PR c++/42758

[Paolo Carlini]

Hi,
> +#include <cassert>
> +#include <set>
>   
I have a side comment about including whole library facilities in C++
testcases, I'm not sure if we ever discussed the issue in the open.
Anyway, in my opinion, it makes sense to avoid doing that as much as
possible: it implies that library people can confidently regression test
only the library and be 100% sure to not cause regressions.

Paolo.