Normal distribution: copy constructor doesn't yield intuitive correct behavior

[Stefan Maagh <s.maagh@…>]

Purging the cache in the copy constructor of normal_distribution is not fine. The compiler generated copy constructor yields the correct behavior.

In the attached example a generator (A) using normal_distribution is called once before a generator (B) is created as a copy of (A). The expected behavior is that calling (A) and (B) now produces identical sequences. But they don't.

The attached example shows that using uniform_int instead normal_distribution yields the expected behavior.

Applying the compiler generated copy constructor worked fine for me. Well, I expect that there was a reason to purge the cache although I can't see it. But for consistence and to provide intuitive correct behavior I suggest to apply the compiler generated copy constructor.

[Steven Watanabe]

This has already been fixed in the trunk, since the current draft standard mandates the behavior that you want. The fix should be in 1.47.