Without proper annotations, clang thinks that mutexes are still held for the
duration of a reverse_lock. This could lead to subtle bugs as
EXCLUSIVE_LOCKS_REQUIRED(foo) passes when it shouldn't.
As mentioned in the docs [0], clang's thread-safety analyzer is unable to deal
with aliases of mutexes, so it is not possible to use the lock's copy of the
mutex for that purpose. Instead, the original mutex needs to be passed back to
the reverse_lock for the sake of thread-safety analysis, but it is not actually
used otherwise.
[0]: https://clang.llvm.org/docs/ThreadSafetyAnalysis.html
noexcept is default for destructors as of c++11. By throwing in reverselock's
destructor if it's lock has been tampered with, the likely result is
std::terminate being called. Indeed that happened before this change.
Once reverselock has taken another lock (its ctor didn't throw), it makes no
sense to try to grab or lock the parent lock. That is be broken/undefined
behavior depending on the parent lock's implementation, but it shouldn't cause
the reverselock to fail to re-lock when destroyed.
To avoid those problems, simply swap the parent lock's contents with a dummy
for the duration of the lock. That will ensure that any undefined behavior is
caught at the call-site rather than the reverse lock's destruction.
Barring a failed mutex unlock which would be indicative of a larger problem,
the destructor should now never throw.
