class LockDependencySet

Checks the dependency hash set for the given lock class. This method may

safely race with AddLockClass(), converging on the correct answer by the

next check.

Adds the given lock class id to the dependency set if not already present.

Updates the set using the following lock free approach:

1. The dependency set is fixed size and all entries start out empty.

2. New entries are added using open addressing with linear probing.

3. An entry may only change from empty to holding a lock class id.

4. To add an entry the set is probed linearly until either:

a) The id to add appears in the set already.

b) The first empty entry is found.

c) The entire set has been probed; return max dependencies error.

5. Attempt to compare-exchange the empty entry with the lock class id:

a) If the update succeeds return success.

b) If the update does not succeed but the winning value is the same

lock class id return with success.

c) If the update does not succeed and the winning value is a different

lock class id proceed to the next entry and continue the probe from

step #4.

Iterator accessors.

Clears the dependency set. This method is not used by the main algorithm

but may be useful for unit tests and benchmarking. Until lock sequence

memoization is implemented it is generally safe to call this method at

any time, resulting in dependency set being rebuilt at runtime. Once

lock sequence memoization is implemented it is necessary to clear the

memoization table whenever any dependency set is cleared so that the

dependency set can be rebuilt; failure to do so could result in missing

new lock violations.