Pointers to member functions are very strange animals

re: Pointers to member functions are very strange animals

Mon, 09 Feb 2004 15:24:00 GMT

'Casting a function pointer can change its size!'

Bu..bu.. bu ... my head hurts.

re: Pointers to member functions are very strange animals

Lonnie McCullough — Mon, 09 Feb 2004 15:44:00 GMT

I was just wondering about this very subject and was going to post a comment asking you to discuss it. Wow! Thanks again, I always wondered why my pointers to member functions had a sizeof() == 8 sometimes and 4 other times.

re: Pointers to member functions are very strange animals

Henk Devos — Mon, 09 Feb 2004 16:02:00 GMT

Long ago, when trying to cast a member function pointer to a DWORD, i couldn't believe that the compiler gave me a size mismatch pointer...

re: Pointers to member functions are very strange animals

Curt — Mon, 09 Feb 2004 16:15:00 GMT

> The size of a pointer-to-member-function can change depending on the class!

As if Mondays aren't hard enough! I'm going back to bed. :)

re: Pointers to member functions are very strange animals

Jonathan O'Connor — Mon, 09 Feb 2004 16:55:00 GMT

My friend and former colleague, Martin O'Riordain came up with the idea of using thunks for ptr-to-member-fn about 1990. He wouldn't tell us about them then, because he wanted to use this for his interview with Microsoft. Its funny to see that they aren't used anymore. But then I guess, C++ has moved on a bit since then.

re: Pointers to member functions are very strange animals

Ben Hutchings — Mon, 09 Feb 2004 18:22:00 GMT

The size of a pointer-to-member should not vary as you describe. The C++ standard says it's OK to cast a pointer-to-member-of-derived to pointer-to-member-of-base (though it must be a non-virtual base) so long as you only use the result with pointers-to-base that really point to instances of the derived class. So the "optimisation" of dropping the adjustment field where it's apparently not necessary is actually a bug. The compiler admits:
"warning C4407: cast between different pointer to member representations, compiler may generate incorrect code". Thankfully there is an option that fixes this: "/vmg". However, the documentation for C4407 and for "/vmg" fails to acknowledge that this behaviour is a bug and basically blames the programmer for finding the edge cases where this "optimisation" doesn't work.

re: Pointers to member functions are very strange animals

Raymond Chen — Mon, 09 Feb 2004 21:58:00 GMT

Actually, Ben, if you're referring to 4.11.2 (pointer to member conversions) the standard actually says the opposite. You can convert pointer-to-base-member into pointer-to-derived-member without loss of fidelity. And that's what we're doing here.

I can't find a place in the standard where it says that you can safely cast pointer-to-derived-member to pointer-to-base-member.

re: Pointers to member functions are very strange animals

Woon Kiat — Tue, 10 Feb 2004 00:44:00 GMT

Hi Raymond, this is a very interesting topic. Is there any book or material you would recommend for further understanding on this topic?

re: Pointers to member functions are very strange animals

Raymond Chen — Tue, 10 Feb 2004 00:53:00 GMT

This is all undocumented implementation details, so I doubt there'd be a book on it. You can read Ellis/Stroustrup "The Annotated C++ Reference Manual" to see the design of a completely different way of doing pointers to functions.

re: Pointers to member functions are very strange animals

Woon Kiat — Tue, 10 Feb 2004 01:42:00 GMT

What about reference on multiple inheritance? Is there any book you would recommend? Most books just touch the surface and never go deep into it. The concept of MI always confuse me when it comes to the vtable layout and implementation. And the architect of C# decide not to have MI in C#. Do you think MI is a good topic to blog on?

re: Pointers to member functions are very strange animals

Raymond Chen — Tue, 10 Feb 2004 02:20:00 GMT

I'm not sure what there is to write about on the subject of multiple inheritance. It's a language feature. The language specification leaves a lot of the details to the implementation, so you shouldn't expect to find a book that details implementation since that's all undocumented and subject to change at any time.

re: Pointers to member functions are very strange animals

Stefan — Tue, 10 Feb 2004 06:35:00 GMT

There is an article in MSDN about this:

"C++: Under the Hood", Jan Gray, March 1994

http://msdn.microsoft.com/archive/default.asp?url=/archive/en-us/dnarvc/html/jangrayhood.asp

Answer to exercise: Pointer to member function cast

The Old New Thing — Tue, 10 Feb 2004 18:00:00 GMT

re: Pointers to member functions are very strange animals

Ben Hutchings — Tue, 10 Feb 2004 21:51:00 GMT

Raymond, I'm talking about explicit conversions (aka casts) which are specified in 5.4/7. 4.11/2 is about standard conversions which can be done implicitly (because they're generally safe).

re: Pointers to member functions are very strange animals

Raymond Chen — Wed, 11 Feb 2004 00:20:00 GMT

Ben: Okay I read 5.4/7 and I think you're right. Then again, I'm not a real language lawyer; I just play one on the Internet.

re: Pointers to member functions are very strange animals

Giovanni Bajo — Mon, 16 Feb 2004 20:52:00 GMT

Why anybody should care about Rich Hickey's ten-years-old techniques when we have boost::bind and boost::function which works wonderfully as generalized binders and callbacks?

Pointers to virtual functions with adjustors

The Old New Thing — Thu, 24 Mar 2005 16:03:00 GMT

Two confusing features that explode when you combine them.

Qooiii.com blog » Blog Archive » Hardly Anybody

Qooiii.com blog » Blog Archive » Hardly Anybody — Wed, 06 Dec 2006 07:31:07 GMT

PingBack from http://www.qooiii.com/blog/?p=27

PeousWare - At last, sharing ! » Blog Archive » Member function pointer size in Visual C++

Tue, 29 Jul 2008 02:33:40 GMT

PingBack from http://www.peousware.com/?p=11