On 30/01/2014 15:13, Quoth Ben Pope:

...

// invoke g with empty object g(shared_ptr<CSomeReallyLongClassName>());

g(0);

Note that (at least in Boost 1.53, haven't checked if this has been changed), .reset(0) is less efficient than .reset() because it still allocates a shared_count to the null pointer. I haven't checked if the constructor has the same issue but it wouldn't surprise me. (Also, the above doesn't compile in 1.53 because the raw pointer constructor is marked explicit. Again, maybe that's been changed in the interim.)

On 30/01/14 14:23, Gavin Lambert wrote:

Mere moments ago, quoth I:

...

On 30/01/2014 15:13, Quoth Ben Pope:

...

...

// invoke g with empty object g(shared_ptr<CSomeReallyLongClassName>());

g(0); [...] (Also, the above doesn't compile in 1.53 because the raw pointer constructor is marked explicit. Again, maybe that's been changed in the interim.)

Whoops, You're right. I did test it, but I think I compiled it in C++11 mode!

If you're using a C++11 compiler you should be able to use this though:

g(nullptr);

Yes, of course. Internally boost::shared_ptr uses boost::detail::sp_nullptr_t, which in C++11 is a typedef of std::nullptr_t, there appears to be no alternative for C++98, which is a shame. I don't see why that type couldn't be defined to an empty struct in C++98 mode. Then, removing the #if !defined( BOOST_NO_CXX11_NULLPTR ) guard from the constructor of shared_ptr that accepts a boost::detail::sp_nullptr_t does indeed seem to work. namespace boost { namespace detail { struct sp_nullptr_t {}; } const detail::sp_nullptr_t nullptr = detail::sp_nullptr_t(); } #include int f(boost::shared_ptr<int> i) { return i ? *i : 0; } int main() { f(boost::nullptr); } ...there's probably some reason I'm not thinking of. Ben

On Thu, Jan 30, 2014 at 8:18 AM, Nevin Liber wrote:

you really don't want to use boost::nullptr, given that nullptr is a keyword in C++11.

boost::none might be a candidate for this. In the meantime looks like I'm stuck with g(shared_ptr<CSomeReallyLongClassName>()); Thanks, Chris

On 1/30/2014 1:12 AM, Ben Pope wrote:

Yes, of course. Internally boost::shared_ptr uses boost::detail::sp_nullptr_t, which in C++11 is a typedef of std::nullptr_t, there appears to be no alternative for C++98, which is a shame.

I don't see why that type couldn't be defined to an empty struct in C++98 mode.

That works for my own classes: If nullptr_t is not available on the platform, I define a class by that name in my own detail namespace. When defining a class that has a member function taking a parameter of type nullptr_t, no conditional compilation is needed around it, and it finds the definition in the current namespace. If the user of the class defines his own value of type libraryName::nullptr_t, that function can be called. I have not experimented with providing a templated conversion operator to allow libraryName::nullptr_t to be converted to any T*. That should satisfy at least some uses of nullptr literal, although it counts as a user-defined conversion. —John