[Boost-build] question regarding "requires cxx11_hdr_array" clause in Jamfile

Subject: [Boost-build] question regarding "requires cxx11_hdr_array" clause in Jamfile
From: Robert Ramey (ramey_at_[hidden])
Date: 2016-12-06 14:03:27

In tests of the boost seriaization library here:

https://api.travis-ci.org/jobs/181492138/log.txt?deansi=true

I have a test failure with the test "test_array".  The test includes
boost/serialization/array.hpp.  This header file morphed from 
implementing the serialization of native arrays (which was moved to the 
core of the library) to implementing serialization of <array>.  So the 
test shouldn't be run if the <array> header doesn't exist and will fail 
if it does.

To this end, boost/serialization/array.hpp contains the following:

#ifndef BOOST_NO_CXX11_HDR_ARRAY

#include <array>
#include <boost/serialization/nvp.hpp>

...

To implement the above I've included the following in my jamfile:

      [ test-bsl-run_files test_array : A : :  [ requires 
cxx11_hdr_array ] ] # BOOST_NO_CXX11_HDR_ARRAY

Which I understand should condition the invocation of the test on 
config.hpp not defining BOOST_NO_CXX11_HDR_ARRAY.  In general this 
approach seems to work well.  But I'm having a problem with this 
specific case.

Interesting parts of the log file above are:

$ TOOLSET=gcc,clang

The command "TOOLSET=gcc,clang" exited with 0.
$ if [ $TRAVIS_OS_NAME == osx ]; then TOOLSET=clang; fi

The command "if [ $TRAVIS_OS_NAME == osx ]; then TOOLSET=clang; fi" 
exited with 0.
$ ./b2 libs/serialization/test toolset=$TOOLSET
Performing configuration checks

     - 32-bit                   : no
     - 64-bit                   : yes
     - arm                      : no
     - mips1                    : no
     - power                    : no
     - sparc                    : no
     - x86                      : yes
     - symlinks supported       : yes
     - Boost.Config Feature Check: cxx11_hdr_array : yes
     - Boost.Config Feature Check: cxx11_hdr_forward_list : yes
     - Boost.Config Feature Check: hash : no
     - Boost.Config Feature Check: cxx11_hdr_unordered_map : yes
     - Boost.Config Feature Check: cxx11_hdr_unordered_set : yes
     - Boost.Config Feature Check: slist : no
     - Boost.Config Feature Check: std_wstreambuf : yes
...patience...

So this shows that BOOST_NO_CXX11_HDR_ARRAY is not defined. So I would 
expect test_array to run - which it does.  BUT when the test is actually 
run, it fails because the code in <array> is not included.  This leads 
me to believe that when the test is actually run 
BOOST_NO_CXX11_HDR_ARRAY IS in fact defined.

In other words, it looks like boost build checks for the existence of 
<array> while boost config checks for the C++ version to see if <array> 
is supported.  This would explain the behavior I'm seeing.

I would like someone else's informed opinion on this and what I might do 
about it.

Note: I intend to change the name of test_array to test_std_array but I 
don't want to do this until the current problem is sorted out.

Robert Ramey