I personally preferred 0 till nullptr came around. There was no overwhelming reason, it just seemed to read better IMO.

Although NULL IS ugly, it cannot be replaced with 0 (like others have explained). But you can always "rename" stuff in C/C++, so while it has been long since I last came across a compiler without support for nullptr, I would just redefine NULL as nullptr since I would probably also build the source on C++0x/C++11 complaint compilers, Of course, you loose the type of nullptr when compiling for pre-C++0x (std::nullptr_t) but with a little care (only load/store and compare with pointers), I cannot see why it would cause trouble.

#ifdef THIS_COMPILERS_VERSION_IS_WAY_TOO_OLD_FOR_MY_TASTE
#define nullptr NULL
#endif


EDIT: An enhanced version might look like this, since it does not pollute the global namespace and even gives nullptr a type:

namespace SuperLibrary
{
#ifdef THIS_COMPILERS_VERSION_IS_WAY_TOO_OLD_FOR_MY_TASTE
const void *nullptr = (unsigned)NULL;
#endif
...
}


#2: 0 is the only literal constant that is defined by the standard to be implicitly castable to a pointer of any kind,

No, any integer constant expression that evaluates to zero is a valid null pointer constant. This includes 0 but also includes (1-1), (0xful >> 10), and '\0'. The standard specifically lists both 0 and 0L as two legal options for NULL. The latter is used by some compilers where an int is not the same size as a pointer, but a long is. This can give narrowing conversion warnings when trying to assign a NULL to a regular int. See the previously mentioned sections 8.1 in C++98/C++03 and 8.2 in C++11, as well as sections 4.10 and 5.19 (same number for all three).

I would just redefine NULL as nullptr since I would probably also build the source on C++0x/C++11 complaint compilers, Of course, you loose the type of nullptr when compiling for pre-C++0x (std::nullptr_t) but with a little care (only load/store and compare with pointers), I cannot see why it would cause trouble.

Again, as previously mentioned, NULL has an integer type which does not play happily with things like function overloads and template type deduction. If you have an overloaded function void foo(int) and void foo(void *), foo(NULL) calls the first overload but foo(nullptr) calls the second overload. Similarly if you have a vector<int *>, std::fill(vec.begin(), vec.end(), NULL) will fail to compile but std::fill(vec.begin(), vec.end(), nullptr) should.

const void *nullptr = (unsigned)NULL;

This is not a valid replacement for the C++11 nullptr. First, as previously mentioned, a void pointer cannot be assigned to other pointer types without an explicit cast. Basic usage such as int * a = nullptr; would fail. Second, this isn't a constant expression. You've defined nullptr as a non-const pointer to const void. Presumably you wanted some variation with const on the right hand side of the * to make the pointer itself const. But also, even when dealing with void pointers you can still get into a situation where you need a cast because a const void * is not assignable to a volatile void * without a cast.