Should I Dispose() DataSet and DataTable?

Question

Here are a couple of discussions explaining why Dispose is not necessary for a DataSet.

To Dispose or Not to Dispose ?:

The Dispose method in DataSet exists ONLY because of side effect of inheritance-- in other words, it doesn't actually do anything useful in the finalization.

Should Dispose be called on DataTable and DataSet objects? includes some explanation from an MVP:

The system.data namespace (ADONET) does not contain unmanaged resources. Therefore there is no need to dispose any of those as long as you have not added yourself something special to it.

Understanding the Dispose method and datasets? has a with comment from authority Scott Allen:

In pratice we rarely Dispose a DataSet because it offers little benefit"

So, the consensus there is that there is currently no good reason to call Dispose on a DataSet.

Update (December 1, 2009):

I'd like to amend this answer and concede that the original answer was flawed.

The original analysis does apply to objects that require finalization – and the point that practices shouldn’t be accepted on the surface without an accurate, in-depth understanding still stands.

However, it turns out that DataSets, DataViews, DataTables suppress finalization in their constructors – this is why calling Dispose() on them explicitly does nothing.

Presumably, this happens because they don’t have unmanaged resources; so despite the fact that MarshalByValueComponent makes allowances for unmanaged resources, these particular implementations don’t have the need and can therefore forgo finalization.

(That .NET authors would take care to suppress finalization on the very types that normally occupy the most memory speaks to the importance of this practice in general for finalizable types.)

Notwithstanding, that these details are still under-documented since the inception of the .NET Framework (almost 8 years ago) is pretty surprising (that you’re essentially left to your own devices to sift though conflicting, ambiguous material to put the pieces together is frustrating at times but does provide a more complete understanding of the framework we rely on everyday).

After lots of reading, here’s my understanding:

If an object requires finalization, it could occupy memory longer than it needs to – here’s why: a) Any type that defines a destructor (or inherits from a type that defines a destructor) is considered finalizable; b) On allocation (before the constructor runs), a pointer is placed on the Finalization queue; c) A finalizable object normally requires 2 collections to be reclaimed (instead of the standard 1); d) Suppressing finalization doesn’t remove an object from the finalization queue (as reported by !FinalizeQueue in SOS) This command is misleading; Knowing what objects are on the finalization queue (in and of itself) isn’t helpful; Knowing what objects are on the finalization queue and still require finalization would be helpful (is there a command for this?)

Suppressing finalization turns a bit off in the object's header indicating to the runtime that it doesn’t need to have its Finalizer invoked (doesn’t need to move the FReachable queue); It remains on the Finalization queue (and continues to be reported by !FinalizeQueue in SOS)

The DataTable, DataSet, DataView classes are all rooted at MarshalByValueComponent, a finalizable object that can (potentially) handle unmanaged resources

• Because DataTable, DataSet, DataView don’t introduce unmanaged resources, they suppress finalization in their constructors
• While this is an unusual pattern, it frees the caller from having to worry about calling Dispose after use
• This, and the fact that DataTables can potentially be shared across different DataSets, is likely why DataSets don’t care to dispose child DataTables
• This also means that these objects will appear under the !FinalizeQueue in SOS
• However, these objects should still be reclaimable after a single collection, like their non-finalizable counterparts

4 (new references):

• http://www.devnewsgroups.net/dotnetframework/t19821-finalize-queue-windbg-sos.aspx
• http://blogs.msdn.com/tom/archive/2008/04/28/asp-net-tips-looking-at-the-finalization-queue.aspx
• http://issuu.com/arifaat/docs/asp_net_3.5unleashed
• http://msdn.microsoft.com/en-us/magazine/bb985013.aspx
• http://blogs.msdn.com/tess/archive/2006/03/27/561715.aspx

Original Answer:

There are a lot of misleading and generally very poor answers on this - anyone who's landed here should ignore the noise and read the references below carefully.

