Stephan T. Lavavej wrote:

You should be aware that while 2013 has the machinery to respect C++11 minimal allocators (in particular, providing a default for rebind), we missed several places where we should be using that machinery. For example, basic_string and allocate_shared were unintentionally requiring the full C++03 allocator interface.

This is not a problem in our specific use case, which is boost::allocate_shared. We just need std::allocator_traits in order to properly support minimal allocators (along with construct/destroy) from our side. :-)

Glen Fernandes wrote:

I'm much in favor of having BOOST_NO_CXX11_ALLOCATOR no longer defined on VC11.

You're in favor of having it defined, you mean :-) It's not defined now.

John Maddock wrote:

Can you make the changes? I'm not yet in a position to test VC11, though I hope to be soon.

I don't have VC11 either, but the minimum test that would disqualify it would probably entail adding to boost_no_cxx11_allocator.ipp something like: std::allocator_traits at; + std::allocator_traits

::rebind_alloc::value_type ra = &at;

A stricter approach would actually test that std::allocator_traits::rebind_alloc is std::allocator.

John Maddock wrote:

Done in develop. Tested VC10, 11 and 12.

Thanks! What is the plan WRT merging to master?

...

...

Done in develop. Tested VC10, 11 and 12.

Thanks!

Belated thanks for BOOST_NO_CXX11_HDR_ATOMIC as well. I hadn't noticed. But it, too, needs to one day make its way into master.

I was waiting for the regression tests to start up again - but now that they're running and everything looks OK, I've gone ahead and merged to master. John.