John Maddock wrote:

The updated test is passing as well, looks like this may be a case of adding a BOOST_WORKAROUND for Apple's GCC and see if it fixes things.

I added some #if for Apple's gcc and the IBM compiler. A look at the regression tests (http://www.boost.org/development/tests/trunk/developer/msm.html) shows: - all problems with Apple's gcc 4.0.1 are gone (waow cool! Better than expected) - IBM's compiler manages a few test cases more (for others, it just crashes or gives a run-time error status 139 which does not bring me much further) This is in my eyes a proof that for these compilers, BOOST_NO_NESTED_FRIENDSHIP and BOOST_NO_MEMBER_TEMPLATE_FRIENDS should be defined. If these features only work sometimes, then they should not be seen as supported. We now have 2 possibilities: - we change boost.config to reflect this situation and I update my code accordingly - we don't and I have to keep my #if. Solution 2 is not great for me because I'm duplicating boost.config and it litters my code with stuff less clear than BOOST_NO_NESTED_FRIENDSHIP and BOOST_NO_MEMBER_TEMPLATE_FRIENDS. In any case, it'd be good if we could decide on it this week on it so that this makes it to the 1.45. Thanks, Christophe