MFC built using /clr compiler option - CString/System::String conversion

Question

I've got a fairly large MFC application that has just been migrated from VS6.0 to VS2008. It was a pretty painful process, but now I'd like to explore any managed-code options that may be available. I was able to successfully build the project using the /clr switch which seems to give me access to managed types.

I'd like to know if conversion between System::String and CString is automatic or not. The MSDN documentation that I've found suggests that this conversion isn't automatic but I haven't found this to be the case. All three of the examples below work and both 1 and 2 are documented by the MSDN. What I'm curious about is example 3 which also works but I don't know why. The CurrentDirectory property is returning a pointer into the managed heap, String^, but why am I able to assign it to a CString? Exactly what is example 3 doing and what are the memory management implications?

Example 1)

    marshal_context ^ context = gcnew marshal_context();
    String ^env = System::Environment::CurrentDirectory;
    const char* env2 = context->marshal_as(env);
    AfxMessageBox(env2);
    delete context;

Example 2)

    CString s(System::Environment::CurrentDirectory);
    AfxMessageBox(s);

Example 3)

    CString s = System::Environment::CurrentDirectory;
    AfxMessageBox(s);

Answer 1

You can go from a system::String to CString because they share a common conversion (lptstr?) going to a System::String from CString will always need System::String^ test = gcnew System::String( CStringVar );

Answer 2

Option 3 works for pretty much the same reason option 2 does. CString::operator= has an overload for System::String. Don't forget that the assignment operator can do a lot more than copy a reference.

This page: How to: Convert Between Various String Types is very useful for mixed apps. pin_ptr is great.

Be wary of sprinkling managed code around. I also work on a large MFC application that went /clr and I sometimes wish that we had done it in select libraries only. It can make debugging painful when there are a lot of managed-native transitions on the callstack. There are performance considerations as well.