问题
I work on an open source product called EVEMon written in C# targeting the .NET 2.0 platform, I have one user who is suffering from a strange .NET crash that we have been unable to resolve.
Event Type: Error Event Source: .NET Runtime 2.0 Error Reporting Event Category: None Event ID: 5000 Date: 4/29/2009 Time: 10:58:10 PM User: N/A Computer: removed this Description: EventType clr20r3, P1 evemon.exe, P2 1.2.7.1301, P3 49ea37c8, P4 system.windows.forms, P5 2.0.0.0, P6 4889dee7, P7 6cd3, P8 18, P9 system.argumentexception, P10 NIL. Data: //hex representation of the above Description
The application itself crashes with out displaying an error (despite having a error handling UI), the above messages was copied out of the Windows Event log. The end user has re-installed .NET and updated to the latest versions. The .PDB files are distributed with every release version of the program to aid in debugging and testing, the user with the problem in question has the full complement of PDB files for the correct version of EVEMon.
Is there a specific, tried and tested technique to analyse and diagnose this type of crash? and if so what tools and technologies are available to aid in debugging?
Special Thanks
I would like to give special thanks to Steffen Opel and highlight that his answer whilst not directly answering the question I was asking, addressed the bigger issue with my code base that the global error handling was missing an important component.
回答1:
This is how I would tackle the problem for a end user with a crash.
Download and install Debugging Tools for Windows at http://www.microsoft.com/whdc/devtools/debugging/default.mspx
Once the tools are installed (they end up going to C:\Program Files\ by default) start a command line window.
Change to the directory which contains adplus (e.g "C:\Program Files\Debugging Tools for Windows (x86)").
Run the follwing command. This will start the application and attach adplus.
adplus -crash -o C:\debug\ -FullOnFirst -sc C:\path\to\your\app.exe
After the crash dump is created
Once the application crashes start WinDbg and load the .dmp file that is created in C:\debug. (File --> Open Crash Dump)
Execute these commands to see the stack trace and hopefully find the problem.
To load SOS for debugging
- Pre .NET 4.0
.loadby sos mscorwks
- .NET 4.0
.loadby sos clr
To see the stack trace
!clrstack
To see a more useful stack trace
!clrstack –p
To poke inside an object..perhaps see what caused the exception
!do <address>
e.g This is the result from a application that faulted randomly with an IO exception. WinDbg pointed out the path that was being referenced which was incorrect.
0:009> !do 017f2b7c
Name: System.String
MethodTable: 790fd8c4
EEClass: 790fd824
Size: 124(0x7c) bytes
(C:\WINDOWS\assembly\GAC_32\mscorlib\2.0.0.0__b77a5c561934e089\mscorlib.dll)
String: \\server\path\not_here.txt
Fields:
MT Field Offset Type VT Attr Value Name
79102290 4000096 4 System.Int32 1 instance 54 m_arrayLength
79102290 4000097 8 System.Int32 1 instance 53 m_stringLength
790ff328 4000098 c System.Char 1 instance 5c m_firstChar
790fd8c4 4000099 10 System.String 0 shared static Empty
>> Domain:Value 00161df8:790d884c <<
7912dd40 400009a 14 System.Char[] 0 shared static WhitespaceChars
>> Domain:Value 00161df8:014113e8 <<
回答2:
Peeking into your source code (trunk) indicates that your unhandled exception handling seems to be incomplete in regard to Windows Forms applications:
You need to handle both non-UI thread exceptions and UI thread exceptions:
For the former you need to implement a CLR unhandled exception handler via
AppDomain.CurrentDomain.UnhandledException, which is in place already.For the latter you need to implement a Windows Forms unhandled exception handler via
Application.ThreadException, which seems to be missing; this could indeed yield exactly those problems you are witnessing. For an implementation example see MSDN documentation of Application.ThreadException Event.
Please note that right now you explicitly suppress catching unhandled Windows Forms exceptions via Application.SetUnhandledExceptionMode(UnhandledExceptionMode.ThrowException), you'll need to change this to UnhandledExceptionMode.CatchException to enable routing to your handler for Application.ThreadException, as correctly suggested by Jehof already.
回答3:
What OS (Windows XP, Windows Vista, etc.) does the user use?
If Windows Vista try to disable "Problem Reports and Solutions feature" (Control Panel-->Problem Reports and Solutions-->Change Settings-->Advanced Settings-->Turn off for my programs, problem reporting)
Or try to set
Application.SetUnhandledExceptionMode( UnhandledExceptionMode.CatchException );
This will always route exceptions to the ThreadException handler.
回答4:
In a nutshell: there is an unhandled exception in the application.
If you have access to the machine (through remote access, etc.), try installing Visual Studio Express and starting the application. You should see a dialog offering the chance to debug the application with a new instance of Visual Studio.
It may also be that there is something preventing Windows Forms from properly initializing. I've seen forum posts that suggest that font issues can cause this -- ensure that the users have the fonts installed that your application needs plus the usual defaults such as MS SansSerif, Arial, Tahoma, Times and suchlike.
And failing that... try sacrificing a chicken over the PC. Works a charm every time!
回答5:
We've had issues with Exceptions in Thread-Code. If you spawn a new Thread and forget to handle an exception in the thread method, the application just "stops" - no error message, no nothing, but only an entry in the Event Log. Not even then UnhandledExceptionHandler is triggered.
Maybe something like this is the cause?
回答6:
...if you are able to contact that suffering user, here is an
Idea: log pre-execution stages
Instead of making a shortcut to your program.exe, make a shortcut to program.bat, which will
echo "Pre-start" > stage.txt
start program.exe
The first line of Program.cs will therefore be
File.WriteAllLines("stage.txt", "Program execution started.");
In the handler of AppDomain.UnhandledException first line will be
File.WriteAllLines("stage.txt", "Unhandled exception has been caught.");
Also, make sure that the handler does not allocate memory or resources — pre-allocate them on programs start. Handler only trigger the writing to the log.
Comments
It is very likely that the stage.txt (sent by the user) will contain "Pre-start". This happens when an exception is thrown in 3rd party .dll — even before your program has started.
In that case you will need a simple checker program, which will not reference the assemblies that you program.exe does, but will Assembly.Load(...) them.
P.S.
stage.txt should be placed somewhere under %APPDATA%, not in Program Files.
I found an interesting case on Server 2003 and another nice discussion.
回答7:
You should get a more detailed stack trace by sending the .pdb file for that particular release to the user (to be put next to the .exe) and having them reproduce the crash.
回答8:
You should handle AppDomain.UnhandledException in code.
There was a similar question asked. See related ones too.
来源：https://stackoverflow.com/questions/814560/how-to-troubleshoot-net-2-0-error-reporting-messages-in-the-event-log