Another option is to define the macro empty if BOOST_NO_EXCEPTIONS is defined, as the throwing expression would surely call a user defined abort mechanism (BOOST_ASSERT or similar) when BOOST_NO_EXCEPTIONS is defined. Or the user should protect BOOST_TEST_THROWS calls with #ifndef BOOST_NO_EXCEPTIONS macros.
Both look sensible. As the user of the macro, what would you prefer? I'd think that in a BOOST_NO_EXCEPTIONS environment it's better to remove the test so that the rest could still be run. We don't actually test in such an environment so the question is a bit moot though.
I've implemented a test for boost.container and I've found expanding BOOST_TEST_THROWS to nothing if BOOST_NO_EXCEPTIONS is defined is quite useful to maintain my test case clean. I only need to know that the expression won't be executed so if I call:
BOOST_TEST_THROWS(c.reserve(100), std::exception)
I can't write my test supposing "reserve" will be called. I think it's a good idea to avoid forcing users wrapping each BOOST_TEST_THROWS with #ifndef BOOST_NO_EXCEPTIONS/#endif. Patch attached with implementation and more comments. Ok to commit?
That sounds like a good idea to me. BTW, I wrote something similar myself which contains almost all of the Boost.Test testing macros, but depends only on the lightweight test header - I'm attaching what I ended up with - do whatever you want with it ;-) John.