Disposing;
if (handler != null) handler(this, args);
}
catch
{
}
}
}
With this design, the base class' Dispose(bool) method will always be called, regardless of whether subclasses subscribe to the Disposing event or not. The biggest flaw that I can see with this revised setup is that there is no predetermined order for when event listeners are called. This could be problematic if there are multiple levels of inheritance, e.g. SubclassA's listener might be triggered before its child SubclassB's listener. Is this flaw serious enough to invalidate my revised design?
This design dilemma makes me wish there were some sort of modifier for methods that was similar to virtual but which would ensure that the base class' method was always called, even if a subclass overrode that function. If there's a better way to achieve this, I would greatly appreciate your suggestions.
You're using an event here when really you want to use an inheritance mechanism like virtual. For scenarios like this where I want to ensure my implementation is always called but want to allow for base class customization I use the following pattern
private void Dispose(bool disposing)
if (mDisposed) {
return;
}
if (disposing) {
mManagedObject.Dispose();
}
// Dispose unmanaged resources
CloseHandle(mUnmanagedHandle);
mUnmanagedHandle = IntPtr.Zero;
mDisposed = true;
DisposeCore(disposing);
}
protected virtual void DisposeCore(bool disposing) {
// Do nothing by default
}
With this pattern I've ensured my base class Dispose implementation will always be called. Derived classes can't stop me by simply forgetting to call a base method. They can still opt into the dispose pattern by overriding DisposeCore but they can't break the base class contract.
The derived class can simply re-implement IDisposable and thus prevent your dispose method from being called, so you can't ensure that either.
Personally I wouldn't use either pattern. I prefer building on SafeHandle and similar mechanisms, instead of implementing finalizers myself.
Consider making it apparent that Dispose is not being called so someone will catch it. Of course Debug.WriteLine will only be called when the code is compiled with DEBUG compiler directive defined.
public class DisposableBase : IDisposable
{
private bool mDisposed;
~DisposableBase()
{
if (!mDisposed)
System.Diagnostics.Debug.WriteLine ("Object not disposed: " + this + "(" + GetHashCode() + ")";
Dispose(false);
}
public void Dispose()
{
Dispose(true);
GC.SuppressFinalize(this);
}
You can break it down:
A destructor (finalizer) is only needed for unmanaged resources.
Using a Safehandle can turn an unmanged resource into a managed resource.
Ergo: You won't need a destructor. That halves the Dispose pattern.
The reference design uses a virtual void Dispose(bool) to cater for the Base/Derived class problem. This puts the burden on the derived class to call base.Dispose(disposing), the core of your question. I use 2 approaches:
1) Prevent it. With a sealed base-class you won't have to worry.
sealed class Foo:IDisposable
{
void Dispose() { _member.Dispose(); }
}
2) Check it. Like #j-agent's answer but conditional. When performance could be an issue then you don't want the finalizers in Production code:
class Foo:IDisposable
{
void Dispose() { Dispose(true); }
[Conditional("TEST")] // or "DEBUG"
~Foo { throw new InvalidOperation("somebody forgot to Dispose") }
}
The destructor is going to be called no matter if any subclass overrides Dispose() (can be via override or new) but your destructor is going to be called ( ~DisposableBase() ) so i bet putting your logic for cleanup there can be a good starting point.
Here is an intersting article about destructors: http://www.c-sharpcorner.com/UploadFile/chandrahundigam/UnderstandingDestructors11192005021208AM/UnderstandingDestructors.aspx