It might also be beneficial to allow finer grained results within a test file. Let's say I have a test file stl_containers.cpp that tests std::vector, std::list, etc. Now, if a compiler has a problem with one of these (e.g. std::map) - if I understand the regression test system correctly - it will mark this test as failed in the regression results and a user >would assume that all the stl_containers would be unusable!

That's the way it is now. That's why there are so many (too many) tests. About 20 tests for base services, void_cast, iterators, iteratorbase64, demo, etc. About 35 test for serialization facilities e.g. tracking, exporting, registering, stl::list, stl::vector, etc. These later are built/run for each archive type xml, wide xml, text, wide text and native binary. So this works out to about 20 + 35 * 5 = 190 (actually 209) * 5 compilers about 1000 test run/builds. Robert Ramey