The existing spinlock code, used to protect SHAMapInnerNode
child lists, has a mistake that can allow the same child to
be repeatedly locked under some circumstances.
The bug was in the `SpinBitLock::lock` loop condition check
and would result in the loop terminating early.
This commit fixes this and further simplifies the lock loop
making the correctness of the code easier to verify without
sacrificing performance.
It also promotes the spinlock class from an implementation
detail to a more general purpose, easier to use lock class
with clearer semantics. Two different lock types now allow
developers to easily grab either a single spinlock from an
a group of spinlocks (packed in an unsigned integer) or to
grab all of the spinlocks at once.
While this commit makes spinlocks more widely available to
developers, they are rarely the best tool for the job. Use
them judiciously and only after careful consideration.
The `SHAMapInnerNode` class had a global mutex to protect the
array of node children. Profiling suggested that around 4% of
all attempts to lock the global would block.
This commit removes that global mutex, and replaces it with a
new per-node 16-way spinlock (implemented so as not to effect
the size of an inner node objet), effectively eliminating the
lock contention.
A large percentage of inner nodes only store a small number of children. Memory
can be saved by storing the inner node's children in sparse arrays. Measurements
show that on average a typical SHAMap's inner nodes can be stored using only 25%
of the original space.
This commit combines a number of cleanups, targeting both the
code structure and the code logic. Large changes include:
- Using more strongly-typed classes for SHAMap nodes, instead of relying
on runtime-time detection of class types. This change saves 16 bytes
of memory per node.
- Improving the interface of SHAMap::addGiveItem and SHAMap::addItem to
avoid the need for passing two bool arguments.
- Documenting the "copy-on-write" semantics that SHAMap uses to
efficiently track changes in individual nodes.
- Removing unused code and simplifying several APIs.
- Improving function naming.
