How can I determine if I should extend one of my interfaces with IDisposable or implement IDisposable on a class that implements my interface?
I have an interface that does not need to dispose of any external resources, except for one particular implementation. My options seem to be:
1) Implement IDisposable on the interface requiring all of the implementations to implement Dispose, even if only an empty method.
-or-
2) Implement IDisposable on only the classes that have resources needing to be disposed. This will cause problems with "using" because my object is created from a factory and therefore all upstream code works against the interface. Since the interface is not bound to IDisposable, "using" does not see the Dispose method. However, I could cast the factory result to the implementation; however, that then makes the consumer aware of the implementation, defeating the purpose of interfaces.
Any ideas as to best practices?
If you expect callers to only be able to interact with the interface, and never the implementation, then you want to have the interface extend
IDisposable
. If not, they'll need to check if thevalue is IDisposable
anyway to see if it needs to be disposed.If the object responsible for disposing of the object knows of the concrete implementation, and it is only ever objects that are given references to it (but aren't responsible for disposing of it) that use the interface, then consider the second option.
A good example of the first option is
IEnumerator
. ManyIEnumerator
objects don't need to do anything when they're disposed, but some do, and so the interface extendsIDisposable
because the object responsible for the creation/lifecycle of that object will (or should) never have knowledge of the underlying implementation.An example of the second would be something like
IComparer
many objects that need to be compared are disposable, but the sections of code using an object through the interface aren't responsible for it's creation/lifecycle, so it needs no knowledge of whether or not that type is disposable.The $50,000 question is whether responsibility of disposal will ever be passed along with the interface or, to put it another way, whether the last entity to use an implementing object might be something other than the entity which creates it.
The big reason that
IEnumerator<T>
implementsIDisposable
is that implementing objects are created by objects that implementIEnumerable<T>.GetEnumerator()
but are then generally used by other objects. The object that implementsIEnumerable<T>
will know whether the thing it returns really needs disposing, but will have no way of knowing when the recipient is done with it. The code that callsIEnumerable<T>.GetEnumerator()
will know when it's done with the returned object, but will have no way of knowing whether it needs any cleanup. The sensible thing to do is specify that the code which callsIEnumerable<T>.GetEnumerator()
is required to ensure that the returned object will haveDispose
called on it before it is abandoned; theDispose
method will in many cases not do anything, but unconditionally calling a do-nothing method which is guaranteed to exist is cheaper than checking for the existence of a method that does not exist.If the nature of your interface type is such that the the questions of whether an implementing object needs cleanup and when such cleanup should occur will both be answerable by the same entity, then there's no need for the interface to inherit
IDisposable
. If instances will be created by one entity but last used by another, then inheritingIDisposable
would be wise.If you implement
IDisposable
on the concrete class and the interface users know that it might be Disposable; you can doIf
foo
does not implementIDisposable
, theusing
will amount tousing(null)
which will work just fine.The output of the below sample program will be
Sample program