michaelcahill writes:

> I'm testing [the spinlock registration changes in WT-805] with wtperf and the medium-lam config. After a merge completes, some handles are closed, then the next statistics log dump segfaults because we have a pointer to a spinlock in freed memory.

In 610ac6a, michaelcahill added an unregister call, and in 8bd9e8a I rearranged that code slightly to minimize the window, but I think there's still an outside possibility of a race.

The spinlock logging code isn't intended for production builds, but maybe we should fix the race? I'm not entirely sure which spinlocks caused the failure, or if this is worth fixing, so I'm going to wait on this one until I talk to Michael.