Without a doubt, Dispose should be called on any Finalizable objects.

DataTables are Finalizable.

Calling Dispose significantly speeds up the reclaiming of memory.

MarshalByValueComponent calls GC.SuppressFinalize(this) in its Dispose() - skipping this means having to wait for dozens if not hundreds of Gen0 collections before memory is reclaimed:

With this basic understanding of finalization we can already deduce some very important things:

First, objects that need finalization live longer than objects that do not. In fact, they can live a lot longer. For instance, suppose an object that is in gen2 needs to be finalized. Finalization will be scheduled but the object is still in gen2, so it will not be re-collected until the next gen2 collection happens. That could be a very long time indeed, and, in fact, if things are going well it will be a long time, because gen2 collections are costly and thus we want them to happen very infrequently. Older objects needing finalization might have to wait for dozens if not hundreds of gen0 collections before their space is reclaimed.

Second, objects that need finalization cause collateral damage. Since the internal object pointers must remain valid, not only will the objects directly needing finalization linger in memory but everything the object refers to, directly and indirectly, will also remain in memory. If a huge tree of objects was anchored by a single object that required finalization, then the entire tree would linger, potentially for a long time as we just discussed. It is therefore important to use finalizers sparingly and place them on objects that have as few internal object pointers as possible. In the tree example I just gave, you can easily avoid the problem by moving the resources in need of finalization to a separate object and keeping a reference to that object in the root of the tree. With that modest change only the one object (hopefully a nice small object) would linger and the finalization cost is minimized.

Finally, objects needing finalization create work for the finalizer thread. If your finalization process is a complex one, the one and only finalizer thread will be spending a lot of time performing those steps, which can cause a backlog of work and therefore cause more objects to linger waiting for finalization. Therefore, it is vitally important that finalizers do as little work as possible. Remember also that although all object pointers remain valid during finalization, it might be the case that those pointers lead to objects that have already been finalized and might therefore be less than useful. It is generally safest to avoid following object pointers in finalization code even though the pointers are valid. A safe, short finalization code path is the best.

Take it from someone who's seen 100s of MBs of non-referenced DataTables in Gen2: this is hugely important and completely missed by the answers on this thread.

References:

1 - http://msdn.microsoft.com/en-us/library/ms973837.aspx

2 - http://vineetgupta.spaces.live.com/blog/cns!8DE4BDC896BEE1AD!1104.entry http://www.dotnetfunda.com/articles/article524-net-best-practice-no-2-improve-garbage-collector-performance-using-finalizedispose-pattern.aspx

3 - http://codeidol.com/csharp/net-framework/Inside-the-CLR/Automatic-Memory-Management/

You should assume it does something useful and call Dispose even if it does nothing in current .NET Framework incarnations. There's no guarantee it will stay that way in future versions leading to inefficient resource usage.

Even if an object has no unmanaged resources, disposing might help GC by breaking object graphs. In general, if an object implements IDisposable, Dispose() should be called.

Whether Dispose() actually does something or not depends on the given class. In case of DataSet, Dispose() implementation is inherited from MarshalByValueComponent. It removes itself from container and calls Disposed event. The source code is below (disassembled with .NET Reflector):

protected virtual void Dispose(bool disposing)
{
    if (disposing)
    {
        lock (this)
        {
            if ((this.site != null) && (this.site.Container != null))
            {
                this.site.Container.Remove(this);
            }
            if (this.events != null)
            {
                EventHandler handler = (EventHandler) this.events[EventDisposed];
                if (handler != null)
                {
                    handler(this, EventArgs.Empty);
                }
            }
        }
    }
}

I call dispose anytime an object implements IDisposeable. It's there for a reason.

DataSets can be huge memory hogs. The sooner they can be marked for clean up, the better.

update

It's been 5 years since I answered this question. I still agree with my answer. If there is a dispose method, it should be called when you are done with the object. The IDispose interface was implemented for a reason.