Ideally you should run the whole serialization regression test suite which would isolate what part of the system is failing. Or better yet, since (if I remember correctly) you're using boost on a "new" platform (one not usually tested for boost) the best would be to run the whole boost test suite. I realise that this take a while. But it's the only way you can be sure you're on solid ground and pinpoint any failings which might be rippling through to some higher level tests in a non-obvious way.