Actually this kind of approach has interested me for some time. I just haven't found the time to invest in it. In the context of wide character support, I wanted to make the wide character librariy build and tests dependent upon the existence of a boost config macro BOOST_NO_WCHAR (or whatever it is). This would entail making a small test program which fails if the required facility isn't present and making the library builds and tests dependent on that. Actually, I'd like to carry the idea even further and do something like make a test for success with BOOST_STRONG_TYPEDEF and make the higher order tests not even run if the lower order tests fail. Other situations would be something like not even running test of serialization with say - variant, unless all the variant test tests pass, etc. Of course this would be if I had nothing else to do. The whole issue here final tweaks in the testing setup so that serialization passes for the latest borland compiler and msvc 7.1. Note that these tests pass with V1 so the issue isn't with the serialization library itself, its just that that is where some "corner cases" show up. I don't have a problem with just just suppressing the tests in the test matrix and adding a note to the 1.34 documenation that the library is known to work with the inclusion of spirit 1.6x . I would hate to see a very small issue hold up a whole release. Robert Ramey Rene Rivera wrote:
Robert Ramey wrote:
b) I would think that the "best" solution would be for the Jamfile to detect that spirit 1.6 isn't available on a platform which requires it and skip adding the xml_?archives to the library. So the library would build in such a case without xml_support.
Spirit has a "boost/spirit/version.hpp" file which you could check and disable the xml support at the library level.