Boost Fusion foreach works for std::tuple OR boost::tuple depending on inclusion order

[bugs@…]

As my attached repro sample shows can't adapt std::tuple and boost::tuple at the same time.

The headers

#   include <boost/fusion/adapted/boost_tuple.hpp>
#   include <boost/fusion/adapted/std_tuple.hpp>

seem to mutually exclude each other (at least in this code sample). Which ever header of the two gets included _first_, determines which type of tuple will be usable as a fusion sequence in that TU.

In the case of the 'blanket' include boost/fusion/adapted.hpp this means that std::tuple is not usable as a Fusion Sequence - unless one explicitely include boost/fusion/adapted/std_tuple.hpp _ahead_ of that.

This is rather surprising, and obviously causes problems for people mixing std::tuple and boost::tuple within a sinlge TU.

Also, although I haven't looked too deeply into the mechanics of it, I worry a bit that this may be a signal that ODR violations could be lurking when different TUs include the headers in different order.

[bugs@…]

Simple reproducer for ticket #8418

[anonymous]

Hi,

I believe the issue behind this is that the header guards for the following 2 files have the SAME header guard value of BOOST_FUSION_CATEGORY_OF_IMPL_09272006_0726.

1. boost/fusion/adapted/boost_tuple/detail/category_of_impl.hpp
2. boost/fusion/adapted/std_tuple/detail/category_of_impl.hpp

[william@…]

Patch

[anonymous]

Amazing! I had this hunch, but on looking at the header guards, I concluded they looked like they are automatically generated :|

On that ground I dismissed the possibility of a collision... How did you 'make up' a new 'unique' header guard name?

Thanks!

[Joel de Guzman]

(In [84000]) Fixes #8418

[William Lynch <william@…>]

Replying to anonymous:

Amazing! I had this hunch, but on looking at the header guards, I concluded they looked like they are automatically generated :|

On that ground I dismissed the possibility of a collision... How did you 'make up' a new 'unique' header guard name?

Thanks!

The tail end of the header guard appears to be a datetime string, so I changed that to today's date. I imagined that if I was wrong with that assumption, that Joel would have used a different one.