Florian Winter
Thank you for pointing this out. So far, our setup works fine on Windows and Linux using completely different toolchains. The compilation problem that we get doesn't seem to be related to the way we build Boost.Test at first glance, but feel free to tell me I am wrong.
You may find it breaking once you do a step away from situation where it works. Why don't you like using shared library?
Is the problem a general problem on Mac OS X 10.3, i.e. will we also run into this problem when compiling the latest release of Boost.Test the standard way? Or is the problem related to our setup in particular?
No. The problem was in general on Mac.
This would be important for us to know because it determines the next step we take to solve the problem, which would be either upgrade to a newer SVN version or try to build Boost.Test the standard way.
Try svn. You can see the status of platform build on corresponding regression test page. Gennadiy