[boost] Re: boost::variant 0.2 (was: Yet another ...)

From: Itay (itay_maman_at_[hidden])
Date: 2002-07-16 13:30:49

>"Itay Maman" <itay_maman_at_[hidden]> wrote:
>
>An updated version of boost::variant is now available at:
http://groups.yahoo.com/group/boost/files/boost_variant_160702.zip. This
version is
>compatbile with these compilers: GCC 3.0.4, CodeWarrior 7.0, MSVC 6 and
MSVC 7.
>Please note that test4 is currently broken with MSVC 7, since it
instantiates variants with more than 8 types. The other 3 compilers use the
RDSL
>(Reduced Dot-Style Typelist) mechanism, so they don't impose this
restriction. Anyway, I expect this issue to be fixed with MPL's formal
relese code.
>
>Comments and compatibility reports are welcomed.
>-Itay
>

One more thing:

The "bet coversion" algorithm which I had previously suggested
(http://lists.boost.org/MailArchives/boost/msg31659.php)  may terminate with
a wrong result ("Ambigious call") when two of three types checked at every
iteration are integral types. The appearance of this situation is determined
by the order of the types on the specified sequence. This affected MSVC 6/7
which were the only compilers where this algorithm was needed. I could find
no fix for this problem (basically, an is_ambig<S, T1, T2>::value primitive
is needed), so I went to a MACRO-based approach to get the same effect,
similar to the one suggested at Eric Friedman's implementation.

-Itay