Classes and objects

Part of C++ FQA Lite. To see the original answers, follow the FAQ links.

One of the stated goals of C++ is support for object-oriented programming. This page introduces C++ classes and outlines the tactics they use to defeat their purpose.

[7.1] What is a class?

FAQ: In OO software, "the fundamental building block".

A class is a type - a representation for a set of states (much like a C struct) and a set of operations for changing the state (moving from one state to another). Classes are similar to built-in types in this sense (for example, an int holds a bunch of bits and provides operations like + and *).

FQA: That's a correct theoretical definition. It's equally applicable to all OO languages, but they are different when it comes to more specific, practical aspects of their particular implementation of classes.

How do I create objects? And what happens when they are no longer needed? Is it my job to figure out which ones are unused and deallocate them? Bad.

What happens if I have bugs? If I have a pointer to an object, can it be invalid (be a random bit pattern, point to a dead object)? It can? The program will crash or worse? What about arrays of objects and out-of-bounds indexes? Crash or a modification of some other random object? You call that encapsulation? Bad.

What happens if I change/add/remove a private value, without changing the interface? All code using the class has to be recompiled? I bet you call that encapsulation, too. Bad.

I don't like C++ classes.

[7.2] What is an object?

FAQ: A chunk of memory with certain semantics. The semantics are defined by the class of the object.

FQA: They are also defined by the bugs which cause the code to overwrite data of these objects without bothering to use the interface defined by the class. People who think that real programmers write code without bugs need to upgrade to a human brain.

Still, it sounds interesting: the memory of our C++ program is apparently broken into chunks storing objects of various classes, with "defined semantics". Looks very promising, that. For example, we could ask a debugger about the kind of object located at such a chunk and inspect its data (as in "this is a Point with x=5 and y=6"). We could even take this one step further and implement things like garbage collectors, which can check whether an object is used by looking for pointers to it in the places which are supposed to store pointers.

Unfortunately, you can't tell the class of a C++ object given a pointer to it at run time. So if you debug a crashed C++ program and find a pointer somewhere in its guts, and you don't know its type, you'll have to guess that "0000000600000005" is a Point. Which is completely obvious, because that's the way a pair of adjacent integers looks like in hexadecimal memory listings of a little endian 32 bit machine. And two adjacent integers might be a Point. Or some other two-integer structure. Or a part of a three-integer-and-a-float structure. Or they might be two unrelated numbers which just happen to be adjacent.

Which is why you can't automatically collect the garbage of C++ programs.

[7.3] When is an interface "good"?

FAQ: It is good when it hides details, so that the users see a simpler picture. It should also speak the language of the user (a developer, not the customer).

FQA: Um, sometimes you want the interface to expose the many details and speak the language of the machine, although it's probably not very common. The generic answer is something like "an interface is good if it gets the user somewhere".

For example, using OpenGL you can render nifty 3D stuff at real time frame rates. FFTW delivers, well, the Fastest Fourier Transform in the West. With Qt, you can develop cross-platform GUI, and "cross-platform" won't mean "looking like an abandoned student project". Writing that stuff from scratch is lots of work; using the libraries can save lots of work. Apparently learning the interfaces of these libraries is going to pay off for many people.

For a negative example, consider <algorithm>. Does std::for_each get us anywhere compared to a bare for loop, except that now we need to define a functor class? That's a bad interface, because learning it doesn't make it easier to achieve anything useful.

[7.4] What is encapsulation?

FAQ: The prevention of "unauthorized access" to stuff.

The idea is to separate the implementation (which may be changed) from the interface (which is supposed to be stable). Encapsulation will force users to rely on the interface rather than the implementation. That will make changing the implementation cheaper, since the code of the users won't need to be changed.

FQA: That's a nice theoretical definition. Let's talk about practice - the properties of the C++ keywords private and protected, which actually implement encapsulation.

These keywords will cause the compiler to produce an error message upon access to a non-public member outside of the class. However, they will not cause the compiler to prevent "unauthorized access" by buggy code, for example upon buffer overflow. If you debug a crashed or misbehaving C++ program, forget about encapsulation. There's just one object now: the memory.

