zero-as-null-pointer-constant C++ trunk

[David Brown]

On 22/05/2019 16:39, Jonathan Wakely wrote:
> On Wed, 22 May 2019 at 15:28, David Brown wrote:
>>
>> On 22/05/2019 16:12, Jonny Grant wrote:
>>> Hello
>>>
>>> This is a little test case. It is not real code.
>>>
>>> We used gcc-8 to get some warnings, but on the godbolt trunk gcc doesn't
>>> those warnings with this little test case. Am I missing something
>>> obvious? Is this an issue?
>>>
>>
>> The warning is for treating 0 as a null pointer, but you are not doing
>> that - you are using NULL as a null pointer.
> 
> Right.
> 
>> The whole point of nullptr is to have a way to give a null pointer that
>> matches function overloads of pointer type rather than overloads of
>> integer type:
>>
>> int foo(void*) { return 1; }
>> int foo(int) { return 2; }
>>
>> Calling "foo(0)" will give you 2, which is not the result you want if
>> you are wanting a null pointer.  Calling "foo(nullptr)" will give you 1,
>> which is the sensible result.  Calling "foo(NULL)" will give you an
>> error about ambiguous overloads.
> 
> That's only because NULL is defined as __null, which is equivalent to
> 0L. It could equally be defined as 0 and then would be identical to
> foo(0).
> 
> But the point is that "NULL" is not an integer literal (even if it
> expands to one), it's a name for a null pointer. It's not as good a
> name as "nullptr" but it's still not a literal zero.
> 

In that case, I personally would have liked the warning to trigger here 
if NULL is defined in a way that matches an "int" overload without any 
warning or error.  Clearly if a programmer writes "foo(NULL)" here, 
he/she expects the pointer overload to win over the int overload.  If 
the rules of the language disagree, then a warning would be appropriate.

mvh.,

David


> 
>> That is perhaps not great, but it will
>> certainly avoid a quite error in the code.  So there is no need for this
>> warning to trigger on the use of NULL for a null pointer.
>>
>> (This is just my opinion as a user.)