On Sep 19, 2005, at 3:33 PM, Robert Ramey wrote:
I like the look of this - though I'm not really familiar enough with boost::tuple to have a strong opinion. I also liked the test. Has this been tested with msvc 6.5? This platform gave me a lot of headaches with serialization of boost::variant but I went through to maintain the idea that serialization is supported on that platform.
I've put up a new implementation at http://svn.resophonic.com/boost/serialization/tuple.hpp http://svn.resophonic.com/boost/libs/serialization/test/test_tuple.cpp per Jaako's comments. This has been tested on gcc4 only, Robert. I saw how much the variant code had to change on the way in. Sorry. I don't have ms platforms available, and don't plan to. I seem to recall, when sifting through the internals of serialization code, (e.g. when attempting to write archives) finding those MS workarounds really distracting. Just so there's a voice on the other side of to-support-or-not-to-support-old-broken-ms question. Anyhow this one is much simpler than variant.
jarvi wrote: The implementation relies on the cons-list implementation of tuples. It is part of the public interface of Boost.Tuple, but not part of the TR1 specification. So if one wants to define serialization so that it works with any TR1 compatible tuple implementation, one must build the metafunctions around the get<N> templates. The downside of this is, that more compiler resources will be used.
So the tuple at boost/tuple/tuple.hpp doesn't have tuple_size<>. One is forced to resort to using either the cons<> stuff, or lots of preprocessor stuff. I went with cons.
Also, at some point we'll roll in "Fusion" that will largely replace the current code of Boost.Tuple. Then we'll have more powerful means to iterate over tuples, in the style of MPL. But for now, my suggestion would be to consider implement serialization using TR1 compatible facilities. Or be prepared to change the implementation when fusion is included into boost.
Best, Jaakko
fusion::tuple, (which I discovered after reading your mail) however, does provide tuple_size<>, so putting together serialization for boost::fusion::tuple is straightforward, and that's what I did, have a look. One thing I did notice while testing, though, is that one cannot instantiate, for instance tuple<bool,bool,bool,bool,bool, bool,bool,bool,bool,bool> (ten bools), but nine are fine. Am I missing something here? At any rate, the next option is the implementation I just put up, which would only require change of a couple header files and a using declaration if fusion::tuple moves. -t