3

I have a windows application that calls an external .dll. After a while, there were fatal errors being brought to my attention that had to do with user marshaling. There was a source online that with that particular error I was to change my target to x86 rather than AnyCPU. I did so, and now whenever I let the app run, it will exit debug mode and crash the application. But if I set a break point immediately after the .dll call, and step over each line until I receive control of the application again, it doesn't crash. Is there anything specific that could be causing this? Has does one debug this issue?

Thanks!

Jake Smith
  • 2,332
  • 1
  • 30
  • 68
  • Try enabling Unmanaged Code Debugging in the project properties and taking a look at the disassembled code. Good luck... – user541686 Aug 14 '12 at 21:06
  • Could the difference in the way the crash is handled be down to [windows forms handling exceptions differently when debugging](http://stackoverflow.com/a/6734767/113141) ? – Justin Aug 14 '12 at 21:17

3 Answers3

1

Stepping code solving an issue is often a symptom of timing problems in the original code. If an external resource loads asynchronously, it will not show up on the stack of the current thread in the debugger, but will be able to be called. Stepping over code induces a delay in the flow.

HenryZhang
  • 1,318
  • 8
  • 11
0

Thank you all for you suggestions! Fortunately, I ended up getting it to work (with minimal understanding as to why it works) but changing the build target to specifically x86 machines rather than "AnyCPU." This was suggested by a website and can no longer find :\ Hope this helps others than run into a similar issue!

Jake Smith
  • 2,332
  • 1
  • 30
  • 68
0

I consider the most common cause of this sort of thing to be uninitialized variables. They pick up whatever was in memory and the presence of a debugger can easily change what's in the unused part of the stack--the memory that's going to become local variables when the next routine is called. Check the DLLs code.

Note that your "fix" makes me even more suspect that this is the real answer.

(Then there's also the really crazy case of a problem with the debugger. Long ago I hit a case where the debugger had no problem loading an invalid value into a segment register if you were single stepping.)

Loren Pechtel
  • 8,945
  • 3
  • 33
  • 45