Thread: Library Implementation of nullptr

I have been looking at a library implementation of nullptr as shown in A name for the null pointer: nullptr (revision 2) (warning: PDF).

One criticism leveled against it is that:

Experiments with several popular existing compilers show that it generates poor and/or misleading compiler diagnostics for several of the common use cases described in section 2. (Examples include: "no conversion from 'const' to 'int'"; "no suitable conversion function from 'const class <unnamed>' to 'int' exists"; "a template argument may not reference an unnamed type"; "no operator '==' matches these operands, operand types are: int == const class <unnamed>".) We believe that compilers will still need to add special knowledge of nullptr in order to provide quality diagnostics for common use cases.

It seems to be that this can be solved simply by naming the class, e.g., nullptr_t. This allows one to create more nullptr_t objects other than nullptr, but since they are all equivalent anyway, it seemes a small price to pay for far more readable error messages. What other problems do you see with turning this anonymous class into the nullptr_t class?

You lose the second part of this design goal:

The null pointer is a value that has no nameable type, and its type cannot be deduced as a template argument.

I don't see the nullptr proposal as even remotely difficult to implement for compilers - GCC already has most of it - so I think the keyword solution is simply preferable.