FlowDocument Memory Issue in C#

Question

I am currently attempting to deal with an issue with releasing a FlowDocument resources. I am loading an rtf file and putting it into a FlowDocument with TextRange.Load. I noticed that after it does this it holds onto those resources and GC doesn't collect it. I have ran a memory profiler and have seen that this is true. I have also narrowed it down to were I load actually put the rtf into the FlowDocument. If I dont do that, then everything is ok. So I know this is the issue.

I am hoping for some guidance to what how I can solve this problem. Here is the code that loads the rtf and everything. I have commented all of the other code out and even put it in its own scope as well as tried GC.Collect(). Any help is greatly appreciated.

EDIT: Here is my code in full at the moment. I have taken out everything else except for the bare essentials to get it running. The problem is still there. As you can see the FlowDocument and TextRange are not referenced anywhere else.

    public LoadRTFWindow(string file)
    {
        InitializeComponent();

        using (FileStream reader = new FileStream(file, FileMode.Open))
        {
            FlowDocument doc = new FlowDocument();
            TextRange range = new TextRange(doc.ContentStart, doc.ContentEnd);
            range.Load(reader, System.Windows.DataFormats.Rtf);
        }
        GC.Collect();
        GC.WaitForPendingFinalizers();
        GC.Collect();
    }

I found this post, which I was hoping would help me solve my problem but I had no luck with it. Any type of help is greatly appreciated. Thank you.

EDIT: I figure I should mention the major way I am checking this. I have Windows Task Manager open and am watching the memory usage my application's process is using. When I run the above code the application goes from 40,000K to 70,000K while doing the TextRange.Load() (this is a large 400 page RTF) and once that finishes it drops down to 61,000K and stays there. My expectation is that it would drop back down to 40,000K or at least very close to it.

As I mentioned earlier I used a memory profiler and saw that there were LOTS of Paragraph, Run..ect. objects still Alive afterwards.

Answer 1

We had a similar problem in which we were creating flow document in different thread, i noticed in memory profiler that objects were still there.

So far as i know, as described in this link

"When a FlowDocument is created, relatively expensive formatting context objects are also created for it in its StructuralCache. When you create multiple FlowDocs in a tight loop, a StructuralCache is created for each FlowDoc. Let's you called Gc.Collect at the end of the loop, hoping to recover some memory. StructuralCache has a finalizer releases this formatting context, but not immediately. The finalizer effectively schedules an operation to release contexts at DispatcherPriority.Background."

So until the dispatcher operations are completed, Flow document will be in memory. So the idea is to complete the dispatcher operations.

If you are in a thread in which Dispatcher is currently running then try code below, it will force all the operations in queue to be completed, as SystemIdle is the lowest priority:

Dispatcher.CurrentDispatcher.Invoke(DispatcherPriority.SystemIdle, 
    new DispatcherOperationCallback(delegate { return null; }), null); 

If you are in a thread in which Dispatcher is not running, as in my case only single flow document was created in thread, so i tried something like:

var dispatcher = Dispatcher.CurrentDispatcher;
dispatcher.BeginInvokeShutdown(DispatcherPriority.SystemIdle);
Dispatcher.Run();

this will queue a shut down at very end and then will run the dispatcher to clean the FlowDocument and then in end it will shut down the dispatcher.

Answer 2

If I've confirmed that there's a memory leak, here's what I would do to debug the problem.

1. Install Debugging Tools for Windows from http://www.microsoft.com/whdc/devtools/debugging/installx86.mspx#a
2. Fire up Windbg from the installation directory.
3. Launch your application and do the operations that leak memory.
4. Attach Windbg to your application (F6).
5. Type .loadby sos mscorwks
6. Type !dumpheap -type FlowDocument
7. Check the result of the above command. If you see multiple FlowDocuments, for each value of the first column (which contains the address), do

Type !gcroot <value of first column>

That should show you who's holding on to the reference.

Answer 3

Consider releasing that file handle. Also consider using the "using" statement instead of calling IDisposable.Dispose (no pun intended).

Answer 4

FlowDocument uses System.Windows.Threading.Dispatcher to free all resources. It doesn't use finalizers because finalizers will block current thread until all resources will be free. So the user may see some UI freezings and so on. Dispatchers are running in background thread and have less impact on the UI.
So calling GC.WaitForPendingFinalizers(); has no influence on this. You just need to add some code to wait and allow Dispatchers to finish their work. Just try to add something like Thread.CurrentThread.Sleep(2000 /* 2 secs */);

EDIT: It seems to me that you found this problem during debugging of some application. I wrote the following very simple test case (console program):

    static void Main(string[] args)
    {
        Console.WriteLine("press enter to start");
        Console.ReadLine();

        var path = "c:\\1.rtf";

        for (var i = 0; i < 20; i++)
        {
            using (var stream = new FileStream(path, FileMode.Open))
            {
                var document = new FlowDocument();
                var range = new TextRange(document.ContentStart, document.ContentEnd);

                range.Load(stream, DataFormats.Rtf);
            }
        }

        Console.WriteLine("press enter to run GC.");
        Console.ReadLine();

        GC.Collect();
        GC.WaitForPendingFinalizers();

        Console.WriteLine("GC has finished .");
        Console.ReadLine();
    }

I've tried to reproduce the problem. I run it several times and it was working perfectly - there were no leaks (about 3,2Kb all of the time and 36 handles). I couldn't reproduce it until I run this program in debug mode in the VS (just f5 instead of ctrl+f5). I received 20,5Kb at the beginning, 31,7Kb after loading and before GC and 31Kb after GC. It looks similar to your results.
So, could you please try to reproduce this problem running in release mode not under the VS?

Answer 5

I tried to reproduce your problem, but it doesn't happen on my machine.

Task Manager shows working set size, which isn't an accurate representation of a program's memory usage. Try using perfmon instead.

1. Start -> Run -> perfmon.msc
2. Add .NET CLR Memory/#Bytes in All Heaps for your application

Now repeat the experiment and check if that counter keeps going up. If it doesn't, it means you aren't leaking any managed memory.

Answer 6

Make sure that the Parent of the FlowDocument isn't hanging around, see here. "Instantiating a FlowDocument automatically spawns a parent FlowDocumentPageViewer that hosts the content." If that control is hanging around it could be the source of your problem.