Object Oriented ANSI C (1993) [pdf]

http://www.cs.rit.edu/~ats/books/ooc.pdf

by tambourine_man • 12 years ago

44 13 12 years ago

cs.rit.edu

_RPM• 12 years ago

This is interesting literature. I haven't had the patience to sit down and completely read it word for word, but I have skimmed through it a few months ago.

It seems that this would be a great way to learn about how to create an object model in C for your own programming language.

And I might be wrong, but reading this inspires me to create my own toy language and virtual machine for educational purposes just because I am interested in it, not to create the next cool language.

wsc981• 12 years ago

Reading chapter 7 it's quite easy to understand that Objective-C once started as a preprocessor for C.

andyjohnson0• 12 years ago

That's also how C++ started. The original C++ compiler, cfront [1], converted C++ source to C source which it then passed to cc.

[1] http://en.wikipedia.org/wiki/Cfront

pjmlp• 12 years ago

It is also how COM works.

You have a struct that exposes function pointers, which happen to map 1:1 to how C++ compilers create their vtables in Windows. Incidentally how MSVC++ handles vtables, but other vendors need to be compatible with COM, so they follow along.

aikah• 12 years ago

Thanks !This book looks fantastic.I'm not very good at C but it will be an interesting exercise to model a OO system in that language.

readerrrr3• 12 years ago

It has a very high learning curve. And while it has great ideas, the usage of variadic functions is not warranted in my opinion.

I recommend you first try a simple OO project yourself.

ExpiredLink• 12 years ago

This is the old abysmal Schreiner book. I don't know why this is re-posted from time to time. Better look here: http://nullprogram.com/blog/2014/10/21/ (discussion here: http://www.reddit.com/r/programming/comments/2mndod/c_object... )

readerrrr3• 12 years ago

That is a very limited system without real polymorphism.

kazinator2• 12 years ago

Use of void * for information hiding is a very poor idea because you lose type checking. For example, the differ function in section 1.5 can easily be misused without any diagnostic:

  int x, y, a[3];
  /*...*/
  differ("foo", "bar");
  differ(&x, &y);
  differ(a, a);
  differ(&a, &a);

Do not hide so much information that even the compiler is left in the dark!

To not reveal anything about how an object is implemented, use a pointer to an incomplete struct, like

  typedef struct object object;

  int differ (const object *, const object *);

bunderbunder• 12 years ago

"If you lie to the compiler, it will get its revenge."

- someone else, not me

readerrrr3• 12 years ago

A full chapter is dedicated to dynamic type checking.

If that isn't acceptable for your project, don't use it. Successful languages that don't have compile time checks, show that this problem is irrelevant.

kazinator2• 12 years ago

The argument is wrongheaded. Implementations of dynamically typed object systems in C do not benefit from using void * all over the place instead of a typed object * pointer.

C itself isn't dynamic and it behooves you to use it as safely as possible if you don't want your dynamically typed framework to fall apart due to easily avoidable bugs.

Whereas you can have dynamic checking which tells you that some object * pointer is a cons cell, vector, string or whatever, if you use void * instead, the type check itself is threatened with undefined behavior:

   object *stringp(object *obj)
   {
     if (is_heap_object(obj) && obj->typecode == TYPE_STR)
       return t_obj;
     return nil_obj;
   }

   // ..

   {
     int x;

     if (stringp(&x)) ...  // compile-time error!

   }

If stringp has the signature

   void *stringp(void *obj);

then there is no compile time error; &x happily passes into the function. Moreover, the function has to do an ugly conversion to recover a typed pointer so it has to use it.

   void *stringp(void *obj_in)
   {
     object *obj = obj_in; // cast needed here in C++

     if (is_heap_object(obj) && obj->typecode = TYPE_STR)

Dynamic typing is a high level feature that is boostrapped from static typing (whether that typing is checked by a compiler or not!) Static typing means that a given piece of code working on a given piece of data correctly uses that data. For instance if we write assembly code which extracts the bottom three bits of a word, and treats that as a type tag, that is static typing: it's a static fact of the program that that code operates on data with a three bit type tag, and if it is given something else by accident, like a floating-point value without a type tag, it will misbehave.

In short, something has to be rigid at the bottom-most layer. C has enough type muscle to help with that quite a bit without getting in the way too much.

jhallenworld• 12 years ago

I wrote a widget library for X using "object oriented" C. It implements single inheritance using preprocessor tricks: https://github.com/jhallen/joes-sandbox/tree/master/x/notif

Object Oriented ANSI C (1993) [pdf]

Related Stories