Feed Icon RSS 1.0 XML Feed available

Kinda Smart Pointers in "C/C++"

Date: 10-Feb-2016/10:20

Tags: ,

Characters: (none)

If you happen to ever title a StackOverlow question something like "How do I do (whatever) in C/C++", then I guarantee within a few short minutes someone will say "Which language? There is no such thing as C/C++.". I don't suggest arguing with them unless you like downvotes. That said--the C/C++ language does exist, and I'm probably one of the few people who sometimes studies it. :-)
This article is a note about an "augmented" pointer...in the sense that it can add some intelligence to a C program if it can also be compiled as C++. I've not actually used the technique, because it requires you to change the notation for how the pointer is declared. But thinking through how one might do it was interesting enough that I thought I'd make a note about it.

TYPE* vs. TYPE(*) - "AUGMENTED" POINTERS

If a C codebase uses raw pointers for an abstraction that you would like to compile differently in C++, the way pointers are declared has to change. There's just no way around it.
That is to say that if the C code has a line like:
TYPE *value;
You can't come up with a way to #define TYPE that makes that line a raw pointer to TYPE in the C build, and a definition of a class variable in a C++ build. It can only be a raw pointer.
Note
If you want to be pedantic AND ridiculous, you might turn it into something like a statement. :-/
#include <stdio.h>

#define TYPE int temp = 5; int *value = &temp; bar =

int main(int argc, char* argv[]) {
    int bar;
    TYPE *value;
    printf("bar=%d\n", bar); // prints bar=5
    return 0;
}
So if TYPE * cannot become an "augmented pointer" class in a C++ build, what's the best tweak to the pattern that could? One possibility would be to strike the asterisk from the usage sites, instead making the typedef itself carry the pointer:
#ifdef __cplusplus
    class AugmentedPointer { TypeImpl *p; ... };

    typedef class AugmentedPointer TYPE_P; // or PTYPE, etc.
#else
    typedef struct TypeImpl *TYPE_P;
#endif
That doesn't seem terrible. But it has problems in terms of the interactions with const. Just considering the C side, if you typedef a pointer, you can't add a const that goes "inside" the typedef. const TYPE_P would mean the pointer itself would be unchangeable, not the fields of the value it pointed to.
One could just keep making new types to get past it--for instance by naming the const versions as well (TYPE_P vs TYPE_CP). Or a parameterized macro could be used: (P(TYPE) vs. P(const TYPE)).
What I thought was cooler was to make the macro parameterizations TYPE(*) and TYPE(const *). Keeping the asterisk at the callsite helps make it more obvious that it's a pointer. And it's also more clearly not a function call (which was already suggested by context, but it can't hurt to look even less like one):
TYPE(*) value = NULL;

Making the Notation Work

It's easy enough to see how to define this macro to work for C:
#define TYPE(x) TypeImpl x
//
// `TYPE(const *) foo;` => `TypeImpl const *foo;`
// `TYPE(*) foo;`       => `TypeImpl *foo;`
But what about for C++?
As mentioned in the introduction...I didn't go very far with this. But here's a sketch that chooses one of two template specializations based on the const of a dummy parameter. The mutable form is then derived from the immutable one:
template<class T>
class TypePtr;

template<>
class TypePtr<const void*> {
protected:
    TypeImpl *p;

public:
    TypePtr () {}
    TypePtr (const TypeImpl *p) : p (const_cast<TypeImpl *>(p)) {}

    const TypeImpl **operator&() { return &p; }
    const TypeImpl *operator->() { return p; }
    const TypeImpl &operator*() { return *p; }

    void SomeCommonMethod() { ... }
};

template<>
class TypePtr<void *> : public TypePtr<const void *> {
    
public:
    TypePtr () : TypePtr<const void*>() {}
    TypePtr (TypeImpl *p) : TypePtr<const void*> (p) {}

    TypeImpl **operator&() { return &p; }  
    TypeImpl *operator->() { return p; }
    TypeImpl &operator*() { return *p; }
};

#define TYPE(x) TypePtr<void x>
Note that matching the C behaviors for a pointer is one of the very rare occasions you'd want to overload the & operator.
Unfortunately, if your type has a lot of pointer math done with it, there's no good way to get that automatically in C++. Those have to be overloaded manually if they are used. :-/
So what exactly might a C++ build bring to the table for such a "kind of smart" pointer class? The ability to hook dereferences could make it easier to write debugging tools, and it would be possible to keep track of how many pointers you have extant. I'm not exactly sure what it might actually be used for in a "special build" of a C program, I just thought it was interesting!
Business Card from SXSW
Copyright (c) 2007-2015 hostilefork.com

Project names and graphic designs are All Rights Reserved, unless otherwise noted. Software codebases are governed by licenses included in their distributions. Posts on blog.hostilefork.com are licensed under the Creative Commons BY-NC-SA 4.0 license, and may be excerpted or adapted under the terms of that license for noncommercial purposes.