Re: nullptr vs 0

From: Ion Gaztañaga (igaztanaga_at_[hidden])
Date: 2022-05-23 20:18:15

On 23/05/2022 22:06, Peter Dimov via Boost wrote:
> Ion Gaztañaga wrote:
>> On 22/05/2022 2:59, Peter Dimov via Boost wrote:
>>> Compilers apparently are warning on the use of 0 as a null pointer
>>> constant, suggesting we use nullptr instead. But compilers don't know
>>> that we support C++03 where nullptr isn't a thing.
>>>
>>> Case in point: https://github.com/boostorg/throw_exception/pull/22
>>>
>>> ifdef-ing every use of nullptr is unwieldy, so maybe we need
>>> BOOST_NULLPTR added to Boost.Config? That would expand to nullptr
>> when
>>> it's supported, and 0 otherwise.
>>
>> Maybe we need the boost::nullptr_t type, emulated in C++ mode and "typedef
>> decltype(nullptr) nullptr_t;" in compilers that support it...
>>
>> And if nullptr/nullptr_t is provided by Boost.Config, without any heavy-include
>> need, that would be perfect ;-)
> 
> That's not what Boost.Config is for.
> 
> Plus, since nullptr is a keyword, available without qualification in any scope, it
> wouldn't be possible for us to provide a compatible replacement as we aren't
> allowed to define things in the global namespace.
> 
> For our purposes it will probably work most of the time if defined in boost::
> but there's no good way to define a global in a header under C++03 anyway.

Well, I wasn't "literally" trying to define nullptr[_t], but a practical 
alternative that is the same type as the standard one when available.

BOOST_NULLPTR/boost::nullptr and boost::nullptr_t are more than enough 
for portable code. Does it make sense?

Ion