Re: nullptr vs 0

From: Gavin Lambert (boost_at_[hidden])
Date: 2022-05-29 23:22:47

On 27/05/2022 23:00, Andrey Semashev wrote:
> It *is* a useless warning. The code with NULL is explicit enough and
> portable, so what is this warning about? That the code is not
> C++11-only? I know that, and it's not up to the compiler to tell me that.

The problem is that NULL is defined as a plain 0, and there is potential 
ambiguity with comparing a pointer to 0 -- did you actually intend to 
compare the pointer itself or were you intending to compare the value 
being pointed to?  This is obvious when compared to anything other than 
0, but 0 gets this special magic treatment due to historic precedent 
(which is, as usual with C/C++, misguided other than for backwards 
compatibility).

It's actually worse in C++, because C defines NULL as explicitly 
pointer-typed, but C++ doesn't (because it introduces new warnings about 
comparing void pointers with other pointers) -- but this also means at 
the compiler level it can't tell if you did actually write NULL or 0, so 
it has even less idea whether it was intended or not.  As usual, 
backwards compatibility is a comedy of errors.

Strongly suggesting (but not requiring) that all newly-written code 
should use nullptr rather than NULL or 0 is the correct choice.  The 
only reason you might not do so is for C++03 compatibility -- but newly 
written code really shouldn't be trying to do that either any more.  The 
language has moved on, and codebases should too.

I support use of BOOST_NULLPTR for compatibility of old codebases, however.