On Sat, Dec 6, 2008 at 14:48, Roman Yakovenko <roman.yakovenko@gmail.com> wrote:
Sorry for noise. I didn't see "native_generator" typedef "on Windows".
May be it worth to introduce "native_generator.hpp" header, which will select the platform and "include" the native generator?
No, the noise was a good thing. I think I like the native_generator better, in retrospect. I'm thinking of something like this (with the details better flushed out): #ifdef BOOST_HAS_E2FSPROGS #include <uuid/uuid.h> struct native_default { void generate(uuiid &u) { uuid_generate(&*u.begin()); } }; struct native_time_based { void generate(uuiid &u) { uuid_generate_time(&*u.begin()); } }; struct native_random { void generate(uuiid &u) { uuid_generate_random(&*u.begin()); } }; #elif BOOST_HAS_WINDOWS #include <rpc.h> struct native_default { void generate(uuiid &u) { UUID u_; UuidCreate(&u_); u.assign((uint8_t*)&u_, 16+(uint8_t*)&u_); } }; // And others, possibly #endif template <typename NativeStrategy = native_default> struct native_generator : NativeStrategy { uuid operator()() const { uuid u; NativeStrategy::generate(u); return u; } }; Looks like less code duplication overall, too. With SFINAE it should even work for SHA1 and MD5 generators, with an overloaded constructor to pass a uuid to the Strategy's constructor, and an overloaded operator() that takes a message to hash. The one problem is if people wanted more choice. I was thinking that the http://www.ossp.org/pkg/lib/uuid/ could provide another generator in the old scheme, but that may also just be a waste of effort, since it's less common than the e2fsprogs libs. Also, what it provides are the same things that Boost.UUID provides natively, so... ~ Scott