Gennadiy Rozental wrote:
Clearly, the problem here is in Metrowerks' standard library, but Boost.Test needs a work-around. The fact that changing a Boost.Test header file makes xpressive's tests pass seems to prove it. I don't know what the work-around is, and I don't have the platform, so I can't submit a patch. Sorry. But it might be as simple as something like:
#if BOOST_WORKAROUND(__MWERKS__, BOOST_TESTED_AT(0x3206)) # include </usr/include/unistd.h> # include </usr/include/signal.h> # include </usr/include/setjmp.h> #else # include <unistd.h> # include <signal.h> # include <setjmp.h> #endif
Which platforms Metrowerks supply it's compiler for? Does it work on NT or Mac?
Good question. I was under the impression it was an Mac thing, but a closer look at the regression results shows that's not the case. I guess my suggested work-around is not general enough. Ben Artin suggested earlier in the thread that the general work-around is: "Copy and paste the relevant bits of /usr/include headers into your own source code and curse CW. That is what everyone I know does upon encountering some variant of this problem." -- Eric Niebler Boost Consulting www.boost-consulting.com