COM Interop, pure function call on exit

A program written by another company allows for add-ins to be written which communicate via COM interop.

Their program appears to be written in C++, and uses managed code according to Process Explorer.

Our Add-In is written in C#. We compile and register our Add-In as a COM object during compile with Visual Studio.

We found their program will crash on exit with a Pure Function Call error when our Add-In is used.

However, the program will NOT crash if we call Marshal.ReleaseComObject() on every object, and then in the OnDisconnect() method we call GC.Collect(); followed by GC.WaitForPendingFinalizers().

I'm trying to understand why this is.

I did find one comment:

"A common approach for releasing unmanaged objects is to set the RCW reference to null, and call System.GC.Collect followed by System.GC.WaitForPendingFinalizers."

Why would this be necessary to prevent a Pure Function Call error on exit?

Examining the call stack of the crash shows it's attempting to perform some sort of RCW release when it encounters the crash.