As to the cost of changes to the the private parts - they trigger recompilation of all code that #includes your class definition. That's typically an order of magnitude more than "code actually using your class", because everything ends up including everything. "The key money-saving insight", as the business-friendly-looking FAQ puts it, is that every time you change a class definition, you are recompiling the programs using it. Here's another simple observation: C++ compiles slowly. And what do we get now when we put two and two together? That's right, kids - with C++ classes, the developers get paid primarily to wait for recompilation.

If you want software that is "easy to change", stay away from C++ classes.

[7.5] How does C++ help with the tradeoff of safety vs. usability?

FAQ: In C, stuff is either stored in structs (safety problem - no encapsulation), or it is declared static at the file implementing an interface (usability problem - there is no way to have many instances of that data).

With C++ classes, you can have many instances of the data (many objects) and encapsulation (non-public members).

FQA: This is wildly wrong, and the chances that the FAQ author didn't know it are extremely low. That's because you can't use FILE* from <stdio.h> or HWND from <windows.h> or in fact any widely used and/or decent C library without noticing that the FAQ's claim is wrong.

When you need multiple instances and encapsulation in C, you use a forward declaration of a struct in the header file, and define it in the implementation file. That's actually better encapsulation than C++ classes - there's still no run-time encapsulation (memory can be accidentally/maliciously overwritten), but at least there's compile-time encapsulation (you don't have to recompile the code using the interface when you change the implementation).

The fact that a crude C technique for approximating classes is better than the support for classes built into the C++ language is really shameful. Apparently so shameful that the FAQ had to distort the facts in an attempt to save face (or else the readers would wonder whether there's any point to C++ classes at all). The FQA hereby declares that it will not go down this path. Therefore, we have to mention this: the forward declaration basically makes it impossible for the calling code to reserve space for the object at compile time. This means that a struct declared in a header file or a C++ class can sometimes be allocated more efficiently than a forward-declared struct. However, this is really about a different tradeoff - safety vs. efficiency, and there's no escape from this tradeoff. Either the caller knows about the details such as the size of an object at compile time - which breaks compile-time encapsulation - or it doesn't, so it can't handle the allocation.

Anyway, here's the real answer to the original question: C++ helps with the tradeoff of safety vs. usability by eliminating both.

C++ is extremely unsafe because every pointer can be used to modify every piece of memory from any point in code. C++ is extremely unusable due to cryptic syntax, incomprehensible semantics and endless rebuild cycles. Where's your tradeoff now, silly C programmers?

[7.6] How can I prevent other programmers from violating encapsulation by seeing the private parts of my class?

FAQ: Don't bother. The fact that a programmer knows about the inner workings of your class isn't a problem. It's a problem if code is written to depend on these inner workings.

FQA: That's right. Besides, people can always access the code if a machine can. Preventing people from "seeing" the code in the sense that they can access it, but not understand it is obfuscation, not encapsulation.

[7.7] Is Encapsulation a Security device?

FAQ: No. Encapsulation is about error prevention. Security is about preventing purposeful attacks.

FQA: Depends on the kind of "encapsulation". Some managed environments rely on their support for run time encapsulation, which makes it technically impossible for code to access private parts of objects, to implement security mechanisms. C++ encapsulation evaporates at run time, and is almost non-existent even at compile time - use #define private public before including a header file and there's no more encapsulation (correction). It's hardly "encapsulation" at all, so of course it has no security applications - security is harder than encapsulation.

The capital E and S in the question are very amusing. I wonder whether they are a manifestation of Deep Respect for Business Values or Software Engineering; both options are equally hilarious.

[7.8] What's the difference between the keywords struct and class?

FAQ: By default, struct members and base classes are public. With class, the default is private. Never rely on these defaults! Otherwise, class and struct behave identically.

But the important thing is how developers feel about these keywords. struct conveys the feeling that its members are supposed to be read and modified by the code using it, and class feels like one should use the class methods and not mess with the state directly. This difference is the important one when you decide which keyword to use.

FQA: struct is a C keyword. class was added to C++ because it is easier than actually making the language object-oriented. And it does a good job when it comes to the feeling of a newbie who heard that "OO is good".

Check out the emotional discussion about which keyword should be used in the FAQ. The more similar two duplicate C++ features are, the more heated the argument about the best option to use in each case becomes. Pointers/references, arrays/vectors... Yawn.

By the way, the forward-declaration-of-struct thing works in C++, and it's better than a class without virtual functions most of the time.

Copyright © 2007-2009 Yossi Kreinin
revised 17 October 2009