Me vs. The Compiler

[billroper]

I figured out the source of yesterday's problem.

I had put in a forward definition for a class in a header file so that I could declare a pointer to an object of that type inside the class that was defined in the header file. Like this:

class AnotherClass;

class ThisClass
{
...
  AnotherClass*   anotherClassInstance;
};

When I destroy ThisClass, I want to destroy the instance of AnotherClass that it contains. No problem. I'll just call delete on the pointer like this:

delete anotherClassInstance;

That should work just fine.

Except it didn't. Because I hadn't actually included the header file for AnotherClass in the .cpp file for ThisClass. What I would have expected was that this should have produced a compile error telling me that I was operating on a class that hadn't been completely defined.

But there was no compile error. Instead, the Visual Studio compiler just called delete on anotherClassInstance like it was a pointer to a random chunk of memory with no destructor.

Hilarity ensued.

Well, at least I should remember what caused me to waste so much time the next time I run into this problem...

*grumble*

Comments

[bedlamhouse]

A compiler that assumes at run time such things would be properly in place? Never!

[billroper]

Who knows? Maybe it's been fixed in a later version of Visual Studio. But I'm still mired on VS2010, because one of our programmers decided that the best thing to do for his particular project was to use a bunch of undocumented methods that are not in later versions of the libraries.

Oops.

So now we have to fix all of his code before we can migrate forward.