Hartmut Kaiser wrote:

Hi all,

Is there a rational for the different behavior (signature) of the atomic_count::operator++()?

This operator returns long (the new value) on Win32 and Solaris, but has no return type (void) on pthread and gcc based systems.

No reason. In the "specification" in atomic_count.hpp I see that ++a is documented as having no return value. So 'void' is correct and 'long' can be considered correct but misleading.